PWSA Satellite Fault Diagnosis

Stottler Henke is adapting the MAESTRO fault diagnosis system for 3 different Proliferated Warfare Space Architecture (PWSA) satellites. These implementations integrate a set of methodologies based on Model Based Reasoning (MBR), Self-Organizing Maps (SOMs), Case-Based Reasoning (CBR), and a modular system called TRIAD which focuses on intelligent aggregation of time-series anomaly detection methods. The SOM and TRIAD systems are machine learning (ML) systems trained using historical telemetry data to detect faults. The MBR algorithm simulates expected behavior and compares simulated to real sensor values to detect significant deviations. MBR cross-references additional sensor values and uses knowledge of the spacecraft subsystem to trace anomalous sensor values back to their root cause. The combination of algorithms allows for diagnoses that are more accurate than any single approach operating on its own. MBR methods can detect and diagnose never-before-seen or simulated problems and can successfully function in operating states that are also without precedent. Meanwhile the ML systems can learn and use unknown relationships.