introductionv-scheiner.brunel.ac.uk/bitstream/2438/15379/1/main text... · web viewprediction of...
TRANSCRIPT
A Network Level Multi-Input Deterioration Prediction Model for Low and High Capacity Asphalt Roads
Maher Mahmood*, PhD
Lecturer, Civil Engineering Department, Faculty of Engineering, University of Anbar, Ramadi, Iraq
Mujib Rahman, PhD CEng MICE
Senior Lecturer, Department of Civil Engineering, Brunel University, Uxbridge, United Kingdom
Senthan Mathavan, PhD
Visiting Research Fellow, School of Architecture, Design and the Built Environment, Nottingham Trent
University, Nottingham, United Kingdom
*Corresponding Author full name, PhD
[email protected], 009647736389271
Abstract Deterioration prediction model is the vital component of any effective pavement
management system. As the performance of pavement is greatly influenced by climatic
conditions and traffic loading, a generic model may not be the true reflection of actual
deterioration. In addition, despite a positive impact on performance, maintenance activities
are often not included in the prediction model. This paper presents a network level Multi-
Input Deterioration Prediction Model (MID-PM) for flexible pavement specific to four
climatic conditions (wet, wet freeze, dry and dry freeze) and for two classes of roads, high
capacity arterial and low-medium capacity collector, and considers the impact of climate,
maintenance, construction, material properties, age, traffic and surface distress like
cracking. The condition data in the Long-term Pavement Performance Database (LTTP)
were used to determine the changes of Pavement Condition Index (PCI), over a period of
time and then utilised them in a regression analysis. The prediction models showed good
accuracy with high determination coefficient. A sensitivity study showed that whilst the age
of construction, and traffic are largely responsible for pavement deterioration, the area and
length of cracks appeared on the road surface can be effectively used in the prediction
model. Maintenance has positive impact on the model performance, showed in
improvement of PCI. The model is simple and versatile and has the potential to be adopted
to countries with similar climate and traffic conditions.
Keywords: Mathematical modelling, Roads & highways, Management, Pavement deterioration model,
pavement condition index (PCI), pavement management system.
1
2
3456789
10111213
14151617
18
19
2
1 Introduction
Prediction of pavement deterioration at the network level is important for efficient treatment programming,
plan prioritization, and resource allocation. An accurate pavement deterioration model is the model with
the minimum errors in “deterioration prediction”. Whilst accurate and reliable condition data is extremely
important, a good deterioration models should also be capable of incorporating the contribution of the
most important variables such as pavement structure, traffic, and climate (Prozzi & Madanat 2004; Lytton
1987). Many highway authorities have developed pavement deterioration models for their pavement
management systems, some of which are simple and limited in their applications, while others are
comprehensive and suitable for a varied range of applications (Lytton 1987; Haas et al. 1994).
The pavement deterioration models can be categorised into two main groups: deterministic and
probabilistic. The deterministic models predict a specific amount of change in the pavement life, its
distress level, or other measures of its condition, while the probabilistic models expect a different lifetimes
or pavement states distribution of such events (Haas et al. 1994; Lytton 1987). To address uncertainty
and nonlinearity, computational intelligence techniques were also employed in order to predict pavement
deterioration.
As Shown in Table 1, until now, the majority of deterministic such as linear or nonlinear statistical
analysis methods (Fwa & Sinha 1986; Abaza 2004; Al-Mansour et al. 1994; Obaidat & Al-Kheder 2006;
Ahmed et al. 2008; Luo 2013; Kerali et al. 1996; Ningyuan et al. 2001; Prozzi & Madanat 2004) or
probabilistic deterioration models such as Markov chains or Bayesian analysis method at network level
(Bovier 2012; Bandara & Gunaratne 2001; Hong & Wang 2003; Lethanh & Adey 2012; Abaza 2016;
Chipman et al. 2001; Hong & Prozzi 2006; Jiménez & Mrawira 2009; Park et al. 2008; Jiménez & Mrawira
2012; Anyala et al. 2012) are developed to estimate distress progression or to predict overall conditions
without considering all factors contributing to the pavement performance. For example, most models
consider the effect of distresses but not the effect of maintenance, whereas only a limited number of
models consider the maintenance effect but not all major distresses. In addition, although the
computational intelligence techniques have the ability to deal with uncertainty and nonlinearity, there are
limitations on using them in pavement deterioration models because of the huge data set requirement.
