Experimental GPR datasets to characterize multi-layered pavement structures–Tack coat evaluation using hybridization method

The use of non-destructive testing methods (NDT) on newly constructed pavements serves both to verify the proper design of the delivered infrastructure and to support its ongoing monitoring. However, once construction is complete, conventional GPR methods are no longer capable of accurately assessing the thinnest layers, millimeter-scale variations, or the precise amount of bituminous emulsion required to ensure proper bonding at the interfaces. To support this research, a database was created using an impulse Ground Penerating Radar (GPR) with multiple central frequencies on controlled and multi-layered pavement structures according to current French standards [1,2]. Variations in materials and geometries were applied to the two upper courses (wearing course and binder course layers) and tack coat emulsion quantity at the interface. The objective is to obtain a representative database for experimental validation, enabling the characterization of the tack coat located between the two layers that form the surface layer [3]. A recent research study conducted on synthetic data led to a comprehensive parametric analysis [4], validating the methodology of a hybrid Machine Learning/FWI data processing approach to enhance performance and reduce computation times [5]. The current challenge is to distinguish between areas with and without tack coat, as well as to detect subtle geometric variations, using an impulse GPR system equipped with ground-coupled bowtie antennas on a pavement structure fabricated under controlled conditions. The objective is to classify the acquired data and map the results of the Support Vector Machine (SVM) model as an electromagnetic image.