A Data-Driven Model for Bearing Remaining Useful Life Prediction with Multi-step Long Short-Term Memory Network

Abstract

AbstractAccurate prediction of the remaining useful life of rolling bearing is of great significance for making reasonable maintenance strategy and reducing maintenance cost. In order to extract the features with time sequence information better for prediction, we propose a data-driven method using multi-step long short-term memory (MS-LSTM) network for predicting the remaining useful life (RUL) of bearings. Firstly, stacked denoised autoencoder (SDAE) and self-organizing maps (SOM) are combined to construct one-dimensional health index (HI) curve according to the original vibration signal, and then the HI curve is input into the MS-LSTM network to predict the long term future trend. Finally, the remaining useful life is calculated according to the failure threshold. Compared with the existing advanced methods, it achieves an absolute improvement of RMSE by 13.63% in the whole remaining life prediction and 7.43% in the last 840 steps.KeywordsRemaining useful life predicationRolling bearingRNNLSTMFeature extraction