Small Fault Detection for Satellite Attitude Control System Actuators with Stacked Autoencoder Network

[摘要] The fault detection method based on the analytical model can not completely deal with the model uncertainty, disturbance torque, measurement error and other disturbances, and it is difficult to detect small faults similar to disturbances. A detection method of small actuator fault based on stacked autoencoder(SAE) network is proposed. By learning historical data of system, SAE network can reconstruct the state data with stable errors. The relation between states of system will change when a fault occurs, and residual will change too. The variation trend of the residual can be used to detect the fault. Simulation results show that SAE network is more robust to disturbance compared with Elman neural network and nonlinear observer. Small faults under the disturbance can be detected by SAE network.

[发布机构] School of Space Information, Space Engineering University, Beijing; 10416, China^1

[效力级别] 无线电电子学 [学科分类] 航空航天科学

[关键词] Detection methods;Disturbance torque;Elman neural network;Historical data;Model uncertainties;Non-linear observer;Satellite attitude control systems;Stacked autoencoder