Satellite Reaction Wheel Fault Detection Based on Adaptive Threshold Observer

Fault detection for satellite reaction wheels is primarily concerned with increasing the dependability and fault-tolerant capabilities of the system. This article presents an adaptive threshold observer for detecting reaction wheel faults in the attitude and orbit control system of a satellite. A highly accurate model of the reaction wheel was established to provide data for normal and abnormal circumstances. The adaptive threshold observer was developed in two stages. The reaction wheel’s dynamics were initially modeled using a long short-term memory network observer. In the second phase, another long short-term memory network was used to train the adaptive threshold utilizing the difference between the measured and predicted reaction wheel output. Finally, fault detection was achieved by comparing the adaptive threshold and the residual. We developed a high-fidelity mathematical model of the satellite reaction wheel in this research, and we used a fault injection simulation to show the efficacy of our fault detection technique.