prediction of clean-bed head loss in crumb rubber filters
TRANSCRIPT
The Pennsylvania State University
The Graduate School
College of Engineering
PREDICTION OF CLEAN-BED HEAD LOSS
IN CRUMB RUBBER FILTERS
A Thesis in
Environmental Engineering
by
Hao Tang
copy 2008 Hao Tang
Submitted in Partial Fulfillment of the Requirements
for the Degree of
Master of Science
December 2008
The thesis of Hao Tang was reviewed and approved by the following
Yuefeng F Xie Professor of Environmental Engineering Thesis Co-Advisor
John M Regan Associate Professor of Environmental Engineering Thesis Co-Advisor
Shirley E Clark Assistant Professor of Environmental Engineering
Peggy A Johnson Professor of Civil Engineering Head of the Department of Civil amp Environmental Engineering
Signatures are on file in the Graduate School
iii
ABSTRACT
Pilot crumb rubber filters were tested to study their clean-bed head loss under the
influences of three design and operational parameters (media size media depth and filtration
rate) Filter media compressions were observed during the filtration test Statistic analysis of the
field data was used to investigate the applicability of the Kozeny and Ergun equations in
predicting the clean-bed head loss in crumb rubber filters A statistical model was also developed
to evaluate the effects of the three parameters and to better predict the clean-bed head loss Both
the Kozeny and Ergun equations were unacceptable for crumb rubber filters especially when the
compressed media depth and porosity from the filtration tests were not available The statistical
model developed based on the original media depth and porosity was statistically valid and was
able to provide best-fit to the actual head loss data without using the compressed media depth and
porosity obtained in filter tests The results from this study indicated that the statistical model
could be used in place of the Kozeny and Ergun equations for the prediction of clean-bed head
loss in crumb rubber filters
iv
TABLE OF CONTENTS
LIST OF FIGURES v
LIST OF TABLES vi
ACKNOWLEDGEMENTS vii
Chapter 1 INTRODUCTION 1
Statement of Problem 1 Objectives 2 Thesis Organization 3
Chapter 2 BACKGROUND 4
Manufacture of Crumb Rubber 4 Characterization of Filter Media 5 Theory of Crumb Rubber Filtration 7 Head Loss Theory 12
Chapter 3 MATERIALS AND METHODS 16
Filter Setup 16 Filter Media 16 Design and Operational Conditions 18 The Kozeny and Ergun Equations 18 Statistical Modeling 19 Regression Verification 20
Chapter 4 RESULTS AND DISCUSSION 24
Media Sphericity 24 Prediction by the Kozeny Equation 27 Prediction by the Ergun Equation 29 Development of a Statistical Model 33 Effects of Parameters 36
Effect of filtration rate 38 Effect of interaction between filtration rate and media depth 39 Effect of filter depth 40 Effect of media size 41 Effect of interaction between media size and filtration rate 42
Prediction by the Statistical Model 42
Chapter 5 CONCLUSIONS 45
Bibliography 46
v
LIST OF FIGURES
Figure 1-1 Overview of research objectives and scope for model development and verification 3
Figure 2-1 Sieve results of crumb rubber media 6
Figure 2-2 Schematic diagram of filter medium configurations 8
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 9
Figure 2-4 Crumb rubber filter performances under various media size conditions 12
Figure 3-1 Experimental setup of the crumb rubber filter 17
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation 28
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity 29
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation 31
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity 32
Figure 4-5 Actual head loss profiles at all filter configurations 36
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation 38
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of 120 mm and filter depth of 09 m 39
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth 41
Figure 4-9 Prediction of clean-bed head loss by the statistical model 43
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity 44
vi
LIST OF TABLES
Table 2-1 Physical characteristics of crumb rubber media 7
Table 2-2 Typical properties of several filter media 10
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media 26
Table 4-2 Parameters in the statistical model 35
vii
ACKNOWLEDGEMENTS
I would like to thank my advisors Dr Yuefeng Xie and Dr John Regan for many years
of mentoring and for helping me to become the person I was meant to be I would also like to
thank Dr Shirley Clark for being on my committee Mr Joe Swanderski and his staff at
University Park Wastewater Treatment Plant for providing research facilities and my colleagues
at Penn State for their friendship and encouragement Finally I wish to thank my parents my
wife Wen and my sons Shunyu and Jinyu Without them I could not have accomplished this
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
The thesis of Hao Tang was reviewed and approved by the following
Yuefeng F Xie Professor of Environmental Engineering Thesis Co-Advisor
John M Regan Associate Professor of Environmental Engineering Thesis Co-Advisor
Shirley E Clark Assistant Professor of Environmental Engineering
Peggy A Johnson Professor of Civil Engineering Head of the Department of Civil amp Environmental Engineering
Signatures are on file in the Graduate School
iii
ABSTRACT
Pilot crumb rubber filters were tested to study their clean-bed head loss under the
influences of three design and operational parameters (media size media depth and filtration
rate) Filter media compressions were observed during the filtration test Statistic analysis of the
field data was used to investigate the applicability of the Kozeny and Ergun equations in
predicting the clean-bed head loss in crumb rubber filters A statistical model was also developed
to evaluate the effects of the three parameters and to better predict the clean-bed head loss Both
the Kozeny and Ergun equations were unacceptable for crumb rubber filters especially when the
compressed media depth and porosity from the filtration tests were not available The statistical
model developed based on the original media depth and porosity was statistically valid and was
able to provide best-fit to the actual head loss data without using the compressed media depth and
porosity obtained in filter tests The results from this study indicated that the statistical model
could be used in place of the Kozeny and Ergun equations for the prediction of clean-bed head
loss in crumb rubber filters
iv
TABLE OF CONTENTS
LIST OF FIGURES v
LIST OF TABLES vi
ACKNOWLEDGEMENTS vii
Chapter 1 INTRODUCTION 1
Statement of Problem 1 Objectives 2 Thesis Organization 3
Chapter 2 BACKGROUND 4
Manufacture of Crumb Rubber 4 Characterization of Filter Media 5 Theory of Crumb Rubber Filtration 7 Head Loss Theory 12
Chapter 3 MATERIALS AND METHODS 16
Filter Setup 16 Filter Media 16 Design and Operational Conditions 18 The Kozeny and Ergun Equations 18 Statistical Modeling 19 Regression Verification 20
Chapter 4 RESULTS AND DISCUSSION 24
Media Sphericity 24 Prediction by the Kozeny Equation 27 Prediction by the Ergun Equation 29 Development of a Statistical Model 33 Effects of Parameters 36
Effect of filtration rate 38 Effect of interaction between filtration rate and media depth 39 Effect of filter depth 40 Effect of media size 41 Effect of interaction between media size and filtration rate 42
Prediction by the Statistical Model 42
Chapter 5 CONCLUSIONS 45
Bibliography 46
v
LIST OF FIGURES
Figure 1-1 Overview of research objectives and scope for model development and verification 3
Figure 2-1 Sieve results of crumb rubber media 6
Figure 2-2 Schematic diagram of filter medium configurations 8
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 9
Figure 2-4 Crumb rubber filter performances under various media size conditions 12
Figure 3-1 Experimental setup of the crumb rubber filter 17
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation 28
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity 29
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation 31
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity 32
Figure 4-5 Actual head loss profiles at all filter configurations 36
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation 38
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of 120 mm and filter depth of 09 m 39
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth 41
Figure 4-9 Prediction of clean-bed head loss by the statistical model 43
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity 44
vi
LIST OF TABLES
Table 2-1 Physical characteristics of crumb rubber media 7
Table 2-2 Typical properties of several filter media 10
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media 26
Table 4-2 Parameters in the statistical model 35
vii
ACKNOWLEDGEMENTS
I would like to thank my advisors Dr Yuefeng Xie and Dr John Regan for many years
of mentoring and for helping me to become the person I was meant to be I would also like to
thank Dr Shirley Clark for being on my committee Mr Joe Swanderski and his staff at
University Park Wastewater Treatment Plant for providing research facilities and my colleagues
at Penn State for their friendship and encouragement Finally I wish to thank my parents my
wife Wen and my sons Shunyu and Jinyu Without them I could not have accomplished this
