Numerical modeling using gprMax to identify a subsurface tack coat for SVM classification

Abstract

Summary Face to an increasing traffic volumes, a poor adhesion between bitumen layers combined with weather conditions can lead to premature deterioration of pavement structures. Therefore, it is essential to resort a tack coat where the wearing course and the binder course connect, so that they work as a monolithic block. The purpose of this study, carried out by using gprMax software, is to identify a thin millimetric subsurface tack coat from a modelled bilayer of bitumen and to differentiate the signals according to the modifications of some parameters like thickness, permittivity and conductivity. The so generated large database of time signals with diverse geometric and dielectric characteristics will enable to classify the datas by a supervised machine learning method namely, Support Vector Machines (SVM). Among existing methods, the algorithm of Two-Class SVM (TCSVM) allows to split the datas in two distinct classes. One data set is described as the “adhered” class, and another as the “non adhered” class. The supervised machine learning is conducted with a resolution by global approach to use the raw data set, without any pre-processing. Finally, the binary classification appears then as a promising method to identify clearly and automatically the presence of a tack coat.