The use of SVM-FFA in estimating fatigue life of polyethylene terephthalate modified asphalt mixtures

Abstract

To predict fatigue life of Polyethylene Terephthalate (PET) modified asphalt mixture, various soft computing methods such as Genetic Programming (GP), Artificial Neural Network (ANN), and Fuzzy Logic-based methods have been employed. In this study, an application of Support Vector Machine Firefly Algorithm (SVM-FFA) is implemented to predict fatigue life of PET modified asphalt mixture. The inputs are PET percentages, stress levels and environmental temperatures. The performance of proposed method is validated against observed experiment data. The results of the prediction using SVM-FFA are then compared to those of applying ANN and GP approach and it is concluded that SVM-FFA leads to more accurate results when compared to observed experiment data.

Introduction

Deterioration in the asphalt pavement service life has been identified as a result of an increase in weight as well as excessive usage of heavy vehicles. Two design solutions have been proposed for this problem: first is to employ a thicker pavement layer which is not a cost effective solution and second is modifying the characteristics of asphalt mixture which can potentially be more cost effective when compared to constructing a thicker pavement layer [1].

Due to potentially lower construction cost of pavement structure with modified characteristics when compared to thicker pavement, the strategy gained interest by pavement engineers and practitioners. There have been previous studies attempting to evaluate various performance measures of modified pavement mixtures [2], [3], [4], [5], [6], [7]. Researchers have recently focused on using reusable modifier for modified pavement mixture as using virgin materials to improve the characteristic makes the final product more costly which is not of interest to practitioners [8], [9], [10], [11], [12], [13], [14].

To avoid costly test used for determining pavement performance, one might resort to using soft computing-based prediction techniques. These techniques are capable of learning the relationship between inputs and outputs from a limited set of known data and predict the outputs for an input for which they are not trained on.

One of the novel soft computing learning algorithms is support vector machine (SVM) that has recently found wide application in the field of computing, hydrology and environmental researches [15], [16], [17], [18], [19]. Further, it majorly has been used in, forecasting, pattern recognition, regression analysis and classification [26], [27], [28] and its application has proven its superiority over other conventional statistical models e.g. neural network [20], [21], [22], [23], [24], [25]. In the study of Liu et al. [29], soft computing based methods were employed to deal with early construction warning system. The methods were association rule mining, support vector machine classifiers, and variable fuzzy qualitative and quantitative change criterion modes. The results of the study showed that the methods can be used together to provide a safe construction early warning system. In another application Zhou et al. [30] developed a cluster-in-cluster firefly algorithm and applied that to wired and wireless sensor placement problems for structural health monitoring. It was concluded that the proposed method performs more accurately and efficiently compared to Genetic Algorithm (GA) approach.

In this study, a prediction model is developed to predict the fatigue life of unmodified and modified asphalt mixtures using SVM coupled with Firefly Algorithm (FFA). Polyethylene Terephthalate (PET) which has been obtained from waste PET bottles was used as modifier. The prediction accuracy of SVM-FFA is compared with that of ANN and Genetic Algorithm (GA).