Increasing performance of spacecraft active fault-tolerant control using neural networks

Actuator fault poses a challenge to the attitude control of spacecraft. Fault-tolerant control (active or passive) is often used to overcome this challenge. Compared to passive methods, active methods have better performance and can manage a broader range of faults. However, their implementation is more difficult. One reason for this difficulty is the critical reaction time. If the actual reaction time becomes larger than the critical reaction time, the system may become unrecoverable. This paper proposes using a feedforward neural network to reduce the actual reaction time in the active fault-tolerant control of spacecraft. Besides this improvement, using a feedforward neural network has the advantage of increasing the success percentage. Success percentage is defined as the ratio of successful simulations to the total number of simulations. Simulation results show that for 200 simulations with random faults and initial conditions, the actual reaction time decreases by 73%, and the success percentage increases by 25%. Based on these results, the proposed controller can be considered a good candidate for practical applications.