Data Analysis in Pavement Engineering: Methodologies and Applications

By Qiao Dong, Xueqin Chen and Baoshan Huang

Data Analysis in Pavement Engineering: Theory and Methodology offers a complete introduction to the basis of the finite element method, covering fundamental theory and worked examples in the detail required for readers to apply the knowledge to their own engineering problems and understand more advanced applications. This edition sees the significant addition of content addressing coupling problems, including Finite element analysis formulations for coupled problems; Details of algorithms for solving coupled problems; and Examples showing how algorithms can be used to solve for piezoelectricity and poroelasticity problems.

Focusing on the core knowledge, mathematical and analytical tools needed for successful application, this book represents the authoritative resource of choice for graduate-level students, researchers and professional engineers involved in finite element-based engineering analysis.

• This book is the first comprehensive resource to cover all potential scenarios of data analysis in pavement and transportation infrastructure research, including areas such as materials testing, performance modeling, distress detection, and pavement evaluation.
• It provides coverage of significance tests, design of experiments, data mining, data modeling, and supervised and unsupervised machine learning techniques.
• It summarizes the latest research in data analysis within pavement engineering, encompassing over 300 research papers.
• It delves into the fundamental concepts, elements, and parameters of data analysis, empowering pavement engineers to undertake tasks typically reserved for statisticians and data scientists.
• The book presents 21 step-by-step case studies, showcasing the application of the data analysis method to address various problems in pavement engineering and draw meaningful conclusions.

• Language: English
• Publisher: Elsevier Science
• Release date: Nov 6, 2023
• ISBN: 9780443159299

Qiao Dong

Dr. Qiao Dong received the B.S. in civil engineering in 2003 and M.S. degree in roadway and railway engineering in 2006 from Southeast University, Nanjing, China and Ph.D. degree in civil and environmental engineering in 2011 from the University of Tennessee, Knoxville, USA. From 2011 to 2016, he was a research associate in the University of Tennessee. He joint Southeast University since 2016 as a Professor. His research interests include pavement asset management based on data analysis and artificial intelligence, pavement distress non-destructive evaluation, pavement materials multiscale characterization and simulation. Dr. Dong has worked on data driven pavement evaluation and management since 2006 and selected pavement data modelling and mining as the topic of his Ph.D. dissertation. He won the first prize in the American Society of Civil Engineers (ASCE) Long-Term Pavement Performance (LTPP) data analysis contest in 2010. He was the PI or co-PI of several related research projects. He has published more than 100 research papers, and more than 30 of them focus the field of pavement data analysis. He is currently an active member of the Bituminous Materials Committee (BMC) of American Society of Civil Engineers, the Pavement Maintenance Committee (AHD20) of Transportation Research Board (TRB) and the Pavement Performance Evaluation Committee of the World Transportation Congress. He served as a young editor for the Journal of Infrastructure Preservation and Resilience and an editor of Coatings.

Book preview

Preface

Data analysis is a process of extracting useful information from collected data using mathematical knowledge and models. It has been widely applied in the fields of medicine, biology, agriculture, finance, education, psychology, engineering, and other disciplines to effectively deal with data from experiments, tests, and surveys. In pavement engineering, many types of data are obtained during the material design, testing, construction, numerical simulations, operation, maintenance, and preservation of pavement infrastructure. To optimize the mix design, quantify the influence of factors, evaluate and predict pavement performance, estimate the effectiveness of maintenance, and optimize maintenance strategies, it is important to analyze the data using effective methods and models based on the data characteristics.

Historically, in the early stages of pavement engineering, data analysis mainly involved conducting statistical descriptions, significance tests, experimental designs, and analysis of variance using a limited amount of laboratory or field test data. With the development of automatic pavement tests of roughness, rutting, friction, and distress, a large quantity of pavement performance data began to be collected and pavement management systems were developed to help manage the pavement infrastructure. Regressions, survival models, and Markov models were developed to predict the pavement performance and analyze the effects of different pavement designs, traffic, etc. Nowadays, with the development of image processing, point cloud object detection, structural response sensing, crowdsourced monitoring, and the Internet of Things, many types of large-size nonstructural data are available for pavement evaluation. Machine-learning algorithms including unsupervised learning methods, supervised learning methods, and deep learning methods have drawn the attention of many researchers and have been used to deal with those data. Data collection, fusion, cleaning, mining, and training will be the keys to the intelligent infrastructure of the next era.