Moreover, as volume of the traffic and climate have significant impact on the pavement performance, it is
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
3
useful to take a holistic approach, considering various influential parameters, to evaluate the overall
performance of the pavement.
2 Objectives and Scope
There are few models, as shown in Table 1, where single distress or age of construction was used to
calculate changes in pavement condition index, PCI, to predict the deterioration of asphalt pavement.
PCI, is a widely used numerical condition rating system of road segments within the road network, where
0 is the worst possible condition and 100 is the best. PCI is a simple, less time-consuming and easily
adaptable system which then can be incorporated with other asset management programs. The condition
rating identifies the remaining useful life of an asset and assists with developing rational rehabilitation and
replacement strategies for a particular asset (Al-Mansour et al. 1994; Fwa 2006).
As mentioned earlier, all PCI based models found in the literature used a single distress or age of
pavement as an input variable (Al-Mansour et al. 1994; Kerali et al. 1996; Ningyuan et al. 2001; Prozzi &
Madanat 2004; Bandara & Gunaratne 2001; Hong & Wang 2003; Lethanh & Adey 2012; Abaza 2016;
Chipman et al. 2001; Hong & Prozzi 2006; Park et al. 2008; Anyala et al. 2012; Kaur & Tekkedil 2000;
Alsherri & George 1988). However, pavement experience multiple distresses simultaneously generated
from the action traffic and environmental loading. Therefore, it is important to include the impact of traffic,
and environmental impact in the model. In addition, the PCI of a pavement improves as a result of
maintenance, so it is also important to include the maintenance effect for rational approach to prediction
model.
The objective of this research is therefore to develop a network-level deterministic deterioration model
called the Multi-Input Deterioration Prediction Model (MID-PM) that will evaluate the changes in overall
PCI for flexible pavement over a period of time and subsequently combined the model with optimization
algorithm for future maintenance programming. This paper presents the development of MID-PM models
and their validations. The maintenance part can be found in (Mahmood et al. 2016). The MID PM predicts
PCI deterioration over time by utilising pavement defects, age, future traffic and planned maintenance
year. A separate life cycle maintenance algorithm can be used to incorporate construction information to
determine pavement strengthening requirements.
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
Table 1. A summary of different pavement deterioration models.
Author Model name Model type Traffic Age Distress Construction & properties
M&RClimateeffect
Output
Fwa and Sinha 1986 PSI-ESAL loss curve Deterministic × PSI lossAI-Mansour et al. 1994 Linear maintenance-effect Deterministic × × × Roughness/maintenance
Kerali et al. 1996 Linear rutting model Deterministic × × rut depth
Ningyuan et al. 2001 dynamic prediction model Deterministic × × Performance index/treatment
Prozzi & Madanat 2004 recursive non-linear model Deterministic × × × serviceabilityAbaza 2004 deterministic deterioration Deterministic × PSI
Obaidat & Al-kheder 2005 Multiple regression models Deterministic × × × distresses quantitiesJain et al. 2005 Calibrated HDM-4 model Deterministic × × × × Distress progression
Ahmed et al. 2008 Linear model Deterministic × PCIKhraibani et al. 2012 nonlinear mixed-effects model Deterministic × cracking progression
Luo 2013 auto-regression model Deterministic × × PCRAlsherri & George 1988 simulation model Probabilistic × × × × serviceability index
Bandara & Gunaratne 2001 Fuzzy Markov model Probabilistic × degradation rates/distress
Hong & Wang 2003 nonhomogeneous continuous Markov chain Probabilistic × pavement performance
degradationHong & Prozzi 2006 AASHO Model based Bayesian Probabilistic × × × serviceability loss
Park et. al. 2008 Bayesian distress prediction Probabilistic × longitudinal crackingAmador-Jiménez & Mrawira
2009 Markov chain deterioration Probabilistic × PCI
Amador-Jiménez & Mrawira 2011 Multilevel Bayesian method Probabilistic × × × IRI
Anyala et al. 2012 Hierarchical deterioration model Probabilistic × × × ruttingLethanh & Adey 2013 Exponential hidden Markov Probabilistic × × composite index
Abaza 2014 back-calculation of the discrete-time Markov Model Probabilistic × cracking and deformation
Kaur & Tekkedil 2000 Buzz expert model AI* × × × rut depthChang et al. 2003 & Pan et al.