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
iii
ABSTRACT
Pilot crumb rubber filters were tested to study their clean-bed head loss under the
influences of three design and operational parameters (media size media depth and filtration
rate) Filter media compressions were observed during the filtration test Statistic analysis of the
field data was used to investigate the applicability of the Kozeny and Ergun equations in
predicting the clean-bed head loss in crumb rubber filters A statistical model was also developed
to evaluate the effects of the three parameters and to better predict the clean-bed head loss Both
the Kozeny and Ergun equations were unacceptable for crumb rubber filters especially when the
compressed media depth and porosity from the filtration tests were not available The statistical
model developed based on the original media depth and porosity was statistically valid and was
able to provide best-fit to the actual head loss data without using the compressed media depth and
porosity obtained in filter tests The results from this study indicated that the statistical model
could be used in place of the Kozeny and Ergun equations for the prediction of clean-bed head
loss in crumb rubber filters
iv
TABLE OF CONTENTS
LIST OF FIGURES v
LIST OF TABLES vi
ACKNOWLEDGEMENTS vii
Chapter 1 INTRODUCTION 1
Statement of Problem 1 Objectives 2 Thesis Organization 3
Chapter 2 BACKGROUND 4
Manufacture of Crumb Rubber 4 Characterization of Filter Media 5 Theory of Crumb Rubber Filtration 7 Head Loss Theory 12
Chapter 3 MATERIALS AND METHODS 16
Filter Setup 16 Filter Media 16 Design and Operational Conditions 18 The Kozeny and Ergun Equations 18 Statistical Modeling 19 Regression Verification 20
Chapter 4 RESULTS AND DISCUSSION 24
Media Sphericity 24 Prediction by the Kozeny Equation 27 Prediction by the Ergun Equation 29 Development of a Statistical Model 33 Effects of Parameters 36
Effect of filtration rate 38 Effect of interaction between filtration rate and media depth 39 Effect of filter depth 40 Effect of media size 41 Effect of interaction between media size and filtration rate 42
Prediction by the Statistical Model 42
Chapter 5 CONCLUSIONS 45
Bibliography 46
v
LIST OF FIGURES
Figure 1-1 Overview of research objectives and scope for model development and verification 3
Figure 2-1 Sieve results of crumb rubber media 6
Figure 2-2 Schematic diagram of filter medium configurations 8
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 9
Figure 2-4 Crumb rubber filter performances under various media size conditions 12
Figure 3-1 Experimental setup of the crumb rubber filter 17
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation 28
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity 29
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation 31
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity 32
Figure 4-5 Actual head loss profiles at all filter configurations 36
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation 38
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of 120 mm and filter depth of 09 m 39
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth 41
Figure 4-9 Prediction of clean-bed head loss by the statistical model 43
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity 44
vi
LIST OF TABLES
Table 2-1 Physical characteristics of crumb rubber media 7
Table 2-2 Typical properties of several filter media 10
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media 26
Table 4-2 Parameters in the statistical model 35
vii
ACKNOWLEDGEMENTS
I would like to thank my advisors Dr Yuefeng Xie and Dr John Regan for many years
of mentoring and for helping me to become the person I was meant to be I would also like to
thank Dr Shirley Clark for being on my committee Mr Joe Swanderski and his staff at
University Park Wastewater Treatment Plant for providing research facilities and my colleagues
at Penn State for their friendship and encouragement Finally I wish to thank my parents my
wife Wen and my sons Shunyu and Jinyu Without them I could not have accomplished this
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
iv
TABLE OF CONTENTS
LIST OF FIGURES v
LIST OF TABLES vi
ACKNOWLEDGEMENTS vii
Chapter 1 INTRODUCTION 1
Statement of Problem 1 Objectives 2 Thesis Organization 3
Chapter 2 BACKGROUND 4
Manufacture of Crumb Rubber 4 Characterization of Filter Media 5 Theory of Crumb Rubber Filtration 7 Head Loss Theory 12
Chapter 3 MATERIALS AND METHODS 16
Filter Setup 16 Filter Media 16 Design and Operational Conditions 18 The Kozeny and Ergun Equations 18 Statistical Modeling 19 Regression Verification 20
Chapter 4 RESULTS AND DISCUSSION 24
Media Sphericity 24 Prediction by the Kozeny Equation 27 Prediction by the Ergun Equation 29 Development of a Statistical Model 33 Effects of Parameters 36
Effect of filtration rate 38 Effect of interaction between filtration rate and media depth 39 Effect of filter depth 40 Effect of media size 41 Effect of interaction between media size and filtration rate 42
Prediction by the Statistical Model 42
Chapter 5 CONCLUSIONS 45
Bibliography 46
v
LIST OF FIGURES
Figure 1-1 Overview of research objectives and scope for model development and verification 3
Figure 2-1 Sieve results of crumb rubber media 6
Figure 2-2 Schematic diagram of filter medium configurations 8
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 9
Figure 2-4 Crumb rubber filter performances under various media size conditions 12
Figure 3-1 Experimental setup of the crumb rubber filter 17
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation 28
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity 29
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation 31
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity 32
Figure 4-5 Actual head loss profiles at all filter configurations 36
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation 38
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of 120 mm and filter depth of 09 m 39
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth 41
Figure 4-9 Prediction of clean-bed head loss by the statistical model 43
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity 44
vi
LIST OF TABLES
Table 2-1 Physical characteristics of crumb rubber media 7
Table 2-2 Typical properties of several filter media 10
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media 26
Table 4-2 Parameters in the statistical model 35
vii
ACKNOWLEDGEMENTS
I would like to thank my advisors Dr Yuefeng Xie and Dr John Regan for many years
of mentoring and for helping me to become the person I was meant to be I would also like to
thank Dr Shirley Clark for being on my committee Mr Joe Swanderski and his staff at
University Park Wastewater Treatment Plant for providing research facilities and my colleagues
at Penn State for their friendship and encouragement Finally I wish to thank my parents my
wife Wen and my sons Shunyu and Jinyu Without them I could not have accomplished this
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
v
LIST OF FIGURES
Figure 1-1 Overview of research objectives and scope for model development and verification 3
Figure 2-1 Sieve results of crumb rubber media 6
Figure 2-2 Schematic diagram of filter medium configurations 8
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 9
Figure 2-4 Crumb rubber filter performances under various media size conditions 12
Figure 3-1 Experimental setup of the crumb rubber filter 17
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation 28
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity 29
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation 31
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity 32
Figure 4-5 Actual head loss profiles at all filter configurations 36
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation 38
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of 120 mm and filter depth of 09 m 39
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth 41
Figure 4-9 Prediction of clean-bed head loss by the statistical model 43
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity 44
vi
LIST OF TABLES
Table 2-1 Physical characteristics of crumb rubber media 7
Table 2-2 Typical properties of several filter media 10
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media 26
Table 4-2 Parameters in the statistical model 35
vii
ACKNOWLEDGEMENTS
I would like to thank my advisors Dr Yuefeng Xie and Dr John Regan for many years
of mentoring and for helping me to become the person I was meant to be I would also like to
thank Dr Shirley Clark for being on my committee Mr Joe Swanderski and his staff at
University Park Wastewater Treatment Plant for providing research facilities and my colleagues
at Penn State for their friendship and encouragement Finally I wish to thank my parents my
wife Wen and my sons Shunyu and Jinyu Without them I could not have accomplished this
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
vi
LIST OF TABLES
Table 2-1 Physical characteristics of crumb rubber media 7
Table 2-2 Typical properties of several filter media 10
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media 26
Table 4-2 Parameters in the statistical model 35
vii
ACKNOWLEDGEMENTS
I would like to thank my advisors Dr Yuefeng Xie and Dr John Regan for many years