Due to the complex nature of the data in pavement engineering, selecting proper data analysis tools is the key to interpreting the data. This book introduces the theories and methodologies of data analysis in pavement engineering, including the definitions, algorithms, principles, and methods. Recent pavement-related studies utilizing those data analysis methods are summarized and discussed in the introduction part of each chapter. Case studies are presented to help readers understand how to apply those methods to solving problems in the field of pavement engineering. The main contents of each chapter are summarized here.

Chapter 1 introduces pavement performance indicators, pavement performance models, the long-term pavement performance program, and an overview of current applications of data analysis in the field of pavement engineering.

Chapter 2 presents a concise review of the fundamentals of statistics for data analysis, including the definitions of random variables and distributions, descriptive statistics, parameter estimates, and hypothesis tests. A case study of the hypothesis tests is presented.

Chapter 3 focuses on experimental designs and introduces the definitions and methods of the design of experiments, one-way and two-way analysis of variance, and orthogonal designs. Two case studies on the two-way analysis of variance and orthogonal designs and analysis are presented.

Chapter 4 introduces single linear regression, multiple linear regression, ordinary least square and maximum likelihood estimates, linear regression diagnostics, stepwise regression, polynomial regression, and nonlinear regression, among others. Case studies on pavement performance prediction using multiple linear regression, polynomial regression with interactions, and nonlinear regression are presented.

Chapter 5 introduces binary logistic regression, multinomial logistic regression, ordinal logistic regression, and generalized linear regression. A case study on pavement cracking prediction using the logistic regression is presented.

Chapter 6 discusses the Poisson model, the negative binomial model, and zero-inflated count data models. A case study on modeling the development of pavement transverse cracks using the zero-inflated negative binomial model is presented.

Chapter 7 introduces the problem of data censoring, survival functions, and the nonparametric, semiparametric, and parametric survival models. A case study on investigating the survival probability of pavement patching using the parametric survival model is presented.

Chapter 8 discusses the definition, decomposition, smoothing, and autoregressive integrated moving average time series models. Case studies on predicting pavement roughness using three types of time series models are presented.

Chapter 9 introduces the definition of stochastic process, Markov chain, stationarity, and finite-state Markov chain. Two case studies on network-level and project-level pavement performance prediction using the Markov chain are presented.

Chapter 10 focuses on several classic decision tree algorithms and ensemble learning methods, including Hunt’s algorithm, the iterative dichotomiser 3 tree, the C4.5 tree, the classification and regression tree, the random forest, and the gradient boosting tree. A case study on classifying pavement patching serviceability using the classification and regression tree is presented.

Chapter 11 introduces the definition, structure, and training of the neural network and convolutional neural network. Two case studies on pavement performance prediction and climatic region classification are presented.

Chapter 12 discusses the algorithm of the support vector machine, the k-nearest neighbor, and performance measures of machine learning. A case study on the classification of pavement roughness level using the support vector machine, k-nearest neighbor, logistic regression, and decision tree is presented.

Chapter 13 introduces the definition and calculation of principal component analysis. A case study on reducing the dimension of the climatic dataset is presented.

Chapter 14 discusses the model and calculation procedures of factor analysis, factor rotation, and factor score. Two case studies on determining the common factors in pavement climatic data and pavement performance data are presented.

Chapter 15 introduces the definition of cluster analysis, distance metrics, hierarchical clustering, k-means clustering, and variable clustering. A case study on clustering pavement climatic regions is presented.

