Anomaly Detection Configured as a Combination of State Observer and Mahalanobis-Taguchi Method for a Rocket Engine

Abstract

In this paper, anomaly detection that is configured as a combination of state observer and Mahalanobis-Taguchi (MT) method is proposed for real time fault detection of rapid and dynamic phenomena such as rocket engine operation. Real time anomaly detecting is recognized as one of the most important elements to realize advanced reusable space transportation system. Conventionally, bottom-up type anomaly detecting logic based on failure mode and effect analysis (FMEA) is usually used for this purpose, however, it requires large amount of time and labor. The proposed method can improve this process. In the present method, error values between calculated ones through rocket engine simulator constructed on autoregressive moving average model and extended Kalman filter (EKF) and measured ones are standardized with existing normal operation data of the rocket engine so as to compute Mahalanobis' distance, which expresses degree of anomaly. We performed engine hot firing tests in simulated anomaly conditions. The obtained data was processed with the present method, and the simulated anomaly in the tests was detected as expected.