2011 Fuzzy regression model AI × PSI
Bianchini and Bandini, 2010 Neuro-fuzzy model AI × × × ∆PSIKargah-Ostadi et al. 2010 roughness model based ANN AI × × × × × IRI
Shahnazari et al. 2012 ANN based model, GP based model AI × PCI* Artificial Intelligence
75
76
The LTPP program is the most comprehensive, reliable program that satisfy a wide range of pavement
information needs (FHWA 2012).
The historical condition data from the LTPP was divided into seven subgroups representing four climatic
zones as shown in Figure 1 and two different functional classes of road, low and high-volume roads, in
order to create a prediction model for each subgroup. This paper presents a brief overview on different
deterioration prediction models, followed by the development procedure of the MID-PM model together
with a sensitivity analysis in order to demonstrate the influence of different input parameters on the model
performance. Furthermore, a cross-validation method was used to evaluate the accuracy of MID-PM. It is
important to note that whilst the model is based on LTTP database in USA, the configuration of the model
is simple and versatile, thus can be adopted to any road condition dataset. It is also important to note that
the MID-PM utilizes condition data, traffic and age to calculate changes in PCI at any given time as well
as future PCI. However, assigning minimum PCI value to trigger maintenance activity is dependent on
the on the road hierarchy and minimum PCI threshold set by the authority to meet their level of service
requirements.
3 Data Source: LTPP Database
The Long-Term Pavement Performance (LTPP) database is a comprehensive pavement database
developed as part of the Strategic Highway Research Program (SHRP) (FHWA 2012). It is available
online and permitted for public use with prior registration. The LTPP database contains condition
information collected from visual and/or automated pavement surveys for each pavement section. This
information consists of the performance requirements, for example, ride quality, roughness, skid
resistance, and texture; distresses such as cracking, rutting, patching, and edge deterioration; and
structural conditions such as pavement life (FHWA 2012).
Additionally, in the LTPP program, investigation of the in-service pavement sections continues, in order to
understand the behaviour under real-life traffic loading. These pavement sections are categorized in the
LTPP program as General Pavement Studies (GPS) and Specific Pavement Studies (SPS). GPS
comprises a study series on approximately 800 in-service pavement test sections in all parts of the
United States and Canada, while SPS are studies of particular pavement parameters involving new
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
6
construction, maintenance, and rehabilitation actions. The LTPP database is separated into seven
modules: Inventory, Maintenance, Monitoring, Rehabilitation, Materials Testing, Traffic, and Climatic
(FHWA 2012). For this research, the data of asphalt concrete pavements on granular base (GPS-1) are
adopted.
4 Methodology
4.1 Input Parameters
The main challenge in developing pavement deterioration models is the existence and use of different
factors affecting pavement condition, which need to be considered in model development. These factors
are pavement age, traffic loading, climate effect, initial design and construction, and maintenance effect
(Al-Mansour et al. 1994; Fwa 2006). In this research, the following parameters are considered as inputs
for the deterioration models.
4.1.1 Pavement Age
Asphalt stiffness increases with age, making it brittle and prone to cracking. In addition, adverse
environmental effects and their interaction with traffic loads accelerate because of the ageing process.
The pavement age is measured from the construction date or from the date of last rehabilitation (Fwa
2006; Al-Mansour et al. 1994).
4.1.2 Traffic Load
Typically, the traffic effect on deterioration consists of volume, vehicle type, load repetition, and axle load
type. Therefore, the equivalent single axle load (ESAL) and then the cumulative ESAL are adopted to
address the vehicle and axle load type, traffic volume, and number of repetitions (Al-Mansour et al. 1994;
Fwa 2006).
4.1.3 Pavement Design and Construction
Pavement section design and construction have a significant influence on its performance. In general,
pavement design consists of two main parts: pavement type and asphalt layer thickness. In performance
prediction analysis, all pavement sections should be the same pavement type: flexible pavement or rigid
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
7
pavement (Fwa 2006). Arterial roads have the maximum level of service at the highest speed for the
longest distance compared to collector roads. Therefore, their pavement design and construction are
different from that of collector roads. Since pavement structure and construction have a significant effect
on deterioration, highway functional classification is employed to reflect the structural design variation.
Asphalt pavements on granular base (GPS-1) is selected in this research. This type of pavement consists
of a dense-graded hot mix asphalt concrete (HMAC) surface layer, with or without other HMAC layers,
located over an untreated granular base.
