Model-based fault detection with uncertainties in a reusable rocket engine

• Published in: 2022 IEEE Aerospace Conference (AERO) pp. 1 - 8
• Main Authors: Omata, Noriyasu, Tsutsumi, Seiji, Abe, Masaharu, Satoh, Daiwa, Hashimoto, Tomoyuki, Sato, Masaki, Kimura, Toshiya
• Format: Conference Proceeding
• Language: English
• Published: IEEE 05.03.2022
• Subjects: Data models; Fault detection; Firing; Monte Carlo methods; Rockets; Training data; Uncertainty
• DOI: 10.1109/AERO53065.2022.9843212

Considerable attention is being paid to model-based prognosis and health management (PHM) that employs reduced-order modeling because of limited training data. This includes fault modeling in aerospace systems. However, uncertainty is inherent in reduced-order modeling and can significantly impact model-based PHM systems. In addition, there is uncertainty unrelated to modeling, such as differences in system inputs from the environment or measurement errors. This study developed a fault detection method that considers both model uncertainty and system variations. The degree of each type of uncertainty was estimated stochastically from past test data. Using this data, system variation was estimated through Monte Carlo simulations. Model uncertainty was estimated as the error distribution from fitting the simulation model to the previous test data. Thus, the total PHM uncertainty was determined by adding the two uncertainties together stochastically. The sensor values of the experimental data in the actual system were compared to the simulated values. A system alert was issued to indicate a fault if the difference between the two falls outside the expected range. This method was applied to the data from firing tests of a reusable rocket engine for RV-X, an experimental reusable launch vehicle being developed by JAXA. While no anomalies were observed in the test, the experimental and simulated values agreed within the expected range of overall uncertainty for most cases. A few cases were judged anomalous, but it was significant to narrow down the possible abnormalities.