Chapter 16 introduces the methodology of distance-based discriminant analysis, Bayesian discriminant analysis, and Fisher’s discriminant analysis. A case study on classifying weather stations into climatic regions using linear discriminant analysis is presented.

Chapter 17 discusses the definition, parameter estimates, model fitness, and explicit expression of latent variables of the structural equation model. Two case studies using the structural equation model for pavement performance evaluation and pavement latent condition index development are presented.

Chapter 18 introduces the definition of the Bayesian model, the Monte Carlo method, and the Markov chain Monte Carlo method. A case study on pavement performance modeling using logistic regression integrated with the Markov chain Monte Carlo method is presented.

Building a data model requires an in-depth understanding of the problem, the mathematical principles, and experience. A statistics professor will tell you that the knowledge of your specialty is still the key to building a good model, just like the prior knowledge in the Bayesian methods. It should be noted that data analysis models are effective tools, but the nature of the data determines how good a model will be. Models cannot rescue poor-quality data.

This book can be used as a textbook for undergraduate and graduate students in civil engineering or as a reference book for engineers in the field of transportation infrastructure. Courses in calculus, linear algebra, statistics, and pavement materials and designs are prerequisites. This book covers a wide variety of data analysis practices in pavement engineering but does not attempt to teach all potential tools. Most of the data used in the case studies are from the online database of the long-term pavement performance program (LTPP).

The authors want to thank Dr. Hehua Zhu, an Academician of the Chinese Academy of Engineering from Tongji University, for his advice during the preparation of the book. The publication of this book is supported by the National Laboratory and Education Demonstration Center for Road and Traffic Engineering of Southeast University, which the authors greatly appreciate.

Introduction

This book introduces the theories and methodologies of data analysis in pavement engineering, including the fundamentals of statistics such as distributions of variables, descriptive statistics, parameter estimates, and hypothesis tests; the design of experiments and analysis of variance; traditional linear, polynomial, and nonlinear regressions; the logistic regression, the count data model, survival analysis, time series, and the stochastic process for specific dependent variables; unsupervised machine learning methods for multivariate data such as principal component analysis, factor analysis, and cluster analysis; supervised machine learning methods including decision trees, support vector machine, neural networks, and discriminant analysis; the structural equation model and the Markov chain Monte Carlo method. The definitions, principles, and algorithms of these data analysis methods are also covered. Case studies are presented to help readers understand how to apply these methods to solve problems in pavement engineering. This book can be used as a textbook for undergraduate and graduate students in pavement engineering, bridge engineering, civil engineering, and other related majors of transportation infrastructure or as a reference book for engineers in the field of pavement and transportation infrastructure maintenance and operation.

Chapter 1 Pavement performance data

Abstract

This chapter briefly introduces the background of pavement performance data and data analysis methods, focusing on the type, source, and analyses of pavement performance data. It starts by discussing the background and definitions of various pavement performance indices. It then presents the development of pavement management systems (PMSs), which utilize pavement performance data for pavement maintenance decision-making. Various pavement performance models are discussed that can be used to predict pavement performance indices based on pavement service time, traffic level, pavement structure, materials, and weather data. The Long-Term Pavement Performance program (LTPP) database is also introduced as a famous database for pavement performance and other related data. In addition, the chapter includes an overview of applications of data analysis methods in pavement engineering.

Keywords

Pavement performance index; Pavement management system; PMS; Pavement performance model; Long-term pavement performance; LTPP; Data analysis; Machine learning

Chapter outline

1.1Introduction

1.2Pavement performance indices

1.2.1Development of pavement performance indices

1.2.2Pavement performance indices in China

1.3Pavement management system

1.4Pavement performance models

1.4.1Classic pavement performance models

1.4.2Time-performance models

1.5The LTPP database

1.5.1The LTPP program

1.5.2Asphalt pavement performance data in LTPP

1.6Data analysis in pavement engineering

1.6.1An overview

1.6.2Machine learning methods

1.6.3Summary

Questions

References

1.1 Introduction

During the material design, construction, operation, and maintenance of pavement infrastructure, engineers need to deal with many types of laboratory and field survey data to draw conclusions and make decisions. With the completion of the pavement infrastructure network, the focus of pavement engineering shifts from design and construction to maintenance and preservation. How to evaluate the performance data over the pavement longitudinal direction is challenging in both evaluation and preservation. In the first chapter, we briefly introduce the pavement performance indices, the pavement management system (PMS), pavement performance models, the Long-Term Pavement Performance (LTPP) program, and an overview of current applications of data analysis methods in pavement engineering.

