Unsupervised Anomaly Detection for Ground Tests of Rocket Propulsion Systems

Abstract

Detecting anomalous events is an important task for ground tests of rocket propulsion systems in several respects. Firstly, early termination of the experiment in case of an anomaly can prevent costly damage to hardware and ground infrastructure. Secondly, these events, even if not catastrophic, can give valuable insights into the combustion process and the characteristics of the propulsion system. The second aspect is especially important when full manual reviews of the experimental data become impractical, for example in the case of high-speed video. This poster will present ongoing work in applying machine learning algorithms for unsupervised anomaly detection in order to automatically detect these events.