On-board telemetry monitoring via support vector machine with application to Philae solar generator

Recent developments in machine learning for anomaly detection make it possible to use spacecraft status telemetry to produce sophisticated system health monitoring applications, that can run autonomously on-board. Both archived and simulated data are used to train intelligent algorithms to automatically detect and classify anomalous time series of the produced telemetry. The focus of this work is the application of a Support Vector Machine (SVM) based classifier to monitor the status of an interplanetary probe photovoltaic system with a minimum set of available measurements. SVM is a popular machine learning method for classification and regression and it has outstanding generalization performance. The Rosetta lander Philae is considered as test case. A complete model of power production subsystem has been developed to simulate the real telemetry. Training data, generated for nominal and faulty cases, are used to train the SVM model, with the goal of classifying permanent and temporary power loss conditions. The simulated telemetry is then used to test its performance and to identify the minimum number of measurements that is necessary for a successful classification of the failures of interest.