of mentoring and for helping me to become the person I was meant to be I would also like to
thank Dr Shirley Clark for being on my committee Mr Joe Swanderski and his staff at
University Park Wastewater Treatment Plant for providing research facilities and my colleagues
at Penn State for their friendship and encouragement Finally I wish to thank my parents my
wife Wen and my sons Shunyu and Jinyu Without them I could not have accomplished this
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
vii
ACKNOWLEDGEMENTS
I would like to thank my advisors Dr Yuefeng Xie and Dr John Regan for many years
of mentoring and for helping me to become the person I was meant to be I would also like to
thank Dr Shirley Clark for being on my committee Mr Joe Swanderski and his staff at
University Park Wastewater Treatment Plant for providing research facilities and my colleagues
at Penn State for their friendship and encouragement Finally I wish to thank my parents my
wife Wen and my sons Shunyu and Jinyu Without them I could not have accomplished this
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Chapter 1
INTRODUCTION
Statement of Problem
Scrap tire stockpiles cause health and environmental concerns by presenting a potential
fire hazard and providing breeding ground for vectors of disease The properties of scrap tires
such as volume flammability toughness durability etc make their storage and elimination
difficult (USEPA 1993) According to the data compiled by Rubber Manufacturers Association
275 million tires remained in stockpiles across the United States in 2003 and approximately 290
million new scrap tires are generated each year (USEPA 2006) Various recycling technologies
have been recommended for conservation of natural resources and minimization of environmental
impacts of these tires The use of crumb rubber a scrap-tire-derived material as a filter medium
is an innovative technology that has been investigated as a potential ldquogreen engineering solutionrdquo
for wastewater treatment and disposal of scrap tires As a compressible material crumb rubber
could form an ideal porosity gradient in filters because the top layer of the media was least
compressed while the bottom layer was most compressed Crumb rubber filters favored in-depth
filtration and could allow longer filtration time and higher filtration rate which could
substantially increase the filtration efficiency (Graf et al 2000 Xie et al 2001) Their light
weight compact size and capability to remove turbidity phytoplankton and zooplankton also
allowed them to be used in ballast water treatment (Tang et al 2006a)
Since crumb rubber filtration has beneficial engineering applications it is important to
understand the effects of design and operational parameters on the filter performance Clean-bed
head loss is such an important filter performance parameter because water head must be provided
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
2
to accommodate the increase of head loss resulting from the accumulation of particulates in the
filter media (Cleasby and Logsdon 1999) There are numerous models for the prediction of
clean-bed head loss but the Kozeny and Ergun equations are most accepted models (Trussell et
al 1999a) These two equations however were developed for conventional rigid granular media
filters Their applicability in the crumb rubber filter was not clear because of the crumb rubber
media compression which changed the filter configurations during the filtration process and
much higher filtration rates
Objectives and Scope
The principle objective of this research was to evaluate the applicability of the Kozeny
and Ergun equations for clean-bed head loss prediction in crumb rubber filters Statistic analysis
was also conducted to investigate the impacts of three design and operational parameters and to
develop a statistical model for the prediction of head loss in crumb rubber filters
A general flow chart showing an overview of the research scope with regard to model
development and verification is shown in Figure 1-1 and summarized as follows
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
3
Figure 1-1 Overview of research objectives and scope for model development and verification
(1) Conduct filtration tests by varying media size media depth and filtration rates to
obtain actual clean-bed head loss data under various filtration conditions
(2) Estimate the crumb rubber media sphericity based on empirical fitting of filtration test
data
(3) The filtration test data and sphericity estimates are applied to the Kozeny and Ergun
equations respectively to determine whether the two equations can be used in crumb rubber
filtration
(4) Develop a statistical model based on three design and operational parameters
Thesis Organization
The thesis is composed of five chapters Chapter 1 describes the problem and objectives
of the study Chapter 2 reviews previous studies on crumb rubber filtration and head loss theories
Methodologies are described in Chapter 3 Chapter 4 elaborates on the results and discussion of
the study Conclusions are summarized in Chapter 5
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Chapter 2
BACKGROUND
Manufacture of Crumb Rubber
Crumb rubber is made by a combination or application of several size reduction
technologies Typically the crumbing processes involve shredding the tire into progressively
smaller pieces removing the metals (belts and bead wire) using magnets blowing away the
fibers and then grinding to further reduce the material to a given size The manufacturing
technologies can be further divided into two major categories mechanical grinding and cryogenic
reduction
Mechanical grinding is the most commonly used process The method consists of
mechanically breaking down the rubber into small particles using a variety of grinding
techniques such as cracker mills and granulators The steel components are removed by a
magnetic separator (sieve shakers and conventional separators such as centrifugal air
classification and density are also used) The fiber components are separated by air classifiers or
other separation equipment These systems are well established and can produce crumb rubber
(varying particle size grades quality etc) at relatively low cost
The cryogenic process consists of freezing the shredded rubber at an extremely low
temperature (far below the glass transition temperature of the compound) The frozen rubber
compound is then easily shattered into small particles The fiber and steel are removed in the
same manner as in mechanical grinding The advantages of this system are cleaner and faster
operation resulting in the production of a fine mesh size The most significant disadvantage is the
slightly higher cost due to the added cost of cooling (eg liquid nitrogen)
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
5
The crumb rubber media used in this study was produced using the first process No
cooling was applied to make the rubber brittle The tires are first developed into chips of 50 mm
in size in a preliminary shredder The tire chips then entered into a granulator where the chips
were reduced to a size to less than 10 mm at ambient temperature and most of the steel and fiber
were liberated from the rubber granules The steel then was removed magnetically and fiber was
removed using shaking screens and wind sifters The fine grinding process generally was
conducted by cracker mills which was currently the most common and productive method of
producing tire rubber The end product was usually an irregularly shaped particle with a large
surface area varying in size from 475 mm to 045 mm (Hsiung 2003)
Characterization of Filter Media
Crumb rubber has been sieved for three media sizes according to the sieve analysis using
American Society for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of
Fine and Coarse Aggregates (ASTM 2001) Sieve results were plotted in Figure 2-1 for
determination of media characteristics As summarized in Table 2-1 the effective size (for which
10 of the grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range
of 05 ndash 12 mm 12 ndash 20 mm and 20 ndash 40 mm crumb rubber media Their uniformity
coefficients (Ratios of sieve size that will permit passage of 60 of media by weight to the sieve
size that will permit passage of 10 of media by weight) were determined to be 139 153 and
128 respectively (Tang etal 2006a) Density of crumb rubber is not correlated with media size
and the value of 1130 kgm3 is uniform for all sizes crumb rubber media Porosity is defined as
the ratio of volume of the void space to the bulk volume of crumb rubber media (Trussell et al
1999) It is expressed in Eq 1 and was determined by measurements of the dry weight of the
media loaded into the filter column and the height of the media in the column
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
6
Where
= the dry weight of the media loaded into the filter column
Eq 1
= the porosity of crumb rubber media
= the density of crumb rubber
= the cross section area of the filter column
= the height of the dry media in the column
The packed porosity was 062 058 and 054 for the size range of 05 ndash 12 mm 12 ndash 20
mm and 20 ndash 40 mm crumb rubber media respectively
Figure 2-1 Sieve results of crumb rubber media (Regenerated from Hsiung 2003)
