Neural Network Based Active Fault-tolerant Control of Spacecraft via Prescribed Performance

This paper mainly studies the fault diagnosis and fault tolerance control of spacecraft attitude control system (ACS). Firstly, a novel model-free fault diagnosis network (FDN) is constructed based on Bi-LSTM neural networks. It uses only historical data to approximate the ACS mechanism, generate residuals, and combine adaptive thresholds to detect actuator and sensor faults. Secondly, to solve the problem of missing angular rate measurement caused by sensor faults, a fault-tolerant controller based on adaptive tracking differentiator (ATD) to generate virtual angular rate and have prescribed performance is designed to improve the robustness to faults and avoid performance degradation. Finally, simulation results are given to evaluate the effectiveness of the proposed scheme.