A recurrent adaptive time delay neural network for fault detection and isolation for the satellite's attitude control system

This thesis investigates a new Fault Detection and Isolation (FDI) scheme for the satellite's attitude control system by using a recurrent adaptive time delay neural network. The results obtained reveal that the proposed new scheme works quite well for detecting and isolating faults in the reaction wheel which cause the satellite to behave abnormally corresponding to either pitch, yaw or roll axes. Moreover, the promising robustness and insensitivity of the proposed neural network scheme due to external disturbances and noise have also demonstrated.;The results presented do indeed demonstrate the satisfactory capabilities and potential advantages of the proposed neural network based fault detection and isolation methodology. The specific faults considered are due to both voltage and current faults in the reaction wheels employed in the attitude control system of a satellite. Both multiple and simultaneous fault signatures and individual fault patterns have been investigated and the results presented validate the very good performances obtained by the proposed neural network. Furthermore, the recovery natures of these faults have also been investigated in several case studies in which the satellite operates under continuous setpoint operating changes.