AI-based Novelty Detection in Space Operations: Three Years of Operational Experience and Progression at GSOC

Kurzfassung

Anomaly detection in satellite telemetry is critical for ensuring operational reliability and early fault detection. This paper presents the integration of the AI-based Automated Telemetry Health Monitoring System (ATHMoS) into the operational workflows of the German Space Operations Center (GSOC) at the German Aerospace Center (DLR). We discuss the challenges encountered during deployment and the solutions implemented to enhance ATHMoS' effectiveness. Key improvements, informed by engineer feedback, include refined parameter classification—particularly expanded support for highly periodic parameters with little to no noise and certain discrete parameters like counters—as well as a user-driven reclassification workflow to reduce false positives from nominal events such as maneuvers and maintenance activities. Additionally, we introduce a continuous integration (CI) pipeline that automates configuration testing across multiple satellite telemetry datasets, streamlining performance evaluation, optimization, and comparison with the operational ATHMoS system. These advancements enable broader applicability of ATHMoS across diverse satellite missions, including both large-scale scientific and communication missions as well as resource-constrained platforms such as CubeSats. Furthermore, ongoing developments focus on a real-time, onboard version of ATHMoS, laying the foundation for future advancements in AI-driven telemetry health monitoring.