a study on a novel sewer overflow screening system

A Study on a Novel Sewer Overflow Screening System

Thesis

Submitted in Total Fulfilment of the Requirements for the Degree of

Doctor of Philosophy

By

Md Abdul Aziz

Student ID: 6632475

Faculty of Science, Engineering and Technology (FSET)

Swinburne University of Technology

Hawthorn, Victoria 3122

Australia

2016

Dedicated to my parents and wife

Md Abdul Aziz i

Declaration

The author hereby declares that this thesis, submitted in fulfilment of the

requirements for the Degree of Doctor of Philosophy, contains no material which has

been accepted for the award of any other degree or diploma, except where due reference

is made in the text. To the best of my knowledge, this thesis contains no material

previously published or written by another person, except where due reference is made

in the text. In places where the work is based on joint research or publications, this thesis

discloses the relative contribution of the respective workers or authors.

Part of this thesis have been copyedited and proofread by Dr Jillian Graham

(Articulate Writing Solutions), whose services are consistent with those outlined in

Section D of the Australian Standards for Editing Practice (ASEP).

Md Abdul Aziz

March, 2016

Md Abdul Aziz ii

Abstract

After heavy rainfall, sewer overflow spills into receiving water bodies, causing

serious concern for the environment and public health. This has led to a need to conduct

research in order to develop different types of screening devices. Some of the limitations

in existing screening processes include the expense involved in ongoing operation and

maintenance costs, poor capture efficiency, and blinding on the screens. Most sewage

overflow screens use sophisticated electro-mechanical systems. These systems are

expensive and complex, and the possibility of malfunctions is a cause for concern,

particularly in unstaffed remote locations. In this study, novel sewer overflow screening

devices have been developed and evaluated to overcome the limitations of low capture

efficiency and high initial and maintenance costs.

The set-ups of experimental screening devices involve significant cost and time.

To overcome this issue, the proposed sewer overflow screening device was analysed

using a 3D computational fluid dynamics (CFD) model. Plausibility checking of the CFD

model was done using an analytical model. The CFD model helped in designing the

location of perforations, inlet length and orientation. However, a limitation was

encountered in the attempt to reduce blinding of the perforations. To address this, a

revised design using comb separators instead of perforations was proposed.

Laboratory tests of the revised ‘comb separator’ were performed to determine the

trapping efficiencies for common sewer solids under different experimental conditions.

Comparisons of the ‘comb separator’ with the industry standard Hydro-JetTM suggest that

the former can perform better when there are blockages, and that it has a higher capture

efficiency during periods of low flow.

It was important to conduct modelling analysis to overcome physical limitations,

and to visualise a range of experimental conditions using CFD and experimental

analysis. An artificial neural network (ANN) model was adopted in this study, as it has

the capacity to intelligently predict the outcome of complex, non-linear physical systems

with relatively poorly-understood physio-chemical processes. The model successfully

predicted the experimental results with more than 90% accuracy, with an average

absolute percentage error of about 7%.

Md Abdul Aziz iii

It was also necessary to conduct sensitivity analysis of the comb separator to

understand its performance qualitatively and quantitatively. A multiple linear regression

(MLR) model was developed taking five input parameters (runtime, flows, effective comb

spacing, weir opening and layers of combs) and output capture efficiency for model

generation. The MLR model results suggest that the most significant predictor influencing

sewer solids capture efficiency is effective spacing, followed by flow discharge and

runtime.

The proposed ‘comb separator’ screening system shows good application

potential during testing in situations of low flow. Further research is recommended,

including a testing plan for high flow scenarios (such as flooding) and onsite testing of

the ‘comb separator’.

Md Abdul Aziz iv

Acknowledgments

I would like to express my deepest gratitude to Almighty Allah, the most merciful,

the most benevolent and the master of the judgement day, for giving me patience and

helping me through to completion of my study at Swinburne University of Technology.

I owe my deepest thanks to my principle coordinating supervisor, Associate

Professor Monzur Imteaz, for his continuous support and guidance, without which it

would not have been possible to complete this thesis. It was an honour to work with him.

During the latter stage of my research project, I was working full-time and studying part-

time, which meant that my thesis took longer to complete. I am indebted to my supervisor

for understanding my situation, and for continuing to show great faith in me.

It gives me pleasure to acknowledge my industry supervisor, Dr Don Phillips, who

was inspiring and supportive during this project. I am also greatly appreciative of the

contribution of Dr Shirley Gato-Trinidad, my co-supervisor. I am grateful to Dr Don

Phillips and Dr Shirley Gato for helping me to settle into in my research project, and for

their on-going guidance.

I would like to acknowledge Associate Professor Jamal Naser, Dr Nazmul Huda

and Dr Morshed Alam for their consistent support and contributions, especially in

developing ideas in the field of Computational Fluid Dynamic (CFD) Modelling. I am

indebted to Dr Tanveer Ahmed Choudhury for introducing me to the world of Artificial

Intelligence. I thank Dr Sylvia Mackie, Carol Farr, Dr Samsuzzoha, Dr Mahabubur

Mollah, Taha Mollah, Paul Fennell and Dr Jan for their assistance in reviewing my thesis,

journal articles and conference papers. I would also like to thank my exam review

committee, including Dr Arul Arulrajah, Dr Scott Rayburg and Dr Richard Manasseh, who

reviewed my work and provided valuable suggestions that contributed immensely to my

completion of this research project.

I am indebted to Swinburne University of Technology for granting me the

Swinburne University postgraduate research award (SUPRA) to facilitate my research

and to support me financially in the first two years of my candidature. I am also grateful

to the University for their flexibility in allowing me to switch from full-time to part-time

student.

I would like to thank my parents, sisters and brother for their constant motivation

and encouragement. My father was the first to inspire me towards my PhD research

Md Abdul Aziz v

during my childhood, and must now be a proud man in heaven. To my wonderful mother

I am particularly and profoundly grateful. When others doubted my educational ability,

she stood by me, giving me the support and courage to struggle on against the odds.

Special thanks go to my wife. She has made me feel as if all the pleasure in study

were mine and all the pain hers during this tough seven years. I honour Almighty Allah

for making it possible for her to share this journey. Her patience and tolerance during the

more difficult times are noted and deeply appreciated.

I am very proud to work for the Wimmera Catchment Management Authority, the

best working environment one could imagine. I am grateful for the affection and respect

this organisation has bestowed on me. I thank Mr Dave Brennen (CEO Wimmera CMA)

for allowing me to continue my study while working full-time, and also my manager Mr

Paul Fennell, an admirable and genuine gentleman whose friendship I treasure.

Thanks to Dr Amin and his family, Sadi and his family, Tanveer and his family,

Morshed, Maruf, Nazmul and their families, and all our family friends in Horsham and

Melbourne, who have shared lots of laughter and mateship. Many thanks to my nephews

and nieces (Aurna, Purba, Piyal, Shreya, Arpon and Megh) who have been a great source

of inspiration; they are fine examples of peace, tranquillity and innocence. Last but not

least, thanks to all of those friends who may not be aware how much they helped and

inspired me. If I have forgotten anyone, this is a shortcoming on my part, which I hope will

be forgiven.

Table of Contents

Md Abdul Aziz vi

Table of Contents

Chapter 1 INTRODUCTION ....................................................................... 1

1.1 Background ........................................................................................................ 2

1.2 Problem Statement ............................................................................................ 3

1.3 Objective of this Study ...................................................................................... 4

1.4 Research Contributions .................................................................................... 5

1.5 Thesis Structure and Overview ........................................................................ 6

Chapter 2 LITERATURE REVIEW ............................................................. 8

2.1 Introduction ........................................................................................................ 9

2.2 Summary of Current Screening Applications ................................................ 12

2.2.1 Static screening ............................................................................... 13

2.2.2 Mechanical and Electrical screening devices ................................ 14

2.3 Methods to improve Sewage Overflow Screening ........................................ 23

2.3.1 Hydrodynamics Applications .......................................................... 23

2.3.2 Experimental Investigations ........................................................... 25

2.3.3 Artificial Neural Network (ANN) Applications ................................ 26

2.3.4 Sensitivity Analysis to Model Results ............................................ 28

2.4 Identification of Research Needs.................................................................... 29

2.5 Summary ...................................................................................................... 30

Chapter 3 RESEARCH METHODS .......................................................... 32

3.1 Introduction ...................................................................................................... 33

3.2 Research Questions ........................................................................................ 33

3.3 Research Process ............................................................................................ 34

3.4 Research Design.............................................................................................. 36

3.4.1 Computational Fluid Dynamic (CFD) Analysis .............................. 37

3.4.2 Laboratory Experiments .................................................................. 37

3.4.3 ANN model to complement deterministic approach ..................... 38

Table of Contents

Md Abdul Aziz vii

3.5 Analysis Procedure ......................................................................................... 38

3.5.1 CFD and Analytical Modeling .......................................................... 38

3.5.2 Experimental Investigation ............................................................. 39

3.5.3 ANN Modeling .................................................................................. 39

3.5.4 Sensitivity Analysis ......................................................................... 39

3.6 Summary ...................................................................................................... 40

Chapter 4 HYDRODYNAMIC ANALYSIS ................................................. 42

4.1 Introduction ...................................................................................................... 43

4.2 Screening Concept .......................................................................................... 46

4.3 Development of the Analytical Model ............................................................. 48

4.4 Computational Fluid Dynamics (CFD) Model ................................................. 51

4.4.1 Finite Volume Method (FVM) ........................................................... 53

4.4.2 Multiphase Flow Modelling ............................................................. 54

4.4.3 Approaches to Multiphase Modelling ............................................. 55

4.4.4 Euler-Lagrange Approach ............................................................... 55

4.4.5 Euler-Euler Approach ...................................................................... 55

4.4.6 Model Geometry and Computational Methodology ....................... 58

4.4.7 Boundary Conditions ...................................................................... 64

4.4.8 Explaining CFD Results .................................................................. 65

4.4.9 Plausibility check of the CFD model .............................................. 68

4.5 Discussion of Results ..................................................................................... 69

4.5.1 Discussion of Hydrodynamic results ............................................. 69

4.5.2 Discussing Location of Circular holes ........................................... 75

4.5.3 Discussion of the Inlet performance .............................................. 76

4.5.4 Standard Weir orientation ............................................................... 78

4.6 Limitation of Screening Device ....................................................................... 80

4.7 Summary ...................................................................................................... 80

Table of Contents

Md Abdul Aziz viii

Chapter 5 IMPROVEMENT OF THE SCREENING DEVICE ...................... 82

5.1 Introduction ...................................................................................................... 83

5.2 Methodology used in the Experiment ............................................................ 86

5.2.1 Data Collection................................................................................. 86

5.3 Test Procedure ................................................................................................ 93

5.3.1 Experimental Conditions Used ....................................................... 93

5.4 Discussions of Results ................................................................................. 101

5.4.1 Sewage solids more than 10 mm in diameter .............................. 101

5.4.2 Sewage solids less than 10 mm in diameter ................................ 102

5.4.3 Performance comparison of Comb Separator and Hydro-JetTM 104

5.5 Limitations of the Experiment ....................................................................... 105

5.6 Summary .................................................................................................... 106

Chapter 6 ANN MODEL TO COMPLEMENT CFD AND LABORATORY TESTING ......................................................................................................... 108

6.1 Introduction .................................................................................................... 109

6.2 Artificial Neural Network (ANN) .................................................................... 111

6.3 Description of Network Structure ................................................................. 116

6.3.1 Artificial Neuron Model .................................................................. 116

6.3.2 Multi-Layer Feed Forward Neural Network Structure .................. 117

6.4 Network Learning .......................................................................................... 117

6.4.1 Back propagation algorithm ......................................................... 118

6.4.2 Levenberg-Marquardt Algorithm .................................................. 118

6.4.3 Resilient Back Propagation Algorithm ......................................... 119

6.5 Data Collection and Pre-processing ............................................................. 120

6.5.1 Creation of Database ..................................................................... 120

6.6 Result Analysis and Discussion ................................................................... 123

6.7 ANN Model Validations.................................................................................. 126

6.8 Summary .................................................................................................... 127

Table of Contents

Md Abdul Aziz ix

Chapter 7 SENSITIVITY ANALYSIS OF THE COMB SEPARATOR ....... 129

7.1 Introduction .................................................................................................... 130

7.1.1 Objective ........................................................................................ 131

7.2 Background .................................................................................................... 131

7.2.1 Developing a Multiple Linear Regression (MLR) Model .............. 134

7.2.2 Summary of the Model .................................................................. 139

7.2.3 Development of the Dataset Using Sampling Techniques .......... 145

7.3 Results and Discussion ................................................................................ 149

7.3.1 Relative Significance of the Input Parameters ............................. 149

7.3.2 Selection of the Input Parameters ................................................ 150

7.3.3 Impact of Effective Spacing on Capture Efficiency ..................... 152

7.3.4 Impact of Flow on Capture Efficiency .......................................... 152

7.3.5 Runtime Impact on Capture Efficiency......................................... 153

7.4 Summary .................................................................................................... 154

Chapter 8 CONCLUSIONS .................................................................... 157

8.1 Introduction .................................................................................................... 158

8.2 Research Summary ....................................................................................... 158

8.3 Knowledge Contributions ............................................................................. 159

8.4 Limitations .................................................................................................... 161

8.5 Future Research ............................................................................................ 162

References ........................................................................................... 164

Appendix A:Experimental Data ............................................................ 179

List of Figures

Md Abdul Aziz x

List of Figures Figure 2.1: Frequency of Trash and Litter from Various Sources (Source: [156]) ........ 12

Figure 2.2: Static bar screens in operation (source: [3]) .............................................. 13

Figure 2.3: Brush-Raked fine screen static screen ...................................................... 14

Figure 2.4: Typical mechanical bar screens ................................................................ 16

Figure 2.5: Internally Fed Rotary Screen..................................................................... 18

Figure 2.6: Centrifugal screen ..................................................................................... 19

Figure 2.7: The cut-away view of the Rotary-Jet Screen ............................................. 20

Figure 2.8: Schematic representation of the Hydro JetTM ............................................ 22

Figure 2.9: A typical artificial neuron k ........................................................................ 27

Figure 3.1: Flow Chart of the Current Research Plan36 (Where RQ stands for Research

Questions) .................................................................................................................. 36

Figure 3.2: Flow chart of the current research plan ..................................................... 41

Figure 4.1: Schematic diagram of the proposed sewage overflow screening device ... 47

Figure 4.2: Front views of the proposed device under different phases ....................... 47

Figure 4.3: Breakdown of the flow components of the experimental device ................ 50

Figure 4.4: Geometric details of the screener device .................................................. 59

Figure 4.5: Position 1 (condition 1) is the inlet parallel to the ogee weir ...................... 63

Figure 4.6: Position 2 (condition 2) is the inlet perpendicular to the ogee weir ............ 63

Figure 4.7: Boundary conditions used in the CFD model ............................................ 64

Figure 4.8: Water levels over the weir at different locations ........................................ 66

Figure 4.9: Volume fraction of water at inlet parallel (position 1) to ogee weir ............. 67

Figure 4.10: Volume fraction of water at inlet perpendicular (position 2) to ogee weir . 67

Figure 4.11: Comparison of flow velocities over the top of the ogee weir .................... 68

List of Figures

Md Abdul Aziz xi

Figure 4.12: Comparison of flow velocities 6cm downstream of the ogee weir ............ 68

Figure 4.13 Comparison of water level along the flow for condition 1 .......................... 69

Figure 4.14 Comparison of water level along the flow for condition 2 .......................... 70

Figure 4.15: Velocity vector at the inlets parallel (left) ................................................. 71

Figure 4.16: Velocity vector at perpendicular (right) to the ogee weir .......................... 71

Figure 4.17: Comparison of flow velocities along the width for condition 1 .................. 72

Figure 4.18: Comparison of flow velocities along the width for condition 2 .................. 72

Figure 4.19: Pressure variation at condition 1 ............................................................. 73

Figure 4.20: Pressure variation at condition 2. ............................................................ 73

Figure 4.21: Shear stress distributions for the inlet parallel to the weir width ............... 74

Figure 4.22: Comparison of shearing stress along the bottom of the curved surface... 75

Figure 4.23: Comparison of water levels along the flow for condition 1, water level on the

top of the weir, 3cm and 6 cm downstream ................................................................. 76

Figure 4.24: Comparison of water levels along the flow for conditions 2, water level on

the top of the weir, 3cm and 6 cm downstream ........................................................... 76

Figure 4.25: Impact of device inlet position on the wave reflection viewed from the back

of the weir with a lateral inflow .................................................................................... 77

Figure 4.26: The Waterways Experimental Station (WES) standard spillway shapes .. 78

Figure 4.27: CFD results viewed from the back of the weir with a lateral inflow on four

standard inlet orientations as suggested by the U.S. Army Engineers Waterways

Experimental Station ................................................................................................... 79

Figure 5.1: Experimental set ups for the Comb Separator ........................................... 86

Figure 5.2: Flow diagram of the revised screening experimental works ...................... 87

Figure 5.3: Concept diagram of target capture efficiency curve ................................... 88

Figure 5.4: Experimental set-up for the proposed sewage overflow screening device . 89

List of Figures

Md Abdul Aziz xii

Figure 5.5: Operation procedure of new sewage overflow screening device- Phase 1.89

Figure 5.6: Operation procedure of new sewage overflow screening device-Phase 2 . 90

Figure 5.7: The design parameter of the ball valve chamber ....................................... 91

Figure 5.8: Vertical position of the Comb Separator in the device ............................... 94

Figure 5.9: Top view of the position of the Comb Separator ........................................ 94

Figure 5.10: Capture of sewage solids during an experimental run ............................. 95

Figure 5.11: Capture of sewage solids after an experimental run ................................ 95

Figure 5.12: Mixing of sewer solids on to the Comb Separator device ........................ 96

Figure 5.13: Comb Separator is in operation, nappe clear the retention screen .......... 96

Figure 5.14: Sewage solids used in the test ................................................................ 97

Figure 5.15: Comb Separator in operation .................................................................. 98

Figure 5.16: Capture efficiency of sewer solids at different experimental set ups ...... 102

Figure 5.17: Effective comb spacing (mm) against average capture efficiency (%) and

flow (l/s per metre length of weir) .............................................................................. 104

Figure 6.1: Conceptual diagram showing an analogy of the work principal between the

human brain and the ANN model .............................................................................. 111

Figure 6.2: Conceptual diagram of input-output and weight adjustment .................... 112

Figure 6.3: Demonstration of over-fitting for a function approximating ANN .............. 115

Figure 6.4: A Non-linear model of an artificial neuron................................................ 117

Figure 6.5: Block diagram of proposed ANN model. ................................................. 122

Figure 6.6: Comparison of different node in the 1st and 2nd hidden layer ................... 124

Figure 6.7: Comparison of experimental and ANN predicted capture efficiency ........ 125

Figure 6.8: Regression Value for training, validation and test results ........................ 126

Figure 6.9: Comparison of experimental and model results for validation dataset.

Experiment numbers 1 to 4 were conducted on a specific experimental setup and then

List of Figures

Md Abdul Aziz xiii

the flow was changed for experiments 5 to 8. Solids used for different test numbers: 1

and 5 = cigarette butts, 2 and 6 = condoms, 3 and 7 = tampons, and 4 and 8 = wipe

papers. ..................................................................................................................... 127

Figure 7.1: Flow chart of the methodology adopted in the sensitivity analysis. .......... 133

Figure 7.2 Regression Standardized predicted value against residual ...................... 136

Figure 7.3: Normal distribution plot of the Standardized Residual vs Frequency ....... 137

Figure 7.4: Experimental data shows Observed vs Expected Cumulative probability 138

Figure 7.5: P-P plots for the runtime predictor ........................................................... 146

Figure 7.6: Data on both sides of normal for runtime ................................................ 146

Figure 7.7: P-P plots for the runtime predictor ........................................................... 147

Figure 7.8: Data on both sides of normal for flow ...................................................... 147

Figure 7.9 P-P plot of effective spacing ..................................................................... 148

Figure 7.10 Data on both sides of normal for effective spacing ................................. 148

Figure 7.11: Relationship between Effective spacing (mm) and Capture Efficiency (%)

................................................................................................................................. 152

Figure 7.12: Relationship between the Flowrate (l/s) and Capture Efficiency (%) ...... 153

Figure 7.13: Relationship between Runtime (min) and Capture Efficiency (%) .......... 154

List of Tables

Md Abdul Aziz xiv

List of Tables

Table 2.1: Sewer solids available in the urban sewage treatment [156] ...................... 10

Table 2.2: Regulatory requirement for aesthetics [147]............................................... 25

Table 4.1: Calculation of hydrodynamic parameters from analytical results ................ 51

Table 5.1 Different experimental setups used for sewage solid less than 10 mm diameter

................................................................................................................................... 98

Table 5.2 Comb Separator Testing at Experimental Set up 1 ...................................... 99



Table 5.5 Comb Separator Testing at Experimental Set up 4 .................................... 100



Table 5.8 Experimental set ups at five different conditions for sewage solids more than

10 mm in diameter .................................................................................................... 101

Table 5.9 Capture efficiency with different experimental set ups ............................... 103

Table 5.10 Comparative performance of the Hydro-JetTM and the Comb Separator .. 105

Table 6.1 Comparison of different training paradigms ............................................... 125

Table 7.1 Provides descriptive statistics of the Comb Separator experimental data set

................................................................................................................................. 140

Table 7.2 Correlations of different parameters .......................................................... 140

Table 7.3 Summary for the multiple linear regression (MLR) model .......................... 141

Table 7.4 ANOVA table for the MLR model ............................................................... 142

Table 7.5 Coefficient of different parameters ............................................................ 143

Table 7.6 Schematic diagram of SaSAT data generation .......................................... 145

List of Tables

Md Abdul Aziz xv

Table 7.7: Results using the Latin Hypercube Sampling (LHS) method for 10,000 data

................................................................................................................................. 150

Table 7.8 Comparison between initial and final model results ................................... 151

List of Notations

Md Abdul Aziz xvi

List of Notations

Symbol Unit Description

ρ kg/m3 Density

μ m2/s Dynamic viscosity

P Pascal Atmospheric pressure

gx m/s2 Gravitational force in X direction

gy m/s2 Gravitational force in Y direction

q m2/s Flow per unit width

Pascal Shear stress at the boundary

v m/s Average velocity

Q m3/s Total discharge

L m Lateral crest length or width

He m Total head upstream from the crest

l/s l/s Liter/second

l/s/m l/s/m Litre/second/meter

mm mm Millimetre

u m/s Velocity in the x- directions

v m/s Velocity in the y- directions

w m/s Velocity in the z (vertical)- directions

Hd m The depth of water upstream of spillway

C0

Discharge coefficient

List of Notations

Md Abdul Aziz xvii

Symbol Unit Description

N No Number of phases

k Volume fraction

k Density of phase

vk Phase velocity

k Pa.s Molecular viscosity

tkT Reynolds stress

k Kronecker delta function

tk Turbulent viscosity

SItc

, Sato’s viscosity

k Turbulent kinetic energy

Dissipation rate of turbulent kinetic energy

DC Drag coefficient

Reb Bubble Reynolds number

c Kinematic viscosity

Db mm Bubble diameter

d Dispersed phase

∆t sec Time step increment in the model

w Shear stress

Uτ m/s The friction velocity

C.E Capture Efficiency

kpw Synaptic weight

List of Notations

Md Abdul Aziz xviii

Activation function

px Input signal

H Hessian matrix

J Jacobian matrix

zk Old parameter value

R Regression coefficient

MSm Average Improvement in Prediction by Model

MSr Average Difference between Model and Observed data

df Degree of freedom

List of Acronyms

Md Abdul Aziz xix

List of Acronyms

Acronym Description

AMP2 Asset Management Plan 2

ANN Artificial neural network

AVL Fire Computational Fluid Dynamic software

BP Back propagation algorithm

BR Bayesian regularization algorithm

CFD Computation Fluid Dynamics

CMC Center for Marine Conservation

CSO Combined Sewer Overflow

CVs Control Volumes

EPA Environmental Protection Agency

FEM Finite Element Method

FDM Finite Difference Method

FLUENT Computational Fluid Dynamics software

FVM Finite Volume Method

HDVS Hydrodynamic Vortex separation

LHS Latin Hypercube Sampling

LM Levenberg-Marquardt algorithm

MLP Multi-layer perceptron

MLR Multiple Linear Regression

RP Resilient back propagation algorithm

SaSAT Sampling and Sensitivity Analyses Tools

Sewer Refers to a channel or pipe or collection of them which carry liquid waste

Sewage The liquid waste (also containing solids) which flows along the sewer(s)

Sewerage The pipes, pumps, and infrastructure and is frequently be referred to as a sewerage system

List of Acronyms

Md Abdul Aziz xx

SM Spectral Methods

SPSS Statistical Package for the Social Science

SSO Sanitary Sewer Overflow

STAR-CD CD-adapco's legacy CFD package

TM Testing Method

UKWIR UK Water Industry Research

VOF Volume of Fluid

VOF Volume of Fluid

ZPRED Z Predicted

ZRESID Z Residual

Outcome from this Research

Md Abdul Aziz xxi


Journal Papers

1) Aziz, M. A., Imteaz, M., Choudhury, T. A., & Phillips, D., 2013a,

‘Applicability of artificial neural network in hydraulic experiments using a new sewer

overflow screening device’, Australian Journal of Water Resources, vol.17, no.1,

pp.77-86.

2) Aziz, M. A., Imteaz, M., Naser J., & Phillips, D., 2013b, ‘Hydrodynamic

Characteristics of a New Sewer Overflow Screening Device: CFD Modelling and

Analytical Study’, International Journal of Civil and Environmental Engineering, vol. 7,

no.1, pp.71-76.

3) Aziz, M. A., Imteaz, M., Huda M., & Naser J., 2014a, ‘Optimising inlet

condition and design parameters of a new sewer overflow screening device using

numerical modelling technique‘, Journal of Water, Sciency and Technology vol.70,

no.11, pp.1880-1887.

4) Aziz, M. A., Imteaz, M., Rasel, H.M., & Phillips D., 2015a, Development

and Performance Testing of ‘Comb Separator’, A Novel Sewer Overflow Screening

Device. International Journal of Environment and Waste Management, vol.16, No.3,

2015.

5) Aziz, M. A., Imteaz, M., Samsuzzoha, M. A., and Rasel, H.M., 2015b,

Sensitivity Analysis on the Pollutant Trapping Efficiencies of a Novel Sewerage

Overflow Screening Device. Revising (March 2016) Journal of Hydro-informatics

6) Aziz, M. A., Imteaz, M., 2016a, -‘A Literature Review on Research

Methodologies on Sewerage Overflow Screening Devices’. To be submitted

International Journal of Water.

Peer reviewed conference papers

1) Aziz, M. A., Imteaz, M., J. Naser, Nazmul H., & Phillips, D. 2010,

‘Hydrodynamic Characteristics of a proposed sewer overflow screening device’, at

The 5th Civil Engineering Conference in the Asian Region, 8-12 August, Sydney

Australia.

2) Aziz, M. A., Imteaz, M., Choudhury, T. A. & Phillips, D. I. 2011, ‘Artificial

Neural Networks for the prediction of the trapping efficiency of a new sewer overflow

screening device’, 19th International Congress on Modelling and Simulation, Perth,

Australia. Indexed in Scopus


Md Abdul Aziz xxii

3) Aziz, M. A., Imteaz, M., Samsuzzoha, M., & Phillips, D., 2013c,

‘Sensitivity analysis for a proposed sewer overflow screening device’, 20th

International Congress on Modelling and Simulation, Adelaide, 2 to 5 December,

Australia.

4) Aziz, M. A., Imteaz, M., Nazmul H., & Jamal N., 2013d, ‘Understanding

functional efficiency of a sewer overflow screening device using combined CFD and

analytical modeling’, 20th International Congress on Modelling and Simulation,

Adelaide, 2 to 5 December, Australia.

5) Aziz, M. A., Imteaz, M., Rasel, H.M., Phillips, D., 2015c, ‘Performance

Testing of ‘Comb Seperator’ –A Novel Sewerage Overflow Screening Device’,

ASEAN- Australian Engineering Congress on Innovative Technologies for

Sustainable Development and Renewable Energy 11-13 March.

6) Aziz, M. A., Imteaz, M., Rasel, H.M., and Samsuzzoha, M., 2015d,

‘Parameter Sensitivity Using Sampling Technique for a proposed ‘Comb Separator’,

A sewer overflow screening device’, ASEAN- Australian Engineering Congress on

Innovative Technologies for Sustainable Development and Renewable Energy 11-

13 March.

Chapter 1: Introduction

Md Abdul Aziz 1

Chapter 1

Introduction


Md Abdul Aziz 2

1.1 Background

During heavy, long-lasting downpours of rain, the urban sewer system is not able

to carry the excess water; hence some of this excess water flows into the open creek

system, carrying with it a lot of sewer solids. These sewer solids are dispersed,

suspended or washed into the creeks and rivers. They eventually settle, creating odours

and a toxic/corrosive atmosphere in the mud deposits on riverbeds. The solids create

additional problems, either through their general appearance (increasing dirtiness) or

through the actual presence of specific, objectionable items, such as floating debris,

sanitary and faecal matter, scum, or even parts of car tyres. These sewage overflows

have physical, chemical and biological effects on the surrounding environment. The raw

sewage has the potential to carry pathogens, including bacteria, viruses, protozoa,

helminths, inhalable moulds and fungi. These pathogens can cause life-threatening

ailments, including diseases such as cholera, dysentery, infectious hepatitis and severe

gastroenteritis [121]. Sewage overflow also has an adverse impact on the environment

by increasing pollutants, including plastic and paper products. This sewage sludge is

also a key concern for environmental scientists and engineers [65].

Some mitigating options to segregate sewage solids have been adopted include

temporary holding tanks at sewage treatment plants, real-time control of sewerage

systems, enlarged upstream sewers to provide transient storage, separation of storm

and sewage flows, and various screening devices in separate and combined sewage

overflow (CSO) chambers. In most cases screening is the only economically-viable

method [60].

Screening is a process that should be automated in order to ensure operational

safety and to reduce aesthetic pollutants. Moreover, a floatable control is preferred by

most o proposed and existing environment regulation agencies. This requirement has

triggered the need for research into the construction of an efficient and effective

screening devices and screening handling systems. To optimise their use in the actual

environment, especially at unmanned locations where there is a requirement for

minimum maintenance.


Md Abdul Aziz 3

1.2 Problem Statement

Past studies have used a number of different screening systems in sewer

overflow locations. The initial screens used were stationary, which caused problems

because sewage solids clogged the screens. Most ‘conventional’ screening systems

utilise electro-mechanical components to facilitate the screening process [131].

However, given the harsh unmanned remote environment of many sewer overflow

device locations, this is clearly not ideal. Blocking and seizure of moving parts are

common maintenance problems, and in many cases, electrical failure necessitates an

onerous maintenance commitment [10]. Moffa [118] used a rotary screen consisting of a

large rotating angled-drum to maximise dewatering, with the screenings travelling up to

the drum, where they are removed from the unit. Metcalf and Eddy [117] proposed a

series of screens attached to a cage that rotates around a vertical axis creating

centrifugal force. The flow enters from the base and flows upward to a deflection plate,

where pollutants are collected from outside the cage. Hydrodynamic vortex separation

(HDVS) was another popular screening concept developed in the early 1960s. In the first

generation, HDVS was found to be effective in retaining 70% of the pollution load

Smisson [151]. A second-generation HDVS developed by the American Waterworks

association and the Environmental Protection Authority (EPA) was reported by Field [64].

The third-generation device was commercially patented in the 1980s as the Storm King®

overflow. The HDVSs went through a series of performance evaluations in Europe,

North-America, and Japan [20]. Unfortunately aesthetic solids of neutral buoyancy were

not trapped in HDVSs [16].

A state-of-the-art review of different screeners is provided by Saul [144] & [146].

A recent update on this literature can be found in the work of Madhani and Brown [106].

The literature suggests that screens need to have the ‘self-cleansing’ mechanism. To

overcome this challenge, a non-powered self-cleansing screening system that can

capture neutrally-buoyant aesthetic solids greater than 6mm in two dimensions was

tested by Smith and Andoh [152]. Faram et al. [59] tested a hydro jet device installed in

the USA, Australia and mainland Europe. However, in most cases the device was directly

associated with blockages in the sewerage system. The most conventional screening

systems usually utilise electro-mechanical components to facilitate such a process [131].

However, given the harsh unmanned remote operating environment of the sewer

overflow device, this is not ideal [10].


Md Abdul Aziz 4

Some common drawbacks in the available commercial screening devices include

inadequate screening capacity, the requirements for external power and the high cost

[153]. The aim of this research was to create a novel sewer overflow screening device

free of the existing limitations. The focus in this research was to ensure that the proposed

screen is inexpensive, has no moving parts, and is free of sophisticated electrical-

mechanical switching systems.

Two different types of sewer overflow screens are analysed in the current

research. The first is a gross pollutant screening device, which comprises a solid sewage

trapping device with an ogee weir. The ogee weir spillway possesses excellent hydraulic

features in terms of flow efficiency, and relatively good flow measuring capabilities. In

the device studied, a deviation from the traditional weir flow was considered, with

construction distinctions such as upstream flow conditions, reflection waves due to a

short device boundary, and different shapes and sizes of construction that changed flow

properties. The second screen proposed is the Comb Separator, which uses a series of

combs rather than the circular holes of the first device. Both these screens have no

moving parts, and would work more efficiently in harsh environmental conditions. A

detailed description of the screens is provided in Chapters 4 and 5.

1.3 Objective of this Study

The objective of this research was to design an efficient and effective sewer

overflow screening device. This thesis describes the development of a sewer

overflow screen that can overcome most of the key limitations of existing sewage

overflow screens.

The research investigation included design of a concept screen, testing of the

performance of the screen using computer model and laboratory experiments. This

research endeavor to innovate a novel sewer overflow screen with the following

features:

High efficiency in trapping pollutants

Minimal blockages on the screen

Automation for effective use in remote unstaffed locations

Operational safety with a bypass channel


Md Abdul Aziz 5

Floatable control to meet environmental regulations

Low maintenance and operational costs

No need for a sophisticated electro-mechanical switching system

Suitable for unmanned remote locations

1.4 Research Contributions

This research makes the following contributions to the discipline of sewer

overflow screening applications in the urban sewerage system:

The literature review revealed several gaps in the research relating to sewer

overflow screens. The scope of this research was designed to address these issues.

An analytical solution of the Navier-Stokes equation was developed for the

proposed model, which was then compared with the results obtained from computational

fluid dynamic (CFD) analysis. Results of the analytical solution confirmed the plausibility

of the CFD model developed for this study [10]. Due to lack of proper experimental data

for the validation exercise, an analytical model was used instead to check the

performance of the CFD model [11].

