Satellites today rely on ground operators to notice that something has gone wrong. With Trusted Space, Inc., I’m developing machine learning methods that classify spacecraft telemetry in real time, so a vehicle can recognize faults on its own. Target faults include power failures, attitude-control anomalies, and thruster malfunctions.
Using realistic spacecraft simulations, we train classifiers to recognize fault signatures across several subsystems at once. The aim is spacecraft that need less constant human monitoring and respond faster when something breaks, which makes missions safer and cheaper to run.