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
7
Table 2-1 Physical characteristics of crumb rubber media (Source Tang et al 2006a)
Theory of Crumb Rubber Filtration
The crumb rubber filter resembles an ideal filter bed configuration consisting of the
largest size media on the top medium size in the middle and the smallest size at the bottom as
shown in Figure 2-2a (Xie et al 2001) For conventional media filters however the fine grains
tend to settle on the top of the filter while the coarse grains accumulate at the bottom during
stratification after filter backwash (Figure 2-2b) Such media size distribution of conventional
media filters will easily clog the filter bed surface which causes a high head loss Compressibility
of the crumb rubber media results in a favorably porosity gradient (Figure 2-2c) Compression
increases with pressure throughout the filter columns Consequently the crumb rubber filter bed
has the smallest pore size at the bottom and the largest pore size on the top
As a result an ideal porosity profile can be formed in filter bed and the filtration
efficiency can be increased The crumb rubber media results in lower head loss and higher
filtration rate and water production According to a comparative study of a sandanthracite filter
and a crumb rubber filter in tertiary treatment (Hsiung 2003) at the filtration rate of 6 gpmft2
both filters performed similarly on turbidity removal However the crumb rubber filter could
continuously run for 50 hours while the sandanthracite filter could only run 3 hours before the
filter bed was clogged (Figure 2-3) For the crumb rubber filtration therefore backwash
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
8
frequency and head loss development can be reduced while filter run time and water production
rate can be increased
Figure 2-2 Schematic diagram of filter medium configurations (Source Xie et al 2001)
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
9
Figure 2-3 Comparison of performances between a crumb rubber filter and a sandanthracite filter treating secondary effluent at 6gpmft2 (a) Turbidity removal (b) Head loss development (Source Hsiung 2003)
Besides compressibility the crumb rubber has several other physical properties that offer
advantages over conventional media Typical properties of conventional filter media and crumb
rubber media are shown in Table 2-2 The porosity of crumb rubber is higher than other media
which will capture more solids and increase the filter run time The mechanisms of crumb rubber
filtration are interception adsorption coagulation and sedimentation inside the filter bed In
addition small hairs that protrude from sides of crumb rubber particle further help capture solids
in voids (Xie et al 2000)
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
10
Table 2-2 Typical properties of several filter media
Property Garnet Illmenite Sand Anthracite GAC Crumb rubber
Effective size (mm) 02-04 02-04 04-08 08-20 08-20 06-19
Uniformity coefficient 13-17 13-17 13-17 13-17 13-24 12-16 Density (gml) 36-42 45-50 265 14-18 13-17 11-12 Porosity () 45-58 NA 40-43 47-52 NA 54-62
Hardness (Moh) 65-75 50-60 7 20-30 Low Very low
Sphericity was a shape factor of filter media According to the literature related with
filtration theories the grain shape was usually described by either shape factor (ξ ) or sphericity (
ψ ) The relationship between them was shown in Eq 2
grainofareasurfaceactualspherevolumeequivalentofareasurface
=ψ Eq 2
ψξ 6= Eq 3
where
=ψ sphericity dimentionless
=ξ shape factor dimentionless
The literature values for sphericity of common filter materials were calculated from head
loss experiments and therefore many of the sphericity values are empirical fitting parameters for
head loss rather than true independent measurements of shape (Crittenden et al 2005) In this
study the sphericity estimates of crumb rubber media were obtained from empirical fitting of
head loss data based on conventional equations
The Density of crumb rubber is 1130 kgm3 (Tang etal 2006a) which is much lower
than that of conventional media (eg 2650 kgm3 for sand) This can substantially reduce the
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
11
backwash water flow rates required for filter backwash Previous research (Hsiung 2003) has
demonstrated that a backwash flow rate of 214 gpmft2 was required to achieve 30 bed
expansion for a 3-feet-deep sandanthracite filter while a same-depth crumb rubber filter only
required a backwash flow rate of 105 gpmft2 to achieve the same degree of bed expansion
In addition to its use in tertiary wastewater treatment research on crumb rubber filtration
for ballast water treatment indicated that crumb rubber filters could remove up to 70 of
phytoplankton and 45 of zooplankton in addition to the removal of particles (Figure 2-4)
Compared with conventional granular media filters and screen filters crumb rubber filters
required less backwash and developed lower head loss (Tang 2006a) When statistical
approaches were applied to develop empirical models including a head loss model that partially
resembles the Kozeny equation to evaluate the influences of design and operational factors the
results indicated that the behaviors of the crumb rubber filtration for ballast water treatment could
not be described by the theories and models for conventional granular media filtration without
modification (Tang 2006b) Considering the much lighter weight of a crumb rubber filter
compared with conventional filters it was suggested as a competitive solution in ballast water
treatment
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
12
Figure 2-4 Crumb rubber filter performances under various media size conditions (a)Phytoplankton removal (b) Zooplankton removal (Source Tang et al 2006)
Head Loss Theory
Several disciplines have made important contributions to the development of models
characterizing flow through porous media The current theory of creeping flow through filter
media is largely based on experimental work and theoretical analysis published by Henry Darcy
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
13
Darcy observed that the flow through a bed of filter sand was directly proportional to the
hydraulic head acting on the sand bed and inversely proportional to its depth (Darcy 1856) The
following is the equation Darcy proposed
[ ]0HLHL
KAQ minusΔ+Δ
= Eq 4
Where
Q = flow rate through the sand bed
K = a coefficient dependent on the nature of the sand (units of velocity)
A = the area of the sand bed (in plain)
H = the height of surface of the influent water above the top of the sand bed
0H = the height of the surface of the effluent water above the bottom of the sand bed
LΔ = depth of the sand bed
The head loss through the sand could be defined
0HLHH minusΔ+=Δ Eq 5
Darcyrsquos equation could then be expressed in a more commonly used form
⎥⎦⎤
⎢⎣⎡ΔΔ
=LHK
AQ
Eq 6
Following Darcy developments in the prediction of head loss through porous media were
advanced by civil and chemical engineers Kozeny (1927a b) Fair and Hatch (1933) and
Carman (1937) developed a model for predicting Darcy resistance from the characteristics of the
porous media And Ergun (Ergun and Orning 1949 Ergun 1952) combined the linear and
nonlinear models to produce one comprehensive model of porous media flow appropriate for a
wide range of Reynolds numbers
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
14
In the laminar range of flow the Kozeny equation (Fair et al 1968) is used to depict the
head loss
( ) Vva
gk
Lh 2
3
21⎟⎠⎞
⎜⎝⎛minus
=εε
ρμ Eq 7
where
h = head loss in depth of bed
L = depth of filter bed
g = acceleration of gravity
ε = porosity
va
= grain surface area per unit of grain volume = specific surface (Sv) = for spheres
and
d6
eqdψ6
for irregular grains
eqd = grain diameter of sphere of equal volume
V = superficial velocity above the bed = flow ratebed area (ie the filtration rate)
μ = absolute viscosity of fluid
ρ = mass density of fluid
k = the dimensionless Kozeny constant commonly found close to 5 under most filtration
conditions (Fair et al 1968)
The Kozeny equation is generally acceptable for most filtration calculations because the
Reynolds number Re based on superficial velocity is usually less than 3 under these conditions
and Camp (1964) has reported strictly laminar flow up to Re of about 6 (Cleasby and Logsdon
1999)
μρ Re Vdeq= Eq 8
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
15
It was found that the actual head loss was often greater than that predicted from the
Kozeny equation particularly when higher velocities are used in some applications or when
velocities approached fluidization (as in backwashing considerations) In this case the flow may
be in the transitional flow regime where the Kozeny equation is no longer adequate Ergun and
Orning (1949) employed the hydraulic radius approach which Carman and Kozeny used to
characterize linear losses to develop a term for kinetic losses The results were nearly the same as
Eq 7 To create an equation for losses though porous media over a wide range of flow conditions
Ergun and Orning added the Kozeny term for linear losses The Ergun equation is adequate for
the full range of laminar transitional and inertial flow through packed beds (Re from 1 to 2000)
Compared with the Kozeny equation the Ergun equation includes a second term for inertial head
