Spacecraft telemetries analysis for anomaly detection functions

In spacecraft health management a large number of time series is acquired and used for on-board units surveillance and for historical data analyses. The lesson learnt from last decades in space segment Operations at Thales Alenia Space highlights that advanced monitoring systems are required to early detect abnormal behaviours and to prevent failures of satellites. A promising class of techniques is based on machine learning algorithms, which allow to inspect large amount of telemetries collected during years of in-flight spacecraft operational life. Such advanced approach exhibits several interesting capabilities for feature extraction, outliers detection and predictive analysis, based on learning from data. This paper presents the results of the analysis, implementation and testing of a novel feature extraction technique combined with anomaly detection algorithm, leveraging on our background on space segment engineering and operations.