4.1.4 Maintenance and Rehabilitation (M&R)
Maintenance type, frequency, and degree and rehabilitation (overlay) time have a significant effect on
pavement performance. Generally, there are two types or groups of maintenance activities, namely (a)
preventive (periodic) and (b) corrective maintenance. The objective of preventive maintenance is to limit
the deterioration rate of pavement structure, while corrective maintenance could be remedial or
emergency to keep the pavement structure in a serviceable state (Fwa 2006; Haas et al. 1994).
Sometimes, when a pavement contains one or more distresses, it might be sufficiently maintained but
repair costs may be too high and the structural capacity for expected future traffic loads may be
insufficient. Therefore, rehabilitation (overlay) or reconstruction of the pavement is the best solution to
restore or upgrade the pavement to its required condition and serviceability level (Fwa 2006). For this
study, only inlay and overlay rehabilitation options are considered as these are expected to improve
pavement condition. In should be noted that depending on the construction and level of deterioration, the
layer thickness of inlay and overlay will be different and consequently the benefit will be variable. One of
the main limitations of the model is to mathematically incorporate the benefit of preventive maintenance
like crack sealing, patching etc. on pavement performance. For this study, irrespective of the
rehabilitation type and extent, the benefit to the pavement was considered constant. This assumption is
to avoid model complications in the PCI calculation. A further study is underway to quantify the benefit
from different levels of rehabilitation in the model performance.
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
8
4.1.5 Climatic effect
Climatic effect is represented by precipitation quantity and freeze–thaw cycles. The climate is one of the
key contributors to pavement structural distress. Structural damage decreases the load-carrying capacity
of pavement, which leads to failure such as cracking of the pavement (Fwa 2006; Al-Mansour et al.
1994). It is difficult to represent the environmental effects mathematically in the prediction model.
Therefore, the environmental effects are embedded by generating the prediction model for each of the
four climatic zones in the LTPP study area. These are wet freeze, wet non-freeze, dry freeze, and dry
non-freeze zones.
4.1.6 Distress Quantity
Distress is the physical deterioration of the pavement surface such as cracking, potholes, and rutting, and
it is generally but not necessarily visible (Haas et al. 1994). At the project level, pavement performance is
expressed in terms of evaluating the pavement distresses separately, while at the network level, it is
essential to find a composite measure of performance that considers most of the distresses (Litzka 2006).
Based on previous research by the authors (Mahmood et al. 2013), it was found that cracking has the
most severe effect on overall pavement performance or PCI. Furthermore, as shown in Figure 3, the
frequency of occurrence of cracking distress is high compared with other distresses. Therefore, only the
cracking area (alligator, block, and edge) and (longitudinal and transverse) cracking lengths are
considered in developing deterioration prediction models. After collecting and analysing condition data,
the PCI is calculated for each pavement section with the Micro-Paver software.
4.2 Development of Multi-Input Deterioration Prediction Model (MID-PM)
Deterministic models are commonly used to find the empirical relationship between the dependent
variable, which is the distress progression or condition index, and one or more explanatory variables such
as cracking area, age, or ESAL. Subjective indices (skid resistance, ride quality, condition index,
serviceability, etc.) and objective indices (rutting, roughness, cracking, etc.) are utilised as dependent
variables. These performance indices are related to one or more independent variables like structural
strength, traffic loading, and climatic effects (Prozzi & Madanat 2004). The pavement skid resistance
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
9
deterioration and required maintenance is not included in the model. The equations of the predicted PCI
are described as:
PCI=a1+a2 X1+a3 X2+a4 X3+a5 X 4+a6 X5 (1)
where
PCI = Pavement condition index.
X1 = Cumulative ESAL.
X2 = Pavement age.
X3 = Maintenance effect (inlay and overlay thickness).
X4 = Longitudinal and transverse cracking length.
X5 = Cracking area (alligator, edge, and block).
a1 , a2 , a3 , a4 , a5 , a6=¿ Coefficients.
The flow chart of the formulation procedures of empirical pavement performance prediction is presented
in Figure 4.
For the development of empirical models, the collected condition data are separated into four groups,
each representing a different climatic zone (wet freeze, wet non-freeze, dry freeze, and dry non-freeze) to
embed the climatic effect in the prediction model. Then, each group is divided into two subgroups to
consider road functional class (arterial and collector). The five independent variables such as cracking
area, length of crack, pavement age, cumulative ESAL, and maintenance effect (inlay or overlay
thickness) are considered in the development of network-level empirical deterioration models for each
road class. Table 2 shows the number of sections and number of data samples used in the study.
