Constructing Robust and Reliable Health Indices and Improving the Accuracy of Remaining Useful Life Prediction

Abstract

A system health index is a measurement of the health condition of complex systems. However, most of the health indices are developed based on strong assumptions. Consequently, existing health indices are not capable of measuring the actual deterioration behaviors with high accuracy. To address this issue, we introduce a probabilistic graphical model to examine the probabilistic relationships among sensor signals, remaining useful life (RUL), and health indices. Based on the graphical model, three types of conditional probabilistic autoencoders are combined to develop the health indices of a complex aero-propulsion system. The proposed method is demonstrated on an engine dataset. The experimental results have shown that the proposed method is capable of constructing robust health indices as well as improving the accuracy of RUL prediction.