S.A. Kowshik, A. Srinivasa, J.N. Reddy
Texas A and M University,,
United States
Keywords: Gaussian process model , Becker-Wong model , uncertainty quantification, tire-terrain interaction, real time simulation, machine learning, off-road conditions
Summary:
We propose a Gaussian process machine learning model (GPM) for real-time simulation of tire-terrain interactions, particularly under off-road conditions. This approach offers a novel method for assessing tire performance by enabling detailed predictions of interaction parameters such as traction forces, tire sinkage, and rolling resistance—alongside uncertainty estimates for these predictions. Compared to purely empirical models or classical neural networks, the GPM requires significantly less input data for training, provides improved explainability, and quantifies the uncertainty of its outputs, making it a robust tool for tire performance evaluation. The model seamlessly integrates data from physics-based numerical simulations, empirical or semi-empirical models, and experimental results, offering flexibility. The key idea is to use established empirical models, such as the steady-state Becker-Wong model, as a baseline and "learn" the differences arising from the dynamic response of the tire using additional detailed inputs from physics-based models or experimental data or a combination thereof. This hybrid approach enables highly accurate predictions of tire responses under varying conditions in real time. This capability is particularly valuable for testing tire designs, allowing manufacturers to evaluate tire performance across diverse terrains and operating conditions without extensive physical testing. Additionally, the method supports the rapid assessment of design changes, making it an essential tool for optimizing tire performance and validating designs in off-road and challenging environments.