Abstract: To enhance hypersonic vehicles' fault-tolerance capability and mission completion ability under various severity levels of actuator faults, this paper proposes a learning-based hierarchical coordination fault-tolerant method for hypersonic vehicles. Firstly, to achieve online quantitative analysis of the severity of actuator faults, deep neural network-based approaches for predicting vehicle trim capability and reachable region boundaries are proposed. Subsequently, using the above prediction methods as "bridges," a hierarchical coordinated fault-tolerance framework is constructed. The method assesses the severity of current actuator faults in real-time based on prediction results, then selectively coordi-nates fault-tolerant mechanisms at the control, guidance, and planning layers to mitigate the impact of actuator faults on flight performance and mission completion capability as much as possible. Finally, simulations under three different severity levels of actuator faults are conducted to validate the effectiveness of the proposed fault-tolerant method.

Key words: Hypersonic vehicle, Actuator fault, Deep learning, Trim capability, Reachable area, Hault tolerance