Machine Learning Based Fault Detection and Identification

As space system technologies continue to progress, space systems will require more robust, online health monitoring systems that are capable of identifying and compensating for faults or threats to the system. Due to unforeseen circumstances and naturally occurring threats, an on-board fault diagnosis system for space vehicles capable of autonomous Threat Detection, Isolation, and Recovery (TDIR) is necessary to maintain space operations and mitigate operational gaps as mission complexities increase. Data-driven methods are being explored for FDIR in aerospace systems. The ADCL lab has developed a health monitoring system tool that employs machine learning algorithms such as Support Vector Machine, incremental learning and the principle of self-nonself-discrimination to distinguish between nominal and failure data. This methodology addresses the complexity and multi-dimensionality of aerospace system dynamic response in the context of abnormal conditions and is aimed to assist spacecraft with recovery maneuvers in real-time.