Transfer prediction of RUL of rolling bearing under variable operating conditions based on TCN and residual self-attention

Here, aiming at problems of feature distribution differences existing in collected life state data of rolling bearing and poor generalization ability of deep neural network model under variable operating conditions, an end-to-end transfer prediction method for remaining useful life (RUL) of rolling bearing was proposed by combining temporal convolutional neural network (TCN) and residual self-attention mechanism. Firstly, one-dimensional time-domain signals collected by sensors were converted into frequency-domain signals with short-time Fourier transform. Secondly, general feature extraction layer of RUL transfer prediction network could use a residual self-attention TCN network. This network could not only better extract time series information, but also capture local degradation features of bearing with the residual self- attention mechanism to enhance the model's transfer feature extraction ability. Once again, the allied field adaptive strategy proposed could be used to match feature distribution differences of rolling bearing life state data under variable operating conditions, and realizethe transfer prediction of bearing life state knowledge under different operating conditions. Finally, the test verification was conducted on publicly available datasets of rolling bearing full life. The results showed that the proposed method can effectively realize RUL transfer prediction of rolling bearing under variable operating conditions and obtain better predictive performance. © 2024 Zhendong yu Chongji/Journal of Vibration and Shock. All rights reserved.