Interestingly, the number of sections in the LTPP database for collector roads was found to be
significantly lower than that for arterial roads, which may affect the accuracy of the model. It can also be
noted that the model was verified by predicting separate data sets, which were not included in the model
development. The data samples of each subgroup consist of pavement condition data for each pavement
sections and also multi-year condition data for each pavement section. Table 3 shows a summary of
pavement condition data samples with all input parameters and PCI for each subgroup.
Table 2. A summary of pavement condition data samples.
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
10
Climatic Zone Road class Number of sections Number of data samples
Wet Freeze Arterial* 13 224Collector** 3 28
Wet non-freeze Arterial 8 91Collector 3 28
Dry Freeze Arterial 22 314Collector 2 24
Dry non-freeze Arterial 8 111*Arterial roads which are primarily for longer-distance high-speed through-vehicle movements and, hence, provide minimal access to adjacent frontages.
**Collector roads, i.e. the 'middle' group, which are intended to provide for both shorter through-vehicle movements and frontage access (O’Flaherty 1997).
Table 3: The average and standard deviation of of data samples.
Climatic zone Road class X1* X2 X3 X4 X5 PCI
Wet Freeze
Arterial
Min. 16 0 0 0 0 0Max. 5207 39.2 6 874.6 516.4 100Mean 1062.3 17.6 0.2 151.4 45.4 73.2
SD 989.5 8.7 0.8 168.1 96.9 25.4
Collector
Min. 9 0 0 0 0 4Max. 849 21.95 6 910.20 215 100Mean 250.5 8.9 0.2 241.7 17.4 71.2
SD 266.6 5.6 1.0 242.1 46.5 24.0
Wet Non-Freeze
Arterial
Min. 8 0 0 0 0 17Max. 22672 37.4 6 1012.5 557.8 100Mean 2900 14.1 0.32 232.6 45.0 79.2
SD 4908 8.5 1.69 253.8 110.1 18.6
Collector
Min. 1 0 0 0 0 9Max. 460 26.8 1.2 281.5 582.2 100Mean 103.75 9.4 0.04 24.9 44.3 82.6
SD 131.07 6.0 0.21 52.5 119.1 23.7
Dry Freeze
Arterial
Min. 12 0 0 0 0 2Max. 6091 39.3 4.8 717.1 453 100Mean 1021.02 15.73 0.12 124.06 15.50 79.31
SD 1238.83 8.83 0.65 129.48 52.38 21.85
Collector
Min. 44 0 0 0 0 19Max. 769 25.9 2.8 468.7 201.2 100Mean 413.36 11.85 0.15 178.18 12.16 72.76
SD 222.30 7.55 0.59 134.61 44.25 19.81
Dry Non -Freeze Arterial
Min. 30 0 0 0 0 12Max. 5437 39.5 2.3 1111.5 364.3 100Mean 801.2 15.5 0.06 108.2 12.2 84.9
SD 852.7 8.9 0.34 169.5 49.5 20.0*(KESAL)
5 Results
5.1 Deterioration Models
Seven network-level deterioration prediction models for flexible pavement were created by using a
205206
207208
209
210211
212
213
214
11
multiple regression analysis technique using the statistical software SPSS (Field 2009). The models and
their corresponding coefficient of correlation (R2) are shown in Table 4. It can be seen that with the
exception of “Wet Non-Freeze Arterial” and “Dry Freeze Arterial”, the models have very good accuracy,
with R2 values greater than 70%. Despite the relatively poor correlation compared to other models, the R 2
values for “Wet Non-Freeze Arterial” and “Dry Freeze Arterial” were still 52 and 62% respectively. These
results indicate a good correlation for network level prediction model.
Furthermore, the results show that empirical models for collector roads have better correlation than
models for arterial roads. This is likely to be because fewer data are available for collector road sections
compared to the arterial roads. Therefore, these points’ data for collector roads satisfy the properties of
linearity, while the data for arterial roads tend to be non-linear. The model behaviour with few data points
tends to be more linear than nonlinear. Therefore, for collector roads, the fewer data points produce a
good linear relationship with little error in contrast to arterial roads, which have more data points.
Table 4. Empirical pavement performance prediction models for each subgroup.
