Multi-Task Physics Informed Neural Network Based Steering Control Strategy for Steer-by-Wire Systems

The steer-by-wire (SBW) system is recognized as a crucial technology for facilitating autonomous driving. However, system nonlinearities, parametric uncertainties, and external disturbances present significant challenges to steering control. This article proposes a control strategy integrating the multitask physics-informed neural network (MTPINN) with model predictive control (MPC), using the rack position as the tracking target. First, the SBW model, including the motor, synchronous belt, ball screw, and rack, is established, with the rack force being treated as an extended state of the system to construct the extended disturbance observer (EDO). Then, by embedding both the extended disturbance observation equation and the prediction equation into the neural network, the MTPINN model that represents the transient behavior of the SBW system is constructed. This model is integrated with the MPC architecture to improve the performance of rack position tracking control. Finally, the bench test results indicate that the maximum observation error root-mean-squared error (RMSE) and the maximum tracking error RMSE are 0.348 kN and 0.784 mm, respectively, validating the effectiveness of the MTPINN-MPC steering control strategy.