A Modified CCN based on Efficient Channel Attention and Resnet for Bearing Remaining Useful Life Prediction

Prognostics and health management technology is proven to be a useful tool for keep large equipment safe and reliable, which is also true for the most commonly used bearings. With the advantages of new generation of artificial intelligence and large scale industrial data, the condition monitoring and remain useful life (RUL) prediction of bearings are still a hot research topic. To improve the prediction accuracy and generalization ability, a modified convolutional neural network (CNN) based on efficient channel attention (ECA) and Resnet (ER-CNN) is presented in this paper for bearing RUL prediction. With the ECA and a deeper network, the learning ability of the CNN is greatly enhanced. Moreover, the use of phased prediction and service time enable the prediction model to follow the change of degradation stages and make full use of time information during the monotonic bearing degradation. The experimental results and comparisons of three bearings indicate that the proposed ER-based CNN has better prediction performance for bearing RUL.