Integrated Bayesian framework for remaining useful life prediction

Abstract

In this paper, a data-driven method for remaining useful life (RUL) prediction is presented. The method learns the relation between acquired sensor data and end of life time (EOL) to predict the RUL. The proposed method extracts monotonic trends from offline sensor signals, which are used to build reference models. From online signals the method represents the uncertainty about the current status, using discrete Bayesian filter. Finally, the method predicts RUL of the monitored component using integrated method based on K-nearest neighbor (k-NN) and Gaussian process regression (GPR). The performance of the algorithm is demonstrated using two real data sets from NASA Ames prognostics data repository. The results show that the algorithm obtain good results for both application.