1.2 Pavement performance indices

1.2.1 Development of pavement performance indices

Most applications of data analysis in pavement engineering are for pavement performance modeling, followed by pavement nondestructive tests, pavement material tests, and numerical simulations. The data for pavement performance modeling include pavement condition data as well as traffic, structure, material, and climatic data, among other types. Pavement condition data include pavement functional, structural, and distress conditions. Usually, a comprehensive pavement performance index is calculated based on multiple pavement condition indicators. In the 1950s, the American Association of State Highway Officials (AASHO) developed the first comprehensive pavement performance index in the United States: the pavement serviceability index (PSI), which originated from the empirical pavement serviceability rating (PSR) using a 1–5 rating scale. As shown in Eq. (1.1), PSI is a regression function of roughness, cracking length, patching area, and rutting depth (Carey & Irick, 1960; Hall & Muñoz, 1999). The coefficients in Eq. (1.1) have been modified to enhance the effectiveness of the PSI (Liu & Herman, 1996; Scrivner & Hudson, 1964).

Equation

   (1.1)

where SV is the pavement slope variance; RD is the rutting depth (in.); C is the length of cracking per 1000 ft² pavement surface area (ft); P is the patching area per 1000 ft² pavement surface area (ft²). (Note that the American Association of State Highway Officials (AASHO) was renamed the American Association of State Highway and Transportation Officials (AASHTO) in 1973.)

Regarding pavement distress, the United States Army Corps of Engineers (USACE) developed the first pavement condition index (PCI) using a 1–100 rating scale. As shown in Eq. (1.2), PCI equals 100 minus the corrected deduct value (CDV), which is calculated based on the severity levels and extent of different types of distress (ASTMD6433-99, 1999). The weights for calculating the deduct values of different types and severity levels of distress are mainly determined based on experience and many studies have been conducted to modify the weights (Eldin & Senouci, 1995; Jackson et al., 1996; Juang & Amirkhanian, 1992; Koduru et al., 2010; Saraf, 1998; Sun & Gu, 2011).

Equation    (1.2)

where CDV is the corrected deduct value.

Based on the PSI and PCI, highway agencies developed various pavement performance indices, such as the distress score and condition score used by the state of Texas, the pavement quality index (PQI) used in China, and the maintenance condition index (MCI) used in Japan (Gharaibeh et al., 2010). Zhang et al. (2003) developed a structural condition index for pavement. Park et al. (2007) investigated the relationship between pavement performance indicators and found that PCI can explain the variations in roughness. Most of the new or modified pavement performance indices are developed based on experience and practices (Eldin & Senouci, 1995; Jackson et al., 1996; Juang & Amirkhanian, 1992; Koduru et al., 2010; Saraf, 1998; Sun & Gu, 2011). Eldin and Senouci (1995) developed a pavement rating index based on cracking and rutting according to the experience of pavement maintenance engineers. Juang and Amirkhanian (1992) developed a unified pavement distress index for six types of pavement distresses according to the experience of many state departments of transportation. Fuzzy logic has been used to help determine the weights for individual pavement performance indicators (Bianchini, 2012; Golroo & Tighe, 2009; Juang & Amirkhanian, 1992; Karaşahin & Terzi, 2014; Koduru et al., 2010; Pan et al., 2011).

1.2.2 Pavement performance indices in China

Based on current pavement evaluation tests, the Chinese Highway Performance Assessment Standards (JTG 5210–2018) recommend eight pavement performance indices for asphalt pavements.

