Development of Fault Detection and Identification Algorithm Using Deep learning for Nanosatellite Attitude Control System

Satellites should have high reliability because they are required to operate autonomously, while performing a given mission. To realize this, it is necessary to use heritage parts or include redundancy. However, for nanosatellites, where it is difficult to include redundancy due to volume and weight limitations, fault management becomes crucial. In this study, a new method based on deep learning is proposed for detecting and identifying the faults in the reaction wheel, which is one of the satellite actuators. A deep learning model is applied to learn the fault and fault type using the residual between the measured attitude data and the estimated attitude date. In this study, it is assumed that three reaction wheels are installed, and fault detection is designed accordingly. The proposed model enables the satellite to detect faults autonomously, even when it is not communicating with the ground station, and is expected to be highly beneficial for the autonomous operation of the mega-constellation mission using nanosatellite that is to be activated in future.