Multistage Convolutional Autoencoder and BCM-LSTM Networks for RUL Prediction of Rolling Bearings

Remaining useful life (RUL) prediction plays a crucial role in bearing maintenance, as it directly affects the safe operation of equipment. This article presents a rolling bearing RUL prediction method based on multistage convolutional autoencoder (MSCAE) with improved long short-term memory (LSTM). Considering that the accuracy of the final prediction results is constrained by the health index (HI) extraction, introducing the convolutional autoencoder (AE) enhanced by the full life-cycle mechanism and the influence of the prediction model by the large concavity of the input HI is mitigated by introducing the LSTM enhanced by the bias correction mechanism (BCM). First, in the feature acquisition stage, the calculated time-domain features of the vibration signal are filtered using correlation calculations with RUL. Second, in the HI extraction stage, the three degradation stages of the bearing are divided using the Hilbert transform-based method, and then the MSCAE-based HI extraction model is built according to the classified stages. In the RUL prediction stage, the loss function of the LSTM is designed based on the BCM, and then the RUL prediction model is obtained. Finally, experiments are carried out with the XJTU-SY dataset, and the experimental results show that the method proposed in this article can efficiently extract the HI and can build a more effective RUL prediction model compared with other methods.