1.Pavement surface condition index (PCI)

The PCI is used to evaluate the distress condition and is calculated as 100 minus the cumulative deduct value. Specification JTG 5210–2018 provides the deduct value for 21 different types and severity levels of distress in asphalt pavements. The PCI is calculated by Eq. (1.3).

Equation    (1.3)

where a0 and a1 are coefficients, which are 15 and 0.412 for asphalt pavements, respectively; DR is the distress ratio, which is the weighted sum of areas of various distress over the area of the pavement, as calculated by Eq. (1.4).

Equation    (1.4)

where A is the area of the pavement (m²); i is the ith pavement distress considering both types and severity levels (light, medium, and heavy); i0 is the total number of distress types, which is 21 for asphalt pavements; ωi is the weight of the ith distress; Ai is the area of the ith distress (m²), and is calculated by Eq. (1.5) when using automated testing equipment.

Equation    (1.5)

where GNi is the number of grids containing the ith pavement distress. The standard size of one grid is 0.1 m × 0.1 m; 0.01 is the coefficient to convert the distress area.

2.Rutting depth index (RDI)

The RDI is used to evaluate the rutting depth and is calculated by Eq. (1.6).

Equation

   (1.6)

where RD is the rutting depth (mm); RDa is the coefficient of the rutting depth, which is 10 mm; RDb is the upper limit for the rutting depth, which is 40 mm; a0 and a1 are the model coefficients, which are 1 and 3, respectively.

3.Riding quality index (RQI)

The RQI is used to evaluate the driving comfort and is calculated by Eq. (1.7).

Equation    (1.7)

where IRI is the international roughness index (m/km); a0 is a coefficient, which is 0.026 for expressways and first-class roads, and 0.0185 for other classes; a1 is also a coefficient, which is 0.65 for expressways and first-class roads, and 0.58 for other classes.

4.Skidding resistance index (SRI)

The SRI, calculated by Eq.(1.8) is used to evaluate the friction and safety of pavement.

Equation    (1.8)

where SFC is the measured sideways force coefficient; SRImin is a calibration coefficient, which is 35; a0 and a1 are model parameters, which are 28.6 and −0.105, respectively.

5.Pavement bumping index (PBI)

The PBI, calculated by Eq. (1.9), also reflects driving comfort and safety.

Equation    (1.9)

where PBi is the number of the ith severity level of pavement bumping; ai is the deduct value of the ith level of pavement bumping; i is the severity level of pavement bumping, including light, medium, and high; i0 is the maximum number of the severity level of pavement bumping, which is 3.

6.Pavement surface wearing index (PWI)

The PWI is used to evaluate the wearing and abrasion resistance of the pavement, and it is calculated by Eq. (1.10).

Equation    (1.10)

where a0 and a1 are coefficients and are 1.696 and 0.785, respectively; WR is the pavement wearing radio (%) and is calculated by Eq. (1.11).

Equation

   (1.11)

where MPDC is the pavement surface texture depth, measured on the unworn pavement surface in the middle of lanes (mm); MPDL is the pavement surface texture depth in the left wheel path (mm); MPDR is the pavement surface texture depth in the right wheel path (mm).

7.Pavement structure strength index (PSSI)

The PSSI is used to evaluate the pavement structural capacity and is calculated by Eq. (1.12).

Equation    (1.12)

where a0 and a1 are coefficients and are 15.71 and −5.19, respectively; SSR is the pavement structural strength ratio, calculated by Eq. (1.13).

Equation    (1.13)

where l0 is the allowable pavement deflection (0.01 mm); l is the measured pavement deflection (0.01 mm).

8.Pavement quality indicator (PQI)

As shown in Fig. 1.1, the PQI is a comprehensive index calculated based on the six pavement performance indicators defined earlier, as shown in Eq. (1.14). It is noted that the PSSI is not included.

Equation

   (1.14)

Fig. 1.1

Fig. 1.1 Performance evaluation indices for asphalt pavement in China.

