Intelligent and Soft Computing in Infrastructure Systems Engineering

Artificial neural network tools for structural pavement evaluation have been developed to facilitate the determination of the integrity of existing flexible pavements. With the onset of the movement toward more mechanistic pavement design, such as Mechanistic Empirical Pavement Design Guide, nondestructive testing techniques play a major role to determine properties of pavement structures. Conventional methods such as backcalculating the layer properties are complex and either require a significant computational effort and/or frequent operator intervention. Studies are presented that show the power of artificial neural networks to estimate pavement layer properties and allow for capabilities in developing pavement performance curves and for estimating and monitoring remaining life.

A wide range of important problems in pavement and geomechnical engineering can be classified as inverse problems. In such problems, the observational data related to the performance of a system is known, and the characteristics of the system that generated the observed data are sought. There are two general approaches to the solution of inverse problems: deterministic and probabilistic. Traditionally, inverse problems in pavement and geomechanical engineering have been solved using a deterministic approach, where the objective is to find a model of the system for which its theoretical response best fits the observed data. In this approach, it is implicitly assumed that the uncertainties in the problem, such as data and modeling uncertainties, are negligible, and the “best fit” model is the solution of the problem. However, this assumption is not valid in some applications, and these uncertainties can have significant effects on the obtained results. In this chapter, a general probabilistic approach to the solution of the inverse problems is introduced. The approach offers the framework required to obtain uncertainty measures for the solution. To provide the necessary background of the approach, few essential concepts are introduced and then the probabilistic solution is formulated in general terms using these concepts. Monte Carlo Markov Chains (MCMC) and its integration with Neighborhood Algorithm (NA), a recently developed global optimization and approximation algorithm, are introduced as computational tools for evaluation of the probabilistic solution. Finally, the presented concepts and computational tools are used to solve inverse problems in Falling Weight Deflectometer (FWD) backcalculation and seismic waveform inversion for shallow subsurface characterization. For each application, the probabilistic formulation is presented, solutions defined, and advantages of the probabilistic approach illustrated and discussed.

Interest on artificial neural networks (ANN) in infrastructure materials research and practice has increased in recent years. This chapter presents a review of ANN applications in characterization of infrastructure materials focusing on portland cement concrete (PCC) and asphalt concrete (AC) materials. The principles of ANN are briefly introduced and summarized. The strengths and limitations of ANN for modeling behavior of infrastructure materials are discussed. Various applications of the ANN approach in infrastructure materials testing, analysis and design problems are discussed.

Analysis of the mechanical properties of existing road pavements is crucial for pavement rehabilitation and management problems. Numerous studies have focused on developing an efficient method for determining the structural conditions of pavements. Non-destructive testing (NDT) methods can characterize stress-strain behavior of pavement layers at relatively low strain levels. However, the majority of NDT techniques are based on measuring the deflections caused by an applied load to determine the stress-strain behavior. Structural analysis techniques can also calculate deflections using material and loading properties where it is commonly necessary to make an inversion between measured deflections and mechanical properties using a back-calculation tool. Soft computing techniques, i.e. neural networks, fuzzy logic, genetic algorithms, and hybrid systems, have successfully been used to perform efficient and precise back-calculation analyses. This chapter explains the advances in pavement back-calculation methodologies based on soft computing approaches by presenting the concepts behind them and the fundamental advantages of each. An alternative utilization of soft computing techniques for pavement engineering is also presented.

The main goal of this study was to discover knowledge from data about Porous Asphalt Concrete (PAC) roads to achieve a better understanding of the behavior of them and via this understanding improve pavement quality and enhance its lifespan. The knowledge discovery process includes five steps, being understanding the problem, understanding the data, data preparation, data mining (modeling), and the interpretation/evaluation of the results of the models. At the moment, almost 75% of the Dutch motorways network has a PAC top layer. The main damage of PAC is raveling, which is when the top layer of the road loses stones. The SHRP-NL databases provided ten years of material property data from PAC roads. The data for climate and traffic were obtained from databases of the Royal Dutch Meteorological Institute (KNMI) and the Ministry of Transport and Water Management, respectively. Due to the low number of data points (74 data points), an extensive variable selection was performed using eight different methods to determine the four or five most influential input variables and consequently reduce the input dimension. These methods were decision trees, genetic polynomial, artificial neural network, rough set theory, correlation based variable selection with bidirectional and genetic search, wrappers of neural network with genetic search, and relief ranking filter. The modeling step resulted in 8 intelligent models which were developed using two prediction techniques, being artificial neural networks and support vector machines and two rule-based techniques, being decision trees and rough set theory. Taking the low number of data points into account, the prediction models showed a good performance (R2 = 0.95). The rule based models were transparent and easy to interpret but performed less.

The primary purpose of this study is to demonstrate the applicability of the genetic algorithm (GA) to solve nonlinear optimization problems encountered in asphalt pavement design. The fundamentals of the GA are briefly discussed, and four case studies are presented. The first case study is an example showing the backcalculation of layer moduli with deflection data from a falling weight deflectometer and a layered-elastic program. The second case study demonstrates how to construct the master curve, either from a mix flexural frequency sweep test or from a binder rheometer test, and how to apply that master curve in pavement design. The third case shows how to apply the GA to characterize the binder discrete relaxation spectrum with a generalized Maxwell solid model. The last case study illustrates how to apply the GA to define the mix fatigue damage process of a flexural controlled-deformation beam fatigue test and the permanent shear strain accumulation process of a controlled-load repetitive simple shear test with constant height using a three-stage Weibull approach, and how to apply the three-stage Weibull approach in predicting pavement performance. The results indicate that the GA is promising and successful in resolving the nonlinear optimization problem although the GA presents some difficulty in terms of computing efficiency in the case study of backcalulation of layer moduli.

