Hybrid ML/FWI method using GPR data to evaluate the tack coat characteristics in pavements: Experimental validation

Time resolution is one of the limiting factors when using Ground Penetrating Radar (GPR) techniques to characterize thin layers in the subsurface, such as the tack coat in pavements. To evaluate this residual bituminous emulsion at the interface between the wearing course and the binder course, we have developed an inverse method based on a hybrid data processing approach that combines machine learning (ML) algorithms with Full-Waveform Inversion (FWI). Adding the dielectric permittivity of the wearing course (extracted via FWI) as a structural a priori input into the SVM/SVR models has demonstrated the strong potential of this methodology on synthetic time domain signals. This research, proposes extending such a methodology through experimental campaigns. To carry out this study, three distinct campaigns have been planned, namely on: Hot-Mix Asphalt (HMA)-controlled slabs manufactured in the laboratory; a controlled full-scale structure using the Gustave Eiffel University fatigue carousel (Nantes, France); and a new, yet-to-be-used, road in France. These experiments serve to validate the performance improvements of various classification and regression SVM/SVR algorithms when adding the dielectric permittivity of the wearing course. Herein will be compared the results of the global approach, without preprocessing raw time domain signals, with the developed hybrid model. • Pavement Tack Coat Characterization through Experimental Validation. • Hybridization between Machine Learning and Full-Waveform Inversion experimental data inversion. • Enhancing the Decision-Making Process by Integrating an experimental A Priori Physical Information.