Climatic Zone
Road class Prediction model R2
Wet Freeze
Arterial PCI=97.744−0.15 X5−0.064 X4−0.515 X2+3.748 X3 0.70
Collector PCI=99.872−4.8 X 2+0.026 X3−0.015 X 4−0.581 X5 0.86
Wet non-
freeze
Arterial PCI=93.546−0.175 X 2−0.083 X5−0.038 X4+1.073 X3 0.52
Collector PCI=100.73−0.245 X5−0.954 X2+17.852 X3 0.93
Dry Freeze
Arterial PCI=97.252−0.245 X5−0.074 X4−0.359 X2+2.967 X3 0.62
Collector PCI=94.461−0.37 X2−0.005 X5−0.068 X 4+20.196 X3−0.000015 X10.68
Dry non-freeze Arterial PCI=98.861−0.407 X2−0.235 X5−0.065 X 4+3.404 X3−0.003 X1 0.79
5.2 Cross-Validation
The cross-validation technique is used to assess how well models can predict PCI or to assess the model
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
12
accuracy across various data samples. For each subgroup, 80% of data samples are randomly selected
to create deterioration models. The remaining 20% of data samples for each subgroup are used to
evaluate the accuracy of the empirical models.
As shown in Table 5, the loss of R2 value and mean square error (MSE) value for arterial roads in four
climatic zones is not significant, with reductions in accuracy of less than 20% for the R2 value and 35% for
the MSE value. This means that the deterioration models for arterial roads have a good accuracy in
predicting PCI. However, the loss of R2 value and MSE value for collector roads is especially insignificant
in the wet freeze zone and significant in the wet non-freeze zone, but significant improvement in R 2 of the
dry freeze zone. Figure (5, 6, 7, 8, 9, 10 and 11) show the errors and linear relation in each subgroup.
This means that empirical models for all zones have a good level of accuracy.
Table 5. Validation results of empirical deterioration models for each subgroup.
Climatic Zone Road classLinear regression Cross-ValidationR2 MSE R2 MSE
Wet Freeze Arterial 0.69 205.9 0.59 252.2Collector 0.86 61.03 0.80 274.4
Wet non-freeze Arterial 0.52 181.5 0.45 178.5Collector 0.93 29.085 0.50 313.3
Dry Freeze Arterial 0.62 182.1 0.74 134.0Collector 0.68 117.5 0.90 346.2
Dry non-freeze Arterial 0.79 86.6 0.64 136.2
5.3 Validation
To check the significance of estimates of each parameter (a1,…. a i), the F test and t test were used.
Generally, in multiple regression model, the null hypothesis was applied: H0: a i=0, HA: a i≠0. The t test
for each independent parameter and F test for the overall linear model were estimated as shown in Table
6. The results of F test and t test with probability level (α=0.05) show that the null is rejected in all
empirical models. Therefore, the independent variables (X1, Xi) have influence on dependent variable Y.
However, the test shows the traffic parameter does not have effect on PCI in majority of the models.
Table 6. The results of t-test and F-test.
Climatic zone
Road Class
t-testF-test
Constant X1 X2 X3 X4 X5
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
13
Wet FreezeArterial 38.428 0 -3.91 2.676 -9.948 -12.868 94.811
Collector 27.52 0 -5.278 0.018 -1.076 -2.001 17.705
Wet non-freeze
Arterial 29.281 -0.542 -0.822 1.237 -4.965 -5.627 15.122Collector 45.006 0 -4.632 3.855 0 -13.223 72.115
Dry FreezeArterial 52.999 0 -3.409 2.176 -10.143 -11.156 96.204
Collector 14.392 -0.370 -0.286 1.621 -2.891 -0.094 5.617Dry non-freeze Arterial 49.51 1.809 -2.379 1.229 -6.936 -12.975 66.827
5.4 Sensitivity Analysis
A sensitivity analysis was conducted to study the influence of input variables on the efficiency of empirical
prediction models in the calculation of PCI. The results of sensitivity analysis are shown in Table 7. It can
be seen that the “cracking area” and “cracking length” are the most significant variables for prediction
model performance. The pavement age and maintenance also have an influence on some extent, while
the cumulative ESAL has a minor effect on the model performance.
