A knowledge-based expert system for the detection of anomalies in rocket engines

Complex physical systems are subject to failures at any time while operating. Depending on the severity and the speed by which a fault develops, partial or total destruction of a component may occur. When a fault occurs during the testing of a component, extensive post-test analysis is performed on the sensor data, in attempt to identify the time at which the fault started, and to determine the cause of the failure. The research and results presented in this thesis demonstrate a prototype knowledge-based expert system which detects rocket engine anomalies dming their early stages of development. This system was designed to analyze rocket engine ground test data acquired from high frequency sensors, and based on the data, to identify anomalous engine behavior. In order to detect anomalous behavior, IDES (Identification and Detection Expert System) employs the same type of knowledge and reasoning process that rotordynamics experts employ when analyzing post-test high frequency sensor data. Due to the high frequency sampling rate of the monitored sensors, the data is transformed into its fast fourier transform representation. IDES extracts certain frequency features from the sensor data, then it processes and analyzes the extracted information acquired from several sensors in order to determine the health status of a monitored rocket engine component. After processing the sensor data from a sensor, IDES generates a hypothesis which states whether a given sensor has detected anomalous behavior in any of the extracted frequencies. After processing all of the sensors, IDES then combines all of the generated hypotheses into a single overall hypothesis about the health status of the monitored engine component. In this case, the monitored rocket engine component was the High Pressure Oxidizer Turbo Pump of the Space Shuttle Main Engine.