Abstract: In order to solve the problem of low diagnosis accuracy and control allocation efficiency of traditional fault diagnosis and faulttolerant control methods, this paper proposes a new method of fault diagnosis and fault tolerance control for spacecraft attitude control system based on deep neural network. Taking the control moment gyroscopes as actuator, the method can achieve robust attitude control when the actuator fails. First, we use three heterogeneous deep neural networks to achieve the functions of fault diagnosis, attitude control and torque distribution of traditional faulttolerant controllers, and the intelligent adaptive faulttolerant controller architecture of full neural networks is established. Then, the parameters of the three neural networks such as the number of network layers, the number of neurons and activation functions are optimized and adjusted, and the influence of the parameters of the neural network on the performance of the controller is compared and analyzed. Numerical simulation is conducted to prove that the proposed new controller has good control accuracy and robustness when the control moment gyroscopes fail. The simulation results show that for the spacecraft with a redundant control moment gyroscope, the method proposed in this paper can not only achieve highprecision faulttolerant control under single gyro failure, but also ensure a certain attitude stability control when multiple gyroscope failures occur.

Key words: artificial intelligence, deep neural network, attitude control, fault diagnosis, faulttolerant control