The construction of experimental set ups involved significant cost and time;

moreover, a proposed concept for a novel gross pollutant screen had to be proven. To

overcome these problems, a state-of-the-art CFD model was developed using the Euler-

Euler approach to study the hydrodynamic characteristics of the sewer overflow

screening device. The results of the CFD model predicted a number of important design

parameters such as flow distribution, velocity distribution, pressure distribution and

sloshing behaviour across the device. The analysis helped to understand the device

orientation, as well as some other critical design parameters. Results obtained from the

CFD model formed the basis for optimising design parameters of the laboratory scale

experimental set up [11].

The physical experiments allowed for a certain number of trials for

experimental set ups. To visualise a range of different conditions within and outside the

physical limitations of the experiment, it was important to do modelling analysis. In the

current research, the sewage solids tested had different densities, which made it too

complex to model using the CFD model. These challenges were overcome by using an

Artificial Neural Network (ANN) model. The ANN can successfully predict the

experimental results with more than 90% accuracy, with an average absolute percentage

error of around 7% [9].


Md Abdul Aziz 6

A novel screening device called the Comb Separator was developed and

tested. The comb separator can capture larger sewerage solids more than 10mm with

over 95% capture efficiency. Two key improvements were recorded in the comparison

analysis. Firstly, there was minimal blinding effect on the Comb Separator, which is a

key improvement over the previous static screening or electro-mechanical switching

concept. Secondly, the research device produced improved sewage solids capture

efficiency in low flows (up to 70 l/s), compared to the industry standard Hydro-JetTM [12].

Analysing the parameter sensitivity for hydraulic devices such as the ‘Comb

Separator’ is a necessary check to understand qualitatively or quantitatively sources of

variation during practical application. Sensitivity analysis can provide insights that guide

informed decision-making for the management of different sewer overflow events using

this screening device [13].

1.5 Thesis Structure and Overview

The thesis is divided into seven chapters. Chapter 2 to 7 present and discuss the

theories, methods, modelling and results of this research into the usage of proposed

sewer overflow screens.

Chapter 2 reviews the relevant literature regarding sewer overflow screens used

in both separate and combined sewage overflow systems. The limitations of the existing

screens are detailed, and the aims and objectives of this research are established. A

brief discussion about the proposed screening concept and the design considerations in

achieving that concept is also presented in this chapter.

Chapter 3 gives a detailed description of the methodology adopted in this thesis.

The research questions are outlined, and the methodology was adopted based on

answering those questions. A flow chart of the research methodology is also provided in

this chapter.

Chapter 4 details the purpose of the modelling investigation. This chapter

discusses the development of both the analytical and CFD models, and describes the


Md Abdul Aziz 7

plausibility check of the CFD model using the analytical model. Results of the CFD model

are also discussed, and an explanation is given on how to optimise the proposed gross

pollutant device. This chapter also discusses the limitations of the gross pollutant

screening device.

Chapter 5 gives an overall description of the laboratory experiments leading to

the design of the proposed Comb Separator device. The design of this device is based

on previous analysis of CFD results, and on the limitations observed in relation to the

earlier gross pollutant device. This chapter also discusses experimental data collection,

test procedures and the results of the experiments. The limitations of the experimental

device are reported at the end of this chapter.

Chapter 6 describes the development of an Artificial Neural Network (ANN) to

overcome some of the limitations of the experimental work and CFD analysis. A detailed

description of the development of the Artificial Neural Network model is also included in

this chapter. It concludes with a discussion of how the ANN results could improve the

optimisation process of the Comb Separator device to obtain maximum capture

efficiency of sewage solids overflow.

Chapter 7 describes the sensitivity analysis of the Comb Separator. The

performance of the comb separator is compared with the industry standard Hydro JetTM

screening device. To analyse sensitivity, a linear regression model was developed. As

there is only a small set of experimental data available for analysing the sensitivity of the

input parameters of the comb separator, sampling techniques were used to expand the

dataset. Results of the input parameters are also included in the discussion.

Chapter 8 summarises the key findings of the study, and offers suggestions for

further research.

Chapter 2: Literature Review

Md Abdul Aziz 8

Chapter 2

Literature Review


Md Abdul Aziz 9

2.1 Introduction

A review of the scientific literature on sewage overflow screening systems is

outlined in this chapter. The literature was sourced from professional societies, research

organisations, local councils, government agencies, published journals, reports and

conference papers. The review covers the following issues:

Effect of sewage solids overflows on the environment

Types of sewage solids found

Three major types of screening systems available in current sewage

overflow systems

Key methods to improve sewage solids overflow screening

Identification of research needs

Key consideration of the proposed sewage solids overflow screening

device

Scope of this research

During wet weather conditions, urban sewerage systems are not capable of

carrying all the excess water; hence it flows into open creeks or wetlands, often carrying

sewage solids. These sewage solids are dispersed, suspended or washed into rivers,

creeks, wetlands and the ocean. These solids further create an aesthetic nuisance, either

by their general appearance or through the actual presence of specific, objectionable

items such as floating debris, sanitary and faecal matter, scum, and even condoms. Due

to increasing public complaints, the focus of scientists and engineers is on the retention

of entrained sewer solids within sewerage overflow devices.

To overcome the issue of sewage solids overflow, different types of screening

systems are available, most of which use floatable controls at wastewater treatment,

sanitary sewer overflow (SSO), and combined sewer overflow (CSO) locations. The

preliminary treatment step is screening, which helps to segregate visible objectionable

materials, and also protects downstream equipment. In recent years, the frequency of


Md Abdul Aziz 10

sewage overflows has increased, both in separate and combined sewer systems, due to

population growth and impervious areas. The untreated overflow causes concern during

storm events because of the level of pollution. Sewage overflow results in visually

offensive littering of the receiving watercourse, and has negative physical, chemical and

biological effects on the receiving environment. The impacts on human health and on

the environment can be acute and cumulative, and there are also aesthetic

considerations. Screening and handling are unpopular processes for plant staff because

of odours, aesthetics and health concerns. Moreover, historically the screening system

has not been a piece of equipment that functions efficiently for long periods without

maintenance. The remote location placement of these screens typically makes the

operation and maintenance of screening systems costly and labour intensive. Thus it is

important to design an effective and efficient sewer overflow screening system.

While replacement of the pipe with one that has a greater diameter will increase

the capacity of sewerage systems, it is a very costly measure that is not economically

viable. Some earlier drainage plans did not consider the impact of flow pollution. The US

EPA estimated that the cost to overcome the sewage overflow problem in the USA alone

would be around 100 billion dollars. Considering the massive cost to replace and enlarge

the existing sewerage system, most research focus has been on improving the screening

capacity of the sewer overflow by introducing different types of screening systems. The

data summary of a sewage treatment works in central England [156], including the

composition of sewage solids, is shown in Table 2.1:

Table 2.1: Sewer solids available in the urban sewage treatment [156]

Component % by Weight (Dry Basis)

Rags 15 - 30

Paper 20 - 50

Rubber 0 - 5

Plastic 5 - 20

Vegetable Matter 0 - 5

Fecal Matter 0 - 5


Md Abdul Aziz 11

To design a screening system to improve screening efficiency, it is important to

understand what types of sewage solids are being collected in the system. The

composition of sewage solids captured by screening varies from one location to another.

An updated review quantity and type of litter in a drainage system in South Africa

[19] found the following types of litter:

Plastic 62%

Polystyrene 11%

Paper 10%

Cans 10%

Glass 2%

Other 5%

These data were collected over a period of 122 days, and involved a total of 106

cubic metres of litter, transported by 32 separate storm events. Similar data is available

for debris found on beaches, both in terms of mass and number of items observed per

length of beach. These results can be found in the reports of CMC [36], HydroQual [87],

1993, HydroQual [88] and [123].

Varieties of sewage solids reported in Australia include condoms, tampons,

cigarette buds, wrap papers and bottle caps [9]. To segregate these sewer solids,

different types of screening systems are used. Wastewater treatment facilities and CSOs

apply different types of screens. These screens can be classified based on type, function

or size of screen opening.

The aim of this literature review is to understand the gaps in research, and to

identify potential areas for improvement of sewer overflow screening technology. The

methodology followed includes recognition of the reported limitations in hydrodynamic

investigation, followed by experimental or on site investigations, heuristic modelling

approaches such as Artificial Neural Network (ANN) models, and sensitivity analysis of

the different types of model used.


Md Abdul Aziz 12

The frequencies of different sewer solids coming from various sources are shown

below:

Figure 2.1: Frequency of Trash and Litter from Various Sources (Source: [156])

2.2 Summary of Current Screening Applications

Designing sewage solids overflow screens is challenging, considering the masses

and loading rates of different sewage solids. The amount of sewage solids in an overflow

varies widely in a wet weather event. It will be particularly high if significant storm water

events happen after a drought. However, if the storm happens on consecutive days, the

mass of floatables and solids can reduce on the second day. Screening systems for

floatable control are used both in separate and combined sewerage systems. A survey

of screening devices shows that there are three types available in the current market:

2.2.1 Static screening

2.2.2 Mechanical and Electrical screening

2.2.3 Hydraulic screening


Md Abdul Aziz 13

2.2.1 Static screening

A static bar screen is a typical example of a static screening device. This type of

screen is one the cheapest forms of screening technology available. The static bar

screen consists of sturdy bars aligned parallel to one another. These static screens are

fixed and capable of capturing solids and floatable material. Static bars constitute a

stand-alone system without any mechanical, electrical parts or automated cleaning

mechanism. As there is no self-cleaning process available with static screens, it is

important that manual cleaning of solids and floatables is done periodically. So as to

avoid restriction of flow, maintenance crews need to make visits to ensure that the screen

does not become clogged. This is one of the key limitations of this type of screen.

Figure 2.2: Static bar screens in operation (source: [3])

These screens also take up significant space at the points of installation into the

existing sewerage system. There is also no provision for any bypass channels, which

increases the risk of clogging with solids and floatables, eventually leading to failure of


Md Abdul Aziz 14

the screen to serve its purpose. This flow restriction limitation necessitates the

installation of new screening chambers.

Figure 2.3: Brush-Raked fine screen static screen (source: [118])

2.2.2 Mechanical and Electrical screening devices

A number of different types of electrical and mechanical screening devices are

available on the market. Descriptions of the four most common types of electrical-

mechanical bar screens are provided below:

2.2.2.1 Vertical mechanical bar screens

2.2.2.2 Horizontal mechanical bar screens

2.2.2.3 Rotary drum screens

2.2.2.4 Centrifugal screens


Md Abdul Aziz 15

2.2.2.1 Vertical Mechanical Bar Screens

These types of screens have both below and above water surface components.

The mechanical arm is above the water surface, whereas the submerged portion has a

vertical, inclined, static bar screen rack. To clean the bar periodically, the rake arm

moves down below the water surface and onto the bar rack. When the rake arm

continues to move upwards on the screen near the discharge chute, the solids and

floatables are dumped in the loading container. The key benefits of using vertical

mechanical bar screens are:

Well known and well understood technology that has been in practice in

wastewater treatment facilities for over a decade.

When the water level in the chamber is high, the rake arm mechanism stops the

bar screen from blockage.

The bar screens are heavy duty bars that are structurally more sturdy than wire

mesh type screens.

The system can be adjusted with the addition of a flushing water system that

allows flush solids and floatables to get back to the interceptor.

The performance of these screens is based on bar spacing; however these

screens are effective in the removal of sewer solids and floatables up to 12.5mm

in diameter or greater.

However, these bar screens have limitations, some of which are listed below:

The mechanical and electrical components require more options for operational

and maintenance screening.

Maintenance of mechanical screening bars operating in remote unstaffed

locations is expensive.

A high clearance height is involved, which creates problems at some overflow

locations.

The initial cost to build such screens is more than to build concrete or other

structures; hence capital costs are higher.


Md Abdul Aziz 16

2.2.2.2 Horizontal Mechanical Bar Screens

The horizontal mechanical bar screens (refer Figure 2.4) are rigidly constructed

and well mounted, using materials that are free from corrosion such as stainless steel

bars, which are then equally spaced apart. These bars are designed in such a way that

there are no intermediate supports to collect solids. A level sensor is activated

spontaneously when storm water rises to such a level that it overflows the weir of the

screen. The hydraulically driven rake system moves back and forth across the screen to

keep the screen clean. The advantages of horizontal mechanical bar screens are listed

below:

When a high level is detected in the chamber, a programmed instruction to the

mechanical screening is activated.

The rake arm assembly effectively protects the bar screen from blockages.

Heavy duty bars are more durable than mesh type screens; therefore the bar

screens perform better in remote unstaffed locations.

Figure 2.4: Typical mechanical bar screens (Source: [117])


Md Abdul Aziz 17

This type of set up requires less maintenance and personnel costs, as sewer

solids are well managed through being pushed back into the wastewater channel.

Based on bar spacing, the horizontal mechanical bar screens are operative for

the removal of sewer solids and floatables up to 12.5mm and greater in size.

There are certain limitations in using the horizontal mechanical bar screens:

The mechanical and electrical components require more operational and

maintenance necessities than most non-mechanical screening systems.

If placed in remote isolated locations maintenance of the sophisticated electrical

and mechanical system can be difficult and expensive.

2.2.2.3 Rotary Drum Screens

A wide variety of industries apply rotary drum screens (refer Figure 2.5), including

the municipal wastewater, processed flood, and pulp and paper industries. These

screens have wedge wire wrapped around a drum screen that is open on both sides.

The drum screen is adjusted to the carriage of mechanical rollers so that it can rotate

around a horizontal axis. This horizontal axis is parallel to the sewage flow. The

screening occurs inside the drum. This type of screen has the following advantages:

This technology is accepted by a variety of industries; hence it is well-known and

understood.

The revolving action and an inner spray cleaning system prevent the drum screen

from blocking.

This screen is able to remove small sewage solids up to 12.5 mm and greater.

The wedge wire used in the drum screen has crossbars, which means the slots

are smaller than those of mechanical bar screens.

This screen has a clearance height lower than that of bar screens.

Although this screen works well in certain conditions, however it got the following

limitations:


Md Abdul Aziz 18

The water spray systems using mechanical and electrical components have more

operational and maintenance requirements than do the mechanical bar screen

and the non-mechanical screening system.

To house the screening facility, additional concrete or other structures are

required, resulting in increased capital costs.

These devices contain more mechanical parts, which could potentially cause

failure of this device.

The maintenance and personnel costs of collection, transportation and disposal

of this screen are high.

The wedge wires for the drums are not constructed of thick, heavy duty bars.

Thus there is the potential that the wedge wire construction may not withstand

the force of repeated high flows.

Figure 2.5: Internally Fed Rotary Screen (Source: [117])

2.2.2.4 Centrifugal Screens

This type of screen considers a series of screens that is attached to a cage which

revolves around a vertical axis, refer to Figure 2.6.


Md Abdul Aziz 19

Figure 2.6: Centrifugal screen (source: [117])

The sewage overflow enters the inside of the screen cage at the bottom, and

flows upwards to a deflection plate mounted at the top of the unit. The deflected flow

passes through the screens, and pollutants are collected outside the cage as shown

above.

Hydraulic Screening

Hydrodynamic vortex separation (HDVS) dates back to the early 1960s, and is

the most commonly used hydraulic screening device. The full scale device was tested at

Bristol in the UK. One of the key attributes of the HDVS separator is that it creates

tangential flow into a cylindrical vessel (refer Figure 2.7).

Figure 2.7 shows a cut-away view of the Storm King ® Overflow HDVSs with a

number of internal components highlighted. This is an industry standard screening

device and developed through a throw research process.


Md Abdul Aziz 20

Figure 2.7: The cut-away view of the Rotary-Jet Screen (Source: [59])

This tangential flow creates a complex rotating flow regime. In the configuration

of the HDVS, the inlet deflector plate minimises headloss by streamlining the incoming

flow as it enters the main vessel body, and joins with the mass of fluid circulating within

the vessel. The device has proven to be more effective and efficient than conventional

chambers or fixed screens [21] & [20]. Over time, different types of HDVS configurations

have evolved. The development HDVSs is discussed in the following paragraphs.

The first generation HDVS separator could retain up to 70% of the pollution load

[151]. This second generation device was introduced in the 1970s, and is reported in

Field’s work [64]. During the early 1980s, a third generation of the HDVS was developed

in the UK. This device overcame shortcomings that had been identified in the previous

EPA Swirl Concentrator. With this upgrade, the device screening system was patented

and commercialised, carrying the trade name Storm King® Overflow. Although the

device was used commercially, there was a need to reduce turbulence in the Swirl


Md Abdul Aziz 21

concentration at high flow. In the mid to late 1980s, the HDVS-FluidsepTM [30] was

developed in Germany.

Alongside the development and upgrade to commercially available devices during

the 1980s, the Hydrodynamic vortex separation was subject to different performance

assessments in Europe, North America and Japan [30]; [80]; [21]; [20] and [125]. A

performance evaluation of a number of these devices based on influent solids and their

settling characteristics especially highlighted the velocity distribution [167]; [14].

The HDVS separator became more sophisticated by the early 1990s, as the

system was used for water quality control for CSOs (e.g. Storm King® overflow, Swirl

Concentrator and FluidsepTM), and stormwater treatment (eg. Downstream Defender ®

and the Vortechs TM system). In the later years of the 1990s, HDVS technology advanced

further in response to the Asset Management Plan 2 (AMP2) requirement in the UK. This

required a non-powered self-cleaning screening system to address the issue of total

capture of neutrally buoyant solids greater than 6 mm in two dimensions [152]; [16].

More than 1,500 HDVSs have been installed around the world for managing

sewer solids in stormwater, combined and separate sewers, and waste water treatment.

Although many HDVSs are used, confirming the application potential of the device, there

are mixed views regarding their efficiency and effectiveness.

Another thoroughly researched device is the Hydro-JetTM Screen. This device has

a self-cleaning mechanism, and its suggested use is in combined sewer overflows that

utilise a purely hydraulic cyclic backwashing mechanism. The National Rivers Authority

[120] in the UK set the standard for intermittent wet weather discharge and removal of

pollutants. The most stringent condition requires the segregation of solids greater than

6mm diameter in any two dimensions. Previous experience has shown that if the device

does not use a screen self-cleaning mechanism, it will be subject to blinding. Some of

the conventional systems use electro-mechanical components to facilitate such a

process; however, these sophisticated systems are subject to seizure or jamming of

moving parts, which means that maintenance requirements are high.


Md Abdul Aziz 22

Figure 2.8: Schematic representation of the Hydro-JetTM (Source: [59])

To overcome such challenges, the UKWIR [145] developed 15 proprietary

screens. Nine had an external power source; 11 had moving parts. The remaining four

screening systems had neither moving parts nor power requirements. Two were

stationary screens, while the others used a self-cleaning mechanism. Of these devices,

the Hydro-JetTM screen is the only one that features all the key attributes of no moving

parts, no electrical-mechanical switching system, and incorporating a self-cleaning

mechanism. Moreover, the system was developed as a cost-effective contender, with a

hydraulically-sophisticated backwashing mechanism.

The Hydro-JetTM screen has been subjected to a rigorous evaluation process [15];

[16] & [58]. Having studied the available documentation on this evaluation process, the

Hydro-JetTM, is considered the benchmark device for a comparative analysis of the

screening systems in this research. A comprehensive discussion of the performance

analysis of the Comb Separator and Hydro JetTM is provided in section 5.4.3.


Md Abdul Aziz 23

2.3 Methods to improve Sewage Overflow Screening

2.3.1 Hydrodynamics Applications

Experimental measurements are probably the best way to understand the

capture efficiency of sewage solids for any proposed device. However, this process

cannot be undertaken before the device has built [51]; [105]. However, the CFD model,

after validation, could offer an alternative method to predict the performance of the

proposed device. The hydrodynamic investigations of a sewage overflow screening

device can contribute significant knowledge without the need for the physical set up of

the screening system. The key advantages of CFD modelling are suggested by Harwood

and Saul [77]:

CFD can simulate experimental conditions without physical laboratory

facilities.

The structural geometry of the CFD model can be changed quickly, which can

avoid the significant time and costs involved in reconstructing a physical model.

Flow parameters of shear stress, velocity and pressure are calculated at all

points, providing more insight than the physical model [83]; [160].

The recent development of powerful computational facilities allows the CFD

model to simulate complex flow dynamics. Best practice hydrodynamic applications

using the CFD model are well documented in the work of Casey and Wintergerste [33].

The CFD model contributes to improvements in the efficiency of sewage overflow

screening systems, especially CSO chambers [143]. A Storm King ® hydrodynamic

separator was modelled by Svejkovsky and Saul [163] using 3D FLUENT. Pollert [132]

and Hrabak et al. [84] modelled and evaluated the hydraulic performance of CSOs. The

complex hydraulic flow features such as erosion, containment or mobilisation of

pollutants were studied by Harwood [78], and the deposition of solid particles was studied

in by Stovin et al., [160]. Similar work by Stovin [158], Stovin and Saul [159;161], and

Adamsson et al., [1] highlighted difficulties with modelling particles transported in

physical models, and showed the application of the CFD model for particle transport in

sewerage systems. An explanation on how to model suspended solids separation and

vortex separation can be found in the work of Pollert and Stransky [133] and Tyack and

Fenner [167].


Md Abdul Aziz 24

The ogee weir spillway possesses excellent hydraulic features in terms of flow

efficiency, as well as relatively good flow measuring capabilities. In the device studied,

the differences from a traditional weir flow considered were the upstream flow conditions,

reflection waves due to a short device boundary, and the change in shape and size of

the flow properties. These slight changes need to be thoroughly researched to identify

whether they have a negative effect on the evaluation of the performance of the spillway.

A detailed study had been carried out to determine the standard shape and size of the

crest of the overflow spillway; the relative height and the upstream slope of the spillway

were also considered [114]; [41]. Similar objectives have been reported in the work of

the US Army Corp of Engineers [169]; [170].

Most previous investigations were confined to physical models. In recent years,

with the advent of powerful computational advances, research is focused on flow

simulation using numerical modelling. An early attempt to model the spill overflow using

potential flow theory was made by Cassidy [34], and with limited experimental data, good

agreement with experimental data was noted. Better accuracy with experimental data

was found in studies of [90] and [24] using linear finite element approximation. In

addition, 2D irrotational gravity flow over the curved water surface was successfully

modelled. Xie and Chen [103] and Guo et al. [66] extended on the potential flow theory

while applying an analytical functional boundary. A 2D finite volume based numerical

model for flow over a spillway was validated using water level and pressure data on the

physical model [27].

Most of the existing literature reported either experimental works or numerically

simulated flow phenomena over an ogee weir in ideal conditions with no wave reflections

and much larger upstream and downstream boundaries. Although such assumptions

simplify the problem, they cannot be incorporated into the existing sewerage drainage

system, where space constraints in the urban drainage system would be a major issue.

This research gap was identified, and the current research has focused on understanding

the space restriction in experimental set ups and in the analysis.

The current research also takes into account discussion regarding the inlet

orientation, the effect of reflected wave on a small dimension screening system,


Md Abdul Aziz 25

optimisation of the inlet length to maximise device performance, and the best ogee weir

orientation based on a previous study by the US Army Corp [169].

2.3.2 Experimental Investigations

The capture or retention onsite of sewage solids in different gross pollutant

devices has been a common challenge for the water industry. Phillips [129] mostly used

real floating litter items for his proposed screen. Armitage and Rooseboom [19] tried to

capture an artificial pollutant. Overflows occur more often in combined sewer overflows

than in a separate sewer system. An attempt to address this issue has led to active

research on CSOs over the last 50 years. The initial work done Sharpe and Kirkbride

[150], followed by a series of works during the 1960s, 1970s and 1980s, formulated the

report of guidelines [23]. Further laboratory testing on the topic identified some limitations

in the gross retention performance. This work also updated the user guide in the report

published in 1994 [69] that explains how to design the combined sewer overflow

structures. Following a thorough investigation, a report entitled ‘Predicting aesthetic

pollutant loadings at CSOs’ was completed and published for the water industry in 2002.

In addition, full scale research at the National CSO test facility Wigan WwTW highlighted

the need to use screen technology [165]. The regulatory requirement for aesthetics can

be found in the work of Saul and Blanksby [147].

Table 2.2: Regulatory requirement for aesthetics [147]

Amenity Use Category Expected Frequency of Spills Standard

High Amenity > 1 spill per year 6 mm solids separation

<= 1 spill per year 10 mm solids separation

Moderate Amenity > 30 spills per year 6 mm solids separation

<= 30 spills per year 10 mm solids separation

Low Amenity and Non- Amenity Good Engineering design


Md Abdul Aziz 26

A comprehensive assessment of different screen types was conducted by UK

Water Industry Research (UKWIR). The screening performance is a combination of

different parameters, however two key parameters are, design flow rate and how many

pollutant particles are present in the sewage overflow.

When used in isolation, most sewage overflow screens tend to blind. To

overcome this problem, many screens use the electro-mechanical switching system.

However, considering the remote locations and harsh environments of some sewers,

these screens are not ideal. To overcome these problems, an effective and low-

maintenance screening system Hydro JetTM screen was developed. The general

concentration device was subject to a series of tests carried out at the University of

Sheffield, which concluded that the system could capture solids of 6mm in two

dimensions as required. This device was given a long-term site trial at Wessex Water,

and has been further tested at 27 sites with 13 rotary Hydro Jet TM screens.

The Hydro JetTM screening system was developed with a rigorous testing and

evaluation process in order to meet the industry requirement, and represents a high-

class solution to many problems that existed before. The performance of the device

proposed in this research will therefore be compared with the Hydro JetTM screen. Further

discussion in this regard can be found in Chapter 5, Section 5.4.3.

2.3.3 Artificial Neural Network (ANN) Applications

The Artificial Neural Network (ANN) was inspired by biological neural networks,

and has the unique ability to learn and generalise knowledge. An ANN can be considered

as a massive parallel-distributed information processing system. This system has certain

performance characteristics that resemble the biological neural networks of the human

brain [79]. ANNs constitute a non-linear data modelling tool used to model complex

relationships between inputs and outputs without any prior assumptions or any available

mathematical relationship between them. ANNs comprise a group of interconnected

artificial neurons, which are simple and fundamental processing units. A neural network

is characterised by its architecture, which represents the pattern of connection weights

and the activation function [61].


Md Abdul Aziz 27

Artificial Neural Networks have already been successfully used to simulate flood

forecasting in urban drainage systems [31]; [96], real-time flood forecasting [126], annual

run off predictions [67]; [68], rainfall forecasting [100], real time control in combined

sewerage systems in Germany [173], and real time water level predictions of sewerage

systems covering gauged and ungauged sites [47]. As ANNs have been successfully

applied to simulate water quality and flow prediction applications [108]; [42], they have

been adopted in the current study. Some fundamental challenges in developing ANNs

include structure identification, parameter estimation, generalisation performance

improvement with proper choices of algorithms, over fitting, and finally, model validation

[109].

Figure 2.9: A typical artificial neuron k

p

jkjkjk xwy

1 (2.1)

Each artificial neuron is basically a computer processor (refer Figure 2.9), where

the output yk is a function of the weighed sum of the inputs, In Eqn. (2.1)., x1,x2,…,xp are

the input signals; wk1, wk2,…, wkp are the assigned weights; θk is the threshold value and

is the transfer function.


Md Abdul Aziz 28

The experimental work in this research was restricted by the physical limitations

inherent in laboratory studies. These include the limited number of trials that can be

achieved, the limited number of experimental set ups possible, and the significant cost

and time required for running the trials. To overcome these problems, the experimental

results were analysed and used to train an Artificial Neural Network (ANN) model. ANN

had already been used successfully in similar kinds of environmental problems such as

water level predictions, flood forecasting and control in combined sewers [47]; [31]; [173].

Willems and Berlamont [174] have demonstrated a number of uncertainties

involved in deterministic models for sewerage systems. In particular, model

simplifications of the physical system make it difficult to adopt physical law based models

such as CFD. In the problem studied, the features which are difficult to model are: (i)

physical characteristics of different sewage particles; (ii) multi-fluid sewerage systems

with changing velocity due to different viscosity of fluids, and (iii) interaction between

liquid and solid particles. However, an ANN has the capability to effectively extract

significant features and trends from complex systems, even if the underlying physics is

either unknown or difficult to recognise [47]. Moreover, it can reduce computational time

and cost, unless completely new sets of experimental conditions are used [135]. They

also have the capability to predict complex input output relationships with little

understanding of the physio-chemical system. This makes the model the obvious choice

among a wide range of urban drainage systems.

In the case of sewage solid capture efficiency under study, the neural network

modelling was able to learn the existing non-linear input-output relationships. A multi-

layer feed forward artificial neural network using a back propagation algorithm was used.

Such networks have been used almost exclusively in environmental modelling [109]. The

current research implemented ANN modelling to overcome the physical limitations of the

experiments. Further discussion in this regard can be found in Chapter 6.

2.3.4 Sensitivity Analysis to Model Results

In the current research, the sewage overflow screening system required hydraulic

testing and modelling. Analysing the hydraulic device parameter sensitivity has been a

standard practice of hydraulic engineers over the years [97]; [116]; [176]. Sensitivity


Md Abdul Aziz 29

analysis, quantitative or qualitative, provides a guide to different sources of variation

[139]. Some of the key reasons for analysing sensitivity are listed below:

To identify the most influential factors in the test result of the model analysis

To detect factors that require further testing and research to improve confidence

in the model output

To recognise areas in the spacing of inputs where the maximum variation occurs

in the model output

To categorise any factors that interact with each other

To develop a robust understanding of the meaningful input parameters

To comprehend the impact of experimental design parameters on sewage solids

capture efficiency

Sensitivity analysis is a standard, accepted methodology for any modelling

investigation, or for analysing expanded data series. Extended analysis from basic

sensitivity analysis can be found in the work of Hall and Solomatine [70] & [71]. A

comprehensive review of the application of sensitivity analysis in environmental models

is presented by Hamby [74]. Other works discussing sensitivity analysis include [81];

[49]; [95]; [94] and [75].

As sensitivity analysis of the model result is a necessary aspect of responsible

model use, the current research has analysed experimental results using this model.

Analysing the parameter sensitivity for the Comb Separator as a proposed hydraulic

device is an essential check done to understand sources of variation, qualitatively or

quantitatively, during practical application of the screen.

2.4 Identification of Research Needs

This section summarises issues identified as requiring further investigation, and

are the points that frame the scope of this research project. The scope of this research

includes:

Understanding the space restriction in an existing urban sewerage system


Md Abdul Aziz 30

Understanding device performance based on the analysis of screening

orientation, wave reflection on the screen, and different device optimum

dimensions

Hydraulic analysis of sewage solids carry over on the screen

Fineness of screens or openings

Bar and perforated or punched hole openings

Performance of the self-cleaning effect

Overcoming physical limitations of the experimental conditions

Sensitivity analysis to develop meaningful and simplified input models, while

considering how key input parameters have an influence on output capture

efficiency

Comparison of the performance of the proposed screening system with the Hydro

JetTM screen

Wash water and power requirements

2.5 Summary

This chapter has provided a brief overview of the relevant scientific literature.

Three types of screening systems were identified: static screening, mechanical and

electrical screening, and hydraulic screening in existing CSO or separate sewer systems.

The limitations of the earlier static screens include the potential for blockage and high

maintenance costs. Mechanical and electrical screening performs better in overcoming

blockages; however, they are not ideal for use in remote locations. Sophisticated

mechanical electrical switching systems can create more issues, due to the isolated,

rough locations of these screens. Hydraulic screens usually have a self-cleaning effect

which makes them desirable in isolated locations.

There are four key methods found in the literature that improve the understanding

of sewage overflow screens:


Md Abdul Aziz 31

a. Hydrodynamic applications such as CFD models

b. Experimental investigations

c. Heuristic approaches such as the ANN model

d. Sensitivity analysis of the model results

Further details of these key methods are discussed in the Chapter 3.

Chapter 3: Research Methods

Md Abdul Aziz 32

Chapter 3

Research Methods


Md Abdul Aziz 33

3.1 Introduction

Background information, a review of relevant literature, and an analysis of the gaps

in research were provided in Chapters 1 and 2. To address these gaps, a set of fundamental

research questions has been formulated. To answer these research questions, a series of

research studies was designed, and the research methodology was derived based on

answering the research questions.

3.2 Research Questions

The key question of this research is: what is an efficient and effective sewer overflow

screen that overcomes the key limitations of existing sewer overflow screens? Some

common drawbacks with the existing sewer overflow screening include the following:

Inadequate screening capacity

Blockages on the screen

Sophisticated electro-mechanical switching system

High operational and maintenance costs

The need for a self-cleansing device due to location in remote unmanned places

To address the issues described above, the current research was planned with the

aim of answering the following key research questions identified:

RQ1: How to improve sewage solids overflow screening capacity?

RQ2: How to reduce blockage on the screen?

RQ3: How to avoid sophisticated electro-mechanical switching system in the screen?

RQ4: How to make the screen self-cleansing?

RQ5: How to optimise the proposed sewer overflow screening device?

RQ6: How to reduce the operational and maintenance cost?

RQ7: Is the performance of the proposed screen any better than current practice?


Md Abdul Aziz 34

3.3 Research Process

Individual research studies were designed to find answers to each of the research

questions above. These are briefly described below:

Study 1: How to improve sewage solids overflow screening capacity?

Comprehensive computational fluid dynamic modelling was completed to

understand the orientation of the screening device and the inlet pipe diameter, in order to

maximise sewer overflow capture efficiency [10]. The performance of the proposed device

was tested in a series of laboratory experiments [12]. The optimum device set-up was

simulated using an ANN model, and good agreement was achieved with the experimental

results [9]. Finally, sensitivity of the capture efficiency was tested at different inlet conditions

to understand which input parameters have the most sensitivity for the proposed device,

and how to use the device efficiently [11]. A more detailed description of the screening

capacity can be found in Chapters 4, 5 and 6.

Study 2: How to reduce blockage on the screen?

The literature shows that blockage on the screen has been a common problem. To

overcome this, a series of experimental set ups was tested using different comb spacing. It

was found that the blockage on the comb separator was minimal, and that if the device is

run for 10 to 15 minutes, there is hardly any blockage seen on the comb spacing [12]. A

detailed description of these experimental results can be found in Chapter 5.

Study 3: How to avoid sophisticated electro-mechanical switching system in the

screen?

Most recent devices use a sophisticated electro-mechanical switching system,

which at times does not work in remote unmanned locations. To overcome this issue, the

sewage solids overflow screening devices described in Chapter 4 [11] and Chapter 5 [12]

did not include an electro-mechanical switching system. A detailed design of the screening

concept can be found in Chapters 4 and 5.

Study 4: How to make the screen self-cleansing?

As the sewer overflow screen devices are located in remote and unmanned

locations, it is important to design the device with a self-cleansing effect [10]. The self-

cleansing effect also reduces operational and maintenance costs. A detailed CFD

investigation was carried out to understand how to maximise the self-cleansing effect for


Md Abdul Aziz 35

the proposed sewer overflow screening device. The performance of the self-cleansing effect

was also verified in laboratory testing of the device [12]. Detailed descriptions of

investigations aimed at maximising the screening self-cleansing effect are set out in

Chapters 4 and 5.

