Neural Network Nash Game-Based Driver Assistance Fault-Tolerant Control for Steer-by-Wire Systems

Abstract

A driver-assisted fault-tolerant control (FTC) based on a neural network-improved Nash game framework is proposed to maintain vehicle trajectory tracking and stability when the steer-by-wire (SbW) system suffers actuator functional faults. A radial basis function (RBF) neural network with an improved activation function is used to train a driver model online to capture difficult-to-model drivers’ behavior accurately and then combine it with a physical vehicle model to construct a data-physical hybrid driver-vehicle system. The differential drive-assisted steering (DDAS) moment generated by the front axle is used to correct the abnormal steering function. Still, the effect of the direct yaw moment control (DYC) is affected due to the conflict between the trajectory tracking target and the lateral stability target. A distributed model predictive control (MPC)-based noncooperative game model is built by mapping the two control targets as the game players. The data-physical hybrid driver-vehicle model is used to predict the dynamic behavioral changes of the players. Finally, a Nash equilibrium solution is obtained to balance the target conflict. Simulation and hardware-in-the-loop (HIL) tests show that the designed method can improve the stability of faulty vehicles under different styles of drivers by up to 60.48% and the trajectory tracking performance by up to 31.55% while maintaining good real-time performance.