loss
( ) ( )g
VvakV
va
gLh 2
32
2
3
2 11174εε
εε
ρμ minus
+⎟⎠⎞
⎜⎝⎛minus
= Eq 9
Note that the first term of the Ergun equation is the viscous energy loss that is
proportional to V The second term is the kinetic energy loss that is proportional to V2
Comparing the Ergun and Kozeny equations the first term of the Ergun equation (viscous energy
loss) is identical with the Kozeny equation except for the numerical constant The value of the
constant in the second term k2 was originally reported to be 029 for solids of known specific
surface (Ergun 1952a) In a later paper however Ergun reported a k2 value of 048 for crushed
porous solids (Ergun 1952b) a value supported by later unpublished studies at Iowa State
University The second term in the equation becomes dominant at higher flow velocities because
it is a square function of V (Cleasby and Logsdon 1999) As is evident from Eq 7 and 9 the
head loss for a clean bed depends on the flow rate media size porosity sphericity and water
viscosity
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Chapter 3
MATERIALS AND METHODS
Filter Setup
The filter study was carried out in the Kappe Environmental Engineering Laboratory at
the University Wastewater Treatment Plant University Park Pennsylvania USA A pilot filter
(Figure 3-1) was constructed with 152-cm-diameter transparent PVC columns A water pump
was used to supply the influent and backwash flow from a water storage tank An air pump was
connected to the bottom of the filter for air scour before water backwashing Filtration rate was
controlled by a flow meter installed in the filter outlet pipe Influent water was kept constant by
an overflow at 329 meters The head loss through the filter media was measured using the
difference between the water level in the filter and the water level in the glass tube connected to
the bottom of the filter
Filter Media
The crumb rubber media size was determined by sieve analysis using American Society
for Testing and Materials (ASTM) Standard Test C136-01 Sieve analysis of Fine and Coarse
Aggregates (ASTM 2001) As summarized in Table 2-1 the effective size (for which 10 of the
grains are smaller by weight) was 066 mm 120 mm and 190 mm for the size range of 05-12
mm 12-20 mm and 20-40 mm crumb rubber media Their uniformity coefficients were
determined to be 139 153 and 128 respectively and the packed porosity was 062 058 and
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
17
054 for 066 mm 120 mm and 190 mm crumb rubber media respectively The density of all
crumb rubber media was 1130 kgm3
Figure 3-1 Experimental setup of the crumb rubber filter
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
18
Design and Operational Conditions
Factorial design was used in this study The approach reduced the experimental burden
while was effective in seeking high quality results to analyze the effects of factors and
interactions Three levels of crumb rubber media sizes and media depths and 19 filtration rates
were investigated The filter was loaded with 066 mm 120 mm and 190 mm crumb rubber
media at each time For each crumb rubber medium the filter was loaded to a depth of 06 m 09
m and 12 m Before each filter run the filter was backwashed by air scour and then water flow
at the rate of 293 m3m2h For each filter configuration the filter was operated at 19 filtration
rates from 0 to 733 m3m2h Head loss was measured after the media depth was stable Porosity
was determined by the measurement of the dry weight of the media initially loaded to the filter
columns and the media depth (Eq 1)
The Kozeny and Ergun Equations
In the laminar range of flow the Kozeny equation (Fair et al 1968) can be used to depict
the head loss (Eq 7) which is proportional to filtration rate (V) media depth (L) porosity term (
( )3
21εεminus
) and media size term ( 2)6(eqdψ
) The equation is generally acceptable for the flow that
has the Reynolds number (Re) less than 6 (Cleasby and Logsdon 1999)
The Ergun equation (Eq 9) has an additional term which is proportional to V2 in
addition to the Kozeny equation to depict the inert head loss The equation is adequate for the full
range of laminar transitional and inertial flow The first term of the Ergun equation is identical
to the Kozeny equation except for the numerical constant (Cleasby and Logsdon 1999) The
constant of the second term (k2) was reported to be 048 for crushed porous solids (Ergun 1952)
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
19
The sphericity for crumb rubber media is the only unknown parameter in the two equations Their
values were determined by empirical fitting of data from the filtration tests
Statistical Modeling
A total of 171 filtration data sets were collected (3 media sizes times 3 media depths times 19
runs) Based on statistical requirements 70 of data sets were chosen to initiate the regression
and the remaining 30 were used to validate the corresponding regression results Regressions
were performed using a least squares criterion and assuming that the residuals were normally
distributed and independent and had constant variance These assumptions were checked after
the model had been fitted Sigma Plot 100 (SPSS Inc Chicago IL) was used to initiate the
fitting process for media sphericity and to conduct the statistical multiple nonlinear regression for
developing a statistical model
For the regression process for media sphericity each data set consisted of values of an
actual head loss its corresponding filtration rate compressed media depth at this filtration rate
porosity at the compressed media depth and media size Data except actual head loss were used to
calculate head losses based on the Kozeny or Ergun equation Regressions were then performed
by varying the sphericity values in the two equations to obtain the best fit of the estimated head
loss data to the actual head loss data
For the regression process for a statistical model each data set consisted of values of an
actual head loss its corresponding filtration rate media size and initial media depth Data except
actual head loss were used to calculate estimated head losses based on a multiplicative power-law
relationship Then regressions were performed by varying the model coefficients to obtain the
best fit between the calculated head loss data and the actual head loss data
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
20
Regression Verification
For each attempted fit of sphericity to actual head loss data the hypotheses tested were
that the derived sphericity of the equation was significantly different from zero Expressed
mathematically the hypothesis was as follow s
0
0
Similarly the hypotheses tested for each coefficient of the statistical model were as
follows
For numerical constant K
0
0
For exponent of filtration rate a
0
0
For exponent of media depth b
0
0
For exponent of media size c
0
0
By examining the p-value of each coefficient the null hypothesis can be rejected and the
alternative one can be concluded if the p-value was very small (lt005 in this study)
Analysis of Means (ANOM) was performed using two-sample t-test for some cases
where comparison of model coefficients was necessary For example when comparing the
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
21
sphericity estimates from the Kozeny and Ergun equations the hypothesis tested was that the
derived sphericity estimates from the Kozeny equation was significantly different from the one
from the Ergun equation Expressed mat m ical e hypothesis was as follows he at ly th
If the variances were not quite different pooled variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
1 1
With degrees of freedom equal to 2
2 Eq 10
1 1 Eq 11
If the variances were quite different separate variance t-test was applied by using the
following equations to compute the t-statistic (Ott and Longnecker 2000)
With degrees of freedom equal to where frasl
Eq 12
By comparing the computed t-statistic with the corresponding one in the t-table the null
hypothesis can be rejected and the alternative one can be concluded if or
where α=005 in this study
Coefficient of determination (R2) is another criterion to verify these regressions It
represents the proportion of the total variability in the dependent variables that the regression
equation accounts for (Burton and Pitt 2001) as expressed in the following equation
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
22
Where SSerr is the sum of squared errors and SStot is the total sum of squares
R 1ssSS
Eq 13
An R2 of 10 indicates that the equation accounts for all the variability of dependent
variables and it is generally good and indicates a strong relation if R2 is large But it does not
guarantee a useful equation since high R2 can occur with insignificant equation coefficient if
only a few data observations are available Regressions in this study were validated based on the
following steps (1) examining the regression R2 and verification R2 (2) examining the residuals
of the resultant regressions statistically and graphically In addition the standard error of each
estimate which was computed from the variance of the predicted values was given as a measure
of the model variability
Graphical analyses of regressions were performed by examining the following
requirements according to Draper and Smith 1981 (1) residuals are independent (2) residuals
