Life Prediction of Rolling Bearing using Temporal Convolution Network and Attention Mechanism

Abstract

This paper focus on the life prediction of rolling bearing. In consideration of the traditional recurrent neural network (RNN) prediction method occupies a large memory and can not parallel processing data, a method combining temporal convolution network (TCN) with attention mechanism (TCN-Attention) is introduced to carry out life prediction. Convolution layers of TCN can obtain more information and reduce the memory occupancy compared with RNN under the same conditions, and the attention mechanism is used to extract the life related information to improve the prediction accuracy. Finally, experimental results show that the introduced TCN method can effectively predict the characteristics of vibration signals of rolling bearings, and the superiority of the introduced method is verified by comparing with the traditional RNN and long short-term memory (LSTM).