Table 7. Sensitivity analysis of input variables on prediction
Climatic zone Road ClassR2%
Pavement age
Cracking area
Cracking length
Maintenance effect
Cum. ESAL
Wet FreezeArterial 11 40 30 6 0
Collector 74 30 52 15 0
Wet non-freezeArterial 9 25 25 4 0
Collector 14 78 0 3 0
Dry FreezeArterial 13 40 33 3 0
Collector 23 0.8 56 15 14Dry non-freeze Arterial 12 39 41 2 0.1
6 Conclusion
For a flexible pavement, the network-level deterministic deterioration prediction model MID-PM has been
proposed for different climatic zones and road classes using LTTP database by evaluation the changes
of PCI with time. Whilst the model is developed using condition data from USA, it could be adopted to
other countries with similar construction, climatic condition and road classes. The MID-PM incorporates
changes in PCI for five input variables: the cumulative Equivalent Single Axle Load (ESAL), pavement
age, maintenance effect (inlays and overlay thickness), and length and area of pavement cracks which
have been developed. These results are promising, with 52 to 95% correlation for two classes of roads in
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
14
four climatic regions.
Analysis of the pavement condition database used in this research indicates insignificant contribution of
rutting to pavement performance compared with other defect parameters, and hence rutting was
excluded from the MID PM model. Therefore, the model has limitations in other
environments/geographies where rutting is an important pavement deterioration mode.
It appears that the smaller amount of data tends to give a more linear relation (data points for collector
roads, for example). In contrast, a larger amount of data points tends to produce a non-linear relation
(data points for arterial roads, for example).
The sensitivity analysis shows that the distress quantity (cracking), pavement age, and maintenance
have the greatest effect on the model performance. Moreover, while traffic loading and construction type
has significant impact of generating distresses, they have minimal further influence on the model
performance when distresses are directly used as an input.
The cross-validation results show that the deterioration models for the arterial road (high capacity) class
in all climatic zones have very good accuracy in estimating the future Pavement Condition Index (PCI).
The accuracy level for the collector road (low to medium capacity) class is relatively poor due to the
shortage of historical testing data. The accuracy of these prediction models can be improved by using
nonlinear regression or computational intelligence techniques such as artificial neural networks or fuzzy
inference systems that consider nonlinearity and uncertainty. In addition, comparing the MID-PM
condition projection with other practical models like MicroPaver model and IRI model will be considered
as future work.
7 References
Abaza, K. a., 2016. Back-Calculation of Transition Probabilities for Markovian-Based Pavement
Performance Prediction Models. International Journal of Pavement Engineering, 17(3), pp.253–264.
Abaza, K. a., 2004. Deterministic Performance Prediction Model for Rehabilitation and Management of
Flexible Pavement. International Journal of Pavement Engineering, 5(2), pp.111–121.
Ahmed, N.G., Awda, G.J. & Saleh, S.E., 2008. Development of Pavement Condition Index Model for
Flexible Pavement in Baghdad City. Journal of Engineering, 14(1), pp.2120–2135.
Al-Mansour, A.I., Sinha, K.C. & Kuczek, T., 1994. Effects of Routine Maintenance on Flexible Pavement
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288289
290291
292293
294
15
Condition. Journal of Transportation Engineering, 120(1), pp.65–73.
Alsherri, A. & George, K.P., 1988. Reliability Model for Pavement Performance. Journal of Transportation
Engineering, 114(2), pp.294–306.
Anyala, M., Odoki, J.B. & Baker, C.J., 2012. Hierarchical asphalt pavement deterioration model for
climate impact studies. International Journal of Pavement Engineering, 15(3), pp.251–266.
Bandara, N. & Gunaratne, M., 2001. Current and Future Pavement Maintenance Prioritisation. Journal of
Transportation Engineering, 127(2), pp.116–123.
Bovier, A., 2012. Markov Processes. , pp.1–153.
Chipman, H., George, E.I. & McCulloch, R.E., 2001. The Practical Implementation of Bayesian Model
Selection, Lecture Notes-Monograph Series.
FHWA, 2012. LTPP Standard Data Release.
Field, A., 2009. Discovering Statistics Using SPSS 3rd ed., London: SAGE Publications Ltd.
Fwa, T.F., 2006. The Handbook of Highway Engineering, Boca Raton, FL: Taylor & Francis.
Fwa, T.F. & Sinha, K.C., 1986. Routine Maintenance and Pavement Performance. Journal of
Transportation Engineering, 112(4), pp.329–344.
Haas, R., Hudson, W.R. & Zaniewski, J.P., 1994. Modern Pavement Management 2nd ed., Cornell
University: Krieger Pub. Co.
Hong, F. & Prozzi, J.A., 2006. Estimation of Pavement Performance Deterioration Using Bayesian
Approach. Journal of Infrastructure Systems, 12(2), pp.77–86.
