Department of Computer, Control and Management Engineering

In spacecraft health management a large number of time series is acquired and used for on-board units surveillance and for historical data analysis. The early detection of abnormal behaviors in telemetry data can prevent failures in the spacecraft equipment. In this paper we present an advanced monitoring system that was carried out in partnership with Thales Alenia Space Italia S.p.A, a leading industry in the field of spacecraft manufacturing. In particular, we developed an anomaly detection algorithm based on Generative Adversarial Networks, that thanks to their ability to model arbitrary distributions in high dimensional spaces, allow to capture complex anomalies avoiding the burden of hand crafted feature extraction. We applied this method to detect anomalies in telemetry data collected from a simulator of a Low Earth Orbit satellite. One of the strengths of the proposed approach is that it does not require any previous knowledge on the signal. This is particular useful in the context of anomaly detection where we do not have a model of the anomaly. Hence the only assumption we made is that an anomaly is a pattern that lives in a lower probability region of the data space.