Data-driven fault detection in a reusable rocket engine using bivariate time-series analysis

Data-driven fault detection using a bivariate time-series analysis was done for the maintenance of reusable liquid-propellant rocket engines. This method is based on the phase plane trajectory of feature vectors extracted from two sensor data by principal component analysis, and applied to all sensor pairs. System and sensor failures can be estimated by visualizing the state of all sensor pairs. The present method was applied to the static firing test results of a reusable rocket engine developed in Japan. Temperature sensor failure was successfully detected from 59 sensors in the 19 static firing tests. System failure caused by incorrect valve operation was also successfully detected. The ability to detect and estimate system and sensor failures was demonstrated, even if the engine's operational sequence changes dynamically due to ignition or cutoff or if there are deviations of the engine's operational sequence between tests.