Hong, H.P. & Wang, S.S., 2003. Stochastic Modeling of Pavement Performance. International Journal of
Pavement Engineering, 4(4), pp.235–243.
Jiménez, L.A. & Mrawira, D., 2012. Pavement Deterioration Bayesian Regression in Pavement
Deterioration Modeling : Revisiting the AASHO Road Test Rut Depth Model. Infraestructura Vial,
14(25), pp.28–35.
Jiménez, L.A. & Mrawira, D., 2009. Roads Performance Modeling and Management System from Two
Condition Data Points: Case study of Costa Rica. Journal of Transportation …, 135(12), pp.999–
1007.
Kaur, D. & Tekkedil, D., 2000. Fuzzy Expert System for Asphalt Pavement Performance Prediction. In
IEEE Intelligent Transportation Systems Conference. Dearborn (MI), USA: IEEE, pp. 428–433.
Kerali, H.R., Lwrance, A.J. & Awad, K.R., 1996. Data Analysis Procedures for Long-Term Pavement
Performance Prediction. Transportation Research Record: Journal of the Transportation Research
Board, 1524, pp.152–159.
295
296297
298299
300301
302
303304
305
306
307
308309
310311
312313
314315
316317318
319320321
322323
324325326
16
Lethanh, N. & Adey, B.T., 2012. Use of Exponential Hidden Markov Models for Modelling Pavement
Deterioration. International Journal of Pavement Engineering, 14(7), pp.1–10.
Litzka, J., 2006. Performance Indicators for Road Pavements. In The 1st Transport Research Arena
(TRA) conference. Gothenburg, Sweden.
Luo, Z., 2013. Pavement Performance Modelling with an Auto-Regression Approach. International
Journal of Pavement Engineering, 14(1), pp.85–94.
Lytton, R.L., 1987. Concepts of Pavement Performance Prediction and Modelling. In 2nd North American
Pavement Management Conference. Toronto, Canada: TRB.
Mahmood, M. et al., 2013. A Fuzzy Logic Approach for Pavement Section Classification. International
Journal of Pavement Research and Technology, 6(5), pp.620–626.
Mahmood, M., Mathavan, S. & Rahman, M., 2016. Pavement Maintenance Decision Optimization Using
A Novel Discrete Bare-Bones Particle Swarm Algorithm. Transportation Research Board 95th
Annual Meeting.
Ningyuan, L. et al., 2001. Integrating Dynamic Performance Prediction Models into Pavement
Management Maintenance and Rehabilitation Programs. In 5th International Conference on
Managing Pavements. Seattle Washington, United States: TRB.
O’Flaherty, C., 1997. Transport Planning and Traffic Engineering First edit., Oxford, UK:
BU’ITERWORTH HEINEMANN.
Obaidat, M.T. & Al-Kheder, S. a., 2006. Integration of geographic information systems and computer
vision systems for pavement distress classification. Construction and Building Materials, 20(9),
pp.657–672.
Park, E.S. et al., 2008. A Bayesian Approach for Improved Pavement Performance Prediction. Journal of
Applied Statistics, 35(11), pp.1219–1238.
Perera, R.W. & Kohn, S.D., 2001. LTPP Data Analysis : Factors Affecting Pavement Smoothness,
Prozzi, J. a. & Madanat, S.M., 2004. Development of Pavement Performance Models by Combining
Experimental and Field Data. Journal of Infrastructure Systems, 10(1), pp.9–22.
Figure 1:Map of climate regions in the United States and Canada based on LTPP (Perera & Kohn 2001).
Figure 2: Pavement condition index (PCI) calculation procedure (Fwa 2006).
Figure 3: Occurrence frequency of distress type in all climatic zones.
Figure 4. Flow chart for multi input deterioration prediction model (MID-PM).
327328
329330
331332
333334
335336
337338339
340341342
343344
345346347
348349
350
351352
353
354355
356
357
358
17
Figure 5: The accuracy of empirical deterioration model for wet freeze arterial.
Figure 6: The accuracy of empirical deterioration model for wet freeze - collector.
Figure 7: The accuracy of empirical deterioration model for wet non freeze - arterial.
Figure 8: The accuracy of empirical deterioration model for wet non freeze - collector.
Figure 9: The accuracy of empirical deterioration model for dry freeze - arterial.
Figure 10: The accuracy of empirical deterioration model for dry freeze - collector.
Figure 11: The accuracy of empirical deterioration model for dry non freeze - arterial.
359
360
361
362
363
364
365