A physics-informed neural network (PINN) framework for inverse mixed contact-traction tire-load reconstruction and finite-layer pavement response

Tire-pavement interaction produces a spatially heterogeneous boundary traction that cannot be fully represented by uniform vertical pressure, particularly when braking, driving, or lateral slip generates appreciable tangential loading. Existing layered pavement models mainly address forward response prediction and provide limited capability for reconstructing unknown contact tractions and mechanically admissible transient subsurface fields from sparse observations. This study develops a physics-informed neural network (PINN) framework for inverse tire-contact traction reconstruction and dynamic finite-layer pavement response. The tire footprint is represented by a total normal contact pressure and longitudinal and lateral traction components, without imposing a mechanically unidentifiable decomposition of the normal pressure. A three-dimensional transient finite-element reference model provides mechanics-consistent displacement, stress, strain, reaction, and strain-energy fields under a time-dependent moving load. These fields are coupled with neural reconstruction through dynamic equilibrium, constitutive, boundary, interface, force-closure, initial-condition, and regularization constraints. Source realizations are partitioned before training, and held-out contact fields are reconstructed from sparse displacement and stress histories without including their prescribed source values in the inverse loss. The framework therefore provides a unified differentiable setting for contact-traction identification, transient pavement-response reconstruction, source-disjoint evaluation, stability assessment, domain transfer, and robustness analysis.