Bearing Remaining Useful Life Prediction Based on AE-BiLSTM

Abstract

The remaining useful life (RUL) prediction of rolling bearings can avoid unreasonable maintenance and major safety accidents. Considering the non-stationary characteristics, it is difficult to utilize the deep learning-based method to directly extract degradation features from the bearing vibration signal. Therefore, in this paper, a fusion prediction model AE-BiLSTM is proposed. The AutoEncoder (AE) is used to extract degradation features from the frequency-domain signals, and BiLSTM network is used to predict the bearing RUL. The experimental verification is conducted on the FEMTO-ST bearing dataset. Experimental results illustrate that the proposed AE-BiLSTM network can accurately predict the RUL of roll bearings.