Study 5: How to optimise the proposed sewer overflow screening device?

The CFD model was too complex to derive an outcome when sewer solids of

different density were tested. To overcome this challenge, laboratory testing was designed.

The laboratory experiments had limitations in the extent of the trial, meaning that the

experimental set up may not lead to the optimum result. To complement this information,

an Artificial Neural Network (ANN) was designed and tested to find the optimum condition

of the experimental set up [9]. A more detailed description can be found in Chapter 6.

Study 6: How to reduce the operational and maintenance cost?

Most previous sewer overflow screening devices used a sophisticated electro-

mechanical switching system. These systems are expensive to design and maintain in

remote unmanned locations. Moreover, such devices do not perform well in harsh remote

physical environments. The proposed device did not have a sophisticated switching system

or electrical equipment. Instead, an automated valve was used of the type recommended

by most environmental agencies. The device was also tested for self-cleansing effect to

reduce the operational and maintenance cost [11].

Study 7: Is the performance of the proposed screen any better than current

practice?

The last research question is of significant importance in highlighting the contribution

of the current research to the actual performance enhancement of the sewer overflow

screen. The performance of the ‘Comb Separator’ was compared with the industry standard

‘Hydro-JetTM’ [12]. Finally, sensitivity analysis of the laboratory experiments was undertaken

so that the ‘Comb Separator’ can be used more effectively in actual sewer overflow

situations [13].


Md Abdul Aziz 36

3.4 Research Design

The current research was designed based on the research questions. A checklist of

different research strategies for seven key research questions was prepared. The initial

concept design was modified based on research experience. The strategies applied in the

current research are shown in the flow chart below:

Figure 3.1: Flow Chart of the Current Research Plan

(Where RQ stands for Research Questions)

Research Design

Define Research Questions (RQ1 to RQ7)

Concept Sewer Overflow Screen Design

CFD Model Investigation (RQ1, RQ3 & RQ4)

Revised Sewer Overflow Screen Design

Laboratory Experiments (RQ1, RQ2, RQ3, RQ4, RQ6 & RQ7)

ANN Modeling (RQ 5) Sensitivity Analysis

Report Writing


Md Abdul Aziz 37

3.4.1 Computational Fluid Dynamic (CFD) Analysis

The key research questions answered using the CFD model are RQ1, RQ3 and

RQ4. The CFD modelling and lab experiments were complementary to each other, with the

aim of finding the answers to the research questions. However, the CFD investigation was

carried out first to get a better insight into the problem so as to try to find solutions

accordingly. The RQ2 could not be tested under the CFD model, as sewer solids differed in

density, making it too complex to model. Where the CFD model could not provide the

answer to research questions, alternative research methods were adopted.

3.4.2 Laboratory Experiments

The next discussion concerns onsite and experimental work on the sewer solids

screening system. The experimental work for this research is addressed in Chapter 5. Out

of the literature review came the following discussion points to explore further in the current

research:

Fineness of screen or openings

Bar or perforated punched hole opening

Performance of the self-cleansing effect

To be more specific, the laboratory experiment tried to answer most of the research

questions, with the exception of RQ5 regarding device optimisation. The CFD model helps

immensely in designing experimental parameters such as device orientation, input pipe

diameter and weir opening. Lab experiments try to answer most research questions

regarding screening capacity, blockage on the screen, self-cleansing mechanism, and so

on. However, there are physical limitations of the experimental set-ups, which necessitated

the use of a different research method/strategy: ANN modelling.


Md Abdul Aziz 38

3.4.3 ANN model to supplement deterministic approach

To overcome the limitations of experimental investigations, it is important to do ANN

modelling. A more in-depth discussion about the ANN model can be found in Chapter 6. A

review of literature on ANN modelling uncovered the following research needs:

Hydraulic analysis of sewage solids with limited experimental set ups

Overcoming physical limitations of physical experimental conditions

This method helps to answer RQ5 regarding optimisation of the experimental set up

using the ANN method. The final research question regarding comparison with the industry

standard device is presented in Chapter 5. The performance of the ‘Comb Separator’ was

tested against the industry standard ‘Hydro-JetTM’ [12].

Sensitivity analysis is an integral and necessary part of any proposed hydraulic

device, including the sewer overflow screening device. Chapter 7 will discuss in detail the

sensitivity analysis adopted in the current research [13]. The sensitivity analysis of the

model results were important contributions in developing meaningful and simplified input

models while considering the key input parameters.

3.5 Analysis Procedure

Different research methods were adopted in this research to achieve the desired

outcome. Key methods adopted are outlined below:

3.5.1 CFD and Analytical Modeling

CFD is a proven modelling approach to understand dynamic behaviour of flows and

analysis of different design parameters. One of the significant benefits of CFD is that a

series of trials can be done before physical set up of the screening device. As experimental

set-ups involve significant cost and time, the initial approach in testing the performance of

the proposed sewer overflow screening device was to use a CFD model. To justify the use

of the CFD model, an analytical model was developed, which was used to check the

plausibility of the CFD model. The CFD model provides detailed insight into different

experimental parameters, orientation of the device, and so on. However, the CFD model

was not an alternative to the laboratory experiments; rather it complemented them,


Md Abdul Aziz 39

improving the performance of the laboratory experiments. In the current search, sewer

solids of different densities were tasted. This is a very complex task for CFD analysis; hence

alternative methods were adopted. A detailed discussion of the development of the CFD

model and analysis of the results are provided in Chapter 4.

3.5.2 Experimental Investigation

The experimental set-up was developed based on learning from the CFD model.

The screening concept was also revised based on CFD results. The key objective of the

laboratory experiments was to improve capture efficiency of sewer solids. A series of trials

with different set-ups were undertaken to test methodologically. The target efficiency of

sewer solids was 80% for smaller sewer solids less than 10mm diameter with minimal

blinding on the screen, and to perform up to one overflow event in one year. There are some

physical limitations in the experimental set-ups. A further investigation considering all

possible scenarios was carried out using ANN modelling analysis. A detailed discussion of

the laboratory experiments is found in Chapter 5.

3.5.3 ANN Modeling

To supplement the limitations of CFD and laboratory experiments, an ANN model

was developed. The benefit of the ANN model over CFD analysis is that it can accurately

predict the input/output of a process, without having the physics explicitly provided.

Moreover, ANN can overcome physical limitations inherent in laboratory studies. Proper

care was taken in developing structural identification, parameter estimation, network

optimisation and ANN model validation. A detailed description of the ANN development and

an analysis of the results are provided in Chapter 6.

3.5.4 Sensitivity Analysis

Sensitivity analysis of hydraulic devices such as the sewer overflow screening device

is a standard procedure. Sensitivity analysis develops meaningful and simplified inputs for

the model taking into consideration key input parameters. The experimental dataset was

extended using the Latin Hypercube Sampling (LHS) technique, without changing the

relationship between input and output parameters. In-depth discussions about the

development, hypothesis and analysis of results are reported in Chapter 7.


Md Abdul Aziz 40

3.6 Summary

In developing the scope for the current research along with the literature review gap

analysis, it was prudent to follow best practice guidelines for standard screening attributes

as suggested in the literature. These include the following stipulations:

The device should be built so that minimum inspection and maintenance are

required, and so that it can be constructed above ground level.

The treatment capacity required for the drainage system must be able to cope with

a maximum of one overflow event a year of 70 litre/second (l/s), and a minimum of

one flow event in four months of 20 l/s.

It should not use sophisticated electrical or mechanical systems, or signals that may

become ineffective during extreme events such as floods or lengthy droughts, and

which incur significant costs in maintenance and re-running.

In the event of a device failing, there must be a bypass option that causes no serious

damage to the device or the environment.

Taking all these points into consideration, the following flow chart illustrates the plan

in the current research. To address the research gaps identified, this research will follow

the following flow chart.


Md Abdul Aziz 41

Figure 3.2: Flow chart of the current research plan

Chapter 4: Hydrodynamic Analysis

Md Abdul Aziz 42

Chapter 4

Hydrodynamic

Analysis


Md Abdul Aziz 43

4.1 Introduction

Nature cannot make sewage solids disappear; rather, it can transport them from

one system to the next, that is, from the drainage system to the creek system. Most urban

sewerage systems built in the 18th century used a simplified one-way material flow.

These sewerage systems had limited focus on long term environmental outcomes like

floating debris, creating issues with aesthetics and public health. The main consideration

for design was structural durability and to a lesser extent flexibility in response to

changing needs. In recent years the number of sewage solids overflows has increased,

due to exponential population growth and the increase of impermeable areas. Sewage

solids disperse, float or wash into rivers and eventually settle on the bottom, creating

odours and toxic/corrosive conditions.

This sewage solids overflow is more visible after heavy rainfall. Sewage overflows

to receiving water bodies raise serious environmental, aesthetic and public health

concerns. To address these problems, much research has been carried out into different

types of screening devices to remove these pollutants. The screening of sewage solids

is a controlled process that is desirable in the sewerage system. Floatable control is

preferred by most of the proposed and existing environmental regulations as it is effective

for use in un-manned remote locations. In most cases screening is the only economically

viable method, according to Faram [60]. These issues trigger the need to research

different types of screening devices and screening handling systems. A state of art

review of these screeners can be found in the work of Saul [146] & [148] and recent

literature updates can be found in the work of Madhani [107].

In past investigations, a number of different screening systems were used in

sewage overflow locations. One of the most common was the rotary screen proposed by

Moffa [118], which consists of a large rotating drum that is slightly angled to maximize

dewatering. The angle of the drum ensures effective dewatering as the screenings travel

up to the drum where they are removed from the unit. Metcalf and Eddy [117] proposed

a centrifugal screen with a series of screens attached to a cage that rotates around a

vertical axis. The flow enters from the bottom and flows upward to a deflection plate at

the top of the unit and screenings are collected from outside the cage. Hydrodynamic

vortex separation (HDVS) was another popular screening concept developed in the early


Md Abdul Aziz 44

1960s. The first generation HDVS devices were found to be effective in retaining 70% of

the pollution load [151]. A second generation HDVS developed by the American Water

Works Association and EPA was reported by [64]. The third generation device in the

1980s was commercially patented as Storm King® overflow. The HDVSs had been

through a series of performance evaluations in Europe, North-America, and Japan.

Unfortunately aesthetic solids of neutral buoyancy were not trapped in HDVSs according

to Saul [144].

To overcome this challenge a non-powered self-cleaning screening system which

can capture neutrally buoyant aesthetic solids greater than 6mm in two dimensions was

tested [152] and [16]. Despite more than 1,500 HDVSs being installed worldwide for

storm water, there are still mixed views regarding their effectiveness [17]. Faram et al.,

[59] tested hydro jet devices installed in the USA, Australia and mainland Europe.

However, in most cases the devices were directly associated with blockages of the

sewerage system.

Reported literature suggests that screens need to be self-cleaning otherwise if

they are placed in remote un-staffed locations, they are subject to blinding [9]; [154].

Usually most conventional screening systems utilise electro-mechanical components to

facilitate such a process [131]. However, given the harsh unmanned remote environment

of many sewage overflow device locations, this is clearly not ideal [10]. Blocking and

seizure are common maintenance problems of moving parts and electrical failure will

dictate an onerous maintenance commitment in many cases [10]. To overcome such

problems a novel self-cleaning, less expensive, low maintenance sewage overflow

screening device with no moving parts is proposed. The aim of this chapter is to

investigate the optimum inlet and design parameters of the novel sewage solids overflow

screening device using CFD modelling.

With the advancement of computational power in recent years Computational

Fluid Dynamics (CFD) has become a proven technology to investigate hydraulic

behaviour of hydraulic structures and CSO design [161]; [113]; [50]. One key advantage

of the CFD based approach is that three-dimensional solid-liquid two phase flow

problems under a wide range of flow conditions can be evaluated rapidly, which is almost

impossible experimentally [164]. This technique has been successfully used for accurate


Md Abdul Aziz 45

prediction of the flow pattern in storage tanks [51] and the solids separation of CSOs and

storage structures [77]; [10].

The proposed sewage overflow screening device has overcome common

drawbacks in available devices such as: inadequate screening capacity, external power

requirement and high cost [10]. Simon and Phillips [154] developed a sewage solids

overflow screening device with temporary holding tanks which provides transient storage

and real time control of sewerage systems. The device has no moving parts, it has a

robust stop/start operation, it works as self-cleaning device and it has no sophisticated

electrical mechanical circuit. The gross pollutant trapping device is a novel self-cleaning

sewage overflow screening device with a sewage overflow chamber, a rectangular tank

and a slotted ogee weir to capture the gross pollutants (Figure 4.1). The device does not

require any power source containing mechanical or electrical components; moreover the

device is a self-cleaning device which is a key requirement for use in remote un-staffed

locations. The proposed gross pollutant trapping device improves most of the reported

limitations of commercial sewage solids overflow screening devices. Those limitations

are:

Limited screening efficiency

Blockages on the screen of the sewerage system

Use of a sophisticated (electrical/mechanical) switching system

No self-cleaning effect for use in remote un-staffed locations

A comprehensive modelling investigation was of paramount importance. Firstly it

was necessary to check the concept design of the proposed screening device. Secondly,

it was essential to get a detailed understanding of the screening device design

parameters and finally, it was compulsory to test it under a series of different conditions;

which are very difficult to carry out in a physical experimental situation. In addition,

experimental work involves significant cost and time. To overcome these challenges and

to design an efficient, effective and optimised experimental sewage solids overflow

screening device, a 3D computational fluid dynamics (CFD) model was used. The main

purpose of the modelling investigation was to optimise and design the sewer overflow

screening system. To achieve this objective a plausibility check was performed between

an analytical and a CFD model.


Md Abdul Aziz 46

The CFD model has adopted the following method or approach for analysing the

results

1. Once the plausibility check was completed for the CFD model the following

analysis was perform with the CFD model.

2. It is important to identify which inlet position will maximize the capture efficiency.

To understand this phenomena to compare two different inlet orientations (inlet parallel

to the ogee weir and inlet perpendicular to the ogee weir) for the proposed screening

device to obtain the maximum self-cleaning effect.

3. In design different experimental parameter it is important to understand the

pressure, velocity and water level information to optimise the experimental device. To

understand hydrodynamic flow properties such as velocity, water level, shear stress to

obtain the maximum self-cleaning effect and to find an effective location for the screening

device to maximise self-cleansing effect.

4. Due to the small device dimension it is important to understand the reflection of

wave on such device. To get a better understanding on the wave reflection it is important

to get the proper inlet length for the experimental device. To determine the optimum inlet

length to reduce wave reflections and improve functionality of the self-cleaning effect

5. There are different weir designs suggested in the literature so it is important to

understand which of these design will work better in the current setting. To optimise ogee

weir design orientation based on standard U.S. Army Engineers Waterways

Experimental Station design best practice guidelines.

4.2 Screening Concept

Figure 4.1 shows the overflow sewage device in the first phase. As sewage builds

up in the left chamber (A), water pressure will push the floatable ball upward in the right

chamber (B), which is connected to the left chamber via a pipe (C). As the floatable ball

goes upward, it will block the hole on the upper surface. The plan view of the proposed

device shows the left and right box chambers with the vertical dotted lines on the plan

view representing the screening device.The thick horizontal dotted lines on the plan view

represent the pipes connecting the left and right chambers. The thick smaller circle is the

hole at the bottom of the right chamber (B) and the dotted circle is the floatable ball. The

sewage builds up in the left chamber (A) until it becomes full, at which time the sewage


Md Abdul Aziz 47

overflow will pass over a weir type structure. Figure 4.2 shows the second phase of the

scenario with the overflowing sewage.

Figure 4.1: Schematic diagram of the proposed sewage overflow screening device

Figure 4.2: Front views of the proposed device under different phases


Md Abdul Aziz 48

Towards the bottom of the sloping weir, the screen will exclude the solids while

allowing water to pass through the screen, bypassing the right chamber (B) to exit to the

creek or waterway through two bypass channels (D). Figure 4.2 also shows the third

phase of the scenario, when the flow has subsided and the sewage level in the left

chamber (A) recedes. Once the sewage level drops down to a certain level, the buoyancy

pressure on the ball will reduce and the ball will drop, allowing the trapped pollutants to

exit into the right chamber (through the pipe C) and then to be flushed back into the

sewerage system using valve (E). This proposed overflow sewage device is designed to

be installed downstream of an existing sewage overflow location.

4.3 Development of the Analytical Model

Analytical models are sometimes also called mathematical models. This model

attempted to explain the behaviour of a system with mathematical equations. The power

of analytical models is that the mathematical function can provide information without

graphical or tabular presentation. In the current research an analytical solution of the

Navier-Stokes equations were derived [10]. A description of this model development

process is explained below.

To find an analytical solution using the Navier-Stokes equations some simple

assumptions are made [55]. Firstly, the flow is considered to be steady and uniform,

flowing under the influence of gravity and parallel to the surface bottom while the effect

of air viscosity at the free surface is negligible. As the surface is inclined we have to

consider the body force. Therefore with constant viscosity the Navier Stokes equation

becomes,

)()( 2

2

2

2

yu

xug

xP

yuv

xuu

tu

x

(4.1)

)()( 2

2

2

2

yv

xvg

yP

yvv

xuu

tv

y

(4.2)

The variables u, v, and w represent the velocities in the x-, y-, and z-directions; ρ

= density; μ = dynamic viscosity of water; P = defined as pressure; gx , gy are the


Md Abdul Aziz 49

gravitational force in x and y directions. As we consider 1D flow the z-direction does not

need to be considered as the flow is steady δ/δt =0. Moreover the flow is parallel to the

inclined surface, i.e. X-axis, so δux /δx =0 and u=0. As the flow is uniform the flow takes

over a constant depth ‘h’ which leads to pressure gradient δP/ δx =0. If z is the vertical

direction, the potential per unit mass due to body force is gz

Therefore the components of body force in the X and Y directions are:

singxzgx

gX z

(4.3)

cosgyzgy

gY z

(4.4)

After incorporating all assumptions, components of body forces Equations. 4.1

and 4.2 reduce to:

0sin 22

dyudg (4.5)

01cos

yPg

(4.6)

Taking the value in Equation 4.6

δP/δy = -ρg cosθ leads to

CgyP cos (4.7)

at y =h, P = 0 atmospheric pressure i.e. C = ρgh. Therefore the expression for

pressure becomes,

)(cos yhgP (4.8)

Now integrating Eq. (4.5) twice with respect to y yields,

21

2

2sin CyCygux

(4.9)

At boundary condition y = 0, ux = 0 i.e. C2 = 0; again at y = h, duv / dy =0; C1 = gh

sinθ/ v;

)2(sin2

2yhygu x

(4.10)


Md Abdul Aziz 50

Figure 4.3: Breakdown of the flow components of the experimental device

To get the shear stress at the boundary, applying Newton’s law of viscosity-

00)(sin

yy

xxy yhgdy

du

, which gives

singhxy (as y = 0) (4.11)

q is the flow per unit width,

sin31)2(sin

232

00

ghdyyhygdyuqhh

x

(4.12)

and average velocity,

sin31 2ghh

qv (4.13)

In the above equations, substituting inflow and weir surface angle with the

horizontal, unit width of the weir, different hydrodynamic parameters were calculated.

The US Army Corp [168] had developed standard shapes for the downstream profile of

the ogee weir defined by equation,

X1.85 = 2.0 Hd0.85 Y (4.14)

X is the horizontal axis and Y is the vertical axis. The depth of water upstream of

the spillway Hd is calculated from the non-dimensional equation for discharge given by,

Y

X2/A2 + Y2/B2 = 1

X1.85 = 2.0 Hd0.85 Y

cosg sing

g

X

h

Flow

Z


Md Abdul Aziz 51

23

0 232

eLHgCQ (4.15)

Where, Q = total discharge; L = Lateral crest length or Width; He = total head

upstream from the crest; g = gravitational constant; and C0 = discharge coefficient. As

the velocity head is relatively very small with total head Hd considered equal to He which

is 7.16 cm (that is, h/Hd greater than 1.33 and He = Hd, for the approach velocity head is

negligible) [41]. Moreover the effect of slope or roughness did not change the average

value of He. The curve of the ogee weir surface is drawn from the equation

Y = 1.744 X1.85 (4.16)

The position (0, 39.5) is the starting coordinate over the ogee weir and different

parameters are calculated based on different points taken on the curve. The slope angles

are used to calculate the analytical results for velocity using Equation 4.13, flow using

Equation 4.12 and shear stress from Equation 4.11. Table 4.1 provides all the analytical

value derives in this regard.

Table 4.1: Calculation of hydrodynamic parameters from analytical results

4.4 Computational Fluid Dynamics (CFD) Model

CFD modelling is the analysis of a system involving fluid flow [171]. This CFD

modelling technique is very powerful and covers a wide range of application areas. Some

of the applications include:

Aerodynamic shape design of aircraft and vehicles

X Y Slope sin h (cm) )(vVel m/s xy )/( 2mN

0.00 39.50

1.98 39.84 0.17 0.169 3.97 0.87 65.62

3.00 39.62 -0.15 0.129 4.33 0.79 54.97

6.00 39.124 -0.17 0.232 3.57 0.97 81.07

11.98 33.00 -0.67 0.555 2.67 1.295 144.93


Md Abdul Aziz 52

Hydrodynamic behaviour analysis of ships

Distribution of pollutant and effluents analysis

Flow analysis in rivers, estuaries and oceans etc.

The CFD codes are structured around numerical algorithms which can tackle fluid

flow problems. Some of the key steps in developing a CFD model include the CFD

modelling approach, discretization methods, schemes, turbulence modelling and

boundary conditions. CFD models use an iterative approach to obtain the solution of

Navier-Stokes equations. The Navier-Stokes equations are comprised of the

fundamental principle of conservation of mass and momentum. Hence, the cornerstone

of CFD is the fundamental governing equations of fluid dynamics – the continuity,

momentum and energy equations. They are the mathematical statements of three

fundamental principles upon which all of fluid dynamics is based:

The mass of fluid is conserved

Momentum is conserved, i.e. the rate of change of momentum equals the sum of

the forces on a fluid particle (Newton’s second law)

Energy is conserved, i.e. the rate of change of energy is equal to the sum of the

rate of heat addition and the rate of work done on a particle (first law of

thermodynamics)

CFD models describe the behaviour of fluid in terms of macroscopic properties,

such as velocity, pressure, density and temperature, and their space and time derivatives

[86]. From the 1960s onwards, the aerospace industry has integrated CFD techniques

into design, research and development of aircraft and jet engines. Increasingly CFD

became a vital component in the design of industrial products and processes. CFD has

entered into the wider industrial community since the 1990s [171]. With the rapid

improvement of the development of high performance computing facilities, CFD

modelling techniques have evolved as a powerful tool for researchers working on product

designs. CFD models can predict flow phenomena from single phase flows to complex

multiphase flows involving different fluid mixtures. Successful and efficient development

of a CFD model can predict fluid flow behaviour, provide insight into different product

orientations and help efficient and effective product design.


Md Abdul Aziz 53

Fluid flow behaviour in a system can be represented with a series of non-linear

partial differential equations. Analytical solutions of these equations are very difficult to

derive, except in some special cases. To get an approximate solution numerically, a

discretization method is applied. This method approximates the differential equations by

a system of algebraic equations. These equations can then be solved using a computing

scheme which provides a description of the flow field at discrete locations in space and

time. The accuracy of experimental data depends on the quality of tools used, and

similarly the accuracy of numerical solutions is dependent on the quality of discretization

used [62]. There are a number of different computing schemes describing fluid flow which

can be solved using computational methods. Most commercial and research codes rely

on the following:

Finite Volume Method (FVM)

Finite Difference Method

Finite Element Method

Spectral Methods Method

All of these schemes above need the definition of discrete points in space at which

variables such as velocity, pressure and temperature will be computed. Although the

governing equations are always the same, the particular geometry with initial and

boundary conditions determines a unique solution for each particular problem. The

current research is based on the finite volume method. Most of the popular CFD codes

currently available use this scheme.

4.4.1 Finite Volume Method (FVM)

The finite volume method (FVM) is comprised of the integral form of the governing

equations involving surface integrals (e.g. convective and diffusive fluxes) and volume

integrals (e.g. those describing sources and sinks). There is also a rate of change term

which applies in the case of transient flow (i.e. unsteady flow that changes over time).

The FVM represents the integration of the governing equations over contiguous control

volumes (CVs) representing the solution domain. As variable values are computed only


Md Abdul Aziz 54

at discrete points, approximations must be used to express the integrals in terms of

unknowns at discrete locations. The algebraic equations are obtained by linking the

variable value at the centroid of the CV with those at a neighbour CV.

A large system of algebraic equations is obtained for the whole solution domain.

As these equations are in general, non-linear and coupled, the solution must be sought

using iterative process. Which means repeating a sequence of operations over and over,

until changes in computed variables becomes negligible and the process is declared as

converged.

The current research used the academic CFD code of AVL Fire [22], however

most of the main commercial CFD codes, such as FLUENT, STAR-CD are based on the

FVM scheme. One of the key reasons FVM has succeeded over the other methods is

that it is inherently conservative. Although there are some errors in various

approximations, the discretised equations still fulfil the conservation laws exactly. In other

words, the errors introduced through various approximations affect only the distribution

of variables within the solution domain without violating conservation principles.

Moreover, FVM is easier for engineers to understand as other schemes involve more

complex mathematical methods.

4.4.2 Multiphase Flow Modelling

If the flow of interest involves more than one phase or component then the

problem needs to be considered as a multiphase flow in CFD. A phase can be defined

as an identifiable class of material that has a particular inertial response to and

interaction with the flow and the potential field in which it is immersed. For example,

different sized solid particles of the same material can be treated as different phases

because each collection of particles with the same size will have a similar dynamic

response to the flow field.

Two phase flow is the simplest case of multiphase flow. Multiphase flow can be

classified according to the state of the different phases or components. Therefore they

can refer to gas-liquid flows, liquid-solid flows or gas-particle flows or bubbly flows, and


Md Abdul Aziz 55

so on. In the studied problem the interaction of flow in the screening chamber can be

categorised as liquid-solid flows and interaction with the atmosphere as gas-liquid flows.

4.4.3 Approaches to Multiphase Modelling

Computational fluid mechanics (CFD) modelling in recent years has provided the

basis for further insight into the dynamics of multiphase flows. Currently there are two

approaches for the numerical calculation of multiphase flows: the Euler-Lagrange

approach and the Euler-Euler approach.

4.4.4 Euler-Lagrange Approach

The Lagrangian discrete phase model follows the Euler-Lagrange approach. A

fundamental assumption made in this model is that the dispersed phase occupies a low

volume fraction, even though high mass loading is acceptable. This approach is

generally used for highly dispersed flows where the volume fraction of the dispersed

phase is small. The time-averaged Navier-Stokes equation is solved for the fluid phase

which is treated as a continuum, while the dispersed phase is solved by tracking a large

number of particles, bubbles, or droplets through the calculated flow field. There is an

exchange of interfacial momentum, mass, and energy between the dispersed and the

continuous phase. The particle or droplet trajectories are computed individually at

specified intervals during the fluid phase calculation. The model is appropriate for the

modelling of spray dryers, coal and liquid fuel combustion, and some particle-laden flows.

However, it is inappropriate for the modelling of liquid-liquid mixtures, fluidized beds, gas-

liquid flow or any application where the volume fraction of the secondary phases is not

negligible. In the current research we had to consider interaction between gas-liquid

flows hence the current research did not adopt the Euler-Lagrangian approach.

4.4.5 Euler-Euler Approach

In the Euler-Euler approach the fluid phases are treated mathematically as

interpenetrating continua. Fluids are treated in every computational cell using the


Md Abdul Aziz 56

concept of phasic volume fraction. For a two phase flow situation each of the phases is

considered to occupy a fixed volume fraction in a computational cell. This is because the

volume of one phase cannot be occupied by the other phase. These volume fractions

are assumed to be continuous functions of space and time and their sum is equal to one.

Conservation equations for each phase are derived to obtain a set of equations, which

have a similar structure for all phases. These equations are closed by providing

constitutive relations that are obtained from empirical information, or, in the case of

granular flows, by application of kinetic theory.

The present study is based on the Euler-Euler approach, because the research

deals with the interaction between gas and liquid flows. The present research was carried

out by using the commercial CFD package AVL FIRE [22]. The FIRE Eulerian Multiphase

Module allows for the use of the following models based on the Euler-Euler approach.

These are listed below in the order of increasing accuracy:

Homogeneous (Equilibrium) Model

Multi-fluid Model

Volume of Fluid (VOF) Free-Surface Model

Homogeneous Model

The homogeneous model is the least accurate multiphase model based on the

Euler-Euler approach. A volume fraction equation is calculated for each phase. However,

only a single momentum equation is calculated for the phases in momentum equilibrium.

Multi-fluid Model

All conservation equations are solved for each phase. Since the multi-fluid model

requires by default the calculation of the complete set of the conservation equations for

each phase, it represents the basis for the Euler-Euler schemes in the FIRE Eulerian

Multiphase Module. The commercial software AVL FIRE’s user-defined subroutines

(UDF) allow for customizing the calculation of the mass, energy and momentum

exchange.


Md Abdul Aziz 57

VOF Model

The VOF model is a surface tracking technique applied to a fixed Eulerian grid

proposed by Hirt and Nichols [82]. This model is designed for two or more immiscible

fluids where the accurate prediction of the interface between the fluids is of interest. In

the current research we had to consider the interaction between air and liquid flow, hence

the concept of VOF was important to explain the interaction between air and water in the

cell volume. In the VOF model, a single momentum equation is shared by the fluids, and

the volume fraction of each of the fluids in each computational cell is tracked throughout

the domain. Applications of the VOF model include stratified flows, free surface flows,

filling, sloshing, the motion of large bubbles in a liquid, the motion of liquid after a dam

break, the prediction of jet break-up (surface tension), and the steady or transient

tracking of any liquid-gas interface.

From the numerical perspective the volume of fluid model is very similar to the

homogeneous model. A single momentum equation is calculated for all phases that

interact using the VOF model. However, the calculation of volume fraction equations

using the VOF model is considerably more accurate allowing for the sharp resolution of

the interfaces. One of the common defects of the VOF calculation can occur when the

interface is not resolved sharply despite the use of the high-order discretization

techniques for the volume fraction equation – in that case the VOF model degenerates

into the homogeneous model. This is quite common in many practical calculations. It

happens due to the very high resolution requirements of the VOF model that can be often

hard to fulfil.

In the following sections, details of the modelling procedures including model

geometry, solution procedures and governing equations solved for each geometry will

be discussed further.

The Navier-Stokes equations can predict fluid flow behaviour in its general form.

The hydrodynamic characteristics of the overflow sewage screening device were

investigated using a CFD model by adopting the finite volume method in the Euler–Euler

approach. The 3D multiphase flow numerical model was developed using a commercially

available CFD package, AVL Fire [22] to predict the flow over the ogee weir.


Md Abdul Aziz 58

4.4.6 Model Geometry and Computational Methodology

The screening device has a rectangular tank (1m X 0.2m), an ogee weir and an

inclined surface. The height of the ogee weir bottom is 0.75m and the diameter of the

inlet pipe is 0.2m. A design flow rate of 40 litres/second was considered for this analysis.

The outflow was assumed to be free flowing and perpendicular to the outlet surface at

the edge of weir. A 3D CAD model, similar to the schematic diagram shown in Figure

4.4, of the proposed sewage overflow screening device was developed using a CAD tool.

The volume mesh generated by using the CAD model for the CFD analysis is

shown in Figure 4.5 and Figure 4.6. Two different inlet positions were chosen to analyse

water level, flow velocity and shear stress distribution for the proposed screening device.

The model which was developed included the following features and assumptions:

Unsteady state multiphase solution for momentum and continuity was considered

Standard k-ε turbulence model for the turbulence modelling was employed

A cell centred finite volume approach was used to discretise the governing equations

and the resulting discretised equations were solved iteratively using a segregated

approach

Pressure and velocity were coupled using the Semi Implicit Method for Pressure Linked

Equations (SIMPLE) algorithm, [128].

Least squares fit approach was used for the calculation of the derivatives. Some other

important works include [32] and [124] on turbulence modelling and [130] on the self-

cleaning approach.

For momentum and turbulence, a first order upwind differencing scheme was used

whereas a central differencing scheme with second order accuracy was used for the

continuity equation

Screening device walls were treated as standard wall functions with no slip condition


Md Abdul Aziz 59

Air was considered to be the dispersed phase and water was considered to be the

continuous phase.

Figure 4.4: Geometric details of the screener device

4.4.6.1 Governing Equations

Basic Eulerian equations (Equations. 4.17 and 4.18), describing multiphase flow

were expressed by the conservation equations for mass and momentum as follows:

Continuity:

t

kk

+ . kk v

k = 0 k= 1,……,N (4.17)

Where N is the number of phases, k is volume fraction of phase k, k is

density for phase k, vk is phase k velocity and 1

1

kk

Momentum Conservation:


Md Abdul Aziz 60

tkk

kv+ . kk v

k v

k = - k p + . k ( k + t

kT ) + k fk +

kll ,1

Mk

(4.18)

kll ,1

Mkl represents the momentum interfacial interaction between phase’s k and l, f is

the body force vector which comprises gravity (g), p is pressure. Pressure is assumed to

be identical for all phases:

p =kp k = 1, ……., N

The phase k viscous stress integral is divided into non-transposed and

transposed terms. It can be expressed as:

k = k (v k + v Tk ) (4.19)

Where, k is the molecular viscosity. For incompressible flow, Reynolds stress,

tkT , takes into account the effect of turbulence which can be expressed by the

Boussinesq hypothesis:

tkT = - k

kkvv = tk (v k + v T

k ) - 32

kkkk (4.20)

Where k is the Kronecker delta function and t

k is the turbulent viscosity. For

the continuous phase, turbulent viscosity was calculated by adding shear induced

turbulent viscosity with Sato’s viscosity due to bubble induced turbulence [142].

tc = SIt

c, + BIt

c, (4.21)

Where shear induced turbulent viscosity for continuous phase can be expressed as,

SItc

, = Cc

c

2ck

(4.22)

Sato’s viscosity due to bubble induced turbulence can be expressed as [142],


Md Abdul Aziz 61

BItc

, = Csato c D

bvr d (4.23)

where C = 0.09 and C

sato = 0.6 are dimensionless constants, k is the turbulent

kinetic energy and is its dissipation rate which can be obtained by solving equations

for the standard k- turbulence model put forward by Launder and Spalding [102]. The

turbulent kinetic energy (k) equation can be expressed as:

N

kllklkkkkkk

k

tk

kkkkkkkkk KPkk

tk

,1.v.

(4.24)

k = 1, ……… , N

N

kllklK

,1 is the interfacial turbulence exchange between phases and Pk is the production

term due to shear. Turbulence dissipation (ε) equation is,

k

kkk

k

kkk

N

kllklk

tk

kkkkkkkkk

kC

kPCD

t2

2

1,1

.v.