1.3 Pavement management system

A pavement management system (PMS) is a system used to analyze pavement condition information and determine the optimal pavement maintenance decisions for either a pavement network or a pavement segment. The information recorded in the PMS generally includes the inventory data such as pavement classes, structure, and materials; pavement performance data such as performance indicators, indices, ratings, and rankings; pavement construction and maintenance records; and other pavement-related data such as the traffic, environment, and terrain data. The term pavement management system first came into use in the 1970s when systems engineering, engineering economics, operations research, life cycle cost analysis (LCCA), and life cycle assessment (LCA) were developing. Systems engineering helps to design, integrate, and manage pavement systems over their life cycles. Engineering economics uses economic principles in the analysis of pavement engineering decisions. Operations research aims to find the optimal solution under various constraints for pavement maintenance decision-making. The LCCA addresses the life cycle costs of pavements to determine the most cost-effective option among different competing alternatives, while the LCA addresses the life cycle environmental impacts of pavements. These theories, methods, and models are the building blocks of a PMS.

Many PMSs have been developed since the 1980s and have become more widely applied with the development of automatic pavement performance testing equipment, which greatly increased the speed of pavement performance evaluation and the amount of pavement performance data. The Highway Development and Management (HDM) packages developed by the World Bank have been used in many countries (Jorge & Ferreira, 2012). The PAVER management software developed by the USACE is mainly used in airports in the United States (Feighan et al., 1989). AASHTOWare developed by AASHTO integrates pavement, bridge, and tunnel management systems (Khattab et al., 2014). The Deighton Total Infrastructure Management System (dTIMS) developed by Deighton Inc. has been used in the United States, Australia, and Europe (Henning & Roux, 2012). The Highway Pavement Management Application (HPMA) developed by Stantec Inc. has been used in Arizona, North Carolina, Minnesota, and Tennessee (Wong et al., 2003). AgileAssets software developed by AgileAssets Inc. has been used in the United States in New Mexico and North Carolina. Many departments of transportation have also developed their own PMSs, such as California and Georgia. In China, many province departments of transportation and road management agencies use the China Pavement Management System (CPMS) developed by the Research Institute of Highway Ministry of Transport (Wong et al., 2003), while some use the PMSs developed by universities or consulting companies, such as Road Keeper, developed by Southeast University, or SHAPMS, developed by Tongji University (Chen et al., 2012).

Recent developments in PMSs include applications of big data and artificial intelligence (AI), more complicated LCA and resilience analysis, infrastructure sensing, building information modeling (BIM), and digital twinning. With the accumulation of pavement performance and maintenance data, it is feasible to conduct big data analysis and to use AI for pavement performance evaluation and prediction, as well as decision-making. LCA has drawn increasing attention due to the issue of global greenhouse gas emissions and several databases have already been developed. Infrastructure resilience aims to address the long-term performance and vulnerabilities of our pavement systems under extreme climatic conditions. Infrastructure sensing systems aim to give a real-time condition evaluation of our infrastructure, especially for safety concerns such as the retaining walls. BIM is used as the digital representation or platform of physical and functional characteristics of pavement infrastructure for planning, design, construction, operation, and maintenance. The purpose of digital twinning is to create a real-time digital counterpart of the pavement infrastructure for testing, development, and analysis, to improve efficiency, economy, and safety. Those technologies and tools can significantly improve the efficiency and effectiveness of pavement management.

1.4 Pavement performance models

1.4.1 Classic pavement performance models

Using various pavement performance indices, pavement performance models are developed to help pavement design, and then to assist in pavement maintenance decision-making. The AASHO built six test loops in the state of Illinois in the United States between 1956 and 1958 to analyze the structural responses of different pavement designs exposed to highway loadings and then developed the first pavement performance model, as shown in Eq. (1.15). The model is a power regression model and the coefficients have been modified since then. Prozzi and Madanat (2004) revised the model by considering the increment in traffic levels. Hong and Prozzi (2010) added the influence of pavement layer thickness and materials