Modeling and Simulation of Dynamic Roadwheel Relationships for Tracked Vehicles using Machine Learning

Within the context of modeling and simulation (M&S), there is often a trade-off between simulation fidelity and computational run time, with more accurate simulations requiring longer run times, and real-time applications sacrificing fidelity for the sake of speed. This paper presents a method for using the high-fidelity Computational Research and Engineering Acquisition Tools and Environments - Ground Vehicles (CREATE-GV) M&S tool to produce a set of' training data, which is then used with machine learning algorithms to train a fast surrogate model for the prediction of tracked vehicle road wheel displacements. This surrogate model could then be used within applications such as autonomous vehicle simulations with real-time requirements, while maintaining a high level of physics fidelity.