Explainable symptom detection in telemetry of ISS with Random Forest and SpecTRM

Flight controllers of the JEM (Japanese Experiment Module), one element of the International Space Station (ISS), are continuously monitoring ISS status, and it is important for them to detect signs of anomaly of its equipment as early as possible. Automatic symptom detection, in this context, can help flight controllers to assess unusual telemetry trends. To assess the trends efficiently, it is essential to provide the reason of detections. In this paper, we propose a new systemic symptom detection method combining three methodologies: the Functional Resonance Analysis Method (FRAM), the Random Forest Regression (RF), and the Specification Tools and Requirement Methodology-Requirement Language (SpecTRM-RL). The method was verified with data of Low Temperature loop (LTL) of JEM; an actual failure event of pump inverter in LTL was selected as a case study. In this case study, a selected objective variable was successfully predicted based on explanatory variables in normal period, whereas the predicted values showed larger deviation from the actual measured values in off-nominal period. The information for explaining the cause of anomaly was eventually identified with the proposed methods and validated by engineering knowledge. These results show the effectiveness of the new methods as the explainable machine learning-based predictive failure detection. The proposed method can be applied to fields where a single mishap of a system could lead to catastrophic hazard or instantaneous loss of human life due to impossibility of physical access (e.g., deep space explorations and remote medicine).