have zero mean (3) residuals have constant variance and (4) residuals follow a normal
distribution The purpose was to determine if the assumption of the regression error being
independent and normally distributed was valid Verification of the assumption of independent
residuals was performed by graphically plotting the residuals against its variables (filtration rate
media depth and media size) and observed values For this assumption to be valid the residuals
should be randomly distributed in these four plots around a zero line Verification of the
assumption of normal distribution of residuals was performed using a normal probability plot
Residuals that fall along a straight line in the plot and fall into the 95 confidence interval are
considered to be normally distributed If residuals fail to pass any one of these tests then the
regression is not valid for the data
Analysis of Variances (ANOVA) was performed to check constant variances for some
instances of the models if they passed the independence and normality tests The hypothesis
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
23
tested was that the variances were significantly different from each other The mathematical
expression was as follows
0
0
The Levenersquos test was performed with the assistance of MINITAB 15 (The Minitab Inc)
If the p-value was small (lt01 in this case) the null hypothesis can be rejected and the alternative
one can be concluded that the variances were not equal
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Chapter 4
RESULTS AND DISCUSSION
Media Sphericity
Sphericity a physical parameter that defines the roundness of the media shape has to be
provided if the Kozeny and Ergun equations are used For crumb rubber media their values were
estimated by empirical fitting of the actual head loss data to the predicted head loss data
computed by the two equations using the compressed media depth and porosity Two initial
assumptions were made before the regressions (1) Sphericity values for different media sizes
were the same and (2) Sphericity values at different operating conditions were the same Based
on that regressions were first performed using 70 of all data sets without differentiating media
sizes However the residuals of both equations could not pass the normality test and therefore
made the regressions invalid indicating that the assumptions could be partially wrong The first
assumption was then modified as Sphericity values for different media sizes were different With
the modified assumptions regressions were individually performed using 70 of data sets of
each media size and the results were summarized in Table 4-1 The Kozeny equation gave a
sphericity of 067 064 and 062 while the Ergun equation gave a sphericity of 071 075 and
079 for the 066 mm 120 mm and 190 mm crumb rubber media respectively All the p-values
were less than 00001 showing these coefficients were significant Two-sample t-test indicated
that a decreasing trend did not exist for the estimates from the Kozeny equation while there was
an increasing trend for the estimates from the Ergun equation In addition it was statistically
proved that the Ergun equations gave a higher sphericity estimate than the Kozeny equation The
95 confidence intervals gave a range of variability for those estimates It has to be pointed out
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
25
that the regressions for 066 mm and 190 mm media failed the normality test showing that the
second assumption could be wrong for the two sizes That is for the 120 mm crumb rubber
media it was acceptable to assume sphericity did not change due to compression but for the
other two sizes this assumption might be invalid But these estimates still could be applied in the
evaluation process of two equations since they were able to provide the highest regression R2 and
verification R2 which addressed the most variability of dependent variables
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Table 4-1 Sphericity estimates by the Kozeny and Ergun equations for crumb rubber media
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Prediction by the Kozeny Equation
Figure 4-1 compared the actual head loss with the predicted head loss by the Kozeny
equation using the original and compressed media depth and porosity respectively to evaluate
the applicability of the equation in crumb rubber filters The 45-degree-line in the figure depicted
the hypothetic head loss estimates that were precisely equal to the actual values It was found in
Figure 4-1a that the R2 value was only 087 and the Kozeny equation under-estimated head loss
especially at high filtration rates at each filter medium configuration This implied that the
Kozeny equation has limitation for crumb rubber filters if no compressed media depth and
porosity were available to compensate the media compression The under-estimation was likely
due to the decreased porosity resulted from the decreased media depth
When the Kozeny equation was examined by using the compressed media depth and
packed porosity it was found in Figure 4-1b that the Kozeny equation could give better predicted
values using the appropriate sphericity estimates Regression of predicted and actual head loss
data by the 45-degree-line gave a R2 of 097 which indicated that predicted data by the Kozeny
equation was close to the actual head loss data and the equation could be used for crumb rubber
filters However it was also noticed in the figure that some predicted data at high actual head loss
values were under-estimated which indicate that the equation had limitation when the filtration
rate was high at each filter medium configuration
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
28
Figure 4-1 Prediction of clean-bed head loss by the Kozeny equation (a) Prediction using theoriginal media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-2 shows the residuals of the Kozeny equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
29
because they were almost centered at the zero line (Figure 4-2 b and c) but they were dependent
on filtration rate and actual head loss because as filtration rate or actual head loss increased
residuals went from positive to negative values (Figure 4-2 a and d) In addition to relatively
lower R2 shown in Figure 4-1 analysis of residuals is against the criterion of independent
residuals as a good model Therefore the Kozeny equation had limitations and the derived
sphericity estimates could not describe the real shape of grains
Figure 4-2 Residual analysis of the Kozeny equation using compressed media depth and porosity
Prediction by the Ergun Equation
To examine the performance of the Ergun equation in the changing environment of
configurations in crumb rubber filters the original media depth and packed porosity were used to
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
30
calculate the head loss which were then compared with actual head loss as was shown in
Figure 4-3a It was found that the Ergun equation gave a R2 of 095 as well as under-estimation of
head loss when the actual head losses were higher than 100 cm at high filtration rates However
the under-estimation was not as great as that of the Kozeny equation using the same data sets of
original media depth and porosity and the R2 value was also higher than that of the Kozeny
equation This suggested that although the inert head loss included by the Ergun equation could
adjust the predicted data at high filtration rates it was still not sufficient to address the head loss
increase caused by the decrease of porosity due to media compression
Same with the testing procedures for the Kozeny equation the data sets of compressed
media depth and porosity from field study were applied to reflect the effect of media
compression The predicted head loss data were calculated by the Ergun equation and were
compared with the actual head loss data As shown in Figure 4-3b it was found that the R2 value
was 099 which indicated that the Ergun equation might be suitable for crumb rubber filters
Comparing the R2 value with that of the Kozeny equation the Ergun equation had higher R2 and
showed better fit to the actual head loss In addition the Ergun equation did not under-estimate
data at high filtration rates in the way the Kozeny equation did This was because the inert head
loss shown as the second term in the Ergun equation was able to adjust the prediction of head loss
development at high filtration rates by portraying a linear relationship not between h and V but
between h and V2
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
31
Figure 4-3 Prediction of clean-bed head loss by the Ergun equation (a) Prediction using the original media depth and porosity (b) Prediction using the compressed media depth and porosity
Figure 4-4 shows the residuals of the Ergun equation against three variables and actual
head loss It was found that the residuals were independent against filter depth and media size
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
32
because they were centered at the zero line (Figure 4-4 b and c) For the filtration rate however
the equation only displayed good residuals distribution when the filtration rate was high
indicating that the residuals were conditionally independent on filtration rate (Figure 4-4 a) And
