Towards Explainable Anomaly Detection in Safety-critical Systems

Ensuring the reliability and safety of space missions necessitates advanced anomaly detection systems capable of not only identifying deviations but also providing clear, understandable insights into their causes. This paper introduces a novel methodology for the detection of systemic anomalies in the telemetry data of the International Space Station (ISS), leveraging the synergistic application of the Functional Resonance Analysis Method (FRAM) and the Specification Tools and Requirement Methodology-Requirement Language (SpecTRM-RL). Integrated with machine learning-based normal behavior prediction model, this approach significantly enhances the explanatory of anomaly detection mechanisms. The methodology is verified and validated through its application to the thermal control system within the ISS's Japanese Experimental Module (JEM), illustrating its capacity to augment diagnostic capabilities and assist flight controllers and specialists in preserving the ISS's functional integrity. The findings underscore the importance of explainability in the machine learning-based anomaly detection of safety-critical systems and suggest a promising avenue for future explorations aimed at bolstering space system health management through improved explainability and operational efficiency.