Kalman filter is a signal processing technique that estimates the state of a dynamic system from a series of noisy measurements. It is used in a wide range of engineering applications from radar to computer vision. This chapter demonstrates the application of a model identification procedure based on extended Kalman filter (EKF) and weighted global iteration (WGI) technique in pavement engineering. In particular, EKF-WGI is used to perform layer moduli back-calculation from falling weight deflectometer (FWD) data and to identify model parameters for Generalized Maxwell Model for hot mix asphalt using frequency sweep test data. In both cases, EKF-WGI is shown to provide consistent results that are independent of the seed values for both linear and nonlinear problems. It is believed that EKF-WGI provides an efficient, consistent and robust tool for optimization that has many potential applications.

Over the years, several techniques have been proposed for back-calculation of pavement layer moduli which involves searching for the optimal combination of pavement layer stiffness solutions in an unsmooth, multimodal, complex search space. In recent years, researchers are actively deriving inspiration from nature, biology, physical systems, and social behavior of natural systems for developing computational techniques to solve complex optimization problems. Some well-known nature-inspired meta-heuristics, which are basically high-level strategies that guide the search process to efficiently explore the search space in order to find (near-) optimal solutions, include, but are not limited to: Genetic Algorithms (GA), Particle Swarm Optimization (PSO), Simulated Annealing (SA), Shuffled Complex Evolution (SCE), etc. Potential applications of such nature-inspired hybrid optimization approaches to pavement backcalculation are conceptually illustrated in this chapter which take advantage of the combined efficiency and accuracy achieved by integrating advanced pavement numerical modeling schemes, computational intelligence based surrogate mapping techniques, and stochastic nature-inspired meta-heuristics with global optimization strategies using a system-of-systems approach.

A combined laboratory and modeling study was undertaken to develop a database for common subgrade soils in Oklahoma and to develop relationships or models that could be used to estimate resilient modulus (M_R) from commonly used subgrade soil properties in Oklahoma. Sixty-three soil samples from 14 different sites throughout Oklahoma are collected and tested for the development of the database and models. Additionally, thirty-four soil samples from 3 different sites, located in Rogers and Woodward counties, are collected and tested to evaluate the developed models. The routine material parameters selected in the development of the models include moisture content (w), dry density (γ _d ), plasticity index (PI), percent passing No. 200 sieve (P₂₀₀), and unconfined compressive strength (Uc). Bulk stress (θ) and deviatoric stress (σ _d ) are used to identify the state of stress. A total of four, two regression models, namely, Polynomial and Factorial, and two feedforward-type artificial neural network (ANN) models, namely, Radial Basis Function Network (RBFN) and Multi-Layer Perceptrons Network (MLPN) are developed. A commercial software, STATISTICA 7.1, is used to develop these models. The strengths and weaknesses of the developed models are examined by comparing the predicted M_R values with the experimental values with respect to the R² values. An evaluation of the four models indicate that for the combined development and evaluation datasets, the MLPN model is a good model for evaluating M_R from the selected routinely determined properties. In order to illustrate the application of the developed model, the AASHTO flexible pavement design methodology is used to design asphalt concrete pavement sections.

Very often it is difficult to develop mechanistic models for pavement geotechnical engineering problems due to its complex nature and uncertainty in material parameters. The difficulty in mechanistic analysis has forced the engineers to follows certain empirical correlations. The artificial neural network (ANN) is being as an alternate statistical method, mapping in higher-order spaces, such models can go beyond the existing univariate relationships. The applications of ANNs in pavement geotechnical engineering problems is mostly limited to constitutive modeling, with few applications on prediction of soil layer properties using Falling Weight Deflectometer (FWD), prediction of swelling potential and compute the remaining life of flexible pavements. However, ANN is considered as a ‘Black box’ system being unable to explain interrelation between inputs and output. The ANNs also have inherent drawbacks such as slow convergence speed, less generalizing performance, arriving at local minimum and over-fitting problems. Recently support vector machine (SVM) is being used due to its, better generalization as prediction error and model complexity are simultaneously minimized. SVM is based on statistical learning theory unlike ANNs (biological learning theory). The application of SVM in pavement geotechnical engineering is very much limited and to best of the knowledge such methods have not been applied to pavement geotechnical engineering. However, engineering application of numerical methods is a science as well as an art. This juxtaposition is based on the fact that even though the developed algorithms are based on scientific logic and belong to the special branch of applied mathematics, their successful application to new problems is problem oriented and is an art. As no method can be the panacea to solve all problems to the last details, their application to new areas needs critical evaluation. With above in view, an attempt has been made to develop the art of applying the above artificial intelligence techniques (ANN and SVM) to different pavement engineering problems such as prediction of compaction characteristics, permeability, swelling potential, coefficient of subgrade reaction etc. The parameters associated with the model developments are discussed in terms of guide line for its future