Reinforcement Learning-Based Fault-Tolerant Control for Semiactive Air Suspension Based on Generalized Fuzzy Hysteresis Model

The air suspension is an advanced suspension system for vibration suppression of vehicles. However, the real-time controllability of the air suspension is weak due to the time-delay characteristics of the air spring. This study designs a novel magnetorheological semiactive air suspension (MSAS) system and examines the fault characteristics of the MSAS system to improve the performance of vibration suppression. First, the generalized fuzzy hysteresis model is novelly proposed to approximate the hysteresis nonlinearity of the magnetorheological fluid damper. Then, the MSAS model with various fault modes is constructed to study the dynamic performance of the MSAS system under different fault modes. Furthermore, the intermediate estimator is adopted to detect the generation of the sensor and actuator faults. Based on the fault estimation, a reinforcement learning-based fault-tolerant (RLF) controller is proposed to improve the dynamic performance of the MSAS system. Moreover, a double wishbone suspension is built to examine the effectiveness of the proposed RLF controller. Experimental results show that the dynamic performance of the MSAS system with the proposed RLF controller is improved in comparison with the MSAS system with the model-based controllers and the passive suspension.