Rolling Bearing Remaining Useful Life Prediction Research Based On Deep Learning

Predicting the remaining useful life of a machine is an important task of Prognostic and Health Management(PHM),and its purpose is to reduce maintenance costs and ensure safe and reliable operation.Rolling bearings are the core bearing components of rotating machinery and equipment.Giving full play to the powerful data mining capabilities of deep learning and the ability to handle complex regression tasks,research on the residual useful life prediction(RUL)method of rolling bearings based on deep learning is very important for the safe and economical operation of rotating machinery.plays a vital role.The main contents of the paper are as follows:(1)Depthwise separable convolution(DSCN)is lightweight and powerful in processing regression tasks.In order to accurately extract rolling bearing vibration signal features and improve prediction accuracy,an end-to-end wavelet kernel-based depthwise separable convolution-convolution attention is proposed,which predictes the remaining useful life of rolling bearings.The continuous wavelet convolutional layer(CWConv)is used to replace the first convolutional layer of the standard separable convolution,which improves the pertinence of the deep learning model for the feature extraction of the original vibration data,and improves the interpretability of the deep network feature extraction.A downsampling module consisting of two separable convolutional blocks and a convolutional attention mechanism module(CBAM)is further designed.Among them,CBAM can adaptively adjust the features extracted by separable convolution and remove redundant information.The experimental results show that the WKDSCN-CBAM model has high prediction accuracy for the full life data of rolling bearings under different working conditions.(2)Aiming at the complex network structure of Mobile Vision Transformer(Mobile Vi T),insufficient adaptability for vibration signal data processing,and overfitting in rolling bearing life prediction,a prediction model called Separable convolution Mobile Vi T(Prog SVi T)is proposed,which integrates lightweight separable convolution and lightweight Mobile Vi T.The model combines the advantages of separable convolution and Mobile Vi T,which uses separable convolution to extract local features,and Transformer extracts global features.According to the characteristics of life prediction,a smooth L1 loss function is used to optimize the Mobile Vi T loss function to improve the stability of training.Adam W optimizer,cosine annealing learning rate decay and warm-up training are adopted used to improve the convergence speed and training effect.The experimental results show that the Prog SVi T algorithm has a relatively accurate prediction accuracy for the full-life vibration data of rolling bearings under different working conditions.(3)Considering that the Swin Transformer hierarchy and sliding window are more comprehensive than Mobile Vi T in extracting features,the Swin Transformer network architecture is introduced into RUL prediction.In view of the insufficient adaptability of Swin Transformer network structure to extract vibration signal data features,there is a problem that the calculation is too complicated in rolling bearing life prediction.A Prognostic Swin Transformer(Prog Sw T)algorithm for predicting the remaining useful life of bearings is proposed.The proposed Prog Sw T optimizes the number of stages in the Swin Transformer and the number of network layers in each level while inheriting the hierarchical structure and sliding window feature extraction,which improves the flexibility and flexibility of the network structure in extracting vibration signal features.portability,while also reducing the computational complexity of the network.According to the characteristics of life prediction,smooth L1 loss is used to optimize the loss function of Swin Transformer to improve the stability of training.Adam W optimizer,cosine annealing learning rate decay and warm-up training are adopted to improve the convergence speed and training effect.The experimental results show that the proposed residual service life prediction model based on Prog Sw T has good prediction accuracy for the full life data of rolling bearings under different working conditions.