Reinforcement Learning-Based Adaptive Event-Triggered Fault-Tolerant Control for Hypersonic Vehicles with Complex Faults

This study proposes a reinforcement learning-based adaptive event-Triggered fault-Tolerant control approach for hypersonic vehicle systems. This method is designed to handle actuator and sensor failures, constrained communication resources, and unknown control directions, as well as external perturbations. The approach adds an unidentified control direction condition and takes into account both time-varying and constant-value fault kinds of actuators and sensors. To accurately estimate the system's unmeasurable states in the case of a sensor failure, an innovative event-Triggered state observer is introduced. The influence of unknown fault parameters can be reduced by applying the Nussbaum function approach, which takes into account the unknown control direction. The reinforcement learning algorithm designed in this paper uses actor-critic neural networks to estimate uncertainty. Furthermore, a dual-channel event-Triggering mechanism is used in this research to alleviate the communication load on the controller-To-Actuator and sensor-To-controller channels, respectively. The system's stability under the intended control scheme is demonstrated via theoretical analysis.