The OPS-SAT benchmark for detecting anomalies in satellite telemetry

Area of Science:

• Space Operations
• Artificial Intelligence
• Data Science

Background:

• Satellite telemetry analysis is crucial but currently manual, time-consuming, and error-prone.
• Existing anomaly detection models lack verified datasets for performance evaluation.
• Automated, data-driven algorithms are increasingly vital for space mission operations.

Purpose of the Study:

• To address the gap in annotated satellite telemetry datasets.
• To introduce the OPS-SAT Anomaly Detection (OPSSAT-AD) dataset for AI model benchmarking.
• To establish a reproducible and transparent validation framework for anomaly detection techniques.

Main Methods:

• Developed the OPSSAT-AD dataset using telemetry data from the OPS-SAT CubeSat mission.
• Implemented and evaluated 30 supervised and unsupervised machine learning algorithms (classic and deep learning).
• Defined a specific training-test data split and proposed quality metrics for algorithm evaluation.

Main Results:

• OPSSAT-AD is an AI-ready benchmark dataset for satellite telemetry.
• Baseline performance results for 30 diverse ML algorithms are provided.
• A framework for fair and objective comparison of anomaly detection algorithms is suggested.

Conclusions:

• The OPSSAT-AD dataset facilitates unbiased validation of anomaly detection models.
• This work promotes reproducible research and transparent evaluation of AI in space operations.
• The dataset and proposed metrics are essential for advancing automated anomaly detection in satellite missions.