Hybridization of machine learning/GPR Full-Waveform Inversion for pavement tack coat characterization: Numerical validation

The characterization of tack coat layers in pavements using conventional techniques, such as forward processing applied to ground-penetrating radar (GPR), is challenging due to the low temporal resolution of raw signals, among other factors. In this research, we propose a double inverse method based on machine learning algorithms (SVM/SVR) combined with Full-Waveform Inversion (FWI), enabling robust detection and characterization of such thin layers. Machine learning (ML) methods applied to GPR signals have initially demonstrated the potential to detect structural anomalies from time signals (A–scans) through inverse global/local approaches. However, despite their performance, these decision-making algorithms have proven insufficiently robust when faced with various physical phenomena, whose effects are not highlighted in the raw GPR signals used in the training database. Consequently, we propose adding a priori material characteristics using FWI in order to boost the machine learning model. This hybridization between ML and FWI data inversion has been successfully validated on very wide numerical databases issued from an FDTD forward model. This double inverse method has shown an improved decision-making process by integrating a priori physical information. • Pavement Tack Coat Characterization through Numerical Validation. • Hybridization between Machine Learning and Full-Waveform Inversion data inversion. • Enhancing the Decision-Making Process by Integrating A Priori Physical Information.