Entropy based anomaly detection applied to space shuttle main engines

Automated model-free anomaly and fault detection using large collections of sensor suites is vital to increasing safety and reducing maintenance costs of complex aerospace systems, such as the space shuttle main engine. Current anomaly and fault detection methods are deficient in that they either require a huge amounts of laborious expert analysis or rely on models that fail to capture unmodelled anomalies. To overcome these deficiencies, model-free statistical approaches to this analysis are needed that do not require significant user input. This paper presents two general automated analysis methods that detect anomalies in sensor data taken from large sets of sensors. The first approach uses entropy analysis over the entire set of sensors at once to detect anomalies that have broad system-wide impact. The global nature of this approach reduces its sensitivity to faulty sensors. The second approach uses automated clustering of sensors combined with intra-cluster entropy analysis to detect anomalies and faults that have more local impact. Results derived from the application of these approaches to sensor data recorded from test-stand runs of the space shuttle main engine show that they can be effective in finding faults and anomalies. With test-stand data consisting time-series derived from 147 sensors, the system-wide approach was able to reveal an anomalous mixture ratio programmed by the test-engineers, but not revealed to the authors. Using similar data from a different engine test, the localized clustering approach revealed a fault in the high pressure fuel turbo-pump early in the test-run and subsequent cascaded faults later in the test run. In addition the clustering approach was able to separate sensors that contained little analytic value from more important sensors, potentially reducing the burden of subsequent expert analysis