Investigation of change detection methods toward automatic condition monitoring of artificial satellites

Abstract

Although artificial satellites are equipped with fault detection, identification, and recovery systems (FDIR systems), current FDIR systems are designed on a primitive rule basis. As a result, FDIR systems rarely capture predictive signs of fault, such as slight trend changes in status values. Thus, human operators are required to monitor the telemetry data. However, since the number of experts is limited, extensive monitoring and review is only performed for some certain types of telemetry. In this study, we investigate a new change detection method for automatic data-driven satellite health monitoring. Compared to previous research, our algorithm aims to faster practical use. We also report the results of applying the proposed algorithm to actual telemetry data acquired by a JAXA's satellite.