(4.25)

Closure coefficients used in the model are k =1.0, =1.3, C1=1.44, C2=1.92,

C

=0.09

N

kllklD

,1 represents interfacial dissipation exchange between phases.

Momentum interfacial exchange between gas and liquid was modelled by

implementing interfacial momentum source at the interface which includes drag and

turbulent dispersion forces (AVL 2008)

Mc= C

D 81

c iA vr

vr + C

TD c ck d = dM (4.26)

Where c denotes continuous (water) and d denotes the dispersed phase (air).

The first term in Equation (10) represents mean contributions due to drag force and the

second term takes into account the turbulence effect.


Md Abdul Aziz 62

The drag coefficient,DC , is a function of the bubble Reynolds number, Re

b. The

following correlation for drag coefficient,DC , was used [22]

687.0Re15.01Re24

bb

DC Reb 1000 (4.27)

Bubble Reynolds number, Reb, and can be defined as:

c

brb

D

vRe (4.28)

Where c is the kinematic viscosity for the continuous phase relative velocity is

defined as:

vr = v

d - v

c

The interfacial area density for bubbly flow can be expressed as [63]:

b

di DA 6 (4.29)

Where Db = 0.01 mm is the bubble diameter and d is dispersed phase volume

fraction. Bubble dispersion coefficient used in Equation [3.26] was,

CTD= 0.1

The flow was initialised in the simulations with small initial values assigned to k

and ε, which made the initial turbulent viscosity roughly equal to the kinematic viscosity

for water. The fluid properties for air and water were taken as the properties at NTP (T

= 293.15 K, P = 1 atm). Typical turbulence quantities at the inlet of the domain were

calculated from inlet velocities by considering turbulence intensity, I = 0.05 where,

81-

inlet 0.16(Re)UuI / .

The simulation was carried out considering an unsteady state condition with time

steps of ∆t = 0.05 second. Total time period for each run was 180 seconds which was

adequate to obtain time averaged steady state results and numerical stability. Three grid

resolutions were tested for the grid independency test. The purpose of a grid

independency test is to determine the minimum grid resolution required to generate a

solution that is independent of the grid used. Starting with a coarse grid, the number of

cells was increased in the region of interest until the solution from each grid was

unchanged for successive grid refinements. All the cells in the calculation domain were


Md Abdul Aziz 63

polyhedral with a large number of hexahedral cells. As the computational domain

consisted of hybrid unstructured meshes in a curvilinear non-orthogonal coordinate

system with artesian base vectors and refined regions in some locations, calculating the

number of cells in each direction was complicated.

Figure 4.5: Position 1 (condition 1) is the inlet parallel to the ogee weir

A view of the coarse computational grid is shown in Figure 4.5 and Figure 4.6,

which consists of a total of 38619 cells in the whole computational domain. Meshing

procedure was done by Fame Advanced Hybrid meshing technique [22].

Figure 4.6: Position 2 (condition 2) is the inlet perpendicular to ogee weir


Md Abdul Aziz 64

4.4.7 Boundary Conditions

To simulate the numerical model for designed flow, it is important that boundary

conditions accurately represent what is physically happening in the device, refer to

Figure 4.7.

Figure 4.7: Boundary conditions used in the CFD model

Inflow Condition: Inlet mass flow (designed) was 40 kg/s and it was assumed that

flow direction was perpendicular to the inlet surface (parallel to inlet pipe) shown in red

(Figure 4.7).

Atmospheric Boundary: Atmospheric pressure boundary condition was

considered for the upper curved surface as shown in blue in Figure 4.7.

Wall Boundary: Roughness height of the ogee weir surface was assumed as

0.001m and wall functions are based on the assumed logarithmic velocity distribution

(mark with pink in Figure 4.7). The friction velocity (Uτ) is defined by;

wU (4.30)

Where, w- shear stress and -fluid density. For mean velocity the following wall

functions are used:


Md Abdul Aziz 65

63.11, *** <yyU (4.31)

63.11,ln1 *** >yEyK

U (4.32)

Where,

pp UUk

CU

2141* (4.33)

and

pp ykCy

2141* (4.34)

Index “p” denotes the values at the centre point of the wall-nearest the control

cell, k is kinetic energy, μ is viscosity for mean velocity and y is the normal distance from

the wall. Here, coefficient C = 0.09, E = 9 is the constant in the law-of the-wall and K =

0.41 is von Karman constant. The near wall viscosity (μw) can be expressed by;

*

*

Uy

w (4.35)

And the wall shear stress (τw) defined by;

ptwt

p

ppw UU

EykKC

*

2141

ln

(4.36)

Where,

wwpwpwptwt nnUUUUUU (4.37)

The subscript’t’ denotes the tangential direction, parallel to the wall surface.

Outflow Condition: Outflow was assumed to be a free flow and perpendicular to

the outlet surface at the edge of the weir, shown in green in Figure 4.7.

4.4.8 Explaining CFD Results

Time periodic results were taken at the end of every 2 seconds for a total time of

180 seconds for every simulation run. Then, the values obtained for velocities and

turbulence properties were time averaged for all the time steps. To obtain a converged


Md Abdul Aziz 66

solution, the approach used was to reduce the normalised sum of absolute residuals to

a value of 1.0 x 10-4 for each time step. The water level was determined by considering

the flooding and drying concept reported by Stelling [157], which suggested a value ≥

50% of the volume fraction to be considered as water level. Flow reflection dominated

the cross section profile of the sewage overflow device.

Figure 4.8: Water levels over the weir at different locations

To analyse the 3D numerical results for water levels, velocities and to show

changing profiles, three distinct sections across the width of the weir were selected, left,

middle and right. The position of the left, middle and right sections in the simulation result

are shown in Figure 4.8. The first set over the weir, and the second and third sets, are 3

cm and 6 cm downstream of the ogee weir respectively. For the presentation of

numerical results, locations parallel to the weir inlet are numbered as points 1, 2, 3 and

perpendicular to the weir inlet are numbered as points 4, 5, 6.

A multiphase modelling approach was selected to model the air and water.

Volume fraction of the water as phase 2 was selected to determine the free surface

profile. The results of the final time steps volume fraction are shown in the Figure 4.9. In

determining water level, assumptions were made based on the flooding and drying

concept

Figure 4.9 suggest there are some wave reflections on the right sight and the

splashing of water will be within the device. Figure 4.10 suggest wave reflections on both

Left Water Level Middle Water Level Right Water Level

Width of weir opening 1000 mm


Md Abdul Aziz 67

edges of the weir. The wave reflection of such small device is very important aspect in

designing the device to maximise self-cleansing effect at the bottom.

Figure 4.9: Volume fraction of water at inlet parallel (position 1) to ogee weir

Figure 4.10: Volume fraction of water at inlet perpendicular (position 2) to ogee weir


Md Abdul Aziz 68

4.4.9 Plausibility check of the CFD model

Plausibility check of the CFD model was done using a simplified analytical

solution. To find an analytical solution using the Navier-Stokes equations some simple

assumptions were made.

Figure 4.11: Comparison of flow velocities over the top of the ogee weir

Figure 4.12: Comparison of flow velocities 6cm downstream of the ogee weir


Md Abdul Aziz 69

Firstly, the flow was considered to be steady and uniform, flowing under the

influence of gravity and parallel to the bottom surface while the effect of air viscosity at

the top surface was negligible. Analytical equations for shear stress, unit flow, average

velocity and curve of the ogee weir were calculated and reported [10]. Plausibility of the

numerical water levels was checked against the analytical water level. The (one-

dimensional) analytical solution matches the trend of the 3D CFD water level reasonably

well, (Figure 4.11 and Figure 4.12). In the absence of experimental data this provides a

decent plausibility check for the numerical model.

4.5 Discussion of Results

4.5.1 Discussion of Hydrodynamic results

To understand flow reflections, CFD simulated water levels at the left, middle and

right sections along the flow were extracted. These results were compared with the one-

dimensional analytical solution considering steady, incompressible fluid. Analytical

formulation is unable to include flow reflections from the wall. A model result shows the

dominant effect of flow reflection in the relatively small sewage overflow device as shown

in Figure 4.13 and Figure 4.14.

Figure 4.13 Comparison of water level along the flow for Condition 1


Md Abdul Aziz 70

Figure 4.14 Comparison of water level along the flow for Condition 2

Figure 4.15 and Figure 4.16 shows how the velocity changes over the weir. As

the flow propagates downstream of the weir, the velocity at the bottom increases three

times that of the velocity over the ogee weir in both cases. This increase in velocity will

effectively increase the self-cleaning capacity of the device.

If screens are provided near the top of the weir surface at condition1, only the

right-hand side strip will get efficient self-cleansing while the left side holes are likely to

be blocked by larger pollutants in the sewer water.

However as flow becomes uniform (across the width) near the bottom of the weir

surface, the self-cleansing capability can be achieved. Keeping this fact in mind, it is

proposed to provide perforations (circular holes) near the bottom of the weir surface.

Analysis from flow velocity and pressure will explain further in regards to perforations

location.


Md Abdul Aziz 71

Figure 4.15: Velocity vector at the inlets parallel (left)

Figure 4.16: Velocity vector at perpendicular (right) to the ogee weir


Md Abdul Aziz 72

To understand how the velocities change due to the reflection effect at different

sections, comparison of velocities at different inlet orientations are shown in Figure 4.17

and Figure 4.18 for inlet condition 1 and condition 2.

Due to varying water levels (high water level near the right side and low water

near the left side), near the top of the weir level surface, the self-cleaning property will

not be as effective near the top region at condition 1. There is not much variation of

pressure in both condition 1 and condition 2 as both cases atmospheric pressure was

used as boundary condition, refer to Figure 4.19.

Figure 4.17: Comparison of flow velocities along the width for condition1

Figure 4.18: Comparison of flow velocities along the width for condition 2


Md Abdul Aziz 73

Figure 4.19: Pressure variation at condition 1

Figure 4.20: Pressure variation at condition 2


Md Abdul Aziz 74

If screens are provided near the top of the weir surface at condition 1, only the

right-hand side strip will achieve efficient self-cleaning while the left side holes are likely

to be blocked by larger pollutants in the sewage water. However as flow becomes

uniform (across the width) near the bottom of the weir surface, the self-cleaning capability

can be achieved towards the bottom. Keeping this fact in mind, it is proposed to provide

perforations (circular holes) near the bottom of the weir surface.

As the water flows down, its velocity increases due to gravitational acceleration.

CFD simulation shows the formation of flow separation near the ogee weir. The CFD

simulated shear stress is lower than the analytical value as it is unable to consider flow

undulations. Moreover analytical calculation assumed an in-viscid fluid domain without

having any boundary layer effect while CFD simulation considered viscous boundary

effects, refer to Figure 4.21. The additional shear stress will be beneficial for removing

the sewer solids sticking near the perforations. Figure 4.22 shows distribution of shearing

stress on the sewer overflow screening system.

Figure 4.21: Shear stress distributions for the inlet parallel to the weir width

Analysis of the shear stress along the flow path was performed to identify efficient

self-cleaning location. From CFD simulation it was shown that shear stress increases

significantly towards the bottom of the inclined surface, which suggests that, the location

Screening area


Md Abdul Aziz 75

of the screen should be towards the bottom. Moreover, the flow becomes uniform across

the width in this area, which will also help effective screening.

4.5.2 Discussing Location of Circular holes

Due to varying water levels (higher at the right than the left side), near the top of

the weir surface, the self-cleaning property will not be as effective near the top region at

condition 1. If screens are provided near the top of the weir surface as at condition 1,

only the right-hand side strip will have efficient self-cleaning while the left side holes are

likely to be blocked by larger pollutants in the sewage water. However as the wave

becomes uniform (across the width) near the bottom of the weir surface, the self-cleaning

capability can be achieved. Additionally as the water flows down, its velocity and shear

stress increases. Velocity of the sewage flow increased up to three times near the

perforated holes from around 1m/s to 3m/s for an inlet flow of 40 l/s. Shear stress also

increased substantially from around 100 N/m2 to 300 N/m2 close to the perforated holes

[10]. Keeping this in mind, it is proposed to provide perforations (circular holes) near the

bottom of the weir surface.

Figure 4.22: Comparison of shearing stress along the bottom of the curved

surface


Md Abdul Aziz 76

4.5.3 Discussion of the Inlet performance

The CFD simulated results showed that due to wave reflection under condition 1

(inlet parallel to the weir, refer to Figure 4.23), water level on the right side overrode the

water levels in the middle and left sections.

Figure 4.23: Comparison of water levels along the flow for conditions 1, water

level on the top of the weir, 3cm and 6 cm downstream

Figure 4.24: Comparison of water levels along the flow for conditions 2, water

level on the top of the weir, 3cm and 6 cm downstream


Md Abdul Aziz 77

Whereas, under condition 2 (inlet perpendicular to the weir, refer to Figure 4.24),

the reflected wave contributed to elevated water levels towards both the left and right

sides of the device.

Towards the bottom of the inclined surface, the wave became an equally

distributed turbulent flow (across the width).The reflected water level on the right side

reduced as the wave traveled downstream of the ogee weir (condition 1). Due to higher

water level, higher velocity was found downstream near perforations which were a

favorable condition for self-cleaning. Condition 1 can provide a better screening effect

on the right side because high water levels generate higher velocity and shear stress

(Figure 4.25).

Figure 4.25: Impact of device inlet position on the wave reflection viewed from

the back of the weir with a lateral inflow


Md Abdul Aziz 78

4.5.4 Standard Weir orientation

Since ogee weirs provide excellent hydraulic features, in-depth research has

been carried out to determine the standard shape and size of the crest of the overflow

spillway [41]. With extensive physical modelling tested by the U.S. Corp of Engineers,

they suggested a standard design parameter for the ogee weir (US Corp of Army [169]

& [170]. The proposed experimental design conditions were tested using the CFD

modelling technique to maximize uniform flow for better screening. Inlet conditions P &

Q with 0H: 3V and 2H: 3V slope demonstrate similar types of water level variations due

to wave reflections from the wall (Figure 4.26).

Figure 4.26: The Waterways Experimental Station (WES) standard spillway

shapes [41]

P Q

R S


Md Abdul Aziz 79

In fact orientation P produced more wave reflection than the rest. Orientation S

with 3H: 3V slope also demonstrated scanty water level available on the left of the weir.

Inlet condition R with 1H: 3V slope for the rectangle with the ogee showed less wave

reflection compared with others.

Orientation R provided the best self-cleaning effect and maximum screening

efficiency downstream. With orientation R flow become equally distributed across the

weir ensuring more effective self-cleaning (Figure 4.27).

Figure 4.27: CFD results viewed from the back of the weir with a lateral inflow on four

standard inlet orientations as suggested by the U.S. Army Engineers Waterways

Experimental Station

P

Q

R

S

Starting point of Equally distributed

Wave


Md Abdul Aziz 80

4.6 Limitation of Screening Device

The device self-cleansing effect largely depends on falling water but after some

modification it could not reduce blinding of the perforations. This is a major limitation as

the annual maintenance requirement exceed regularly due to blinding of the perforated

screening during CSO events [131].

4.7 Summary

Some of the key finding from the CFD model are summarised below:

The inlet parallel to the ogee weir was considered a better inlet option as water

level over the ogee weir was higher due to wave reflection on the right side which can

provide higher velocity and shear stress. Condition 1 can produce a better self-cleaning

mechanism due to higher velocity and shear stress near the perforated holes.

The flow becomes uniform near the bottom of the inclined surface with higher

velocity and shear stress. This suggests that the perforations should be placed near the

bottom of the inclined surface to achieve an effective self-cleaning capability for the

device. Uniform flow towards the bottom of the inclined surface will help to remove any

pollutants adhered to the perforations.

As the sewage overflow screening device is small, the wave reflection effect

was found to be dominant for this device. It is suggested that a 1.5m long inlet will reduce

wave reflection up to 10% compared to a 0.3m inlet.

Four standard ogee weir orientations were analysed to reduce wave reflection.

Orientation R with an inclined slope 1H: 3V from the rectangular device to the ogee weir

was found to be the most efficient based on the best practice guidelines provided by the

U.S. Army Corp of Engineers.

The study provided valuable insights into designing an efficient and effective

gross pollutant device. The experimental work was restricted by the physical limitations

of cost and time which are inherent in laboratory studies. CFD modelling provided an

excellent opportunity to design the gross pollutant screening device. Study findings will

help to maximize its functionality and its effectiveness in trapping sewage solids with high

efficiency, both in terms of the capture of sewage solids and the development of a self -


Md Abdul Aziz 81

cleaning mechanism. Chapter 5 will discuss an improvement to the proposed self-

cleaning device for it to work efficiently in remote un-staffed locations.

Work presented in this chapter has been published in the following journal and

conference papers:

Aziz, M. A., Imteaz, M., Huda M., & Naser J., 2014a, ‘Optimising inlet

condition and design parameters of a new sewer overflow screening device using

numerical modelling technique‘, Journal of Water, Sciency and Technology vol.70,

no.11, pp.1880-1887

Aziz, M. A., Imteaz, M., Naser J., & Phillips, D., 2013b, ‘Hydrodynamic

Characteristics of a New Sewer Overflow Screening Device: CFD Modelling and

Analytical Study’, International Journal of Civil and Environmental Engineering, vol. 7,

no.1, pp.71-76.

Aziz, M. A., Imteaz, M., J. Naser, Nazmul H., & Phillips, D. 2010,

‘Hydrodynamic Characteristics of a proposed sewer overflow screening device’, at The

5th Civil Engineering Conference in the Asian Region, 8-12 August, Sydney Australia.

Aziz, M. A., Imteaz, M., Nazmul H., & Jamal N., 2013d, ‘Understanding

functional efficiency of a sewer overflow screening device using combined CFD and

analytical modeling’, 20th International Congress on Modelling and Simulation, Adelaide,

2 to 5 December, Australia.

Chapter 5: Improvement of the Screening Device

Md Abdul Aziz 82

Chapter 5

Improvement of the

Screening Device


Md Abdul Aziz 83

5.1 Introduction

Under wet weather conditions, sewage overflows are of serious concern to the

environment, aesthetics and public health. Sewage solids are dispersed, suspended or

washed into rivers. They eventually settle, creating odours and a toxic/corrosive

atmosphere in bottom mud deposits. The solids also are aesthetically unpleasant either

in their general appearance or because of the actual presence of specific, objectionable

items, such as floating debris, sanitary discards/faecal matter, scum or even parts of car

tyres. In order to reduce these sewage solids, different types of screening devices are

used in existing networks. According to Faram et al., [60] screening is the only

economically viable method to segregate sewer solids in most cases.

Although the sewage solids overflow screener, discussed in Chapter 4, worked

well and had reasonable capture efficiency, the self-cleaning mechanism only gave

reasonable results. Blockages on the screener reduced its screening capacity which

reduced the capture efficiency of the gross pollutant device. Similar findings [9] report

that hairs from fibrous material wrap around the wires of bar screens and reduce

screening efficiency. This blockage effect occurred more often with small sewage solids

less than 10 mm in diameter.

This triggered the need to research efficient and effective self-cleaning screening

devices and screening handling systems. Past studies used a number of different

screening systems in sewage overflow locations. Hydrodynamic vortex separation

(HDVS) was a popular screening concept developed in the early 1960s. First generation

HDVSs were found to be effective in retaining 70% of the pollution load [151]. To improve

the pollution load second generation HDVSs developed by the American Water Works

Association and EPA were reported by Field [64] and the third generation device in the

1980s was commercially patented as Storm King® overflow. The HDVSs went through

a series of performance evaluations in Europe, North-America, and Japan [20].

Unfortunately lightweight sewer solids of neutral buoyancy were not trapped in HDVSs

[16]. A comprehensive review of different screeners was provided by Saul [144] & [146].

The recent update of these research papers can be found in the work of [106] and [107].


Md Abdul Aziz 84

Reported literature suggests that screens need to have a self-cleaning mechanism to

work efficiently.

To overcome this challenge a non-powered self-cleaning screening system that

can capture neutrally buoyant light weight sewage solids greater than 6 mm in two

diameters was tested by [152]. Later [59] tested six hydro jet devices installed in USA,

Australia and mainland Europe. However, in most cases the devices were directly

associated with blockages of the sewerage system. Usually the most conventional

screening systems utilise electro-mechanical components to facilitate such a process

[131]. However, given the harsh, unstaffed, remote operating environment of many

sewage overflows this device is not ideal [10]. Some of the common drawbacks in the

available commercial sewer overflow screening system include inadequate screening

capacity, requirements of external power and high cost [153]. To overcome such

problems a novel self-cleaning sewage overflow screening system is proposed [11].

Some of the key attribute of the proposed sewer overflow screening device include: less

expensive, has low maintenance and contains no moving parts, no moving parts and

robust stop/start operation. The proposed screening system consists of temporary

holding tanks that provide a transient storage and real time control of sewerage overflow.

The aim of this proposed device was to overcome key limitations in existing

commercial screening devices. The key attributes of the proposed Comb Separator

device are listed below:

To minimise blockage on the screen of solid sewage materials at the Comb

Separator the device should be built so that minimal inspection and maintenance

are required. The device should be able to be constructed above ground level.

The treatment capacity required for the drainage system must be able to cope

with a maximum one year overflow event i.e., 70 l/s per meter width of the device

and a minimum of a one in four months flow of 20 l/s per meter width.

The screening device should capture more than 90% of solids exceeding 10 mm

in diameter and return them to the sewerage system for all events up to and

including the one year overflow event (70 l/s). It should provide high capture

efficiency (more than 80%) for solid sewage materials less than 10 mm in


Md Abdul Aziz 85

diameter. If the proposed screening device can capture such high efficiency than

the ongoing operation and maintenance cost will be minimum.

It should not use sophisticated electrical, mechanical systems or signals which

may become ineffective during extreme events, such as floods or lengthy

droughts. In the event of a device failing there must be a bypass option causing

no serious damage to the device or the environment.

The present research focused on long-term operational performance, durability

and low maintenance requirements. In comparison to the existing electro-mechanical

screeners, the proposed Comb Separation device has negligible blinding during

screening and better applicability for remote locations. As there are no moving parts,

there will less maintenance and operational cost for the device as these screening

located in harsh environmental conditions [8]. This research also focused on optimising

the performance of the proposed comb separator to ensure high capture efficiency.

A series of laboratory tests were carried out to evaluate steady state, short

duration (varying from 6 minutes to 32 minutes) flow conditions with flow variations from

20l/s to 70l/s. Instead of using static screens which have a high maintenance cost, a self-

cleaning comb separation device was used (Figure 5.1). Analysis of the result will

discuss the key experimental conditions of flow, comb spacing, effective spacing

(spacing between different layers of comb), comb layers, weir opening, and runtime.

The key objectives of the experiments are listed as follows:

To analyse capture efficiency of sewage solids more than 10 mm in diameter

To comprehend the impact of experimental design parameters (flow discharge,

weir opening, comb spacing and layers) on sewage solids capture efficiency

To check reproduction of the experimental results and ensure generation of

consistent results

To compare performance of the proposed Comb Separator with the standard

Hydro JetTM screen under low flow (up to 70 l/s) conditions.


Md Abdul Aziz 86

5.2 Methodology used in the Experiment

5.2.1 Data Collection

A series of experimental trials were conducted to monitor capture efficiency of the

experimental device.

Figure 5.1: Experimental set ups Comb Separator

AA = Width of the sharp crested weir,

BB = Distance between 1st and 2nd comb spacing,

EE= Distance between sharp crested weir to the outlet chamber,

FF = Height of the sharp crested weir from sewer screening chamber,

GG= Distance from the face of the sharp crested weir to the angle screener,

HH = Distance from the sharp crested weir to the angle screener,

JJ = Height of the angle screener, KK = Diameter of the screening chamber,

MM = Distance of the angle screener from the outlet chamber,

NN= Diameter of the valve ball, PP = Fall distance of the valve ball,

QQ= Height of the sharp crested weir from bottom of the chamber,

RR = Height of the device screening chamber from setup location,

TT= Height of the outlet chamber.


Md Abdul Aziz 87

Six different experimental set ups with twenty two different experimental

conditions were tested in the experimental facility at Swinburne University of Technology,

Melbourne, Australia. The flow diagram of the experimental process is shown in Figure

5.2.

Figure 5.2: Flow diagram of the revised screening experimental works

The experimental conditions were varied by changing the critical experimental set

ups such as flow discharge, weir opening, comb spacing and comb layers. A total of 51


Md Abdul Aziz 88

sets of experimental data were collected. The schematic diagram of the experimental

set-up is provided in Figure 5.2. To test and validate the performance of the Comb

Separator, critical model parameters, such as runtime, flow discharge, weir opening,

spacing and layers of combs were varied. The aim was to achieve capture efficiency at

or above 80% with minimal blockage on the combs. Overflow event was on a one in one

year overflow for sewer solids less than 10mm like cigarette butt, refer to Figure 5.3. The

robustness of the optimum experimental conditions was validated by repeating the

experiment several times to ensure reasonable consistency of results.

Figure 5.3: Concept diagram of target capture efficiency curve

5.2.2 Screening Mechanism

The laboratory device was connected to an inlet pump and inlet pipe. Two outlets

were mounted on the device - one to convey overflow water away and the other to drain

the sewage water remaining in the storage chamber [9]. A series of combs, to segregate

sewage solids from the sewage overflow, were mounted next to the sharp crested weir

(Figure 5.4).


Md Abdul Aziz 89

Figure 5.4: Experimental set up for the proposed sewage overflow screening device

Phase 1: After the start of precipitation overflow, the storage chamber fills with

sewage. A floating ball, at the bottom of the sewage solids holding chamber, closes at

this point as shown in Figure 5.5. As the overflow continues, the storage chamber

overflows above the sharp crested weir. The captured sewage solids are intercepted by

the parallel combs and fall into the holding chamber (pollutant capture chamber).

Figure 5.5: Operation procedure of new sewage overflow screening device- Phase 1


Md Abdul Aziz 90

Figure 5.6: Operation procedure of new sewage overflow screening device-Phase 2

Phase 2: After cessation of precipitation, the water level within the storage

chamber falls below the valve level. The low pressure of the liquid in the sewage solids

holding chamber allows the ball to drop and flushes the entire captured sewage solids

back into the storage chamber which is shown in Figure 5.6.

Ball Valve Details

Efficient functioning of the lower ball valve chamber was the key aspect of the

device design. The device needs to work in unstaffed remote locations so self-cleaning

activity was the key functionality of the device. The detailed designs of ball valves used

in the sewage chamber, are shown in Figure 5.7.

Common sewage solids, like condoms, tampons, cigarette butts, wraps, cotton

balls and bottle caps, were used in the experiment. The experimental conditions were

varied with different flow volumes and the number spacing of comb layers. The sharp


Md Abdul Aziz 91

crested weir effectively responded to the device failed condition and bypassed the

outflow chamber.

Figure 5.7: Showing the design parameter of the ball valve chamber

To obtain the optimum comb spacing, an experiment was set up using various

comb spacing’s was varied from 25 mm centre to centre to 10 mm centre to centre. Weir

openings were adjusted from 460 mm to 970 mm across the width of a 1000 mm weir.

To confirm blinding at the comb, separation was minimal. Two and three comb layers


Md Abdul Aziz 92

were tested. In addition to these, flow volumes varied from a one in four month sewage

overflow condition (20 l/s/m) to a one in one year overflow condition (70 l/s/m). Sewage

solids tested in the experiment are classified as sewage solids more than 10mm in

diameter (toilet paper, bottle tops, dish wipes, tampons, condoms, cotton balls) and

sewage solids less than 10 mm (cigarette butts and artificial cigarette butts). Capture

efficiency of the sewage solids was calculated using Equation 5.1.

𝐶𝑎𝑝𝑡𝑢𝑟𝑒 𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦 (𝐶. 𝐸) =𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑜𝑙𝑖𝑑 𝑚𝑎𝑡𝑒𝑟𝑖𝑎𝑙𝑠 𝑟𝑒𝑡𝑎𝑖𝑛𝑒𝑑

𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑜𝑙𝑖𝑑𝑠 𝑖𝑛𝑠𝑒𝑟𝑡𝑒𝑑 𝑤𝑖𝑡ℎ 𝑖𝑛𝑓𝑙𝑜𝑤∗ 100 (5.1)

Sample Data Collection

Data was collected for a series of different experimental conditions. Data was

collected using two specific diameter criteria. These were sewage solids more than

10mm in diameter and sewage solids less than 10mm in diameter. A sample data

collection processes for sewage solids less than 10mm in diameter is shown below which

helped to improve the cigarette butt capture efficiency in the Comb Separator screening

device.

This data collection exercise was conducted on 20th October 2010 at Swinburne

University Hydraulic Experimental Laboratory. For each data collection experimental run

the following standard reporting process was followed see appendix for further

information.

Aim: To test the cigarette butt capture efficiency of a two-comb arrangement.

Test set-up:

Two overlapped comb screens, the first with wires 25mm centre to centre

The second with wires 10mm centre to centre, combs 20mm centre to centre

Front comb 65mm from crest of weir

Retention screen 140mmm behind the crest of the weir

Concrete block placed on the floor of the model sewage chamber to distribute flow

Crest length reduced to 470mm

Retention screen screwed to the floor

19 cigarette butts wrapped in duct tape with a mean sample diameter of 8.82 mm.


Md Abdul Aziz 93

Test 1

Head on weir 50mm giving a flow of 20L/m/s.

Nappe easily cleared the retention screen, refer Figure 5.13.

Test items:

10 artificial cigarette butts

Test run commenced 11:50 AM, Finished at 12:05 PM.

Capture efficiencies:

Cigarette butts: First pass, 1/10, second pass 0/1. 2 butt in down pipe.

Hence = (8/9) x100 = 89%.

Comments:

It was intended to close the bar spacing of comb two to 9 mm but time constraints

prevented this for that days tests. To offset this, the butts were wrapped in duct tape to

increase their diameters accordingly in order to simulate accurately the closer bar

spacing.

Further discussions about experimental data process can be found from Table 5.2 to

Table 5.7.

5.3 Test Procedure

A series of experimental test runs were completed at Swinburne University

Hydraulic Laboratory. The valve operation on the previous design was effective; hence

no alteration of the valve operation was suggested in the revised design. The blockage

on perforations was the key issue which was updated in the revised design.

5.3.1 Experimental Conditions Used

A series of different sewage solids were tested. Figures 5.8 and 5.9 show the

arrangement of the comb separator in the sewage overflow screening device. How the

sewer solids are capture during and after an experimental run is shown in Figure, 5.10

and Figure 5.11. Figure 5.12 shows inserting of sewer solids in the ‘comb separator’


Md Abdul Aziz 94

whereas Figure 5.13 shows the comb separator in operation with three layers of combs

to capture sewer solids.

Figure 5.8: Vertical position of the comb separator in the device

Figure 5.9: Top view of the position of the Comb Separator


Md Abdul Aziz 95

The device tested the following sewage solids: condoms, tampons, wrap papers,

tissue papers, cigarette butts, cotton buds, cotton balls, bottle tops, cans etc. The main

improvement achieved with this device was on blinding or blocking of perforations.

Figure 5.10: Capture of sewage solids during an experimental run

Figure 5.11: Capture of sewage solids after an experimental run


Md Abdul Aziz 96

Figure 5.12: Mixing of sewer solids on to the Comb Separator device

Figure 5.13: Comb Separator is in operation, nappe clear the retention screen


Md Abdul Aziz 97

The concept of the comb separator worked well against blinding or blocking of the

screening device. After some initial runs, it was found that the device could trap more

than 90% of sewage solids more than 10 mm in diameter. There was no reported

blockage on the comb separator. Figure 5.14 shows sewer solids used in the experiment.

Figure 5.14: Sewage solids used in the test

5.3.2 Optimising Experimental Results

Most of the sewage solids such as condoms, tampons, bottle tops, cans, wrap

papers etc. were trapped relatively easily. However, trapping cigarette butts was not

achieved at a satisfactory level. To improve the trapping efficiency of this relatively small

diameter sewage solid a series of adjustments were made in the experimental set up.

Primarily four key experimental parameters were changed to improve capture efficiency.

The four key parameters were:

Number of comb separator layers

Spacing of comb separator

Inflow volume; and

Weir opening


Md Abdul Aziz 98

A series of six different experimental setups were used in the current investigation

for sewage solids less than 10mm in diameter as reported below:

Table 5.1 Different experimental setups used for sewage solid less than 10 mm diameter

Exp. Set-Up

Max Flow

(L/s/m)

Effective Spacing

(mm)

Spacing of 1st comb

Spacing of 2nd comb

Spacing of 3rd Comb

Layer of combs

Weir opening

(mm)

1 46 3.50 25 25 25 3 970 2 46 1.50 20 20 20 3 470 3 71 1.50 20 20 20 3 970 4 30 4.50 15 15 N/A 2 970 5 67 3.25 15 12.5 N/A 2 470 6 67 4.80 25 10 N/A 2 470

To obtain the optimum comb spacing a trial and error methodology was adopted

where comb spacing was varied from 25 mm centre to centre to 10 mm centre to centre.

The effective spacing is the spacing between two layers of comb as shown in Figure

5.15. Weir openings were adjusted from 460 mm to 970mm across the width of a 1000

mm weir length. Results of the trial and error methods and summaries are in the Tables

5.2 to Table 5.7:

Figure 5.15: Comb Separator in operation


Md Abdul Aziz 99

Experimental Sewage Solids Materials less than 10 mm in diameter Table 5.2 Comb Separator Testing at Experimental Set up 1

Experimental Set-Up

Name of material

Flow (Litre/s)

Spacing of 1st comb

Spacing of 2nd comb

Weir opening

mm

Layer of

combs

Capture efficiency

%

1 Cigarette butts 45 25 25 970 3 60

Cigarette butts 45 25 25 970 3 85



Art Cig butts 45 25 25 970 3 67

Key Learning Capture efficiency increases but not up to desired level, artificial cigarette butts need to soak in water to simulate actual conditions, also used dry cigarette butts

Proposed Revision Reduce comb spacing to improve capture efficiency

Table 5.3 Comb Separator Testing at Experimental Set up 2

Experimental Set-Up

Name of material

Flow (Litre/s)

Spacing of 1st comb

Spacing of 2nd comb

Weir opening

mm

Layer of

combs

Capture efficiency

%




Key Learning Surprisingly, despite using 20mm centre to centre combs, the capture efficiency remains much less than expected.

Proposed Revision

Important to know each comb spacing layer performance on capture efficiency. Further reduce first comb spacing to 12.5mm and second comb spacing to 15mm


Experimental Set-Up

Name of material

Flow (Litre/s)

Spacing of 1st comb

Spacing of 2nd comb

Weir opening

mm

Layer of

combs

Capture efficiency

%





Md Abdul Aziz 100

Key Learning Cigarette butts are very buoyant, soaking for 24 hours provided decent improvement, no blockage in the screener

Proposed Revision

As blockage in the screener further reduces, the comb spacing also reduces. It was suggested to use two comb layers instead of three combs for further testing.


