Automated Satellite Fault Detection using Machine Learning
Kendra Lang, Bruce Xu, Michelle Simon, Benjamin Seibert · ASCEND 2022 · 2022
View Video Presentation: https://doi.org/10.2514/6.2022-4297.vid The US Space Force (USSF) requires pervasive and game-changing technologies to improve its concept of operations. One recognized need is for increased satellite survivability through local situational awareness and fault attribution which lead to faster decision-making. The following paper discusses the process the USSF uses to identify technology needs within its enterprise and identification of solutions to those technology needs. We then provide a specific instance of a potential technology solution in the form of machine learning-based fault detection. The solution presented uses machine learning models trained on actual satellite telemetry to predict future anomalies encountered by the satellite. The detailed proof-of-concept methodology highlights a case study in how new technologies can be shown to apply to identified USSF needs, which in turn allows for rapid decision-making for future investments. The specific fault detection proof-of-concept was successful in providing fault alerts up to two weeks in advance of when those faults were recorded by the satellite operators, with minimal false alarms. This early measure of success enabled continued investment and further advancement of the concept towards a deployable software solution for fault management.