they were dependent on actual head loss because when actual head loss was high residuals went
negatively higher (Figure 4-4 d) Although the R2 of the Ergun equation was relatively higher as
shown in Figure 4-3 analysis of residuals is against the criterion of independent residuals as a
good model Therefore the Ergun equation also had limitations although not as severe as the
Kozeny equation
Figure 4-4 Residual analysis of the Ergun equation using compressed media depth and porosity
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
33
Development of a Statistical Model
Both the Kozeny and Ergun equations had deficiencies for the prediction of clean-bed
head loss in crumb rubber filters and they may provided better fit when the data of compressed
media depth and porosity were used to reflect the media compression Their predictions using the
data of original media depth and porosity however could not provide the same degree of
comparison at all to the actual head loss data
The benefits of developing a statistical model for crumb rubber media are obvious (1)
No filtration tests need to be conducted to obtain the compression situations on media depth and
porosity (2) No hassle to derive a sphericity estimate since this value varies to the filtration
conditions and (3) the conventional equations are not good models for crumb rubber filtration
The statistical model development used a multiplicative power-law relationship which
resembled the Kozeny equation The model expressed in Eq 14 consisted of the three factors
examined in this study as independent variables filtration rate media depth and media size
ceq
ba dLVKh )()()(= Eq 14
Where
K = dimensionless constant for the statistic model
a = exponent of filtration rate
b = exponent of media depth
c = exponent of media size
Exponents of each factor and the numerical constant were obtained via regression using
70 of data sets The regression results were summarized in Table 4-2 The initial assumption of
the regression was that the model coefficients were the same for all media sizes However the
resultant group of coefficients failed the normality test indicating a probably wrong assumption
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
34
The assumption was then modified as the model coefficients were different among media sizes
The model then had a form as follows
ba LVKh )()(= Eq 15
Regressions were then performed individually using 70 of data sets for each media
size and all the three groups of coefficients passed the normality tests The p-values were less
than 00001 indicating that the coefficients were significant Standard errors and 95 confidence
intervals gave the ranges of the variability Regression R2 and verification R2 of the model were
higher than 0996 which demonstrated a potential to be a good model
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Table 4-2 Parameters in the statistical model
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
Effect of Parameters
Three design and operational parameters (Media size media depth and filtration rate)
were investigated for their effects Figure 4-5 shows the actual head loss profile at all filter
configurations Factorial calculations were conducted by using these data to explore the effects of
factors and possible interactions
Figure 4-5 Actual head loss profiles at all filter configurations
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
37
Although three-level factorial design was conducted in this experiment only the
results of medium and high level factorial calculation were shown in Figure 4-6 because the
results would be more meaningful since crumb rubber filters were usually designed to operate
at medium and high levels of filtration rates Four of seven effects were judged significant by
Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was
denoted by the vertical line and factors filtration rate media depth media size interaction
between filtration rate and media depth and interaction between filtration rate and media size
were determined to have significant effects The interaction between filtration rate and media
depth could be explained by the compression as the filtration rate increases which
correspondingly decreases media depth and therefore confounds head loss The interaction
between filtration rate and media size although very small was likely due to the change of
sphericity during compression because the assumption of equal sphericity at different
filtration conditions were proved to be wrong for some media sizes It is obvious that the
Kozeny and Ergun equations could not be able to address these distinctive effects of crumb
rubber media The statistical model developed by the multiplicative power-law relationship
could be used to analyze these effects on clean-bed head loss in crumb rubber filters
Interaction between filter depth and media size and interaction between all three factors were
statistically insignificant therefore they were not included in the following discussion
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
38
Figure 4-6 Pareto chart of the standardized effects showing the results of medium and high level factorial calculation (Response is observed head loss in centimeters Significance level = 005)
Effect of filtration rate
The exponents of filtration rate shown as 155 151 and 175 for 066 mm 120 mm
and 190 mm crumb rubber media in the statistical model were all higher than 1 found in the
Kozeny equation This could explain the under-estimation of the Kozeny equation because the
filtration rate in crumb rubber filters had a larger impact than that in other conventional granular
media filters This caused the actual head loss to increase faster when the filtration rate was high
The faster increase was due to the rapid development of inert head loss that could not be
addressed by a linear relationship between head loss and filtration rate
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
39
Effect of interaction between filtration rate and media depth
The larger exponent of filtration rate was also believed to be caused by the interaction
between filtration rate and media depth due to compression of the filter media which was shown
in Figure 4-7 The filter depth was decreased as the filtration rate increased It was found that the
crumb rubber filter that had media size of 12 mm and filter depth of 09 m was compressed by
9 as the filtration rate increased from 0 to 73 m3m2h However for conventional granular
filters the filter depth and filtration rate were independent parameters In crumb rubber filters the
phenomenon could not be addressed by the conventional head loss models
Filtration rate (m3hourm2)
0 20 40 60 80
Filte
r dep
th (c
m)
102
104
106
108
110
112
114
116
Figure 4-7 Impact of filtration rate on filter depth for a crumb rubber filter that has media size of120 mm and filter depth of 09 m
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
40
Effect of media depth
The exponents of media depth shown as 135 097 and 121 for 066 mm 120 mm and
190 mm crumb rubber media in the statistical model showed the influence of media compression
on clean-bed head loss because the exponent was found to be 1 in both the Kozeny and Ergun
equations The effect of media compression in crumb rubber filters was complicated according to
the head loss theories The compression decreased the media depth which could decrease head
loss but it also decreased the porosity at the same time which could increase head loss Figure 4-
8 showed the media depth and porosity terms change at one filter configuration It was found that
for a crumb rubber filter loaded with 066 mm crumb rubber media to a depth of 06 m as the
filtration rate gradually increased from 0 to 733 m3m2h the media depth was steadily decreased
by 136 However the decrease in media depth did not cause a corresponding decrease in its
exponent which was used to be 1 as shown in the Kozeny and Ergun equations In fact the
exponent of media depth was increased to 135 because the porosity term 3
2)1(εεminus
was increased
by 695 due to the compression The porosity term 3
)1(εεminus
in the second part of the Ergun
equation did not increase as much as the term 3
2)1(εεminus
in Kozeny equation which only
increased by 464 But it still played an important role when the filtration rate was high and the
inert head loss was dominant Therefore the exponent of media depth implied the overall
influence of media depth decrease and porosity terms increase in crumb rubber filters
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
41
Figure 4-8 Impact of media compression for the filter configuration of 066 mm crumb rubber media and 06 m media depth
Effect of media size
The exponents of media size shown as -154 in the statistical model which did not
differentiate media sizes was between -2 in the Kozeny equation and -1 in the second term of
Ergun equation indicating that the Ergun equation might be able to address the effect of this
factor while the Kozeny equation could not
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-
42
Effect of interaction between media size and filtration rate
Filtration rate affected crumb rubber media by changing their sphericity The assumption
of equal sphericity at one filter configuration was unacceptable for some media sizes when their
