Interacting Nonlinear Observers for Sensor and Actuator Fault Diagnosis in a Satellite System

In this paper, the problem of real time sensor and actuator fault diagnosis is studied in a satellite system fusing sensor information. Measurements of inertial sensors are fused with auxiliary sun, earth and star sensors as well as magnetometers and global positioning systems (GPS). All of these sensors are prone to faults, failures and, noises. Thrusters, as actuators, are also employed in satellite attitude control system and are subjected to different faults. In this paper, it is assumed that all sensors are calibrated and the only possible faults are hard faults (failures) and noises. Due to this fact that when a hard fault occurs in a sensor, the measurement model changes to a new one, interacting multiple models (IMMs) are employed for diagnosis of sensor faults. Unscented Kalman filters (UKFs) are used in IMMs due to nonlinear and Gaussian translational and attitude models. In order to reduce the number of parallel filters in the proposed IMM method, the translational and attitude subsystems are decoupled and separate IMMs are designed for each subsystem. Unknown input observers (UIOs) are used to estimate amplitude of faults in thrusters. The efficiency of the method is finally evaluated through simulation and compared with similar approaches.