Remaining Useful Life Estimation Based On Feature Reconstruction And Variational Bayesian Inferences

Abstract

The prediction of remaining useful life (RUL) plays an important role in prognostics and health management (PHM) to improve the reliability of machines and reduce the cycle cost of mechanical systems. In recent years, deep learning (DL) for RUL prediction has become increasingly popular with the dramatic increase in computational power and has yielded a large number of results in research. However, most DL learning prediction frameworks tend to provide only a point estimate, but there is relatively less research on the uncertainty of the prediction and the confidence interval of the prediction results. This paper proposes a variational inferential Bayesian method to enhance the study of prediction result uncertainty, consequently, the output of prediction result changes from a point estimate to a confidence interval output. To improve the prediction accuracy, the feature are extracted and reconstructed, which make the feature degradation more recognizable. Furthermore, an attention mechanism is considered to improve the performance of RUL prediction by assigning weights to the input features. The effectiveness of our proposed method is validated with a publicly available dataset and compared with the-state-of-the-art methods.