Real-time satellite telemetry anomaly detection for on-board autonomy

As the number of satellites in orbit grows rapidly, with constellations ranging from dozens to thousands of spacecraft, continuous human monitoring does not scale; ground contact is limited and quick autonomous responses are essential to protect mission assets and space safety. The project aims to create a fast, reliable method to automatically detect anomalies in small-satellite telemetry, enabling a spacecraft to monitor itself in real time. We will design simple yet informative indicators that condense streams from key sensors and subsystems (power, attitude control, thermal, communications) into compact features suitable for on-board processing. These features will feed a lightweight machine-learning model capable of flagging both known and previously unseen behaviors with minimal computing and power. Validation will use simulated and historical flight data across nominal, degraded and fault scenarios, verifying strict limits on latency, memory and energy. Evaluation will emphasize accuracy, low false-alarm rates and clear operator explanations. Expected benefits include safer, more reliable constellations, reduced ground workload, faster diagnosis and recovery, and reusable tools adaptable across platforms.

Faculty Supervisor:

Jesus Gonzalez Llorente

Student:

Partner:

Universidad Mayor

Discipline:

Engineering

Sector:

Aerospace; Artificial Intelligence

University:

École de technologie supérieure

Program: