A Bearing Remaining Useful Life Prediction Method based on Residual Convolutional Network and LSTM

Abstract

Rolling bearing is an important component for equipments, and failure caused by bearings may cause heavy casualties and realistic losses. Therefore, remaining useful life prediction for bearings has important practical significance. The degraded vibration signals is taken as the research object, and the end-to-end life prediction under different working conditions is taken as application background. In order to improve the prediction accuracy of remaining useful life, the prediction method based on residual convolutional network and long short-term memory is proposed. This method makes training of deeper convolutional network more effective by introducing skip connections in the network to construct different residual unit modules. It can avoid the disappearance of gradient or the degradation of network caused by too many layers, and effectively extract deep-level features of data. In view of time series features representation for degradation process, the deep long short-term memory network is used to construct the trend features of bearing degradation signal. Finally, simulation results indicate the superiority in life prediction.