Experimental Set-Up

Name of material

Flow (Litre/s)

Spacing of 1st comb

Spacing of 2nd comb

Weir opening

mm

Layer of

combs

Capture efficiency

%




Key Learning Positive improvement in capture efficiency, no blockage observed

Proposed Revision

Flow decreases lead to high capture efficiency. However, it is important to perform on high flow. Weir opening reduces to increase flow capacity.


Experimental Set-Up

Name of material

Flow (Litre/s)

Spacing of 1st comb

Spacing of 2nd comb

Weir opening

mm

Layer of

combs

Capture efficiency

%

5 Cigarette butts 18 12.5 15 470 2 94

Cigarette butts 25 12.5 15 470 2 53



Key Learning High efficiency which is consistent with previous results except with cotton buds. Cigarette butt capture efficiency increases but remains unacceptably low at high flow.

Proposed Revision

Reduce second comb spacing and increase first comb spacing as velocity increases when solids pass first comb spacing which reduces capture efficiency.


Experimental Set-Up

Name of material

Flow (litre/s)

Spacing of 1st comb

Spacing of 2nd comb

Weir opening

mm

Layer of combs

Capture efficiency

%


Md Abdul Aziz 101



Art Cig butts 29 25 10 470 2 100

Art Cig butts 34 25 10 470 2 100

Key Learning No blockage in the screener observed

Observation As the second comb spacing is much closer than first comb spacing, capture efficiency increases to desired level.

5.4 Discussions of Results

5.4.1 Sewage solids more than 10 mm in diameter

Five different experimental set ups (considering set up 2 and 3 as same) were

used and sewage solids capture efficiency was recorded very high. The experimental

conditions used are shown in Table 5.8.

Table 5.8 Experimental set ups at five different conditions for sewage solids more than 10 mm in diameter

Exp. Set ups

Max Flow

(L/s/m)

Effective Spacing

(mm)

Spacing of 1st comb

Spacing of 2nd comb

Spacing of 3rd comb

Layer of

combs

Weir opening

(mm)

1 53 3.50 25 25 25 3 510

2 46 3.50 25 25 25 3 970

3 46 1.50 20 20 20 3 970

4 50 3.25 12.5 15 N/A 2 470

5 67 4.00 25 10 N/A 2 470


Md Abdul Aziz 102

Analysing the performance of the proposed Comb Separator device, for sewage

solids larger than 10mm in diameter, a high capture efficiency of more than 95% was

obtained. The sewage solids used in the testing included toilet paper, bottle tops, cans,

dish wipes, tampons, condoms, cotton balls. Blinding on the combs was negligible during

the tests. The results were consistently capture sewer solids over 95% in all different

experimental set-ups (Figure 5.16), further experiments only considered smaller sewage

solids less than 10 mm in diameter.

Figure 5.16: Capture efficiency of sewer solids at different experimental set ups

5.4.2 Sewage solids less than 10 mm in diameter

5.4.2.1 Increasing the effective comb spacing

Blinding on the screener was a key limitation in designing the sewage overflow

screening devices in most reported literature studies [131]. The comb separator spacing

was reduced to make sure minimal blinding occurred on the combs. As the effective

comb spacing was reduced to less than 2mm, minor blinding occurred on the combs.

Minimal blinding was observed when effective screening was more than 3.5mm. It was

found that the flow had actually taken all the sewage solids either into or outside the

capture chamber without anything sticking against the comb separator bars. There were

90

91

92

93

94

95

96

97

98

99

100

Set up 1 Set up 2 Set up 3 Set up 4 Set up 5 Set up 6

Aver

age

Cap

ture

Effi

cien

cy

Experimental Set ups (Sewer Solids more than 10mm diameter)


Md Abdul Aziz 103

no blockages at the end of most experimental runs, to get optimise consider more trials

were given till blockage were found by reducing comb spacing. Two layers of combs

were found to be more efficient than three layers.

Two different experimental set ups satisfied the design criteria of minimal

blockage on the comb and more than 80% capture efficiency. To confirm robustness of

the experimental set ups, the same experiment was repeated on high flow to validate

minimum variations of the experimental results. Setup 4, which performed well on low

flows, could not perform well on high flow. Setup 6 produced consistent results in

robustness checking within reasonable variations of 2% to 5% in total capture efficiency.

The optimum design criteria was achieved for the device set-up which is an effective

comb spacing of 4.8mm where the first comb spacing was 25mm, whereas the second

comb spacing was 10mm. The optimum design set-up is shown as experimental Setup

6 in Table 5.9.

Table 5.9 Capture efficiency with different experimental set-ups

Experimental Sewage Solids Materials less than 10 mm in diameter

Exp.

Setup

Max

Flow

(L/s/m)

Effective

Spacing

(mm)

Spacing

of 1st

comb

Spacing

of 2nd

comb

Spacing

of 3rd

Comb

Layer

of

combs

Weir

opening

(mm)

Average

Capture

Efficiency

%

1 46 3.50 25 25 25 3 970 68.4

2 46 1.50 20 20 20 3 470 66.5

3 71 1.50 20 20 20 3 970 71.0

4 30 4.50 15 15 N/A 2 970 83.7

5 67 3.25 15 12.5 N/A 2 470 68.7

6 67 4.80 25 10 N/A 2 470 90.7


Md Abdul Aziz 104

As the effective spacing gap reduced to 2mm or less there was minor blinding

occurs on the ‘Comb Separator’ which reduces the capture efficiency of sewer solids

[12]. As the effective spacing increases the capture efficiency also increases. It is

important to note that flow volume also plays an important role in explain capture

efficiency. Figure 5.17 compiling average capture efficiency (%) and average flow (l/s) in

Y axis to give a better understanding about the performance of the effective comb

spacing (mm) in X axis.

Figure 5.17: Effective comb spacing (mm) against average capture efficiency (%) and

flow (l/s per metre length of weir)

5.4.3 Performance comparison of Comb Separator and Hydro JetTM

The Hydro JetTM screen had been subjected to a series of development, testing

and evaluation programs [15]; [17]. The screen used in the Hydro JetTM was a static one

with a specific focus on capturing sewage solids 6mm in diameter. The performance of

the Hydro JetTM screen focussed around screening capacity and effectiveness. The

proposed Comb Separator device used comb spacing which had minimal impact on


Md Abdul Aziz 105

screening capacity during most of the experimental runs with different sewage solids.

The comparative performances of both the devices are tabulated in Table 5.10

In comparison to the Hydro JetTM, the Comb Separator performed with higher

efficiency and negligible blockage on the combs. In the current study, the experiments

were conducted on low flows (up to 70 l/s/m). Further improvement in the experimental

set up to generate high flows (90 l/s to 120 l/s) could provide better insight into this

screen’s suitability for managing higher flows [12].

Table 5.10 Comparative performance of the Hydro-JetTM and the Comb Separator

Hydro-JetTM Screen Mesh Aperture Size (mm)

Static 6 4 2

Average total Efficiency (%) 51 67 69

Number of Observations 17 20 12

Flow Range (l/s) 17 - 60 18 - 60 17 - 45

Average Flow (l/s) 45 43 33

Comb Separator Effective Comb Spacing (mm)

Effective Gap between combs 4.8 4.5 1.5

Average total Efficiency (%) 90.7 83.7 66.5

Number of Observations 6 14 6

Flow Range (l/s/m) 20 - 67 15 -30 30-70

Average Flow (l/s/m) 48 22 47

5.5 Limitations of the Experiment

The experimental facility can generate only up to 70 l/s flow hence the facility

would not be able to test on higher flows, for example 120 l/s. There are always some

physical limitations in the experimental set up. In this research the closest spacing

between combs that could be drilled was 10mm, therefore, the test could not be

conducted for spaces smaller than 10mm. It was recommend that further analysis and

sensitivity testing of the input parameters needs to be performed using standard


Md Abdul Aziz 106

sensitivity testing. Chapter 6 will discuss in detail an ANN application in the current

research to overcome physical limitations of experiments, data generation, training and

testing of ANN model. Chapter 7 will discuss in detail the sensitivity analysis of the input

parameters of the proposed Comb Separator.

5.6 Summary

A new sewage overflow device with improved capture efficiency, low

maintenance, and a self-cleaning mechanism was tested at the hydraulic laboratory of

Swinburne University of Technology. The proposed device overcame most of the

common limitations in existing screening systems such as blinding (minimal blockage),

high maintenance requirements and the installation of an electrical-mechanical switching

system [154]. A series of trials with different numbers of combs, the spacing of combs,

flow volume and weir opening were tested. Some of the key findings from the

experiments are summarised below:

The proposed device can capture larger sewage solids of more than 10mm

diameter with greater than 95% capture efficiency

Capture efficiency is dependent on selected experimental parameters such as

flow, weir opening, comb spacing and layers which all vary sewage solids capture

efficiency. Two layers of combs were found to be more efficient than three layers

Increasing comb spacing improves capture efficiency. Robustness of the

optimum set up was tested to generate consistent results

Comparisons with the Hydro-JetTM suggest that the Comb Separator shows

minimal blockage and higher capture efficiency on low flows

The hydraulic experiments suggested good application potential of the proposed

device in urban sewerage systems. Further experiments with an improved pumping

device to generate higher flow (up to 120 l/s) could provide better insight especially with

smaller solids with a diameter less than 10 mm.


conference papers:


Md Abdul Aziz 107

Aziz, M. A., Imteaz, M., Rasel, H.M., & Phillips D., 2015a, Development and

Performance Testing of ‘Comb Separator’, A Novel Sewer Overflow Screening Device.

Accepted with minor review International Journal of Environment and Waste

Management, Vol 15, No 3, 2015.

Aziz, M. A., Imteaz, M., Rasel, H.M., Phillips, D., 2015c, ‘Performance Testing

of ‘Comb Seperator’ –A Novel Sewerage Overflow Screening Device’, ASEAN-

Australian Engineering Congress on Innovative Technologies for Sustainable

Development and Renewable Energy 11-13 March

Chapter 6: Developing an Artificial Neural Network Model

Md Abdul Aziz 108

Chapter 6

ANN Model to

Complement CFD

and Laboratory

Testing


Md Abdul Aziz 109

6.1 Introduction

It is too complex to model different density sewer solids using the CFD modelling.

The laboratory experiments also has its limitations due to limited number of experimental

set ups are physically possible to conduct the experiments. To overcome these

limitations it was important to adopt an approach which can simulate the complex input

output relationship without knowing the underline physical characteristics. An ANN

model was adopted to perform this task. ANN models were suitable considering mixing

of different solids in the sewer system which leads to different density fluid flow

conditions. Moreover, the nonlinear relationship between input and output variables, and

complex physio-chemical interaction leads to difficulties in formulating mathematical

model based on physical laws.

The experimental work discussed in Chapter 5, the physical experiments allow

for a certain number of trials for experimental set ups. To visualize a range of different

conditions with and outside physical limitations of an experiment, it was important to do

modelling analysis. Moreover, experimental work involves significant cost and time. To

overcome these problems, current research set up a computational model using

experimental results. The initial focus was on the Computational Fluid Dynamics model

(CFD), which also provided reasonable results in developing and optimising the earlier

proposed gross pollutant device as discussed in Chapter 4. However, the studied

problem had some unique challenges which are difficult to model using physical law

based models such as CFD. These include:

Physical characteristics of different sewage solids particles

Multi-fluid sewerage systems with changing viscosity of fluid

Interaction between liquid and solid particles.

Considering these limitations, it was important to adopt an alternative method for

analysis. There are alternative heuristic approaches such as multivariate regression,

fuzzy logic and the artificial neural networks (ANNs) model which could provide

satisfactory results. Adamowski and Karapataki, [2] performed a comparative

investigation between ANN and multivariate regression forecasting of peak urban water


Md Abdul Aziz 110

demand. They found that the ANN model approximations are better for peak weekly

water demand compared with multiple linear regression models. Moreover, ANNs have

the following model attributes:

ANNs can approximate the complex relationship between the input and output

parameters without knowing their physical characteristics

At times noise cannot be avoided in the simulations, even so the ANN can

produce good results

The ANN follows an adaptive approach to solutions over time to compensate

for changing circumstances

Once the model is trained it is easy to use

ANNs have already been used successfully in similar kinds of environmental

research like water level predictions, flood forecasting and control in combined sewers

as presented by [47]; [31] and [173]. Willems et al. [174] demonstrated a number of key

challenges in the use of an ANN model for a sewerage system. Chiang et al. [47]

demonstrated that ANNs have the capability to effectively extract significant features and

trends from complex systems even if the underlying physics is either unknown or difficult

to recognize. Moreover, ANNs greatly reduce the computational time and cost, unless

completely new sets of experimental conditions are used [135].

In the current research there are sixteen input parameters and one output

parameter. Moreover, a few input parameters can change their physio-chemical

behaviour once mixed with sewage output. Considering all these complexities; the ANN

model was the obvious choice in analysing the current Comb Separator sewage overflow

screening device. In this research of sewage solid capture efficiency, neural network

modelling is capable to predict any nonlinear input-output relationship. A multi-layer feed

forward artificial neural network, using a back propagation algorithm was used. Such

networks have been widely used for several environmental problems and modelling

[109]. A series of 47 sets of experimental data were collected to develop the ANN model.

A separate validation set of 8 experimental data were collected to validate the trained

ANN model.


Md Abdul Aziz 111

6.2 Artificial Neural Network

An ANN was designed to simulate the functionality of a human brain where

millions of neurons are connected to each other. The reasoning used in ANN is more like

human brain which than learns the attitude and stores the information based on a

representative dataset. The power of ANN is the ability to process that has a massively

parallel distributed information processing system. This activity is same as human brain

has millions of neurons and trillions of interconnection between neurons so the ANN

resembling biological neural networks of the human brain [79]. It is inspired by the

structural, functional and computational aspect of a biological neural network as shown

in Figure 6.1.

Figure 6.1: Conceptual diagram showing an analogy of the work principal between the

human brain and the ANN model (Source: [149])

The development of ANN model is based on the following rules:

The information processing units are known as neurons also known as nodes,

units or cells

Neurons are interconnected using signals known as connection links


Md Abdul Aziz 112

Connection strength is represented with associated weight between two neurons

All neurons are applied with an activation function to determine its output signal.

The inter-connection between the two neurons is represented by the term weight [122].

A basic concept in evaluating the parameter relationship can be derived by these weights

as shown in 6.2.

Figure 6.2: Conceptual diagram of input-output and weight adjustment (Source: [40])

Proper optimisation of the weight matrix is the key aspect to forming relationships

between neurons. One of the important aspects of optimization involves modification of

the synaptic weights to generate minimal error between actual outputs and predicted

from the model. This modification process is known as a learning algorithm. The most

potent algorithm, used in this work and which is well accepted, is the back propagation

neural network [56]. Werbos [172] developed the back propagation algorithm in his PhD

dissertation at Harvard University. However, work from [137] made the algorithm popular

as he demonstrating how to train the hidden neuron for a complex mapping problem.

The parallel computational process and the ability of the network to learn and generalize

a process, makes ANNs highly popular in solving problems that are currently intractable.

In addition, an ANN provides the following benefits and capabilities [79].


Md Abdul Aziz 113

Nonlinearity: An ANN is a non-linear computational tool. This feature is useful in

modelling non-linear problems that are challenging to model by existing mathematical

techniques.

Input-output mapping: From a given dataset containing input and output samples;

the network demonstrates the ability to learn examples to do input-output mapping. This

is achieved by weight optimization using a learning paradigm during the network training.

Adaptation process: An ANN can adopt its weights with changes in the

surrounding environment. With small changes in the operating condition the ANN can be

easily retrained. This feature is particularly useful to change weights in real time for

implementation of an on line control system using the ANN.

Evidential response: An ANN can be designed to generate information on both

the choice and reason behind the selection of a particular pattern for a pattern

classification problem. This helps in improving the classification performance through the

rejection of ambiguous patterns.

Contextual information: The parallel distributed structure allows every neuron in

the network to hold some knowledge of the problem and be influenced by the global

activity of the other neurons in the model.

Fault tolerance: An ANN is integrally fault tolerant due to the presence of massive

parallelism. In case of a fault, instead of a catastrophic failure, the network’s performance

reduces [28]. Two other key functions of ANNs are categories of recognition and function

approximation [138]. One of the previous seen inputs are trained in the recognition

category. For the function approximation, network need to approximate unseen inputs

from training to model complex input output relationships.

Neural networks have a wide variety of applications and have been implemented

in many practical applications, it is more powerful when the physical law based

mathematical techniques are difficult to implement or it become too complex to develop

input and output relationship between parameters. ANNs have been successfully used


Md Abdul Aziz 114

for solving a wide range of practical problems in hydrology and water quality such as

forecasting the rainfall [45]; [46], operation of a reservoir [37]; [43], forecasting flow in the

stream [6]; [44], water quality forecasting in rivers from predicted pesticide leaching

through turf grass-covered soil [4]. Starrett et al.[155]; Sandhu and Finch [140]

investigated the flow condition and gate positions in the Sacramento San Joaquin Delta

salinity levels between the interior and along the boundary of the delta using ANN.

The current research concentrates on a function approximation network, where

the term, generalization, indicates the capability of the network to acquire the complex

input output relationships and predict capture efficiency of sewage solids. Generalization

is the key strength of the ANN that makes it desirable from other methods of

approximation by having the network trained to be responsive to an unseen environment.

However, it is important to make sure the trained ANN model is protected from over

fitting, which is a noticeable problem which could led to poor generalization for a function

approximation. In these cases, the output generalization fails to respond well with an

unseen dataset. Instead of learning the data the network actually memorizes the

samples which led to cause this issue.

One common aspect of over fitting is when we use few dataset for training in

comparison to the total number of network parameters. In the current research we have

used 60% of the dataset for training. On the flip side if the dataset is too big it creates

more complex functions. Thus, it is important to use adequate data to improve the

generalization ability of the model; the dataset should not be too small not too big [40].

An example of a typical over fitting problem is shown in Figure 6.3. Here the blue boxes

represent the data of a noisy sine function, whereas the red dashed line represents the

response of a trained ANN. The result indicates that the network has over-fitted the data

and, thus, the network would not generalize well in an unknown environment. The

network actually memorizes each data point instead of trying to map the input-output

relationship. The trained network, without over-fitting, should be able to ignore the noise

and learn the underlying function, which is the sine function for this case. The black line

represents the ideal output of such a type of network without over-fitting.

Several authors [18]; [112]; [177] have tested how to improve the generalization

ability of a trained ANN by injecting noise into the inputs [38]. Karystinos and Pados [99]


Md Abdul Aziz 115

tried by expanding training samples which were generated randomly in agreement with

probability distribution function (PDF) generated by the Parzen-Rosenblatt estimate [98].

The intent of this procedure was to overcome the problem of over-fitting and to improve

the generalization ability of the ANN. The generalization performance of the trained

network was evaluated from the error generated by the network on the data that are

training dataset which is known as the generalization error. Cross validation [101]; [104]

and early stopping [134]; [175] are other statistical techniques to overcome the problem

of over fitting; which helps to reduce the generalization error that led to better

performance of the ANN. The database is divided, into training, testing and validation

datasets. In the current problem we have used 60% of the data for training, 20% for

testing (which is similar to calibration of hydrologic model) and 20% for validation (also

known as cross training which is similar to model validation in hydrologic models).

Figure 6.3: Demonstration of over fitting for a function approximating Artificial

Neural Network (Source: [40]).


Md Abdul Aziz 116

During the training process the connection weights of an ANN are consistently

and continuously stimulate by the environment where the network is embedded. The

training goal of the model was to minimize the error function by continuously update the

connection strengths and threshold values to make the output close to the targets [5].

The training set all the time trying to minimise the error gradient and update the network’s

weights and biases. Cross training (validation) dataset is recommended to prevent

overfitting. The network’s error on the validation set is calculated and monitored during

the training process. This set is not used to update the network’s weights and biases.

The validation error generally decreases when the network’s training starts. A rise in the

validation error for a certain number of iterations (also described as epochs) indicates

over fitting of the network. When the network realise the overfitting happening it led to

stop the training and the weights and bias values are saved at the minimum validation

error. It is important to test with a separate test set to check the performance of the

trained network through calculation of the generation error. Few different independent

data splits are performed, followed by extensive training to get good statistical results.

6.3 Description of Network Structure

6.3.1 Artificial Neuron Model

An artificial neuron is the fundamental non linear information processing unit.

Three key components of a neuron are given below:

Weights: Weights represent the strength or value assigned to each of the

connecting links or synapses. An input signal px to the neuron k is multiplied by the

synaptic weight kpw . The synaptic weight kpw defines the strength of the connection

between the input px and the neuronk .

Adder: A linear adder was used to gather the weighted input signal.

Activation function: The summed output of the weighted input signal kv is limited

to some finite value within a permissible amplitude range of the output signal set for the


Md Abdul Aziz 117

model. This is done by the activation function . The function defines the neuron

output ky relative to the activity level at the function’s input kv .

Figure 6.4: A Non-linear model of an artificial neuron k (Source: [40])

6.3.2 Multi-Layer Feed Forward Neural Network Structure

The structure is made up of three main sections: the input layer, the hidden layers

and the output layer. Current research consider multilayer feed forward neural network

structure as it is most common network structure to use in environmental modeling, a

detail description of the model can be found in the work of Churchland & Sejnowski [48].

6.4 Network Learning

How the ANN optimises the weights within the network parameters is referred to

as a paradigm. The neural network operates using a learning paradigm which refers to

a model of the environment [162]. There are three classes of learning paradigms:


Md Abdul Aziz 118

supervised learning, reinforcement learning and self-organized or unsupervised learning.

Further details about these learning paradigms can be found in the work of [48]; [26].

6.4.1 Back propagation algorithm

Back propagation algorithm is the most commonly used neural network algorithm

used in the environmental model. In the current research a model considering multi-layer

perceptron, based on back propagation algorithm was used [9]. A detail description of

the algorithm can be found in the work of [56]; [40]; [5] & [7].

6.4.2 Levenberg-Marquardt Algorithm

The Levenberg-Marquardt (LM) algorithm use Newton’s method to approximate

and is selected to reach the second order training speed without calculating the Hessian

matrix. The Hessian matrix (H) approximation and error gradient (g) is calculated as per

Equation 6.1 and Equation 6.2.

TH J J (6.1)

Tg J e (6.2)

J denotes the Jacobian matrix designed with the first derivatives of the network

errors, e, on the training set so that the network’s weights and biases and can be

considered using the typical back propagation method [76]. JT is the rearrange of the

Jacobian matrix, J.

The LM algorithm uses the estimated calculation of the Hessian matrix to adjust

and alter the parameters. If zk signifies the old parameter value; whereas the new

parameter value after calculation of the network errors is given by zk+1 (Equation 6.3)

1

1T T

k kz z J J I J e

(6.3)


Md Abdul Aziz 119

The parameter µ is set to a precise value at the beginning of the training. After

each epoch, the performance function is calculated. When the performance function is

less than the previous epoch, the value of µ is decreased by an exact value. On the other

hand, if the performance function rises, the value of µ is also increased by an exact value.

The value of µ equal to zero turns Equation (6.3) into Newton’s method. The aim is to

return to Newton’s method quickly since it is faster and more accurate near when the

error is less.

A highest value of µ is set beforehand the training. If µ touches its highest value,

the training ends. This activity suggests the network has failed to get a converging

solution. The training is also ends when the error gradient (Equation 6.2) drops below a

precise set value or when the goal set for the performance function is met.

The network training steps using the LM algorithm are as follows:

1. All the inputs to the network are presented. The corresponding network

outputs, errors and the sum of square errors over all inputs are computed.

2. The Jacobian matrix, J, is computed.

3. Equation 6.3 is computed to obtain the new parameter values.

4. The sum of squares of errors is recomputed with the updated parameter

values.

5. If the new sum of squares is smaller than the previous value, μ is reduced by

a specific factor β and the process is re-started from step 1.

6. If the new sum of squares is increased, the value of μ is increased by α and

the process is re-started from step 3.

The network is assumed to have converged when the error gradient is less than

some predetermined value or when the sum of squares has been reduced to some

specific error goal. Another popular algorithm used in the environmental model is known

as Resilient back propagation algorithm.

6.4.3 Resilient Back Propagation Algorithm

The Resilient Back Propagation (RBP) algorithm eliminates the detrimental

effects of the levels of the partial derivatives. The sign of the derivative governs the

direction of weight update. The size of the weight change is determined by a discrete


Md Abdul Aziz 120

parameter update. The update parameter value for each weight and bias is enlarged by

a specific factor. This increase occurs if the derivative of the performance function, with

respect to that of weight, has the identical sign for two consecutive iterations. On the

other hand the update value is reduced by a factor if the derivative, with respect to that

of weight, changes sign from the previous iteration. When the derivative is zero, the

update value remains the same. As the weights fluctuate, the change in the weight is

reduced. When the weight continues to change in the same direction for little iteration,

the levels of the weight change rises. A detailed study of the algorithm is provided in

[136].

6.5 Data Collection and Pre-processing

6.5.1 Creation of Database

The database for ANN analysis was created from the hydraulic experimental data

of the Comb Separator, a sewage solids overflow screening device. In total there were

sixteen input parameters which had an impact on the output sewage solids capture

efficiency. The input parameters were considered with all the logically possible input

conditions or materials that could have an impact on the output capture efficiency. Out

of these sixteen parameters there were nine parameters related to the experimental

setup and seven parameters related to the experimental materials used. The input

parameters considered in the device setup were:

1. Number of comb layers

2. Spacing of the first comb layer

3. Spacing of the second comb layer

4. Centre to centre spacing of combs

5. Position of comb from crest of weir

6. Length of the crest over the weir

7. Height of water over the weir head

8. Flow (measured rate per meter length of the weir)

9. Position of the retaining screen

The input parameters considered in the experimental sewage solids testing were:


Md Abdul Aziz 121

1. Cigarette butts (actual)

2. Cigarette butts (artificial used for experimental purpose)

3. Cotton buds

4. Condoms

5. Tampons

6. Wrap papers

7. Toilet paper

All this experimental data was collected and normalised using feature scaling

method also known as unity based normalization that bring all values into the range [0,1]

before they were ready for analysis using the ANN method. MATLAB [111] software

package was used for data analysis and reporting.

6.5.2 Network Architecture

6.5.2.1 Selection of Input Output Parameters

To ensure a good model approximation the number of input-output data pairs

used for training should be equal to or greater than the number of parameters (weights)

in the network [35]. Like most environmental modelling approaches, some of the key

steps in ANN modelling include data collection (from experimental results), pre-

processing of the data (normalizing experimental data) and assessment of the output. A

robust and sufficiently large database is essential to construct a network which

generalizes well. Moreover, a clear understanding of the hydraulic process is required

for successful modelling of this nature. For instance, physical insight into the problem

being studied can lead to better choices of input variables for proper mapping [7]. This

will lead to effective and efficient modelling, avoiding loss of information due to

inappropriate choice of input parameters. In the present study 40 sets of experimental

data were collected.

Eight different types of experimental conditions were tested by changing the

volume of flow, the number and spacing of combs, weir opening and different sewage

solids materials to improve the capture efficiency of sewage solids. Based on these

experimental results an input-output relationship was assumed, which showed that 16


Md Abdul Aziz 122

input parameters influence the pollutants capture efficiency [8] & [9]. While conducting

the experiment, one of the experimental conditions (inflow volume, number of combs,

spacing of combs etc.) was varied, while the others were kept constant to their reference

values. This allows examining how different experimental conditions could affect the

output pollutant capture efficiency, referring to Figure 6.5.

Figure 6.5: Block diagram of proposed ANN model.

6.5.2.2 Parameter Estimation and Network Optimisation

Optimal network architecture should retain a simple and compact structure while

providing best performance in terms of error minimization. The network should be neither

too small; as it will have insufficient degrees of freedom to capture the underlying

relationship in the data, nor too large; as it may fail to generalise, memorising fluctuations

in the training data that are not representative of the system being modelled. A model

considering multi-layer perceptron (MLP), based on the back propagation algorithm, is

used in this work. The multi-layer ANN architecture comprises three main parts: the input

layer, the output layer and the layer in-between termed the hidden layer. The number of


Md Abdul Aziz 123

neurons required to describe each parameter is dependent on the nature of the

parameter. A real valued parameter requires one neuron to represent the value, while x

neurons are required to describe 2x categories for parameters representing

classifications. The flexibility lies in selecting the number of hidden layers and in

assigning the number of neurons to each of these layers. Maren [110] suggested two

hidden layers when the outputs need to be continuous functions of the inputs. Two

hidden layers were thus used in this study and the number of neurons in each layer was

ascertained from the network training and optimization process.

In the back propagation algorithm, all the weights within a network are adjusted

simultaneously. Every neuron learns a feature, defined by the back propagated error

signal, through weight changes. The weight changes in each neuron are independent

from one another. The parameters are constantly redefined until a set error minimum is

reached [42]. The studied network optimization uses the experimental database to fix

the number of neurons in the hidden layers as well as optimising the weight population

to produce a minimum output error (Table 6.1). The training process for ANNs can be

considered to be similar to the idea of calibration which is an integral part of most

hydrologic modelling studies. The purpose of training is to determine the set of

connection weights and nodal thresholds using the back propagation algorithm that

cause the ANN to predict outputs that are sufficiently close to target values [7]. To avoid

the problem of over-fitting and to improve the generalization ability of the trained network,

the method of cross-validation and early stopping were implemented. The available

dataset was divided into three parts: the training set, the validation set and the test set.

For the current study, we used 60% of the data for the training set and 20% for each of

the validation and test sets. A detailed description of the data split technique is given by

Sarle [141].

6.6 Result Analysis and Discussion

In the studied case 40 weight trials were tested for all the nine algorithms. Error

performances for these algorithms are listed in Table 6.1. Based on the error

performance and regression coefficient (R) value the Levenberg Marquardt and the

Resilient Back Propagation algorithms were chosen as suitable for this study. A trial and

error approach had to be taken as the error back propagation required that the number


Md Abdul Aziz 124

of hidden layers needed to be specified prior to network training [42]. The number of

neurons in the first and second hidden layer was varied from a combination of 5/4

neurons to 23/22 neurons to identify the optimum hidden layer for the problem, refer to

Figure 6.6. Standard regression analysis was performed between the predicted and

experimental pollutant capture efficiency values. From the performed regression

analysis, it was found that the ANN structure with 5/4 neurons in the first and second

hidden layer respectively.

Figure 6.6: Comparison of different node in the 1st and 2nd hidden layer

Nine different standard algorithms were tested to ensure that the Levenberg

Marquardt algorithm and the Resilient Back Propagation algorithm were the optimal

training algorithm for the problem of modelling sewage solid capture efficiency. Forty (40)

different network weight trials were given in each step, where different random initial

weights were used in each trial and the best values for the regression co-efficient (R)

were collected. The performance of different algorithms was judged with respect to the

root mean square error using the following equation.

R2= Sum of Squared Errors / Total Sum of Squares (6.4)

The proposed model responded well to both the Levenberg Marquardt and the

Resilient Back Propagation algorithm with regression values close to unity, refer to Table

0

0.2

0.4

0.6

0.8

1

5.4. 8.7. 11.10. 14.13. 17.16. 20.19. 23.22.

Reg

ress

ion

Valu

e 'R

'

Combination of Neuron in 1st and 2nd Hidden Layer

Levenberg-Marquardt Paradigm Resilient Backpropagation Paradigm


Md Abdul Aziz 125

6.1. The choosing of the paradigms are not limited to the regression value as minimum

and maximum error terms also consider in considering the best paradigms as shown

below, Table 6.1. It is important to note Levenberg-Marquadt and Resilient

Backpropagation algorithm showing less variation between minimum and maximum

error.

Table 6.1 Comparison of different training paradigms

Order Name of Paradigms Tested

Minimum Error

Maximum Error

Regression Value 'R'

A Levenberg-Marquardt 0.08 0.21 0.87 B BFGS Quasi Newton 0.06 0.76 0.61 C Resilient Backpropagation 0.08 0.21 0.87 D Scaled Conjugate 0.11 0.71 0.66 E Conjugate Gradient 0.10 0.40 0.70 F Fletcher-Powell 0.10 0.40 0.70 G Polak-Ribiere 0.15 0.16 0.72 H One Step 0.07 0.65 0.72 I Variable Learning 0.08 0.52 0.63

The regression analysis was performed on the training, validation and test sets

for the ANN with five and four neurons in the first and second hidden layer respectively;

and then trained with the Resilient Back Propagation algorithm. Both the Levenberg-

Marquardt and the Resilient Back Propagation algorithms revealed a promising

regression co-efficient (R) value of 0.862. Comparison of the trained network and

experimental capture efficiency showed good agreement between model results and

experimental data refer to Figure 6.7.

Figure 6.7: Comparison of experimental and ANN predicted capture efficiency.


Md Abdul Aziz 126

Figure 6.8: Regression Value for Training, Validation and Test results

The regression analysis performed on the training, validation and test sets for

the ANN with five and four neurons in the first and second hidden layer respectively

and trained with Resilient Back propagation algorithm is shown in Figure 6.8

6.7 ANN Model Validations

A separate set of validation data was used to judge the overall performance of

the trained network. In the studied problem eight new sets of experimental data were

collected and validated against the trained network to obtain the predicted values. The

generalization performance of a trained network is measured on the error it produces


Md Abdul Aziz 127

with the new dataset. The smaller the error, the better the generalization ability of the

trained network is. It was found that the trained ANN successfully predicted the new

experimental results with an average absolute percentage error of about 7%, which

reveals that the network was trained properly with good generalization ability, refer to

Figure 6.9.

Figure 6.9: Comparison of experimental and model results for validation dataset.

Experiment numbers 1 to 4 were conducted on a specific experimental setup and then

the flow was changed for experiments 5 to 8. Solids used for different test numbers: 1

and 5 = cigarette butts, 2 and 6 = condoms, 3 and 7 = tampons, and 4 and 8 = wipe

papers.

6.8 Summary

Limited understanding of physics of non-Newtonian fluids made it quite

complicated and time consuming to model the physio-chemical interaction of different

sewage particles with water. In order to assess the trapping efficiency of the developed

device under different experimental conditions, a neural network modelling approach

was proposed. The empirical knowledge gained through a series of experiments helped

the formulation of various assumptions which led to a successful ANN model. ANNs

routinely simulate the non-linearity of the physical process without solving complex

partial differential equations. Unlike any other form of mathematical or regression based


Md Abdul Aziz 128

modelling there is no need to make assumptions about the mathematical form or the

relationship between input and output parameters.

The ANN’s powerful modelling approach, when trained with input-output data (a

separate set of test data), shows that the model can mimic the underlying complex

physio-hydrodynamic processes (involving different types of materials) that otherwise

would be extremely difficult to model. Such a model also overcomes some common

experimental drawbacks such as different scales and structures and provides a

significant reduction in the time and cost involved in the experimental processes [135].

