Case Studies: Use of Big Data for Condition Monitoring

This paper describes two case-studies that exercised a big data framework, data mining, machine learning (ML) algorithms, and predictive analytics to improve the accuracy of aircraft engine health monitoring and HUMS-based condition indicators’ (CI) ability to predict gearbox removals. In the first example machine learning algorithms operating on multiple data sources produce useful insights to increase our ability to predict engine fuel controller failures prior to the in-flight auto-shutdown. In the second example, we demonstrate the use of aggregated HUMS CIs to predict an intermediate gearbox removal event. Using statistical hypothesis testing, we found three identifiable distributions of aggregate (Super) CI values: normal/good condition, anomalous with more than 100 flight hours remaining, and anomalous with less than 100 flight hours remaining. From these results, we conclude that the value of the condition indicator cannot provide high-resolution condition information; for example, the amount of remaining useful life down to flight-hour accuracy. However, it does provide potentially valuable low-resolution information, such as when an intermediate gearbox has less than 100 hours of remaining useful life.