A novel method for online prediction of the remaining useful life of rolling bearings based on wavelet power spectrogram and Transformer structure

Abstract

The vibration signal characteristics of rolling bearings are closely related to the performance decay process, predicting the remaining useful life (RUL) of rolling bearings by vibration signals can effectively prevent the occurrence of bearing failures. In this paper, a deep learning-based method for rolling bearing RUL prediction is proposed. The convolutional neural network (CNN), which is more effective in extracting local information, is combined with the Transformer structure, which is specialized in extracting global information, to deeply explore the complex mapping relationship between signal features in wavelet power spectrogram and bearing RUL. Meanwhile, the method of detecting the first prediction time of rolling bearings based on 3 criteria is improved. The proposed method is validated with the XJTU-SY rolling element bearing accelerated life test datasets, as well as compared with other methods to prove its superiority. The results show that the proposed method can effectively extract bearing degradation information and realize the accurate prediction of rolling bearing RUL. The performance-improved rolling bearing RUL prediction model is highly robust and generalizable, which applies to other mechanical parts performance prediction and can be realized for practical applications in industrial fields.