Coupling a Dynamic Linear Model with Random Forest Regression to Estimate Engine Wear

Abstract


 
 
 
 Mechanical systems wear or change over time. Data collected over a system’s life can be input to statistical learning models to predict this wear/change. A research effort at The Boeing Company has modeled gas turbine engine exhaust gas temperature (EGT) as a function of other recorded parameters. Our investigation chose gas turbines, but these techniques could also be used for other systems that slowly change (degrade) over time. Previous work trained a flexible empirical regression model at a fixed point of wear, and then applied it independently at time points over the life of an engine to predict wear. However, wear typically occurs slowly and smoothly. This paper describes the benefit of relating wear predictions over time using a dynamic linear model, which is an example of a state space method. The combined model predicts wear with dramatically reduced variability over both our previous effort and a baseline method. The benefit of reduced variability is that engine wear is more evident, and it is possible to detect operational anomalies more quickly. In addition to tracking wear, we also use the model as the basis for a Bayesian approach to monitor for sudden changes and reject outliers, and adapt the model after these events. Experiments compare methods and give some guidance in applying the methodology.