All of these attributes, along with the nonlinear nature of the activation function,

truly enhance the generalization capabilities of ANNs in the studied problem. Special

attention was given to the generalization of errors during test cases with different

algorithms which significantly contributed to the ANNs performance in predicting

experimental pollutant capture efficiency as shown in Figure 6.9. Separate validation

datasets were used to judge the robustness of the trained network. Promising results

were observed with more than 90% accuracy in eight different experimental results with

an average absolute error of about 7%. This demonstrates the ability of the model to

predict sewage solid capture efficiencies of the device in real-world conditions.


conference papers:

Aziz, M. A., Imteaz, M., Choudhury, T. A., & Phillips, D., 2013a, ‘Applicability

of artificial neural network in hydraulic experiments using a new sewer overflow

screening device’, Australian Journal of Water Resources, vol.17, no.1, pp.77-86.

Aziz, M. A., Imteaz, M., Choudhury, T. A. & Phillips, D. I. 2011, ‘Artificial Neural

Networks for the prediction of the trapping efficiency of a new sewer overflow screening

device’, 19th International Congress on Modelling and Simulation, Perth, Australia.

Indexed in Scopus

Chapter 7: Sensitivity Analysis of the Comb Separator

Md Abdul Aziz 129

Chapter 7

Sensitivity Analysis

of the Comb

Separator


Md Abdul Aziz 130

7.1 Introduction

Analysing the parameter sensitivity of a hydraulic device such as a Comb

Separator has been a standard practice by hydraulic engineers for many years [97];

[116]; [127]; [176]. Such analysis qualitatively or quantitatively explains different sources

of variation [139]. Extended analysis from basic sensitivity analysis can be found in the

works of Hall [70] and Hall and Solomatine [71]. A comprehensive review regarding

application of sensitivity analysis in environmental models is presented by Hamby [73].

Sensitivity analysis of the input parameters, of the Comb Separator device provides a

better understanding of those input parameters, This understanding includes their

influence on the outcome capture efficiency, identification of which parameter is the most

important, the relative importance of each input parameter and identification of which

parameter requires further research.

Our proposed Comb Separator, a sewer overflow screening device, consists of a

rectangular tank and a sharp crested weir. In front of the weir are a series of vertical,

parallel combs to separate entrained sewer solids from the overflow. The studied device

was tested with a series of sewer solid materials including condoms, tampons, cigarette

butts, cotton buds, bottle caps, wrap papers, etc.

A detailed discussion of the experimental work is provided in Chapter 5. Larger

sewer particles (greater than 10mm in diameter) can be captured relatively easily with a

capture efficiency of more than 90%. This capture efficiency was tested in different input

conditions, such as flow, effective spacing, weir opening, comb layers and run time. The

output of this testing is the sewer overflow capture efficiency.

Comparison with the industry standard Hydro JetTM screen shows the capture

efficiency of ‘Comb separator’ performs better on low flow. However there was a variation

of output capture efficiency. Hence the focus of this chapter is to understand parameter

sensitivity on the capture of smaller sewer solid particles.


Md Abdul Aziz 131

7.1.1 Objective

The key objectives of the sensitivity analyses of the Comb Separator device are

listed as follows:

To develop a robust understanding of the meaningful input parameters

To undertake a performance comparison of the proposed Comb Separator with

a standard Hydro JetTM screen under low flow (up to 60 l/s) conditions

To comprehend the impact of experimental design parameters (runtime, flow

discharge, effective comb spacing, weir opening and comb layers) on sewer

solids capture efficiency

To understand the relative significance of the input parameters and to identify

which parameter is the most influential in development of output results.

7.2 Background

A total of 42 sets of experimental data were collected on six different sets of

experimental setups for Comb Separator. Based on the experimental experience, five

input parameters (runtime, flow volume, effective comb spacing, weir opening and layers

of combs) were identified as being influential on output sewer capture efficiency. A

sensitivity analysis of such a device is necessary to ensure optimisation and validation

that also serve as a guide to future improvement opportunities for other proposed

experimental devices. The sensitivity analyses of the experimental data can help to

understand the following:

Which input parameter can be neglected and removed from the model

Which input parameter requires additional research for strengthening knowledge

and understanding to reduce output uncertainty

Which input parameter contributes the most to the output variability

Parameters correlation with output capture efficiency

Once the device is in practical use, what would be the best approach to manage

effectively and efficiently the device performance


Md Abdul Aziz 132

The experimental work was restricted by the physical limitations of conducting

experiments with limited trial combinations of the experimental conditions. This is

inherent in any laboratory study. In addition, the device performance was tested against

industry standard Hydro JetTM screen against industry standard device also identified the

importance of input parameters sensitivity testing. Both ‘Comb Separator’ and Hydro

JetTM were developed to improve screening efficiency of sewer overflow screening

device. The Hydro JetTM screen had subjected to a series of development, testing and

evaluation program [15], [17]. The screen used in Hydro JetTM was a static screening with

specific focus to capture sewer solids 6mm in dimensions. Performance of Hydro JetTM

screening focussed around screening capacity and effectiveness. As the proposed

‘Comb Separator’ device used comb spacing which had minimal impact on screening

capacity during most of the experimental runs with different sewer solids.

The Comb Separator can produce better screening efficiency in capturing sewer

solids with low flows. There was hardly any blinding effect on the Comb Separator which

is a key improvement from the previous static screening concept. However, the capture

efficiency for Comb Separator varies more than the Hydro JetTM screen. It is important

to understand the performance of the input parameters influencing the capture efficiency.

To investigate this issue it was important to perform parameter sensitivity testing.

Sensitivity of these parameters is of paramount importance in considering the ability of

this device to function property in remote, unstaffed locations. Understanding and

analysing model sensitivity and uncertainty has been an active theme of research for

hydraulic engineers for many years [72]. Sensitivity analysis is predominantly used in

design for hydraulic experimental parameters.

In the current investigation a model was developed using Multiple Linear

Regression (MLR) method. As the dataset is small for detail sensitivity analysis, the Latin

Hypercube Sampling (LHS) technique was used to expand the data series without

compromising the input-out relationship. Methodology used for this purpose is

schematised in Figure 7.1.


Md Abdul Aziz 133

7.2 Methodology used in Sensitivity Analysis

The methodology adopted in the sensitivity analysis is shown in the flow chart below:

Figure 7.1: Flow chart of the methodology adopted in the sensitivity analysis.

Developing regression

Model for input output

Relationship Yes

No

No

Improving Screening Efficiency

Comparison between Comb

Separator and Hydro Jet TM

Selection of input Parameters

Latin H sampling for data generation

Check for assumptions

Remarks on input parameters sensitivity


Md Abdul Aziz 134

Some of the key points in developing the methodology are given below:

Define the research problem regarding the sewer overflow screening device and

why we need to do the sensitivity analysis using a comparative analysis between

the Hydro JetTM and Comb Separator

Develop meaningful and simplified inputs to the model considering the key input

parameters influence on the output capture efficiency

Develop a regression model and check for the necessary assumptions

Use sampling techniques to develop large input data sets. In this case the Latin

Hypercube sampling (LHS) technique which is highly recommended by scientific

literature for parameter sampling will be used. LHS was used to generate 10,000

units of data of the input parameter considering their distribution type taken from

experimental results.

7.2.1 Developing a Multiple Linear Regression (MLR) Model

Multiple linear regressions (MLR) are a statistical method which uses some

explanatory (independent) variables to predict the outcome of a response (dependent)

variable. So MLR is to model the relationship between independent (input or predictor

variables) and dependent variables.

The model for MLR, given total ‘n’ observations, is:

Yi = (b0+b1X1i +b2X2i+ ………….. +bnXni) +ᶓ (7.1)

Y is the dependent variable, b1 is the coefficient of the first input X1, b2 is the

coefficient of the second input X2, bn is the coefficient of the nth input (Xn), and ᶓ is the

difference between the input and the observed value of Y for the ith participant. In our

studied case the model become

(CaptureEfficiency)i=b0+b1(runtime)i+b2(flow)i+b3(effectivespacing)i+b4(weiropening)i+b5

(comb layers)i + ᶓ (7.2)


Md Abdul Aziz 135

This model is based on equation (7.2) and includes a b-value for both inputs and

the constant. If we calculate the b-values, we could make predictions about capture

efficiency based on all five input (predictor) variables. It is important to efficiently design

the model input (predictors) parameters as the values of the regression coefficient

depend upon the variables in the model. Predictor (or input) variables should be selected

based on past research to make sure all the key inputs variables are included in the

model generation.

7.2.1.1 Validity of Model Assumptions

There are a few assumptions that need to be satisfied to use MLR model [25]. They are:

Variable Type: All input parameters must be quantitative or categorical and the

output parameter (outcome variable) must also be quantitative and continuous. For

example, in the studied case the three input parameters, runtime (minutes), flow (l/s/m)

and effective spacing (mm) are all quantitative predictor variables. The outcome variable

is the capture efficiency (%) which is also a quantitative variable. Hence the model

satisfies this assumption.

Non-zero variance: The input parameters should vary in value and not have

variances of zero. The experimental data suggest that all the input parameters are non-

zero values so the predictor variables satisfy this criterion.

No perfect multicollinearity: Among the three predictor variables, they should not

have perfect linear relationship two or more of the predictors. For the current study the

predictor variables did not show strong correlation as shown in table 7.3.

Predictors are uncorrelated with ‘external variables’: External variables such as

weir opening and comb layers that were not included in the regression model had little

influence on other predictors or the outcome variable. This assumption means that weir

opening and comb layers should not correlate with runtime, flow and effective spacing

predictors or influence capture efficiency. Both weir opening and comb layers did not


Md Abdul Aziz 136

influence the other predictor variables or outcome variable, hence this assumption was

satisfied.

Homoscedasticity: At each level of the input variables the variable of the residual

terms must to be constant. At every level the residual predictors must have the same

variance (homoscedasticity). However, if these variances are not unequal the data said

to be heteroscedasticity. During the model run two parameters were plotted. One was

ZRESID or residual along Z and the other was ZPRED which is prediction along Z. The

ZRESID and ZPRED should look like dots points are randomly dispersed around zero

as shown in the graph below in figure 7.2:

Figure 7.2 Regression Standardized predicted value against residual

When there is some sort of curve in this graph that indicates the data has broken

the assumption of linearity. On the other hand when graph funnels out there are

possibilities that the data have heteroscedasticity. The points from experimental data are


Md Abdul Aziz 137

randomly and located evenly distributed across the plot. The pattern satisfies the

assumptions of linearity and homoscedasticity have been met.

Independent Errors: The residual terms should not be correlated between any two

observations which suggest a lack of autocorrelation. This assumption is tested using

the Dubin-Watson test [53 & 54]: a very conservative rule of thumb, values less than 1

or greater than 3 are definitely cause for concern. In the studied case the value was

observed at less than 2 which show a positive correlation but not a cause for concern.

Normally Distributed Errors: The residuals used in the model are random and

normally distributed with a mean of zero. This assumption assumes that most of the time

the observed data are zero or very close to zero with only occasional differences much

greater than zero. The histogram and normal probability plot as shown in figure 7.3 was

assessed to examine this assumption.

Figure 7.3: Normal distribution plot of the Standardized Residual vs Frequency


Md Abdul Aziz 138

The bell shape curve on the histogram shows the shape of the distribution. The

curve follows a satisfactory bell shape which suggests the experimental data for the

Comb Separator satisfies the normality assumption in the data set.

Figure 7.4: Experimental data shows Observed vs Expected Cumulative probability

The line in the plot in figure 7.4 represents a normal distribution where the dot

points are observed residuals. In a perfect normally distributed data set, all points will lie

on the line. The data sets are reasonably close to the straight line so the normal

distribution assumption is reasonably satisfied with this data set.

7.2.1.2 Test of Sample Size and Input Parameters

There are a number of different rules of thumb available about how big a data set

should be to obtain a reasonable representation between input (predictor) and output

(outcome) variables. The simplest rule of thumb would be the bigger the data set, the


Md Abdul Aziz 139

better the model. A sampling data set of 10,000 was chosen on this basis. However, for

any practical experiment there are limitations of resources and time. This is a common

restriction embedded in most experimental research. In the current research comb

spacing was varied in the comb layers and up to six different set ups were possible to

test.

The reason is that the regression coefficient ‘R’ is linked with a number of input

parameters ‘k’ and the number of datasets ‘N’ as in following relationship. In the MLR it

was used five input (predictor) variables than expected ‘R’ (regression coefficient) for

random data is k/(N-1) for example with five predictors with 21 sets of data can appear

to have a strong effect, R =5/(21-1) = 0.25. Obviously ‘R’ is expected to be close to zero

so that random data can have no effect. The two most common are 10 cases of data for

each predictor in the model, or 15 cases of data per predictor. This reduces the value of

randomness within 0.1 to 0 which was a reasonable reduction of the effect of

randomness on the dataset. In the studied case we tried two different models. In the first

model we used 5 input parameters with 42 sets of data. However, the data set showed

a randomness of more than 0.1 and with the elimination of two input parameters it had

only a minor variation to the model output. In the second case, 3 input parameters with

42 sets of data were used with the results showing the data randomness effect within 0

to 0.1. This was reasonable under the model assumption. Hence the data set used in

the model was sufficient to produce a decent result using the MLR model.

7.2.2 Summary of the Model

The first table shows the mean and standard deviation of each variable in our

data set, so we understand that average capture efficiency of sewer solids was

72.95~73%. This table was not necessary to interpret the regression model, but it was a

useful summary of the data. The table 7.2 shows three things, firstly, the value of

Pearson’s correlation coefficient between every pair of variables. For example, capture

efficiency had positive correlations with runtime and effective spacing but has a negative

correlation with flow. This also explains that effective spacing has the most significant

positive correlation with capture efficiency, r =0.538. Secondly, the one-tailed

significance of each correlation was displayed. For example, both effective spacing and


Md Abdul Aziz 140

flow are significant (as p<0.001). Thirdly, the number of experimental data used in the

analysis was 42.

Table 7.1 Provides descriptive statistics of the Comb Separator experimental data set

Descriptive Statistics

Sensitivity Parameters Mean Standard

Deviation No of data (N)

Unit

Capture Efficiency 72.95 14.49 42 %

Run time 16.67 6.90 42 Minutes

Flow 43.38 15.83 42 Litre/Sec

Effective Spacing 3.13 1.27 42 mm

Table 7.2 Correlations of different parameters

Correlations

Capture

Efficiency (%)

Runtime (minutes)

Flow (Litre/sec)

Effective Spacing(mm)

Pearson Correlation

Capture Efficiency (%) 1.00 0.32 -0.50 0.54

Run time (minutes) 0.32 1.00 -0.77 -0.71

Flow (litre/sec) -0.51 -0.77 1.00 -0.24

Effective Spacing (mm) 0.54 -0.71 -0.24 1.00

Sig. (1-tailed)

Capture Efficiency (%) - 0.02 0.00 0.00

Run time (minutes) 0.02 - 0.31 0.33

Flow (litre/sec) 0.00 0.31 - 0.07

Effective Spacing (mm) 0.00 0.33 0.07 -


Md Abdul Aziz 141

The correlation matrix is extremely useful for getting a rough idea of the

relationships between inputs (predictors) and output (outcome), and for a preliminary

look for multicollinearity. There was no multicollinearity in the data set; hence there was

no substantial correlation (r> 0.9) between input parameters. If we look at the input

parameters or predictors the highest correlation was between predictors flow and

effective spacing (r=-0.235, p>0.005). This correlation was not significant and the value

was also small, so little collinearity exits in the predictors. The predictor effective spacing

correlates best with the outcome (r=0.538, p<0.001); so it is likely this predictor will best

predict capture efficiency.

The model summary section is the key section to discuss about the overall model

performance of the MLR model. Column ‘R’ lists the values of the multiple correlation

coefficients between inputs (predictors) and output (outcome). The model suggests that

three input parameters or predictors account for 54.9% of the variation in capture

efficiency.

Table 7.3 Summary for the multiple linear regression model

Model Summary

Model R R Square

Adjusted R Square

Std Error of the Estimate

Durbin- Watson

1 0.741 0.549 0.513 10.11 1.14

Change Statistics

R Square Change

F Change df1 df2 Sig. F Change

0.549 15.419 3 38 0

a. Predictors (input parameters): (Constant), Effective spacing (mm), Runtime (min), Flow (litre/sec)

b. Dependent variable: Capture Efficiency (%)


Md Abdul Aziz 142

The adjusted R2 gives us understanding of the model performance and preferably

its value is very close to R2. In the studied dataset the difference between R square and

adjusted R square for the final model was small (0.54-0.52 =0.03 or 3%). This shrinkage

suggests that if the model were developed from the population rather than a sample it

would account for approximately 3% less variance in the outcome. The significance of

R2 can be tested using an F-ratio. Fchange was 15.419 which is significant (p<0.001) as

shown in the table 7.3. The Durbin-Watson statistic in the last column verifies the

assumption of independent errors. As a conservative rule Durbin-Watson value must be

within 1 to 3 and close to 2 is better. The model results satisfy this criterion.

The next table contains an ANOVA, which tests whether the model is significantly

better at predicting the outcome than using the mean as a ‘best guess’. The chosen

model has three predictors and one constant so the model has 38 degrees of freedom.

The F ratio is calculated by dividing the average improvement in prediction by the model

(MSm).

𝐹 =𝑀𝑆𝑚

𝑀𝑆𝑟=

Average Improvement in Prediction by Model

Average Difference between Model and Observed data

Model results greater than 1 suggest that the improvement due to fitting the

regression model is much greater than the inaccuracy. For this model F is 15.419, which

is also highly significant (p<0.001).

Table 7.4 ANOVA table for the MLR model

ANOVA

Model Sum of Squares df Mean

Square F Sig

Regression 4731.19 3 1577.06 15.419 0

Residual 3886.71 38 102.28

Total 8617.91 41

a. Dependent Variable: Capture Efficiency (%)

b. Predictors: (constant), Effective spacing (mm), Runtime (mins), Flow (Litre/sec)


Md Abdul Aziz 143

The next discussion table concerns Model parameters. The first part of the table

gives us an estimate for which b-values (Equation 7.2) and these values indicate the

individual contribution of each predictor to the model. The other two input parameters,

weir opening and comb layers were not considered in the final model as they provided

an insignificant contribution to the model outcome.

The final regression equation is given below:

Capture Efficiency=59.312+0.69(runtime)-0.339(flow) +5.3(effective spacing) (7.3)

The ‘b’ values show the relationship between the capture efficiency and model

input parameters.

Table 7.5 Coefficient of different parameters

Runtime (b =0.69): This value indicates that in the experiment runtime increases

by one unit and capture efficiency increases by 0.69 of a unit. For example, if the Comb

Separator device operates/runs for one additional minute the sewer overflow capture

efficiency may increases around 0.69%. This explanation is true only if effects of flow

discharge and effective spacing are held constant and the device operates till it captures

all sewer solids.

Flow (b = -0.339): This value indicates that flow discharge has a negative

correlation with sewer overflow capture efficiency, so if one more unit of flow passes the

Comb Separator device it means that there is around 0.339 unit reduction in sewer solid

Standardized Coefficients

B Std. Error Beta(Constant) 59.312 8.250 7.189 0.000

Runtime (min) 0.690 0.230 0.329 2.994 0.005

Flow (Litre/sec) -0.339 0.103 -0.370 -3.288 0.002

Effective

Spacing (mm)5.390 1.280 0.474 4.211 0.000

Lower Bound Upper Bound Zero-order Partial Part Tolerance VIF

(Constant) 42.611 76.014

Runtime (min) 0.223 1.157 0.324 0.437 0.326 0.986 1.015

Flow (Litre/sec) -0.548 -0.13 -0.507 -0.471 -0.358 0.936 1.069

Effective

Spacing (mm) 2.799 7.982 0.538 0.564 0.459 0.937 1.068

95.0% Confidence Interval for B Correlations Collinearity StatisticsModel

ModelUnstandardized

Coefficients t Sig.


Md Abdul Aziz 144

capture efficiency. For example, if flow is increasing by 1 cubic meter the sewer overflow

capture efficiency will reduce by around 0.339%. This phenomenon is true when the

other two parameters of runtime and effective spacing are constant.

Effective Spacing (b = 5.390): Effective spacing of the comb has a positive

correlation with sewer overflow capture efficiency. In other words, if the effective spacing

increases, capture efficiency also increases. For example one unit increase in effective

spacing may increase capture efficiency up to 5.39%. Therefore one millimetre increased

in effective spacing will crease 5.39% sewer solids capture by the screen. This

interpretation is true only when runtime and flow discharge are constant.

Each of these beta values has an associated standard error showing how much

these values would vary across different data series or sample. These standard errors

were used to determine whether or not the b-value differed significantly from zero. A t-

statistic test can be performed, which tests whether a b-value is significantly different

from zero. Rule suggests if in the t-test associated with a b-value is significant (if the

value in the column labelled sig. is less than 0.05) then the predictor is making a

significant contribution to the model.

In this case all three predictors are significant at less than p<0.05. In addition, the

smaller the value of significance (and the larger the value of t), the greater contribution it

has on the predictor variables. For example effective spacing has the highest t value of

4.2 and also lowest significance of 0.000 so effective spacing is the most significant

predictor in the discussed model, followed by flow and runtime.

Multicollinearity: Tolerance and VIF statics are the key guidelines to check for

multi-collinearity. If the largest VIF is greater than 10 then there is cause for concern [29];

[119]. If the average VIF is substantially greater than 1 then the regression may be

biased. In this study the VIF values are all well below 10 and the tolerance statistics all

well below 0.2; therefore there was no collinearity within the experimental data set. The

average VIF is close to 1 and this confirms that collinearity is not a problem for the data

set.


Md Abdul Aziz 145

7.2.3 Development of the Dataset Using Sampling Techniques

One of the key challenges for model sensitivity analysis where there are few

parameters which involves the multi-dimensional space in an equitable and

computationally efficient manner. All model input parameter for this model can be defined

in a way that each input parameter have an approximate probability density function

associated with it. The next step would be to simulate by sampling a single value from

each parameter’s distribution. A sample and sensitivity analysis tools (SaSAT) was used

in the model [85]. SaSAT produces input parameter samples for an external model.

These samples, in conjunction with outputs (responses) created from the outer model

(for example regression model), perform the sensitivity analysis. The data generation

process using SaSAT is shown in the table below:

Table 7.6 Schematic diagram of SaSAT data generation

In the studied case the input parameters distribution checks are given below:

It was important to check the input parameter for the sampling model using LHS

technique.

Runtime: The P-P plots suggest that the model data set is close to the inclined

horizontal line, so it was a reasonable consideration to consider Runtime as having a

linear distribution.


Md Abdul Aziz 146

Figure 7.5: P-P plots for the runtime predictor

Figure 7.6: Data on both sides of normal for runtime


Md Abdul Aziz 147

Flow: The P-P plots suggest that the model data set is lying close to the diagonal

line, so it was a reasonable to assume that the flow predictor has a linear distribution,

refer to Figure 7.7 and 7.8.

Figure 7.7: P-P plots for the flow predictor

Figure 7.8: Data on both sides of normal for flow


Md Abdul Aziz 148

Effective Spacing: The P-P plots suggest that the model data set lying close to

the diagonal line, so it was a reasonable assumption that the effective spacing predictor

has a linear distribution, refer to Figure 7.9 and Figure 7.10.

Figure 7.9 P-P plot of effective spacing

Figure 7.10 Data on both sides of normal for effective spacing


Md Abdul Aziz 149

7.2.3.1 Random Sampling: In random sampling each parameter’s distribution is

used to consider N values randomly. This is usually superior to drawing N values

arbitrarily. This is usually better to univariate approaches to sensitivity analysis; however

this is not the most efficient way to sample the parameter space.

7.2.3.2 Latin Hypercube Sampling (LHS): LHS is more efficient and refined

statistical technique being a type of stratified Monte Carlo sampling [91& 92]. Monte

Carlo analysis is an extension of Latin square sampling as proposed by McKay et al.

[115] and further developed and familiarized [91]; [92]; [93]. For each parameter a

probability density function is defined and stratified into N equal-probable serial intervals.

A solitary value is then selected arbitrarily from every interval and for every parameter.

This process provides an input value from each sampling interval and is used only

once in the analysis. This ensures that the whole parameter space is equally sampled in

an efficient manner. The outcome variables can be derived by running the model N times

with each of the sampled parameter sets. A more insight about the Latin Hypercube

sampling methodology can be found in the work [91]; [93]; [94].

7.3 Results and Discussion

7.3.1 Relative Significance of the Input Parameters

Pearson’s correlation coefficient suggests that effective spacing has a large

positive correlation (r=0.684) with the outcome, capture efficiency. Flow discharge has a

negative correlation (r =-0.548) with capture efficiency and runtime has a positive

correlation, (r =0.476) and all these results are statistically significant with p<0.001, refer

to Table 7.7.

The final regression model equation after simulating the individual dataset for

10,000 times using the LHS method shows:

CaptureEfficiency=59.442+0.702(runtime)-0.341(flow)+5.317(effectivespacing) (7.4)


Md Abdul Aziz 150

Table 7.7: Results using the LHS method for 10,000 data

Correlations

Capture

Efficiency (%)

Runtime (minutes)

Flow (Litre/sec)

Effective

Spacing(mm)

Pearson Correlation

Capture Efficiency (%) 1.00 0.49 -0.53 0.68

Run time (minutes) 0.49 1.00 0.01 0.01

Flow (litre/sec) -0.53 0.01 1.00 0.01

Effective Spacing (mm) 0.68 0.01 0.01 1.00

Sig.

(1-tailed)

Capture Efficiency (%) 0.00 0.00 0.00

Run time (minutes) 0.02 0.12 0.31

Flow (litre/sec) 0.00 0.12 0.18

Effective Spacing (mm) 0.00 0.31 0.18

7.3.2 Selection of the Input Parameters

For this research, SPSS Version 22 [89] was used as a tool for MLR modelling.

In the current research forced entry (known as Enter in SPSS) was used as the method

so that all input parameters or predictors are forced into the model concurrently. Unlike

the hierarchical method, the forced entry method makes no decision regarding how the

variables are entered. Table 7.8 shows the results for two MLR models with five and

three input parameters. In developing the MLR model, initially all input parameters that

could have any influence on the output capture efficiency were considered. Trials were

done in the MLR analysis in such a way as all predictors are entered into the model and

their outputs are examined to see which input parameters or predictors contributed

significantly to the model’s capability to predict capture efficiency. In the initial model all


Md Abdul Aziz 151

the five input parameters being run time (minutes), flow (litre/second), weir opening

(millimetre), effective spacing (millimetre) and layers of combs (number) were

considered.

After few different trials with the input parameters, it was found the ‘weir opening’

and ‘comb layers’ were insignificant input (predictor) parameters as they had little

influence on the output sewage solid capture efficiency. ‘R’ is the value of the multiple

correlation coefficients between the predictors and the outcome. ‘R’ values vary from

0.753 to 0.741, from the first model to the second model, which is an insignificant

difference between the two datasets. For the first model ‘R square’ had a value of 0.567

and in the second model with three parameters used its value was 0.549. Therefore two

input parameters; weir opening and comb layers account for only 1.8% of the prediction

accuracy. So, the final MLR considered three input parameters; runtime (minutes), flow

(litre/second) and effective spacing (millimetre), and excluded comb layers (number) and

weir opening (mm). The ‘adjusted R square’ indicates the performance of the model and

in an ideal condition its value will be very close or same as ‘R square’.

Table 7.8 Comparison between initial and final model results

Model 1: Considering Five Parameters Change Statistics

Predictors R R

Square Adjusted

R R Square Change

F Change df1 df2 Sig.F

Change

Layers of combs,

runtime, flow,

weir opening and

effective spacing

0.753 0.567 0.507 0.567 9.438 5 36 0

Model 2: Considering Five Parameters Change Statistics

Predictors R R

Square Adjusted

R R Square Change

F Change df1 df2 Sig.F

Change

Runtime, flow,

effective spacing

0.741 0.549 0.513 0.549 15.419 3 38 0

In this case the difference of ‘R square’ between the final model and the initial

model was small (0.549 – 0.513 = 0.036). This reduction highlights that if the model were


Md Abdul Aziz 152

derived from the population rather than a sample it would account for approximately 3.6%

less variance in the outcome.

7.3.3 Impact of Effective Spacing on Capture Efficiency

Effective spacing was the most significant predictor (input parameter) to influence

capture efficiency. The value indicates that for one unit increase in ‘effective spacing’ the

‘capture efficiency’ increases by 5.317 units. The effective spacing is measured in

millimetre; whereas capture efficiency is measured in percentage. Therefore one

millimetre increased in effective spacing will crease 5.317% sewer solids capture by the

screen. This relation of effective spacing with capture efficiency is valid from 1mm to

6mm and also when ‘runtime’ and ‘flow’ are constant, refer to Figure 7.11.

Figure 7.11: Relationship between Effective spacing (mm) and Capture Efficiency (%)

7.3.4 Impact of Flow on Capture Efficiency

Flow has a negative correlation to sewage solids capture efficiency. As the flow

increases in the sewerage overflow system it produces a higher flow velocity over the

ogee weir and hence more sewage solids are likely to escape the traps/combs causing

lower rate of capture efficiency. Flow is measured in litre/second, whereas capture

efficiency is measured in percentage. An increase in flow by 1 litre/second causes to


Md Abdul Aziz 153

reduce the capture efficiency by 0.341%. It is to be noted that for this analysis, other two

parameters i.e. ‘runtime’ and ‘effective spacing’ are kept constant during the analysis.

Flowrate is one of the key understandings in the sewer overflow screening device

as the flow increases in the device (with fixed weir openings), the flow velocity also

increases which leads to a higher velocity of the sewage solids. Faster movement of

sewage solids near the comb separator is likely to reduce trapping efficiency, refer to

Figure 7.12.

Figure 7.12: Relationship between the Flowrate (l/s) and Capture Efficiency (%)

7.3.5 Runtime Impact on Capture Efficiency

The ‘runtime’ of sewage overflows varies positively with sewer solid ‘capture efficiency’

i.e. the longer the device runs, the higher the capture efficiency of the ‘comb separator’.

The runtime is measured in minutes, whereas capture efficiency is measured in

percentage. It is found that if the device ‘runtime’ increases one unit, the capture

efficiency increases by 0.702%. For example if the device runs for 16 minutes instead of

15 minutes, during the additional minute 0.702% more sewage solids are likely to be

trapped. This relation of device ‘runtime’ (minutes) is valid only on experimental cases

where sewer solids are present in flow while ‘effective spacing (mm)’ and ‘flowrate

(litre/second)’ are kept constant.


Md Abdul Aziz 154

Figure 7.13: Relationship between Runtime (min) and capture efficiency (%)

7.4 Summary

A series of trials with different runtimes, flow, effective spacing, layers of combs

and weir openings were tested. Sensitivity analyses of these input parameters were

performed to identify the influence of them on sewage overflow capture efficiency. The

sensitivity analysis was aimed at developing a robust understanding of the relationships

between the input (predictors) and output (outcome) variables. The MLR model was

initially considered using five input parameters. After significant trial and error it was

found that two input parameters, the weir opening and the comb layers, could be

excluded because these two parameters only contribute to 1.8% of the output.

MLR model Equation 7.3 and LHS sampling technique Equation 7.4 are almost

identical which ensured that the model retained the underlying input and output

relationship when expanding the dataset from 42 sets to 10,000 sets. Sensitivity analysis

delivered a cleaner understanding of the relative importance (rank) of the input

parameters. It was found that ‘effective spacing (mm)’ is the most influential parameter

followed by ‘flowrate (litre/second)’ and ‘run time (minutes)’. The sampling technique also

provided better understand of the input output relations; for example a 1 unit increase in

‘effective spacing’ could increase output ‘capture efficiency’ by 5.31%.


Md Abdul Aziz 155

These sensitivity analysis results will be immensely valuable for developing a

practice manual for the proposed device. Further effort to understand the performance

of the proposed ‘comb separator’ could focus mainly on three parameters: effective

spacing (mm), flow (litre/second) and runtime (minutes). These sensitivity analysis

results inform decisions made by the device operator and manager in managing different

sewerage overflow events. Further experiments are suggested to improve the

understanding of the input parameters on high flows.

Work presented in this chapter has been published in the following conferences and

journal papers:

Aziz, M. A., Imteaz, M., Samsuzzoha, M., & Phillips, D., 2013c, ‘Sensitivity

analysis for a proposed sewer overflow screening device’, 20th International Congress

on Modelling and Simulation, Adelaide, 2 to 5 December, Australia.

Aziz, M. A., Imteaz, M., Rasel, H.M., and Samsuzzoha, M., 2015d, ‘Parameter

Sensitivity Using Sampling Technique for a proposed ‘Comb Separator’, A sewer

overflow screening device’, ASEAN- Australian Engineering Congress on Innovative

Technologies for Sustainable Development and Renewable Energy 11-13 March.

Aziz, M. A., Imteaz, M., Samsuzzoha, M. A., and Rasel, H.M., 2015b,

Sensitivity Analysis on the Pollutant Trapping Efficiencies of a Novel Sewerage Overflow

Screening Device. Revising (March 2016) Journal of Hydro-informatics

Conclusions

Md Abdul Aziz 157

Chapter 8

Conclusions

Conclusions

Md Abdul Aziz 158

8.1 Introduction

During continuous downpours, the urban sewer system is not able to carry all the

excessive water; hence some of this excess water flows into the open creek system,

carrying a lot of sewer solids. These sewer solids are dispersed, suspended or washed

into the rivers. They eventually settle, creating odours and toxic/corrosive atmospheres

in the mud deposits on riverbeds. The solids also create aesthetic problems, either

through their general appearance (increasing dirtiness) or through the actual presence

of specific, objectionable items, such as float debris, sanitary discards/faecal matter,

scum or even parts of car tyres. The current study took a holistic view in order to

understand the limitations of existing sewage overflow screens, and to analyse the

research gaps in this field. Some common limitations in existing screening devices

include high on-going maintenance and operational costs, low capture efficiency of

sewage solids, and the use of sophisticated electrical-mechanical switching systems. To

overcome these issues, the objective of the current research was to develop a new

sewage overflow device with improved capture efficiency, low maintenance, and a self -

cleansing mechanism.

8.2 Research Summary

This research focussed on the innovation of a novel sewer overflow screening

device, a ‘Comb Separator’ that can overcome common limitations such as lower sewer

solids capture efficiency and blockage on the screen. The initial concept was tested using

a CFD model analysis, which included no sophisticated electrical switching system and

designing the device so that it is self-cleansing. This also reduces the operational and

maintenance costs of the proposed screen. To prove the concept of the functionality of

this screen, a CFD model analysis was performed. The CFD model provides much better

insight into the design of the device for laboratory testing, with better understanding

about the design parameters.

