Surrogate Modeling of Suspension Mechanisms for Driving Simulator Using Machine Learning

A suspension system has a complex geometric arrangement of link elements. Furthermore, the force characteristics of suspension elements, such as a tire, a damper and a rubber bush, have nonlinearity. Multibody analysis methods are utilized to obtain accurate characteristics for suspension system. On the other hands, the computational cost for a multibody analysis is high, especially when elastokinematic elements are considered. In this study, a machine learning model of a suspension system for real-time calculation is investigated, which is used as an alternative model for a multibody suspension model with considering the bush deformation. To evaluate the performance of the machine learning model, vehicle dynamics simulations for two kinds of steering inputs were conducted, and the simulation results of the machine learning model were compared to those of the multibody model. The machine learning model was applied to our driving simulator system. It was verified that the whole vehicle dynamics calculation process could be executed in real-time with a 1 ms step time.