Transient tire-soil contact model using enhanced grid-based data-driven approach for virtual testing of off-road mobility systems

Abstract This study proposes an enhanced grid-based data-driven transient tire-soil contact model to enable quick and accurate simulation-based assessment of off-road mobility systems. In this model, the compact tire-soil contact kinematic description in the classical terramechanics model is generalized to account for three-dimensional non-uniform transient contact stress distributions on a grid contact patch by leveraging machine learning techniques. Adaptive piecewise functions are introduced to the grid contact model, such that only a small number of control points are learned by neural networks to predict transient contact stress responses within the contact patch, considering the granular soil material flow. By doing so, the learning data can be effectively compressed, and the number of neural network calls in the online mobility simulation can be significantly reduced while maintaining accuracy comparable to the physics-based tire-soil interaction model. The effect of soil eruption, which causes the top surface to rise when a heavy wheel load is applied to cohesive soil, is also incorporated in the proposed model. The accuracy and computational improvement by the proposed approach are examined using transient traction and cornering simulations, involving large soil deformation, as well as bulldozing effects induced by significant sideslips. Furthermore, the proposed model is applied to mobility simulations of an all-terrain vehicle performing obstacle avoidance at high speed on sandy terrain, and the simulation results are validated against field test data in a scenario not considered in the training data. It demonstrates the potential of the proposed approach to support the development of off-road mobility systems through rigorous simulation-based assessment.