A series of laboratory tests was completed as part of this project to achieve the

objectives of high capture efficiency, minimal blockage on the screen and effective

function of the device in remote unstaffed locations. It was found that the proposed

Conclusions

Md Abdul Aziz 159

device was free of most of the common limitations in existing screening systems, such

as blinding (minimal blockage), high maintenance requirements and an electrical-

mechanical switching system [154]. Physical limitations of the experimental work were

overcome by doing an ANN model analysis. ANN has the potential to understand the

complex relationship between input and output parameters without having knowledge of

the physical characteristics of the sewer solids. The ANN model supplements the

limitations of the CFD and laboratory testing. Finally, detailed sensitivity testing of the

hydraulic device was performed: a standard check for any hydraulic device. The

sensitivity analysis also provides a guideline for future application of the device in the

physical environment. The ‘Comb Separator’ showed good application potential for

further research in actual sewer overflow systems.

8.3 Knowledge Contributions

In this section, the contributions made in the context of industry practice and

academic research on sewer overflow screening devices are listed.

8.3.1 CFD application

Physical experimental set-ups involve significant cost and time. To overcome

this issue, a thorough modelling analysis was performed using the CFD model. The

modelling investigation optimised the design parameters and the inlet orientation for the

novel gross pollutant device. Two different inlet conditions were analysed. One was an

inlet parallel to the ogee weir, and the other perpendicular to it. The inlet parallel to the

ogee weir was considered the better option, as the water level over the ogee weir was

higher due to wave reflection, which can provide higher velocity and shear stress. The

location of the perforations was found to work more efficiently at the bottom, as there

was higher velocity and shear stress in that position.

To reduce the reflected wave on the small screening device, the model showed that

providing longer inlet pipes was a solution. An inlet 1.5m long will reduce the reflected

wave by 10% more than a 0.3m inlet. The four standard ogee weir orientations proposed

by the US Army Corp of Engineers [169] were investigated, and it was found that an

inclined slope of 1H: 3V from the rectangular device to the ogee weir was the most

efficient.

Conclusions

Md Abdul Aziz 160

The CFD model analysis was very helpful for designing the experimental parameters,

as well as providing the concept for the gross pollutant device.

8.3.2 Laboratory experiments

Once the design parameters and inlet orientation using CFD models were

understood, a second experimental facility was set up with minor adjustments to the

initial device. This novel device is called the ‘Comb Separator’. The ‘Comb Separator’

was subjected to a series of trials with different combinations of numbers of combs,

spacing of combs, flow volume and weir openings being tested. The proposed device

can capture larger sewage solids of more than 10mm diameter with over 95% capture

efficiency. The capture efficiency is dependent on selected experimental parameters

which can vary the sewer capture efficiency. These parameters are flow (discharge),

weir opening, comb spacing and layers. Two layers of combs were found to be more

efficient than three layers. Increasing the comb spacing improves capture efficiency.

Robustness of optimum set-ups were tested to generate consistent results. The

laboratory testing was especially beneficial in helping to understand the sewer solids

capture efficiency and the level of blockage on the screen.

8.3.3 ANN application

The experimental work was restricted by the physical limitations inherent in

laboratory studies. As sewage solids vary in density, they can be difficult to study using

physical laws based on deterministic models such as CFD. The Artificial Neural Network

(ANN) model has the capacity to accurately predict the outcome of complex, non-linear

physical systems where physico-chemical processes are relatively poorly understood. A

series of laboratory tests was conducted with 55 different sets of data (i.e. varying flows

and combs spacing conditions). Forty-seven sets of experimental data were used with

60% being for training, and 20% each for testing and validation of the model. Separate

validation data sets were used to judge the overall performance of the trained network.

The model is able to successfully predict the experimental results with more than 90%

accuracy, with the average absolute percentage of errors varying from 4% to 7%.

The application of ANN supplemented the CFD and laboratory experiments.

This method is especially beneficial when other deterministic physical modelling or

experimental results are not convincing enough to derive a conclusion.

Conclusions

Md Abdul Aziz 161

8.3.4 Sensitivity analysis

Once an ANN model was established and ready to use for experiments on

capture efficiency, it was tested with different options to identify the best one for the

current settings. Input parameters such as flowrate, effective comb spacing, device run

time, weir opening and the number of comb layers were considered. It was important to

compare with the industry standard device Hydro-JetTM screen. It was found that, in low-

flow situations, the Comb Separator has better capture efficiency and avoidance of

screen blockage than the Hydro-JetTM screen. The comparison highlighted the

importance of understanding parameter sensitivity in the Comb Separator. It is strongly

recommended that sensitivity testing be undertaken for any hydraulic device before

application in the natural environment [97].

The sensitivity analysis provided a more precise understanding of the relative

importance of the input parameters. For example, with the application of the Latin

Hypercube sampling technique using 10,000 data [85], it was found that effective spacing

was the most influential parameter, followed by flow discharge and device run time. The

sampling technique also provided a better understanding of the contribution of each input

parameter. For example, with one unit increase in effective spacing, the capture

efficiency of sewage solids in the Comb Separator can increase by 5.31 units. This

sensitivity information provided better insight about the relative importance for the input

parameters. This information would be highly valuable in managing the device in actual

sewer overflow conditions.

8.4 Limitations

A potential limitation of this research was the small sample size in the laboratory

experiments. This is due to the physical limitations in creating more data realistically from

the experimental series. The smallest spacing tested was 10mm, as it was difficult to

provide set-ups smaller than 10 mm. There was also one limitation with the CFD model.

As the sewer solids were materials of different density, this was difficult for the CFD

model to simulate. To overcome these limitations, the ANN model was used and

statistical sensitivity testing was done in the later phase of this research. The use of such

Conclusions

Md Abdul Aziz 162

methods is recommended in order to overcome the above-mentioned challenges

experienced in this research.

8.5 Future Research

Future study on this topic should consider the following key points:

The ANN model could be revised based on the understanding gleaned from

the sensitivity analysis, which initially considered all 16 input parameters. As the

laboratory data set is small, this could provide better insight about the optimisation

conditions for the laboratory experimental set-ups.

Sensitivity of the different algorithms while using ANN also needs to be tested.

Changing the number of hidden layers or splitting the data set difference between input

and output parameters could provide a better understanding of the best case scenario

for the ANN.

Further experimentation with the Comb Separator device is recommended,

especially when flow rates are higher (up to 120 l/s). It is important to understand the

performance of the device in high-flow conditions such as floods and heavy rains. These

results need to be compared with industry standard devices such as the Hydro-JetTM.

Further experiments should consider small-diameter sewage solids such as

cigarette butts. It was observed that the small-diameter sewage solids showed higher

variation in capture efficiency. Further trials could contribute to a better understanding of

this issue. The current sensitivity analysis will provide a useful guideline for analysis.

Future investigation should also consider onsite testing of the proposed Comb

Separator in actual sewage overflow conditions.

The ‘Comb Separator’ showed good application potential at low flows for

improving sewage solids capture efficiency in the urban sewerage system.

References

Md Abdul Aziz 163

References

References

Md Abdul Aziz 164

References

1. Adamsson A, Stovin V, Bergdahl L., 2003, ‘Bed shear stress boundary

condition for storage tank sedimentation’, Journal of Environmental

Engineering, vol. 129, no.7, pp.651–658.

2. Adamowski J., & Karapataki C., 2010, ‘Comparison of multivariate regression

and artificial neural networks for peak urban water demand forecasting:

evaluation of different ANN learning algorithms’, Journal of Hydrologic

Engineering, Vol. 15, No. 10.

3. Alex Dick, 2013, ‘Campaigners storm parliament and tell them: UK Surfing is

worth British Pound 1.8 Billion’, the surfer’s path,

http://surferspath.com/news/campaigners-storm-parliament-and-tell-them-

uk-surfing-is-worth-1-8billion.html#kY3MTAA8J6dl4W2l.97

4. Alp, M., Cigizoglu, H., 2007, Suspended sediment load simulation by two

artificial neural network methods using hydro-meteorological data.

Environmental Modelling and Software, vol.22, no. 1, pp.2-23.

5. ASCE Task Committee on Application of Artificial Neural Networks in

Hydrology, 2000a, ‘Artificial neural networks in hydrology. I: preliminary

concepts.’ Journal of Hydrologic Engineering, ASCE vol.5, no. 2, pp.115-123.

6. Akhtar, M.C., Corzo, G.A., van Andel, S.J., & Jonoski, A., 2009, ‘River flow

forecasting with artificial neural networks using satellite observed precipitation

pre-processed with flow length and travel time information: case study of the

Ganges river basin,’ Hydro. Earth Syst. Sci., vol.13, pp.1607-1618.

7. ASCE Task Committee on Application of Artificial Neural Networks in

Hydrology, 2000b, ‘Artificial neural networks in hydrology. II: hydrologic

applications.’ Journal of Hydrologic Engineering, ASCE vol.5, no.2, pp.124-

137.

8. Aziz, M.A., Imteaz, M.A., Choudhury, T.A., & Phillips, D.I., 2011, ‘Artificial

Neural Networks for the prediction of the trapping efficiency of a new sewer

overflow screening device’, 19th International Congress on Modelling and

Simulation, Perth, Australia.

9. Aziz, M. A., Imteaz, M., Choudhury, T. A. & Phillips, D. 2013a, ‘Applicability of

artificial neural network in hydraulic experiments using a new sewer overflow

screening device’, Australian Journal of Water Resources, Vol. 17, No. 1,

pp.77-86.

http://surferspath.com/news/campaigners-storm-parliament-and-tell-them-uk-surfing-is-worth-1-8billion.html#kY3MTAA8J6dl4W2l.97

http://surferspath.com/news/campaigners-storm-parliament-and-tell-them-uk-surfing-is-worth-1-8billion.html#kY3MTAA8J6dl4W2l.97

References

Md Abdul Aziz 165

10. Aziz, M. A., Imteaz, J. Naser & Phillips, D. 2013b, ‘Hydrodynamic

Characteristics of a New Sewer Overflow Screening Device: CFD Modelling

and Analytical Study’, International Journal of Civil and Environmental

Engineering, vol.7 no.1, pp 71-76.

11. Aziz, M. A., Imteaz, M. Huda., & J. Naser 2014a, ‘Optimising inlet condition

and design parameters of a new sewer overflow screening device using

numerical modelling technique‘, Journal of Water, Science and Technology,

vol.70, no.11,pp.1880-1887

12. Aziz, M. A., Imteaz, M., Rasel, H.M., & Phillips, D., 2015a, ‘Development and

Performance Testing of ‘Comb Separator’, A Novel Sewer Overflow

Screening Device’, International Journal of Environment and Waste

Management, Vol. 16, No.3, 2015.

13. Aziz, M. A., Imteaz, M., Rasel,H.M., & Samsuzzoha, M. 2015d, ‘Parameter

Sensitivity Using Sampling Technique for a proposed ‘Comb Separator’, A

sewer overflow screening device’, ASEAN- Australian Engineering Congress

on Innovative Technologies for Sustainable Development and Renewable

Energy 11-13 March 2015.

14. Andoh, R.Y.G., and Smisson, R.P.M., 1994, ‘High Rate Sedimentation in

Hydrodynamic Separators’. Proceeding of 2nd International Conference On

Hydraulic Modelling Development and Application of Physical and

Mathematical Models, Stratford, UK, pp. 341. 358.

15. Andoh R.Y.G., Smith B.P. & Saul A.J., 1999, ‘the screen efficiency of a novel

self-cleanshing CSO.’ 8th International Conference on Urban Storm Drainage,

Sydney, Australia, 30 August – 3 September, pp.205-212.

16. Andoh, R. Y. G., and Saul, A.J., 2000, ‘Field Evaluation of Novel Wet-Weather

Screening Systems’, Proc.WEF Speciality Conf., Collection Systems Wet

Weather Pollution Control: Looking into Public, Private and Industrial Issues,

New York, USA, 7-10 May.

17. Andoh, R.Y.G. and Saul, A.J., 2003, ‘The use of hydrodynamic vortex

separators and screening systems to improve water quality’, Water Science

& Technology, Vol. 47 No. 4, pp. 175–183.

18. An, G., 1996, ‘The effects of adding noise during backpropagation training on

a generalization performance’, Neural Computation, vol. 8, pp. 643-674.

19. Armitage, N & Rooseboom, A., 2000, ‘the removal of urban litter from

stormwater conduits and streams: paper 1 the quantities involved and

References

Md Abdul Aziz 166

catchment litter management options’. Water SA (South Africa), vol.26, no.2,

pp.181-188.

20. Arnett, C. J. & Gurney, P. K., 1998, ‘High rate solids removal and chemical

and non-chemical UV disinfection alternatives for treatment of CSO.s’,

Innovation 2000, Conference on Treatment Innovation for the Next Century,

Cambridge, UK.

21. Averill, D., Mack-Mumford, D., Marsalek, J., Andoh, R. & Weatherbe, D.,

1997, ‘Field Facility for Research and Demonstration of CSO Treatment

Technologies’, Water Science & Technology, vol.36, no.8-9,pp. 391 . 396.

22. AVL 2008, AVL Fire CFD Solver v8.5 manual, A-8020 Graz, Austria.

23. Balmforth, D.J., & Henderson, R.J., 1988, ‘ A guide to the design of storm

overflow structures’, Water Research Centre Report, ER304E, WRc,

Swindon, Wilts, England.

24. Betts, P. L., 1979, ‘A variation principle in terms of stream function for free

surface flows and its application to finite element method’, Comp. and Fluids,

vol.7, no.2, pp.145-153.

25. Berry, W.D., 1993, ‘Understanding regression assumptions. Saga university

paper series on quantitative applications in the social sciences’, 07-092.

Newbury Park, CA: sage.

26. Becker, S., 1991, ‘Unsupervised learning procedures for neural networks,’

International Journal of Neural Systems, vol. 2, pp.17-33.

27. Bhajantri, M. R., Eldho, T. I., & Deolalikar, P. B., 2006, ‘Hydrodynamic

modelling of flow over a spillway using a two-dimensional finite volume-based

numerical model’, Sadhana Vol. 31, Part 6, December, pp. 743-754.

28. Bolt, G., 1992, ‘Fault tolerance in artificial neural networks,’ Advanced

Computer Architecture Group, Department of Computer Science, University

of York, Heslington, York, YO1 5DD, U.K.

29. Bowerman, B,L., & O’Connell, R.T., 1990, ‘Linear statistical models: An

applied approach’ (2nd ed.). Belmont, CA: Duxbury.

30. Brombach, H., 1992, ‘Solids Removal from Combined Sewer Overflows with

Vortex Separators’. NOVATECH 92, International Conference on Innovative

Technologies in the Domain of Urban Water Drainage, Lyon (France),

November 3-5, pp. 447-459.

31. Bruen, M., Yang, J., 2006, ‘Combined hydraulic and black-box models for

flood forecasting in urban drainage systems’, Journal of Hydrologic

Engineering, vol.11, no.6.

References

Md Abdul Aziz 167

32. Burgisser, M. F. & Rutschmann, P., 1999, ‘Numerical solution of viscous 2DV

free surface flows: Flow over spillway crests’, Proc., 28 th IAHR Congr.,

Technical University Graz, Graz, Austria.

33. Casey, M. & Wintergerste, T., 2000, ‘Special interest group on quality and

trust in industrial CFD: best practice guidelines’. Switzerland: European

Research Community on flow turbulence and combustion (ERCOFTAC).

34. Cassidy, J. J., 1965, ‘Irrotational flow over spillways of finite height’, Journal

of Engineering and Mechanical Division., ASCE, vol. 91, no.6, pp.155-173.

35. Carpenter, W.C., & Barthelemy, J., 1994, ‘Common misconcepts about neural

networks as approximations.’ J.Comp. in Civ. Engrg. ASCE, vol.8, no.3,

pp.345-358.

36. Center for marine conservation (CMC),1998,‘Coastal clean-up- united states’,

www/cmc-ocean.org/cleanupbro.

37. Chandramouli, V. & Deka, P., 2006, ‘Neural network based decision support

model for optimal reservoir operation,’ Water Resources Management,

vol.19, no.4, pp.447-464.

38. Choudhury, T. A., Hosseinzadeh, N. & Berndt, C. C., 2011, ‘Artificial Neural

Network application for predicting in-flight particle characteristics of an

atmospheric plasma spray process’, Surface and Coatings Technology, vol.

205, pp. 4886-4895.

39. Choudhury, T. A., Hosseinzadeh, N. & Berndt, C. C., 2012, ‘Improving the

Generalization Ability of an Artificial Neural Network in Predicting In-Flight

Particle Characteristics of an Atmospheric Plasma Spray Process,’ Journal of

Thermal Spray Technology, vol. 21, pp. 935-949.

40. Choudhury T.A., 2013 ‘Artificial Neural Networks Applied to Plasma Spray

Manufacturing’. Ph.D. thesis. Swinburne University of Technology, Australia.

41. Chow, V. T., 1959, Open-channel hydraulics, McGraw-Hill, New York,

pp.365-380.

42. Claire E, Imrie, Sevket Durucan, 1999, ‘River flow prediction using the

cascade-correlation neural network learning architecture’. Water 99 Joint

Congress- Brisbane, Australia 6-8 July 1999.

43. Chang, L.C. & Chang, F.J., 2001, ‘Intelligent control for modelling of real time

reservoir operation,’ Hydrol. Process., vol.15, no.9, pp. 1621-1634.

44. Chang, F.J. & Chang,Y.T., 2006, ‘Adaptive neuro-fuzzy inference system for

prediction of water level in reservoir’, Adv. Water Resources., vol.29, no.1,

pp.1-10.

References

Md Abdul Aziz 168

45. Chang, L.C. and Chang, F.J., 2009, ‘Integrating hydrometeorological

information for rainfall-runoff modelling by artificial neural networks,’

Hydrol.Process., vol.23, no.1, pp.1650-1659.

46. Chiang, Y.M., Hsu, K.L., Chang, F.J., Hong, Y., & Sorooshian, S., 2007,

‘Merging multiple precipitation sources for flash flood forecasting’, Journal of

Hydrology, vol. 330, no.3-4, pp.183-196.

47. Chiang, Y.M., Li-Chiu, Chang., Tsai, M.J.,Wang, Y.F., & Chang, F.J., 2010,

‘Dynamic neural networks for real-time water level predictions of sewerage

systems-covering gauged and ungauged sites’, Journal of Hydrology and

Earth System Sciences, vol.14,pp.1309-1319.

48. Churchland, P. S. & Sejnowski, T. J., 1992, ‘the computational brain:’ The

MIT press.

49. Crick, M.J., Hill, M.D. & Charles, D.,1987, 'The Role of Sensitivity Analysis in

Assessing Uncertainty.’ In: Proceedings of an NEA Workshop on Uncertainty

Analysis for Performance Assessments of Radioactive Waste Disposal

Systems, Paris, OECD, pp. 1-258.

50. Dewals, B.J., Kantoush, S.A., Erpicum, S., Pirotton, M., & Schleiss, A.J.,

2008, ‘Experimental and numerical analysis of flow instabilities in rectangular

shallow basins’, Environ. Fluid Mech., vol.8, pp.31-54.

51. Dufresne, M., Vazquez, J., Terfous, A., Ghenaim, A., & Poulet, J.,

2009, ‘CFD Modeling of Solid Separation in Three Combined Sewer Overflow

Chambers’, J. Environ. Eng., vol.135 no.9, pp. 776–787.

52. Dufresne, M., Vazquez,j., Terfour, A., Ghenaim, A., & Poulet, J.-B., 2009,

‘Experimental investigation and CFD modelling of flow, sedimentation, and

solids separation in a combined sewer detention tank.’ Computational Fluid,

vol.8, no.5, pp.1042-1049.

53. Durbin, J., Watson, G. S., 1950, ‘Testing for Serial Correlation in Least

Squares Regression, I’. Biometrika vol.37 (3–4), pp. 409–428.

doi:10.1093/biomet/37.3-4.409. JSTOR 2332391

54. Durbin, J., Watson, G. S., 1951, ‘Testing for Serial Correlation in Least

Squares Regression, II’, Biometrika vol.38 (1–2), pp.159–179.

doi:10.1093/biomet/38.1-2.159. JSTOR 2332325

55. Eldho, T.I., 2008, Lecture 26, Lecture Series on Fluid Mechanics, Department

of Civil Engineering IIT Bombay, URL: http://nptel.iitm.ac.in

https://en.wikipedia.org/wiki/Biometrika

https://en.wikipedia.org/wiki/Digital_object_identifier

https://dx.doi.org/10.1093%2Fbiomet%2F37.3-4.409

https://en.wikipedia.org/wiki/JSTOR

https://www.jstor.org/stable/2332391

https://en.wikipedia.org/wiki/Digital_object_identifier

https://dx.doi.org/10.1093%2Fbiomet%2F38.1-2.159

https://en.wikipedia.org/wiki/JSTOR

https://www.jstor.org/stable/2332325

http://nptel.iitm.ac.in/

References

Md Abdul Aziz 169

56. Fahlman, S. E., 1988, ‘Faster-learning variations on back propagation: an

empirical study,’ Proceedings of the 1988 Connectionist Models Summer

School, pp. 38-51.

57. Faram, M.G., Andoh, R.Y.G. and Smith, B.P. 1999, ‘The Screen Efficiency of

a Novel Self-cleansing CSO’, 8th International Conference on Urban Storm

Drainage, Sydney, Australia, 30 August - 3 September 1999, pp. 205-212.

58. Faram, M. G. & Andoh, R. Y. G., 2000, ‘Application of Simulation and

Predictive Techniques for the Evaluation of Hydrodynamic Separators’,

Wastewater Treatment: Standards and Technologies to Meet the Challenges

of the 21st Century, CIWEM/AETT Millennium Conf., Leeds, UK, 4-6 April, pp

223-230.

59. Faram, M.G., Andoh, R.Y.G. & Smith, B.P., 2001, ‘Optimised CSO screening:

A UK perspective’, 4th Novatech International Conference on Innovative

Technologies in Urban Drainage, Lyon, France, 25-27 June 2001, pp.1031-

1034.

60. Faram, M.G., 2001, ‘Developments in screening technology’ CIWEM

Conference-Combined sewer overflow: The challenge and latest innovations

61. Fausett, L., 1994, Fundamentals of neural networks, Prentice Hall,

Englewood Cliffs, N.J.

62. Ferziger, J.H. and M. Peric, 2002, ‘Computational Methods for Fluid

Dynamics’. 3rd ed: Springerlink.

63. Floyd, J., 2005, ‘Converting an Idea into a Worldwide Business

Commercializing Smelting Technology.’ Metall. Trans. B., vol.36 (B), pp.557-

575.

64. Field, R. 1972, ‘The Swirl Concentrator as a CSO Regulation Facility’, USEPA

Report no. R2-72-008, USA.

65. Gonzalesz-Ubierna, S., Jorge-Mardomingo, I., Cruz, M.T., Valverde, I. &

Casermeiro, M.A., 2013, ‘Sewage Sludge Application in Mediterranean

Agricultural soils: Effects of Dose on the Soil Carbon Cycle’, International

Journal of Environment and Research, vol .7, no.4, pp 945-956.

66. Guo, Y., Wen, X., Wu, C. & Fang, D., 1998, ‘Numerical modelling of spillway

flow with free drop and initially unknown discharge’, J. Hydr. Res., Delft, The

Netherlands, Vol. 36, no.5, pp. 785-801.

67. Guanghua Q, Wang, S., 2011, ‘Annual runoff prediction with a sensitive

artificial neural networks model’, 34th IAHR World Congress – Balance and

References

Md Abdul Aziz 170

Uncertanity, 33th Hydrology and Wagter Resources Symposium, 10th

Hydraulics Conference, 26th -1st July , Brisbane, Australia.

68. Guanghua Qin, Hongxia Li, Xin Wang, Qingyan He,& Shenqi Li, 2015,

‘Annual runoff prediction using a nearest-neighbour method based on cosine

angle distance for similarity estimation’, ICGRHWE14, Guangzhou, China.

69. Guide to the design of combined sewer overflow structures, 1994, Report No

FRO488.

70. Hall, J. W., 2003, ‘Handling uncertainty in the hydroinformatic process’

Journal of Hydro informatics., vol.5, no.4, pp. 215–232.

71. Hall, J. W., & Solomatine, D., 2008, ‘A framework for uncertainty analysis in

flood risk management decisions’ J. River Basin Manage., vol.6, no.2,pp.85–

98.

72. Hall, J.W., Boyce, S.A., Wang, Y., Dawson, R.J., Tarantola, S., Saltelli, A.,

2009, ‘Sensitivity analysis for hydraulic models’ Journal of hydraulic

engineering @ ASCE/November 959.

73. Hamby, D.M., 1995a, 'A Numerical Comparison of Sensitivity Analysis

Techniques', scheduled to appear in Health Phys. 68.

74. Hamby, D.M., 1995b, ‘A review of techniques for parameter sensitivity

analysis of environmental models’, Environmental Monitoring and

Assessment vol.32, pp. 135-154.

75. Hamby, D.M., 1993, 'A Probabilistic Estimation of Atmospheric tritium Dose',

Health Phys. vol.65, pp.33-40.

76. Hagan, M. T. & Mehnaj, M. B., 1994, ’Training feedforward networks with the

Marquardt algorithm,’ IEEE Transactions on Neural Networks, vol. 5, pp. 989-

993, November.

77. Harwood, R. & Saul. A.J., 1999, ‘The influence of CSO chamber size on

particle retension efficiency performance.’ In Proceedings of the 8th

International Conference on Urban Storm Drainage, 30 August-3 September,

Sydney, Australia, pp.1-9.

78. Harwood, R., 2002, CSO modelling strategies using computational fluid

dynamics, In Eric W. Strecker and Wayne C. Huber (Eds.), Urban Drainage

2002: Proceedings of 9th International Conference on Urban Drainage -

9ICUD. Portland, Oregon, USA: American Society of Civil Engineers (ASCE),

pp.8-17.

79. Haykin, S., 1994, Neural networks: a comprehensive foundation, Mac Millan,

New York.

References

Md Abdul Aziz 171

80. Hedges, P.D.,1993, ‘The Relationship between Field and Model Studies of

an Hydrodynamic Separator Combined Sewer Overflow’, 6th International

Conference on Urban Storm Drainage, Niagara Falls Canada

81. Helton, J.C., Iman, R.L. & Brown, J.B., 1985, 'Sensitivity Analysis of the

Asymptotic Behaviour of a Model for the Environmental Movement of

Radionuclides’, Ecol. Modelling. Vol.28, pp. 243-278.

82. Hirt .C.W., & Nichols.B. D.,1981, ‘Volume of fluid (VOF) method for the

dynamics of free Boundaries.’ Journal of Computational Physics, vol.39, no.1,

pp. 201-225.

83. Hilgenstock, A. & Ernst, R., 1996, ‘Analysis of installation effects by means of

computational fluid dynamics CFD vs experiments’, Flow measurement and

instrumentation, vol.7, no.3-4, pp.161-171.

84. Hrabak D, Pryl K, Richardson J, Zeman E., 1999, ‘3-Dimensional modelling –

a new tool for the evaluation of CSO hydraulic performance’, In Proceedings

of the 8th ICUSD, pp. 928–933, Sydney, Australia.

85. Hoare, A., Regan, D.G., & David, Wilson.,D.P., 2008, ‘Sampling and

sensitivity analyses tools (SaSAT) for computational modelling’, Theoretical

Biology and Medical Modelling. doi:10.1186/1742-4682-5-4

86. Huda. M.N., 2012 ‘Computational Fluid Dynamic Modelling of Zinc Slag

Fuming Process.’ Ph.D. thesis. Swinburne University of Technology.

87. HydroQual, 1993, city-wide floatables study, final report-sources, fate and

control of floatable materials in New York Harbor; prepare for the New York

city department of environmental protection, bureau of environmental

engineering, Division of water quality improvements.

88. HydroQual, 1995, City Wide Floatables study, floatables pilot program final

report, evaluation of non-structural methods to control combined and storm

sewer flotable materials; prepared for the new York city department of

environmental protection, bureau of environmental engineering, divisions of

water quality improvements.

89. IBM Corp. Released 2013. ‘IBM SPSS Statistics for Windows’, Version 22.0.

Armonk, NY: IBM Corp.

90. Ikegawa, M. & Washizu, K., 1973, ‘Finite element method applied to analysis

of flow over a spillway crest’, International Journal of Numerical Methods in

Engineering, vol.6, pp. 179-189.

91. Iman RL, Helton JC, Campbell JE, 1981a, ‘An Approach To Sensitivity

Analysis Of Computer-Models. Introduction, Input Variable Selection and

References

Md Abdul Aziz 172

Preliminary Variable Assessment’. Journal of Quality Technology, vol.13,

no.3, pp.174.

92. Iman, R. L., Helton, J. C. & Campbell, J. E., 1981b, ‘An approach to sensitivity

analysis of computer models: Part II- Ranking of input variables, response

surface validation, distribution effect and technique synopsis’. Journal of

Qual. Technol. Vol.13, pp. 232-240.

93. Iman RL, Helton JC, 1988, ‘An Investigation of Uncertainty and Sensitivity

Analysis Techniques for Computer Models’, Risk Analysis vol.8, no.1, pp.71-

90.

94. Iman, R.L., & Helton, J.C., 1991, 'The Repeatability of Uncertainty and

Sensitivity Analyses for Complex Probabilistic Risk Assessments', Risk

Analysis. Vol.11, pp. 591-606.

95. International Atomic Energy Agency (IAEA), 1989, ‘Evaluating the reliability

of predictions made using environmental transfer models.’ Vienna: Safety

Series No. 100. Report No. STI/PUB/835; 1-106.

96. Jayawardena, A. W., & Fernando, D. A. K., 1996, ‘Comparison of multi-layer

perceptron and radial basis function network as tools for flood forecasting’,

Proc., North Am. Water and Envir. Conf., ASCE, New York, pp.457–458.

97. Johnson, P. A., 1996, ‘Uncertainty of hydraulic parameters’, J. Hydraul. Eng.,

122(2), 112–114.

98. Kai, W. Jufeng, Y., Guangshun, S. & Qingren, W., 2008, ‘An expanded

training set based validation method to avoid overfitting for neural network

classifier,’ in Fourth International Conference on Natural Computation (ICNC),

pp. 83-87.

99. Karystinos, G. N. & Pados, D. A. 2000, ‘On overfitting, generalization, and

randomly expanded training sets,’ IEEE Transactions on Neural Networks,

vol. 11, pp. 1050-1057.

100. Kin Choi Luk, James E. Ball and Ashish Sharma, 1999, ‘Integration of

Artificial neural networks and geographical information systems for rainfall

forecasting’, Water 99 Joint Congress- Brisbane, Australia 6-8 July 1999.

101. Kohavi, R., 1995, ‘A study of cross-validation and bootstrap for accuracy

estimation and model selection,’ in International Joint Conference on Artificial

Intelligence (IJCAI), pp. 1137-1145.

102. Launder, B.E. & D.B. Spalding, 1974, ‘The Numerical Computation of

Turbulent Flows.’ Com. Meth. App. Mech. Eng.,. vol.3, no.2, pp. 269-289.

References

Md Abdul Aziz 173

103. Li, W., Xie, Q., & Chen, C. J., 1989, ‘Finite analytic solution of flow over

spillways’, J. Engrg. Mech., ASCE, vol.115, no.12, pp. 2635-2648.

104. Liu, Y., 2006, ‘Create stable neural networks by cross-validation,’ in

International Joint Conference on Neural Networks (IJCNN), IEEE, pp. 3925-

3928.

105. Luyckx, G., Vaes, G., & Berlamont, J., 1999, ‘Experimental investigation

on the efficiency of a high side weir overflow’, Water Science and Technology,

vol. 39 no.2, pp. 61–68.

106. Madhani, J. T. & Brown, R. J., 2011, ‘A literature review on research

methodologies of gross pollutant traps’, The Proceedings of the First

International Postgraduate Conference on Engineering, Designing and

Development the Built Environment for Sustainable Wellbeing, Queensland

University of Technology, Brisbane, Qld.

107. Madhani. J.T., 2010 ‘The Hydrodynamic and Capture/Retention

Performance of A Gross Pollutant Trap’, Ph.D. thesis, Queensland University

of Technology.

108. Maier, H. R., & Dandy, G.C., 1996, ‘The use of artificial neural networks

for the prediction of water quality parameters’, Journal of Water Resources

Research, vol.32, no.4, pp.1013-1022.

109. Maier, H. R., & Dandy, G.C., 2001, ‘Neural network based modelling of

environmental variables: asystematic approach’, Journal of Mathematical and

Computer Modelling, vol.33, pp.669-682

110. Maren, A., Harston, C., & R. Pap R., 1990, ‘Handbook of neural

computing applications,’ academic, San Diego, Cardiff, 448.

111. MATLAB (R2012a: The MathWorks Inc., Natick, MA, USA)

112. Matsuoka, K., 1992, ‘Noise injection into inputs in back-propagation

learning,’ IEEE Transactions on Systems, Man and Cybernetics, vol. 22, pp.

436-440.

113. Matthieu D., Jose V., Abdelali T., Abdellah G & Jean B.P., 2009, ‘CFD

Modeling of Solid Seperation in Three Combined Sewer Overflow Chambers.’

Journal of Environmental Engineering (ASCE), September, pp. 776-789.

114. Maynord, S. T., 1985, ’General spillway investigation’, Tech. Rep. HL-85-

1, U.S. Army Engineer Waterways Experiment Station, Vicksburg, Miss.

115. McKay MD, Beckman RJ, Conover WJ, 2000, ‘A comparison of three

methods for selecting values of input variables in the analysis of output from

a computer code’, Technometrics vol.42, no.1, pp.55.

References

Md Abdul Aziz 174

116. Melching, C. S., 1995, ‘Reliability estimation’, Computer models of

watershed hydrology, V. P. Singh, ed., Water Resources, Littleton,Colo.,

pp.69–118.

117. Metcalf and Eddy, 1991, ‘Wastewater Engineering, Treatment, Disposal

and Reuse’, Third Ed., Boston, MA: McGraw-Hill, Inc. pp. 448-451.

118. Moffa, P.E. 1997, ‘Scarborough CSO project-review of CSO screen

application’, report prepared for the City of Scarborough, Municipality of

Metro, Toronto, ON.

119. Myers, R., 1990, Classical and modern regression with applications (2nd

ed.). Boston: Duxbury Press.

120. National Rivers Authority, 1993, General guidance note for preparatory

work for AMP2, Version 2, October, UK.

121. Natural Resource Management Ministerial Council 2004, ‘National Water

Quality management strategy’, Guidelines for sewerage systems bio solids

management, Australia.

122. Nelson, M. and Illingworth, W., 1991, ‘A practical guide to neural nets,’

ed. United States.

123. Newman, T.L., Leo, W.M., & Gaffoglio, R., 2000, ‘Characterization of

urban-source floatables. In collection system wet weather pollution control

conference’, WEF; Rochester, New York, May 2000.

124. Olsen, N. R. & Kjellesvig, H. M., 1998, ‘Three-dimensional numerical flow

modelling for estimation of spillway capacity’, J. Hydr. Res., Delft, The

Netherlands, Vol. 36, no.5, pp. 775-784.

125. Okamoto, Y., Konugi, M. & Tsuchiya, H., 2002, ‘Numerical Simulation of

the Performance of a Hydrodynamic Separator’, 9ICUD conference, Portland,

Oregon, USA.