sphericity estimates were verified None of conventional equations could address this problem
yet since they both assumed that sphericity did not change However it was not necessarily the
case for crumb rubber media Compression could change the shape of the media especially when
the filters were operated at higher filtration rates which correspondingly exerted higher pressure
to change the media shape
Prediction by the Statistical Model
Because the statistical model was developed using actual head loss initial media depth
filtration rate and media size it had included the effect of media compression that changes the
filter configurations Also it addressed the problem encountered by the Kozeny and Ergun
equations that is test must be conducted at all filtration rates to obtain the data of compressed
media depth and porosity Figure 4-9 showed the comparison of actual head loss and predicted
head loss data by the statistical model at all filter configurations The statistical model did not
under-estimate head loss at high filtration rates in the way the Kozeny and Ergun equation did
The R2 value was as high as 0998
43
Figure 4-9 Prediction of clean-bed head loss by the statistical model
Figure 4-10 shows the residuals of the statistical model against three variables and actual
head loss It was found that the residuals were independent against all variables since they were
all centered at the zero line In addition when the hypothesis of equal variances of residuals
against actual head loss was tested the p-value for Levenersquos test was 0122 Comparing to
01 the default choice for rejecting the hypothesis of equal variances the p-value was
higher and therefore no conclusion could be made that they were not equal Because normality
independence and constant variances all applied the statistical model was valid for head loss
prediction of crumb rubber filters
44
Figure 4-10 Residual analysis of the statistical model using initial media depth and porosity
Chapter 5
CONCLUSIONS
(1) Both the Kozeny and Ergun equations were unacceptable for clean-bed head loss
prediction in crumb rubber filters especially when the test data of compressed media depth and
porosity were unavailable to compensate the media compression
(2) The effects of filtration rate media depth media size and their interactions in crumb
rubber filters were different from conventional granular media filters due to the media
compression
(3) A statistical model developed by the multiplicative power-law relationship is able to
predict clean-bed head loss in crumb rubber filters using the data of original media depth
filtration rate and media size The model is statistically valid and can be applied for crumb rubber
filtration
BIBLIOGRAPHY
Abdul-Wahab SA Bakheit CS and Al-Alawi SM (2005) Principle component and multiple
regression analysis in modeling of ground-level ozone and factors affecting its
concentrations Environmental Modeling and Software 20 1263
American Society for Testing and Materials (1993) Concrete and Aggregates 1993 Annual Book
of ASTM Standards Vol 0402 Philadelphia Pennsylvania ASTM
Bear J (1972) Dynamics of fluids in porous media Dover New York
Chen P (2004) Ballast water treatment by crumb rubber filtration ndash a preliminary study
Graduate Thesis Environmental Pollution Control Program at Penn State Harrisburg
Pennsylvania USA
Cleasby J L and Logsdon G S (1999) Granular Bed and Precoat Filtration Chap 8 in R D
Letterman (ed) Water Quality and Treatment A Hankbook of Community Water
Supplies McGraw-Hill New York
Ergun S and Orning A (1949) Fluid flow through randomly packed columns and fluidized
beds Inds and Engrg Chem 41(6) 1179-1184
Carman P (1937) Fluid flow through granular beds Trans Inst Of Chemical Engrg 15 150
47
Drapner NS Smith H (1981) Applied regression analysis 2nd ed John Wiley and Sons Inc
New York
Ergun S (1952) Fluid flow through packed columns Chem Eng Prog 48 2 89-94
Fair G Geyer J and Okun D (1968) Water and wastewater engineering Vol2 Water
purification and wastewater treatment and disposal Wiley New York
Fair G M and Hatch L P (1933) Fundamental Factors Governing the Streamline Flow of
Water through Sand J AWWA 25 11 1551-1565
Graf C and Xie Y F (2000) Gravity downflow filtration using crumb rubber media for tertiary
wastewater filtration Keystone Water Quality Manager 33 12-15
Crittenden JC et al (2005) Granular filtration Chap 11 in 2nd ed Water treatment principles
and design John Wiley amp Sons Inc Hoboken New Jersey
Hsiung S ndashY (2003) Filtration using a crumb rubber filter medium preliminary studies using
secondary wastewater effluent as feed MS thesis the Pennsylvania State University
University Park PA USA
Kozeny J (1927a) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Sitzungsbericht Akad Wiss 136 271-306 Wein Austria
48
Kozeny J (1927b) Ueger Kapillare Leutung des Wassers im Boden (On Cappillary Conduction
of Water in Soil) Wasserkraft Wasserwirtschaft 22 67-78
Lang J S Giron J J Hansen A T Trussell R R and Hodges W E J (1993) Investigating
Filter Performance as a Function of the Ratio of Filter Size to Media Size J AWWA 85
10 122-130
Lenth RV (1989) Quick and easy analysis of unreplicated factorials Technometrics 31 469-
473
Ott RL Longnecker MT (2000) An introduction to statistics and data analysis 5th ed
Duxbury Press Florence Kentucky USA
Rubber Manufacturer Association (2002) US Scrap tire market 2001 Washington DC
Sunthonpagasit N amp Hickman HL Jr (2003) Manufacturing and utilizing crumb rubber from
scrap tires MSW Management 13
Tang Z Butkus M A and Xie Y F (2006a) Crumb rubber filtration A potential technology
for ballast water treatment Marine Environmental Research 61(4) 410-423
Tang Z Butkus M A and Xie Y F (2006b) The Effects of Various Factors on Ballast Water
Treatment Using Crumb Rubber Filtration Statistic Analysis Environmental
Engineering Science 23(3) 561-569
49
Trusell R Chang M (1999) Review of flow through porous media as applied to head loss in water
filters J Environ Eng 125 11 998-1006
Trussell R R Chang M M Lang J S Guzman V and Hodges W E Jr (1999) Estimating
the Porosity of a Full-Scale Anthracite Filter Bed J AWWA 91 12 54-63
Trussell R R Trussell A Lang J S and Tate C (1980) Recent Developments in Filtration
System Design J AWWA 72 12 705-710
United States Environmental Protection Agency (USEPA) (1993) Scrap tire handbook EPA905-
K-001 USEPA Region 5 USA October
United States Environmental Protection Agency (USEPA) (2006) Scrap tire cleanup guidebook
EPA-905-B-06-001 USEPA Region 5 and Illinois EPA Bureau of Land USA January
Xie Y Bryon K Gaul A (2001) Using crumb rubber as a filter media for wastewater filtration
2001 Technical Meeting of American Chemical Society Rubber Division Cleveland
OH USA
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
- MS_Thesispdf
-
- Chapter 1INTRODUCTION
-
- Statement of Problem
- Objectives and Scope
- Thesis Organization
-
- Chapter 2BACKGROUND
-
- Manufacture of Crumb Rubber
- Characterization of Filter Media
- Theory of Crumb Rubber Filtration
- Head Loss Theory
-
- Chapter 3MATERIALS AND METHODS
-
- Filter Setup
- Filter Media
- Design and Operational Conditions
- The Kozeny and Ergun Equations
- Statistical Modeling
- Regression Verification
-
- Chapter 4RESULTS AND DISCUSSION
-
- Media Sphericity
- Prediction by the Kozeny Equation
- Prediction by the Ergun Equation
- Development of a Statistical Model
- Effect of Parameters
-
- Although three-level factorial design was conducted in this experiment only the results of medium and high level factorial calculation were shown in Figure 4-6 because the results would be more meaningful since crumb rubber filters were usually designed to operate at medium and high levels of filtration rates Four of seven effects were judged significant by Lenthrsquos method (Lenth 1989) with the significance level of 005 The threshold value was denoted by the vertical line and factors filtration rate media depth media size interaction between filtration rate and media depth and interaction between filtration rate and media size were determined to have significant effects The interaction between filtration rate and media depth could be explained by the compression as the filtration rate increases which correspondingly decreases media depth and therefore confounds head loss The interaction between filtration rate and media size although very small was likely due to the change of sphericity during compression because the assumption of equal sphericity at different filtration conditions were proved to be wrong for some media sizes It is obvious that the Kozeny and Ergun equations could not be able to address these distinctive effects of crumb rubber media The statistical model developed by the multiplicative power-law relationship could be used to analyze these effects on clean-bed head loss in crumb rubber filters Interaction between filter depth and media size and interaction between all three factors were statistically insignificant therefore they were not included in the following discussion
- Effect of filtration rate
- Effect of interaction between filtration rate and media depth
- Effect of media depth
- Effect of media size
- Effect of interaction between media size and filtration rate
-
- Prediction by the Statistical Model
-
- Chapter 5CONCLUSIONS
- BIBLIOGRAPHY
-