Performance Evaluation and Improvement of Anomaly Detection Algorithms on Satellite Telemetry

Abstract

The goal of this thesis is to create a basis for future work in the field of evaluation of satellite telemetry. In particular, this work is concerned with the development of a framework for the benchmarking of different algorithmic methods for anomaly detection in satellite telemetry. For this purpose, common and newly developed algorithms from many different areas of scientific applications are presented. For many years now, in-depth research has been conducted in many areas of anomaly detection, for example to detect anomalies in power supplies or to support sensor technology on ships. The algorithms will be presented and examined with respect to their suitability in the field of satellite telemetry. Furthermore, telemetry data as such will be examined in detail. Satellite telemetry data represent a special case of data, because they can differ depending on the mission and have special characteristics. Typical problems in telemetry data include gaps in the data, occurrence of noise and unknown events and thus unexpected changes in the recordings. Benchmarking should be considered as the third fundamental area of research. For this purpose it has to be examined how far the development in the evaluation of anomaly detection algorithms has progressed and the use of possible approaches for the evaluation of telemetry data has to be discussed. Finally, the results are summarized and an outlook on future developments is given.