126. Pannell, J.C., Daniell, T.M., Walker,D.J., 1996, ‘Rainfall Based Real-Time

flood forecasting using artificial neural networks’, 23 Hydrology and Water

Resources Symposium, Hobart, Australia, 21-24 May.

127. Parsinejad, M., Feng, Y., & Ghanbarian, B., 2006,‘Sensitivity Analysis of

Hydraulic Parameters in the Simulation of Unsaturated Soil Water Dynamics’,

Iran Agricultural Research, Vol. 24, no 2 and Vol. 25, no 1.

128. Patankar, S.V. & D.B. Spalding, 1972, ‘A Calculation Procedure for Heat,

Mass and Momentum Transfer in Three-Dimensional Parabolic Flows.’ Int. J.

Heat. Mass. Trans., vol.15,pp.1787-1806.

References

Md Abdul Aziz 175

129. Phillips, D.I., 1999, ‘A new litter trap for urban drainage system’, Water

Science and Technology, vol.39, no.2, pp.85-92.

130. Phillips, D. I. & Simon, M., 2007, ‘A sewage solids screening system for

CSO chambers. Novatech’, Proc 6th International Conference, Sustainable

techniques and Strategies in Urban Water Management, Lyon, France, Vol 2,

pp. 697-704.

131. Phillips, D.I., Simon, M., 2010, ‘An improved method of screening sewer

solids during CSO events.’ proceedings of the 7th International Novatech

conference, Lyon, 2010, abstracts pp.175.

132. Pollert J., 1999, ‘Free surface modelling in sewer system’, In Proceedings

of the 5th Fluent User Conference, Prague, Czech Republic, pp. 61-69.

133. Pollert J, Stransky D., 2003, ‘Combination of computational techniques –

evaluation of CSO efficiency for suspended solids separation’, Water Science

and Technology, vol.47, no.4, pp.157–166.

134. Prechelt, L., 1998, ‘Automatic early stopping using cross validation:

quantifying the criteria,’ Neural Networks, vol. 11, pp. 761-767.

135. Raduly, B., Genaey, K.V., Capodaglio, A., Mikkelsen, P., & Henze, M.,

2007, ‘Artificial neural networks for rapid WWTP performance evaluation:

methodology and case study’, Journal of Environmental Modelling and

Software, vol. 22, no.8, pp.1208-1216.

136. Riedmiller, M. & Braun, H. 1993, ‘A direct adaptive method for faster

backpropagation learning: the RPROP algorithm,’ in IEEE International

Conference on Neural Networks, vol.1, pp. 586-591.

137. Rumelhart, D.E., Hinton, G.E., & Williams, R. J. 1986, ‘Learning internal

representations by error propagation’, Parallel distributed processing, Vol. I,

MIT Press, Cambridge, Mass., 318-362.

138. Rumelhart, D. E., 1990, ‘Brain style computation: learning and

generalization,’ in An introduction to neural and electronic networks, ed:

Academic Press Professional, Inc., pp. 405-420.

139. Saltelli A., 2004, ‘Sensitivity analysis in practice: a guide to assessing

scientific models’, Wiley, Hoboken, NJ.

140. Sandhu, N., and Finch, R. 1996, ‘Emulation of DWRDSM using artificial

neural networks and estimation of Sacramento River flow from salinity’, Proc.,

North American Water and Environment Conference, ASCE, New York,

pp.4335-4340.

References

Md Abdul Aziz 176

141. Sarle, W.S 1995, ‘Stopped Training and Other Remedies for Overfitting’

in Proceedings of the 27th Symposium on the Interface of Computing Science

and Statistics, pp.352-360.

142. Sato, Y. & Sekoguchi, K. , 1975, ‘Liquid Velocity Distribution in Two Phase

Bubble Flow.’ Int. J. Multiphase Flow,. Vol.2.no.1, pp.79-95.

143. Saul AJ, Ellis DR., 1992, ‘Sediment deposition in storage tanks’, Water

Science and Technology, vol. 25, no.8, pp.189-198.

144. Saul, A.J., 1998, ‘CSO state of the art review: a UK perspective.’ UDM’98,

Fourth International Conference on developments in urban drainage

modelling. September, London, UK.

145. Saul A J, 2000, ‘Screen Efficiency (Proprietary Designs)’, UK Water

Industry Research Limited (UKWIR), Report 99/WW/08/5.

146. Saul, A.J., 2003, ‘CSO: State of the Art Review’ Urban Water

Management: Science Technology and Science Delivery, NATO Science

Series, vol.25, pp. 179-190.

147. Saul, A J, Blanksby, J., 2007, ‘CSO Aesthetics and Static Screening

Technologies’, World Environmental and Water Resources Congress:

Restoring our natural Habitat.

148. Saul, A. J., (2008). CSO: State of the Art Review, ICUD Conference

Edinburgh.

149. Science Clarified, 2015, ‘Mind Versus Metal’,

http://www.scienceclarified.com/scitech/Artificial-Intelligence/Mind-Versus-

Metal.html

150. Sharpe DE and Kirkbride TW, 1959, ‘Storm Water Overflows: The

Operation and Design of a Stilling Pond’, Proc. ICE, Vol. 13.

151. Smisson, B., 1967, ‘Design, Construction and Performance of Vortex

Overflow’, Institute of Civil Engineers, Symposium on Storm Sewage

Overflow, London, UK, pp.99-110.

152. Smith, B. P., and Andoh, R. Y. G., 1997, ‘New generation of

hydrodynamic separators for CSO treatment’. Proc. 2nd Int. Conf. on the

Sewer as a Physical, Chemical and Biological Reactor, Aalborg, Denmark,

25-28 May.

153. Simon. M., 2003, ‘Vorthbach km 2+335 bis 3+500 Hydraulische

Bestandsuntersuchung and Vorschlag zur Veranderung’, Report prepared for

the Emschergenossenschaft, Essen, Germany.

http://www.scienceclarified.com/scitech/Artificial-Intelligence/Mind-Versus-Metal.html

http://www.scienceclarified.com/scitech/Artificial-Intelligence/Mind-Versus-Metal.html

References

Md Abdul Aziz 177

154. Simon. M. and Phillips, D.I. 2008, ‘The development of a sewer solids

screening system for CSO cAhambers’ 11th International Conference on

Urban Drainage, Edinburgh, Scotland, UK.

155. Starrett, S.K., Najjar, Y.M., & Hill, J.C. 1996, Neural networks predict

pesticide leaching. Proceeding American Water and Environmental

Conference, ASCE, New York, pp.1693-1698.

156. Stephenson, J., Gall, B., Mroczek, C., Newbigging & Parker, M. J, 2002,

‘Assessment of Technologies for screening, floatable control, and screening

handling’, Water Environment Research Foundation

157. Stelling, G.S. 1984, ‘On the construction of computational methods for

shallow water equations’, Rijkswaterstaat communication No. 35.

158. Stovin VR., 1996, ‘The prediction of sediment deposition in storage

chambers based on laboratory observations and numerical simulation’. Ph.D.

thesis, University of Sheffield

159. Stovin VR, Saul AJ., 1998, ‘A computational fluid dynamics (CFD) particle

tracking approach to efficiency prediction’ Water Science and Technology,

vol.37, no.1, pp.285–293.

160. Stovin, V., R., Saul, A., j., Drinkwater, A. & Clifforde,I.,1999,’ Field testing

CFD-based predictions of storage chamber gross solids separation

efficiency’, Water Science and Technology, vol.39, no.9,pp.161-168.

161. Stovin VR, & Saul AJ., 2000, ‘Computational fluid dynamics and the

design of sewage storage chamber’, J. CIWEM, vol.14, no.2, pp.103–110.

162. Sutton,R.S.,1984,‘Temporal credit assignment in reinforcement learning,’

8410337 Ph.D., University of Massachusetts Amherst, Ann Arbor.

163. Svejkovsky, K., & Saul, A.J., 1993, ‘Computational modeling of the

stormKing™ hydrodynamic separator using 3D mathematical model fluent’,

Department of Civil and Structural Engineering, University of Sheffield, UK.

164. Thinglas T. & Kaushal. D.R., 2008, ‘Three-Dimensional CFD Modeling for

Optimization of Invert Trap Configuration to Be Used in Sewer Solids

Management.’ Particulate Science and Technology, vol.26, pp.507-509.

165. Thomson. B., 2012, Combined sewer overflows (CSOs) and CSO

screens

166. Tyack, J.N., Hedges, P.D., & Smisson, R.P.M., 1992, ‘The Use of Sewage

Settling Velocity Grading in Combined Sewer Overflow Design’. NOVATECH

92, International Conference on Innovative Technologies in the Domain of

Urban Water Drainage, Lyon (France), November 3-5.

References

Md Abdul Aziz 178

167. Tyack, J. N. & Fenner, R. A., 1999, ‘Computational fluid dynamics

modelling of velocity profiles within a hydrodynamic separator’, Water

Science and Technology, vol.39, no.9, pp.169-176.

168. U.S. Army Corps of Engineers, Waterways Experiment Station,

Vicksburg, Miss., 1952, Corps of Engineers Hydraulic Design Criteria,

prepared for office of the chief of engineers, revised in subsequent years

169. US Army Corps of Engineers, 1970, Corps of Engineers Hydraulic Design

Criteria, prepared for office of the chief of engineers, waterways experiment

station, Vicksburg, MS, USA.

170. U.S. Army Corp of Engineers, 1990, ‘Hydraulic design of spillways’, EM

1110-2-1603, Dept. of the Army, Washington, D.C.

171. Versteeg, H.K. and W. Malalasekera, 2007, ‘An introduction to

Computational Fluid Dynamics.’ Second ed2007: Pearson Education Limited.

172. Werbos, P. 1994, ‘Beyond regression: new tools for prediction and

analysis in the behavioral science,’ PhD dissertation, Harvard University,

Cambridge, Mass.

173. Weyand, M., 2002, ‘Real-time control in combined sewer systems in

Germany-some case studies’, Journal of Urban Water, vol.4, pp.347-354.

174. Willems, P., & Berlamont, J., 1999, ‘Probabilistic modelling of sewer

system overflow emissions’, Water Sci. Tech. vol.39, no.9, pp.47-54.

175. Wu, H. & Shapiro, J. L., 2007, ‘Parameter cross-validation and early-

stopping in univariate marginal distribution algorithm,’ in Proceedings of the

9th annual conference on Genetic and evolutionary computation, ACM, pp.

632-633.

176. Yeh, K. C., & Tung, Y. K., 1993, ‘Uncertainty and sensitivity analyses of

pit-migration model’, Journal of Hydraulic Engineering, vol.119 no.2, pp. 262–

283.

177. Yulei, J., Zur, R. M., Pesce, L. L. & Drukker, K., 2009, ‘A study of the

effect of noise injection on the training of artificial neural networks,’ in

International Joint Conference on Neural Networks (IJCNN), pp. 1428-1432.

Appendix A: Experimental Data

Md Abdul Aziz 179

Appendix A:

Experimental

Data


Md Abdul Aziz 180

Comb Separator testing program, Swinburne University of Technology

Run 5 Monday PM 9th August 2010

Test results and conclusions:

Test 1.

Crest length reduced to 510mm due to incomplete combs and backup.

3 No combs.

Spacing’s as in above diagram.

Flow 27L/s, equivalent to 53L/m/s. Cleared retention screen satisfactorily.

Test items: 20 No strips of toilet paper

20 No cigarette filters

No bottle tops

1 No drink can.

Test run commenced 3.07PM, completed 3.18PM.

Captured:

20 No strips of toilet paper or 100%

6 No cigarette filters or 20%

4 No bottle tops or 100%

1 No drink can or 100%.

Test 2

Maximum flow test 32.7L/s.


Md Abdul Aziz 181

Conclusions:

Maximum attainable flow probably 35L/s before backing up drowns retention

screen.

Perfect results again for toilet paper, bottle tops and drink cans.

Slight blinding of the retention screen noted, about 5%.

The drink can readily passed over and down between the weir and first comb.

Cigarette filter capture rate still low.

Comments:

Maximum equivalent flow possible will be about 80L/m/s in present configuration,

or the once-a-year Vorthbach CSO overflow.

The present configuration is continuing to give excellent toilet paper capture

rates.

The cigarette filters will be enlarged to represent cigarette butts and retested but

may not result in a significant capture rate improvement.

Blinding of the retention screen is negligible.

Drink cans will be captured, but may not pass through the valve, unless valve

clearances are increased.

The retention screen was tilted forwards by 25mm without any problems, as the

flush-water chamber is full to above the exit weir crest level. It enhanced the

screen cleaning by the impacting nappe.

The retention screen, for some distance on both sides of the ball valve, should

be solid (no holes) for a height of about 100mm. This would improve flushing by

forcing flows less than 100mm deep, ie, towards the end of flushing, to pass

around the ends of the retention screen. This measure would obviate the need

for segment walls along the filtered water chamber and the possibility of solids

build-up on them.

The tilting forward of the retention screen allows for a larger and heavier ball to

be used (refer to above figure).

Future tests:

To improve cigarette butt capture, the spacing of the wires in the combs could be

reduced to 20mm centre to centre, thus reducing the effective gap from 3.5mm to 2mm.

This could be tried later and its effect on the nappe checked.


Md Abdul Aziz 182

Conclusion:

There is little further progress to be made by laboratory testing alone, and the

development of the comb separator has reached the stage where testing in real

situations is essential.


Md Abdul Aziz 183


Present: Dr D Phillips; Mr A Aziz

Run 6 Monday PM 23rd August 2010

Test 1.

The comb screens were carefully straightened and aligned so as to accurately

overlap.

Crest across full length of test box ie, 970mm. Configuration as previously.

3 No full length combs straightened and fixed in position.

Spacing’s as in above diagram.

Flow 45L/s. Cleared retention screen satisfactorily.

Test items:


Test run commenced 12:20PM, completed 12:30PM.

Capture:

12 No cigarette filters or 60%

Comments:

The artificial butts were a single filter wrapped in a tape cover and smaller than actual

butts.

The flow now passes over the weir at right angles with the full length 970mm crest.

Test 2

Real 20 butts were collected at random and their mean diameter and lengths

measured and recorded. The mean of these were calculated as, width =

7.375mm and length, = 35.75mm.

Test flow 45L/s

Other conditions as per test 1


Md Abdul Aziz 184

Test items

20 representative butts to the above mean dimensions were prepared

Test run commenced 2:18PM, completed 2:24PM

Capture rate:

17 captured, 3 passed or 85%

Test 3

Repeated the above Test 2 at 40L/s with a 75% capture efficiency.

This result is invalid due to rupture of valve housing seal, leading to loss of 4 cig butts.

Comments

Valve housing to be sealed off and tests repeated later.


Md Abdul Aziz 185

Run 7 Wednesday 25th August 2010

The valve housing was sealed off, pressure tested, and found to be secure.

The interception screen was refixed with its top 150mm only, instead of 160mm

from the back of the weir.

30 plastic bottle tops of varying sizes were obtained from the recycling bins.

Test1

Commenced 4:54PM and finished at 5:06PM:

Flow-rate, 45L/s.

Test items Number in Number retained capture efficiency%

Dish wipes 20 20 100

Assorted bottle tops 20 20 100

Tampons 20 20 100

Balloons 20 20 100

Ersatz cig butts (t1) 20 11 55

Ersatz cig butts (t2) 9 6 67

Comments:

The capture efficiency for all of the above items is 100% with the exception of cigarette

butts that averaged about 61%.

The test for cigarette butts used relatively dry butts and so, to better represent actual

conditions, will be repeated after allowing them to soak for some time.

Blinding of the retention screen was zero with the full width weir crest nappe now

impacting along the length of the retention screen.

Bringing the top of the interception screen to within 150mm of the back of the weir, and

within 40mm of the last comb, did not appear to influence the results.


Md Abdul Aziz 186

Future tests:

To improve cigarette butt capture, the spacing of the wires in the combs could be

reduced to 20mm centre to centre, thus reducing the effective gap from 3.5mm to 2mm.

As a first step the front comb could be so modified, as it operates in the sub-

critical zone and so should not affect the hydraulics of the nappe.

Conclusion:

The full weir produced overflow normal to the weir and maintained all captured

materials in slow motion in the holding chamber without any suggestion of adherence to

the retention screen. Some improvement in cigarette butt capture may result from prior

soaking.


Md Abdul Aziz 187

Swinburne University of Technology

Comb Separator testing program

Wednesday PM 2nd September 2010

Present: Dr D Phillips; Mr A Aziz.

All these tests were conducted using the full length weir of 970mm and tree combs.

Test 1. Cigarette butts

Both real and substitute butts were soaked in water for 24 hrs. The 20 butts of

each type were then drained and collectively weighed.

The substitute butts were squeezed and trimmed until they weighed almost the

same as the real butts.

Test items: 20 No substitute cigarette filters

10 No real cotton balls

29 No real cotton buds

The latter two items were added at request of Herr M Simon.

Test run commenced 10:58AM, completed 11:30AM.

The ruler attached to the box read 65mm below the top of box so that the head on the

weir was 150-65 = 85mm. Francis formula gave an overflow of 45L/s. The flow meter

mass flow over one minute gave 2.56 cu m or 43L/s.

Air bubbles entrained in the flow caused major turbulence in the box. A test

showed that the air was entraining as the flow fell from the overhead tank.


Md Abdul Aziz 188

Capture rates:

The technique adopted was to readmit to the test item port, all those items that

had escaped capture. This was repeated until all items had been captured.

Cigarette butts

1st admission; 12/20: 1st repeat7/8: 2nd repeat 1/1

Hence capture efficiency = 20/29 or 69 per cent.

Cotton balls

1st admission 10/10. Hence capture efficiency 100 percent.

Cotton buds

1st admission; 6/20: 1st repeat 9/14: 2nd repeat 3/5: 3rd repeat 2/2.


Comments:

The artificial butts were trimmed in length to match the weight of the real butts

and so were a little shorter than the 35.75mm representative length, contributing to their

escape. It was noted that the cotton buds were very buoyant and that the cotton balls

ended up as single cotton mass in the holding chamber.

Test 2

The flow was calculated as before as 45L/s and read on the meter as 41.5L/s.

Test items, 20 cotton buds and 20 cigarette butts. All had been left soaking in water

between tests.

Test procedure conducted as a previously.

Testing commenced at 1:15PM and finished around 1:30PM.


Md Abdul Aziz 189

Capture rates:

Cigarette butts

1st admission; 14/20: 1st repeat3/6: 2nd repeat 3/3


Cotton buds

1st admission; 13/20: 1st repeat 6/7: 2nd repeat 1/1.


Comments:

It was noticeable that while five of the cigarette butts still floated, all of the cotton

buds remained highly buoyant.

The other observation was that while the cigarette butt capture efficiency

remained the same, the cotton bud capture both within and between the two tests

dramatically increased.

Test 3

For the third test; to better represent real world conditions, all 40 items were kept

submerged for 45 minutes and the 20 substitute butts squeezed under water to expel

trapped air. This reduced their buoyancy without affecting their weight.

For the third and final test of the current series, the following items were tested:

20 No wipes (toilet paper), 20 No tampons, 20 No balloons (condoms), 20 No drink

bottle tops (various sizes), 20 No cigarette butts (representative), 20 No cotton balls, 20

No cotton buds. The 140 items were pre-soaked.

The flow was again adjusted to 45L/m/s, the maximum currently possible.

Testing commenced at 2:15 and finished at 2:40PM.


Md Abdul Aziz 190

The items were randomly added one at a time over about six minutes.

Test items No 1st cap 2nd cap. 3rd cap No repeats % trapped

Wipes 20 20 20/20 100

Tampons 20 20 20/20 100

Balloons 20 20 20/20 100

Assorted bottle tops 20 19 1 20/21 95

Subst. butts 20 14 6 20/26 77

Cotton balls 20 20 20/20 100

Cotton buds 20 16 1 3 20/27 74

Total 140/154 92.3

Comments:

As in previous tests, the capture efficiency for the wipes, tampons, balloons and

bottle tops was virtually 100 per cent.

The capture efficiencies for cigarette butts and cotton buds increased with

subsequent tests, presumably as they became saturated.

The overall capture efficiency of the above 140 items was 92.3% with that of

cigarette butts and cotton buds 77% and 74 % respectively.

The severe turbulence, due to the entrained air at one end of the overflow

chamber, tended to toss some of these lighter items over the weir, thus reducing their

chance of capture and so contributed to the lower, though acceptable, capture rates.

Blinding of the retention screen was negligible with the full width weir nappe

impacting along the full length of the retention screen.


Md Abdul Aziz 191

Future testing:

This is the last test run pending further modifications to the rig and model. These

include a 100mm ball valve, a captured solids strainer basket at the tank and the drilling

and tapping of 3mm holes, 5mm apart, in two of the comb holders. This will allow the

testing of different comb wire spacing’s.

Conclusions:

The results of the latest series of full-length weir tests indicate that, at the tested

flow rate, the three-comb configuration produces high capture rates for all the items

tested.

The test flow-rate was similar to a typical once-yearly peak overflow for Europe.

Further testing will be conducted with a two-comb configuration and a varied per

metre flow-rate to optimize the performance and economy of the comb separator.


Md Abdul Aziz 192


Supplementary tests conducted Thursday 16th September 2010

Present: Dr D Phillips and Abdul Aziz

Test setup:

Two comb screens, wires 15mm centre to centre, combs 25mm centre to centre.

Front comb 75mm from crest of weir, or 63mm from back of weir.

Retention screen 120mmm behind back of weir. Concrete and brick blocks placed

on floor of overflow chamber to better distribute flow.

Crest full length of test box ie, 970mm. Configuration as per tests 09092010.

Test 1.

The 2 No comb screens were checked, accurately overlapped and fixed

in position.

Flow 15L/m/s. Cleared retention screen satisfactorily.

Test items:


20 No cotton Q buds

Test run commenced 11:48AM, Finished at 12:15PM.

Capture:

Cigarette butts

1st pass 2/20, 2nd pass 0/2. 4No butts remained in overflow chamber.

Hence efficiency = (20-4)/ (22-4) x100 = 91%.

Cotton Q buds


Md Abdul Aziz 193

1st pass 2/20, 2nd pass 0/1. 1 discarded, 1 remained in chamber.

Hence efficiency = (20-1-1)/ (21-1-1) x100 = 95%.

Comments:

The discarded cotton bud had lost a bud, reducing it to a diameter of about 3mm

Test 2

Test flow 30L/m/s

Other conditions as per test 1

Test items

22 No cigarette butts

Test run commenced 1:30PM, Finished 1:45PM

Capture:

1st pass 4/22, 2nd pass 3/4, 3rd pass 0/3. 1 remained in overflow chamber

Hence efficiency = 21/ (22+4+3-1) x100 = 75%.

Test 3

Test flow 20L/m/s.

Other conditions as per Test 1

Test items




Md Abdul Aziz 194

Capture:

1st pass 4/22, 2nd pass 0/4,

Hence efficiency = 22/ (22+4) x100 = 85%.

Comments:

Capture efficiency of cotton Q buds was 95 per cent at 15L/m/s, the only flow

tested.

Capture efficiency of cigarette butts decreased with increasing flow.

At the Vorthbach MWSS six-monthly flow of 30L/m/s, the capture rate was 75

per cent, indicating that a two-comb setup would only achieve some 58 percent

capture at the once-annual overflow of 70L/m/s. (See graph below).

This appears inadequate and suggests that a three-comb setup would be

required to achieve acceptable capture rates at higher flows.

Proposal for further testing:

A three-comb set-up for testing cigarette butt capture at higher flows, consisting

of wires at 20m centre to centre within combs and combs spaced 20mm behind

one another. The effective gap would be 3.33mm compared to 4.50mm for the

above two comb arrangement.


Md Abdul Aziz 195


High flow 2-comb tests, Monday 27th September 2010


Aim:

To test the efficiency of the two-comb 15mm centre to centre set-up, for

cigarette butts at higher per metre flows based on extrapolation of the supplementary

test results 160902010.

Test setup:

Two comb screens, wires 15mm centre to centre, combs 25mm centre to centre.


Retention screen 120mmm behind back of weir.

Concrete and brick blocks placed on floor of overflow chamber to distribute flow.

Crest length reduced to 460mm.

Test 1.

The 2 No comb-screens were checked, accurately overlapped and fixed in

position.

Head on weir 120mm giving a flow of 35L/s or an equivalent flow of 76.2L/m/s.

Nappe easily cleared retention screen.

Test items:

22 No cigarette artificial butts

12 No wipe clothes (toilet paper)



Md Abdul Aziz 196


Cigarette butts

1st pass 13/22, 2nd pass 3/13, 3rd pass 2/3, 4th pass 0/2. 1No butt not found.

Hence Ѯ = ((22-13)+(13-3)+(3-1)+(2-2)) / (22+13+3+2) = (21/40-1) x100

= 55%.

Wipe clothes

1st pass 12/12.

Hence efficiency = (12)/ (12) x100 = 100%.

Comments:

The missing butt may be stuck under a brick that was knocked over by flow. To

be checked.

Test 2

Head on weir 110mm giving equivalent flow of 67L/m/s


Test items



Capture:

1st pass 11/21, 2nd pass 3/10, 3rd pass 2/3, 4th pass 0/2.

Hence Ѯ = 21/ (21+11+3+2+1) x100 = 57%.

Comments:

Capture efficiency of cloth wipes was 100 per cent at 76L/m/s, the only flow

tested, predicting perfect toilet paper capture efficiencies up to the Vorthbach Q1

overflow.


Md Abdul Aziz 197

The tests confirmed that the capture efficiency of the 2-comb set-up for cigarette

butts decreases with increasing flow to unacceptably low values.

This allows the extrapolated plot of supplementary test results 160902010 to be

adjusted and completed for cigarette butt capture with a 2-comb set-up. (See

graph below).

This appears inadequate and suggests that a three-comb setup would be

required to achieve acceptable capture rates at higher flows.


A three-comb set-up for testing cigarette butt capture at higher

flows, consisting of wires at 20m centre to centre within combs and combs

spaced 20mm behind one another. The effective gap would be 3.33mm

compared to 4.50mm for the above two comb arrangement.


Md Abdul Aziz 198


3-comb tests, Monday 11th October 2010


Aim: To test the capture efficiency of three-combs for cigarette butts.

Test setup:

Three comb screens, wires 20mm centre to centre, combs 25mm centre to

centre.


The comb screens were checked, accurately overlapped and fixed in position.

Retention screen 120mmm behind back of weir.

3rd comb 30mm in front of retention screen.


New, 100mm dia. ball valve installed

Test 1.

Head on weir 113mm giving an equivalent flow of 71.4L/m/s.


Test items:

20 No cigarette artificial butts

Test run commenced 3:15PM, Finished at 3:30PM.



Hence Ѯ = 19/38 = 50%.

Comments:

Most of the day was spent sealing the new valve housing, leaks around model

sewer chamber and fixing retention screen after causing initial test to be abandoned.

Test 2

Head on weir 80mm giving equivalent flow of 41.5L/m/s



Md Abdul Aziz 199

Test items



Capture:


Hence Ѯ = (17/33) x100 = 52%.

Test 3


Other conditions as per Tests 1 and 2.

Test items



Capture:

1st pass 4/16, 2nd pass 1/4.

Hence Ѯ = (15/20) x100 = 75%.

Comments:

Two butts remained in model sewer chamber due to the low flow.

Conclusions:

The tests surprisingly showed that despite using three 20mm centre to centre

combs, the capture efficiency of cigarette butts remained unacceptably low.

The tests confirmed previous observations that capture efficiency decreases

with increasing flow rate.

Further testing:

The unexpectedly poor results showed that the mean gap of 3.67mm was

ineffective at intercepting 7mm dia. cigarette butts, suggesting that another

approach to this problem is needed.

The next tests will return to a two comb set-up with more closely spaced wires.


Md Abdul Aziz 200


2 comb tests, Wednesday 13th October 2010

Present: Dr D Phillips, Dr M Imteaz, Herr M Simon, Mr A Aziz.

Aim: To test the cigarette butt capture efficiency of a two-comb close-wire

arrangement.

Test setup:

Two overlapped comb screens, 1st with wires 12.5mm centre to centre, 2nd

with wires 15mm centre to centre, combs 20mm centre to centre.

Front comb 70mm from crest of weir.

Retention screen 140mmm behind crest of weir.

Concrete block placed on floor of model sewer chamber to distribute flow.


Retention screen screwed to floor.

Test 1.



Test items:

18 No artificial cigarette butts

20 No wipe clothes (toilet paper)+ 3 quarter pieces

10 tampons

10 bottle tops

19 cotton buds



Cigarette butts

1st pass 1/16, 2No butt in overflow chamber. Hence Ѯ = (15/16) x100 = 94%.

Wipe clothes

1st pass 0/23. Hence Ѯ = (23/23) x100 = 100%.

Tampons


Md Abdul Aziz 201

1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.

Bottle tops

1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.

Cotton buds

1st pass 6/19. Hence Ѯ = (13/23) x100 = 63%.

Comments:

The high efficiencies of both the larger and smaller items are consistent with previous

results.

Actual sheets of toilet paper were tested with virtually 100 per cent captured. The paper

tended to disintegrate in the holding chamber but readily passed to the outlet valve.

Test 2



Test items:


20 No quarter pieces of wipe cloth

10 tampons

10 bottle tops

Test run commenced 12:12PM to 12.23PM, Finished 12:35PM


Wipes, 1st pass 0/20. Hence Ѯ = (20/20) x100 = 100%.

Tampons, 1st pass 1/10. Hence Ѯ = (9/ 10) x100 = 90%.

Bottle tops, 1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.

The flow was increased to 58L/m/s with a head on the weir of100mm.

Cigarette butts

1st pass 7/15, 3No butts in model sewer chamber. Hence Ѯ = (8/15) x100 = 53%.

Comments:


Md Abdul Aziz 202

Capture efficiency of cloth wipes, tampons and bottle tops was very high but

that of cigarette butts was low and the same as for the previous tests 11102010 at the

same flow.

Test 3:

The combs were reversed with the 15mm spaced comb 60mm from weir crest

and the 12.5 mm spaced comb 15mm further behind.

Head on weir 110mm giving an equivalent flow of 67L/m/s


Test items:



10 cotton buds


Capture:

Cigarette butts, 1st pass 8/20, 2nd pass 2/8, 3rd pass 2/2, 4th pass 1/2.

Hence Ѯ = (19/32) x100 = 59%.

Wipes, 1st pass 2/18, 2nd pass 0/2. Hence Ѯ = (18/20) x100 = 90%.

Cotton buds, 1st pass 7/10, 2nd pass 5/7, 3rd pass 4/5. Hence Ѯ = (6/22) x100 =

27%.

The flow was then reduced to determine the minimum overflow to pass over the

retention screen and was found to be about 15L/m/s at an overflow depth of 40mm.

The retention screen crest was 440mm below the weir crest and 140mm downstream

of it.

Conclusions:

Cigarette butt capture improved by reversing the screens but remains

unacceptably low at higher overflows.

Cotton bud capture is also very low at high overflows.

It was observed that the nappe profile was unaffected by the closer spacing of

the wires in the two combs.


Md Abdul Aziz 203

Real toilet paper capture is virtually 100 per cent but disintegrates and binds

with other captured solids.

The various comb arrangements trialled to-date are all highly effective at

intercepting the larger common sewer solids, but are less effective with thin items such

as cigarette butts and cotton Q buds.


From the above tests, the overflow nappe did not appear to be effected by the

closely spaced wires of the two comb arrangement. Hence it appears feasible to

reduce the wire spacing to 9mm, leaving a 6mm clear gap to physically intercept

cigarette butts and cotton Q buds.

A possible arrangement would be to locate this comb immediately downstream

of two 25mm centre to centre wire spaced combs. These would intercept and remove

the larger items that could otherwise foul, or blind the 9mm centre to centre wire

spaced comb.


Md Abdul Aziz 204


Cigarette butt tests, Wednesday 20th October 2010

Present: Dr D Phillips, Mr A Aziz.

Aim: To test the cigarette butt capture efficiency of a two-comb arrangement.

Test setup:

Two overlapped comb screens, 1st with wires 25mm centre to centre, 2nd with

wires 10mm centre to centre, combs 20mm centre to centre.

Front comb 65mm from crest of weir.

Retention screen 140mmm behind crest of weir.

Concrete block placed on floor of model sewer chamber to distribute flow.


Retention screen screwed to floor.

19 No butts wrapped in duct tape giving mean sample diameter of 8.82 mm.

Test 1.



Test items:




Cigarette butts

1st pass, 1/10, 2nd pass 0/1. 2No butts in down pipe. Hence Ѯ = (8/9) x100 = 89%.

Comments:


Md Abdul Aziz 205

It was intended to close the bar spacing of comb 2 to 9mm but time constraints

prevented this for today’s tests. To offset this, the butts were wrapped in duct tape to

increase their diameters accordingly so as to simulate accurately the closer bar spacing.

Test 2



Test items:

19 No artificial wrapped cigarette butts

9 No real butts, mean diameter 8.0mm

Test run commenced 12:35PM. Finished 12:55PM


Wrapped cigarette butts

1st pass, 0/11, 8No butts in model sewer chamber Hence Ѯ = (11/11) x100 = 100%.

Real cigarette butts:

1st pass 5/9, 2nd pass, 2/5, 3rd pass, ½ 4th pass, 1/1 Hence Ѯ = (8/17) x100 = 47%.

Comments:

Capture efficiency of wrapped butts affected by low number that passed over

weir, possibly as they were heavier with the extra wrapping. Real cigarette butt

capture was low. This was because they were not all physically intercepted by

the combs, as some were less than7mm in one dimension, having been

squashed underfoot by the smoker.

Test 3:

Head on weir 110mm giving an equivalent flow of 67L/m/s


Test items:



10 No full pieces of wipe cloth.


Md Abdul Aziz 206


Capture:

Cigarette butts, 1st pass 3/19, 2nd pass 1/3, 3rd pass 0/1.

Hence Ѯ = (19/23) x100 = 83%.

¼ Wipes, 1st pass 1/20, 2nd pass 0/1. 1 on 2nd comb. Hence Ѯ = (20/21) x100 =

95.2%. The wipe caught on the 2nd comb included as captured.

Full wipes, 1st pass 0/10. Hence Ѯ = (10/10) x100 = 100%.

Overall wipes capture efficiency = 96.8%.

Comments: Wrapped cigarette but capture acceptable while wipes capture,

representing different sized toilet sheets, was very good.

Conclusions:

Cigarette butt capture rates are now satisfactory but tests using a 9mm wire

spacing needed to confirm the above results.

Two 25mm combs are needed in front of the 9mm comb to prevent matting.

It was observed that the nappe profile was little affected by the close wire spacing

of the second comb.

Hence a 9mm comb, having a mean clear spacing of 6mm should intercept

virtually all cigarette butts.


Test for cigarette butt capture using two 25mm and one 9mm combs. This is the

optimum setting developed from the experimental set ups. Further updates in this regard

are included in Chapter 4.

a study on a novel sewer overflow screening system

Documents