A Bayes-LSTM-Based Model for Remaining Useful Life Prediction of Rolling Bearings

Abstract

In this paper, a remaining useful life (RUL) prediction method for rolling bearings based on Bayesian optimized LSTM is proposed. Firstly, the time domain, frequency domain and time-frequency domain features of rolling bearings are extracted to form the original feature set. And then the most prominent features are selected to form the feature subset by using the monotonicity and Laplacian Score for assessment. Finally, the selected features are used as the input of LSTM to construct the health index (HI) of rolling bearings, so as to predict the RUL. In the process of LSTM training, Bayesian method is introduced to optimize the hyperparameters, so as to improve the prediction performance. Experimental results show that the prediction accuracy of the proposed model is higher than that of SVM and unoptimized LSTM.