Full-Wave gprMax Simulation of Top-Down and Bottom-Up Surface Cracks in Reinforced Asphalt Pavements: Descriptor-Based Discrimination Within a Common Three-Hyperbola Regime

Ground-penetrating radar assessment of reinforced pavements remains limited because top-down and bottom-up cracks produce similar radargram signatures and exhibit comparable diffraction patterns. This study investigates this ambiguity using full-wave gprMax simulations and a two-level interpretation framework that combines processed radargram morphology with descriptors extracted from a central trace of the simulated B-scan at the crack axis. Thirty-two deterministic reinforced asphalt pavement models were analysed by varying crack width, crack depth, surface-layer thickness, and granular-layer thickness. Analytical travel-time checks showed that the extracted surface and intermediate events were consistent with the imposed geometry, with deviations of +0.016 ns (+2.51%) and +0.006 ns (+0.40%) for the representative top-down case and +0.028 ns (+3.02%) and +0.002 ns (+0.10%) for the representative bottom-up case. All simulations preserved three accepted hyperbolic paths, establishing a common 3H regime for the full parametric matrix. Crack width reorganised mainly the intermediate branch; crack depth acted on the intermediate branch in top-down models and on the shallow branch in bottom-up models; variation in surface-layer thickness at a constant relative crack depth (d/hb = 0.4) produced the largest temporal reorganisation; and granular-layer thickness acted as a control variable. Descriptor-based analysis further revealed configuration-dependent differences between the two crack families within the common 3H regime using temporal spacing and amplitude partitioning, with classification accuracies of 69.23% and 84.62% for the waveform-based and radargram-based descriptors, respectively, under leave-one-geometry-out cross-validation. The study provides an interpretation framework for distinguishing coupled surface and internal crack responses when the hyperbola count alone is non-discriminative.