3D tire–pavement contact stresses: physics-informed prediction approach

Accurate prediction of tire–pavement contact stresses is essential for pavement structural analysis and performance evaluation. Traditional finite element analysis (FEA) methods, while may be accurate, they remain computationally intensive. This study proposes a physics-informed generative adversarial network, phyContactGAN, to predict three-dimensional (3D) non-uniform contact stresses under varying wheel loads, tire inflation pressures and rolling conditions (free rolling, braking, and acceleration). The model incorporates physics-based constraints and leverages multimodal input encoding with skip connections to enhance spatial fidelity. A dataset of 1,852 FEA-generated stress fields spanning various tire conditions was constructed for model training and evaluation. phyContactGAN demonstrated reliable accuracy across vertical, longitudinal, and transverse stress components; with a mean absolute error (MAE) of 0.0036 MPa and a mean absolute percentage error (MAPE) of 0.17%. Robustness and generalization were confirmed through fivefold cross-validation. Moreover, the model required only 1.56 sec to predict a full 3D stress field, enabling rapid inference with physical consistency. This study demonstrated the use of physics-informed deep learning framework to produce reliable and computationally efficient surrogate of validated FEA tire loading for large-scale engineering analysis.