Intelligent Prediction of Bearing Remaining Useful Life Based on Data Enhancement and Adaptive Temporal Convolutional Networks.

The bearing is one of the most important components of rotating machinery. Predicting the remaining useful life (RUL) of bearings is of great importance to avoid unforeseen failures in the mechanical system and reduce the maintenance cost of the system. In conventional RUL prediction methods, the influence of noise and timing information on the prediction accuracy is often ignored. To address this problem, a method for RUL prediction of bearings based on data enrichment and adaptive temporal convolutional networks (TCN) is proposed in this paper. First, the sample entropy is used to optimize the singular spectrum decomposition (SSD) and combine with the correlation coefficient, and the data enhancement method based on the enhanced SSD and correlation coefficient (ISSD–CC) is proposed to reduce the signal noise; then, the time domain information is extracted to construct the feature vector. Finally, the Nadam optimizer and adaptive parametric rectified linear unit activation function (APRelu) are introduced, and an improved temporal convolutional network (ITCN) is proposed to improve the adaptive learning ability of TCN for time and time–domain information. IEEE PHM2012 dataset and the XJTU-SY dataset are used to verify the feasibility of the proposed method. The experimental results show that the accuracy of ISSD correlation coefficient and ITCN prediction method in RUL prediction of bearings is better than other comparison methods. [ABSTRACT FROM AUTHOR]