Development of a vehicle occupant body control model using multibody dynamics and deep learning

Abstract

Since the occupant is not negligibly light relative to the weight of the vehicle body, it is required to consider the occupant's dynamic characteristics at the design stage. For that purpose, a model capable of simulating the physical behavior of the occupant during traveling is required. In this study, we first conduct vibration test on humans to investigate the characteristics of body behavior. After that, principal component analysis by singular value decomposition is performed to evaluate the characteristics of body behavior and to extract the primary motion modes. Then, the vehicle occupant is approximated as a two-link model in which one point is subjected to displacement excitation, and the equations of motion are derived using multibody dynamics. As a body control model, a model using both nonlinear PD control and deep learning will be examined. Finally, the effectiveness of the proposed occupant body control model is verified by comparing the behavior predicted by the proposed model with the behavior obtained by the experiment.