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

maher.mahmood@uoanbar.edu.iq, 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