Fault Diagnosis And Remaining Life Prediction Of Bearing Based On Deep Learning Algorithm

As a critical component of mechanical equipment,the performance of rolling bearings to a certain extent determines the reliability and safety of mechanical equipment.Research on more effective methods for bearing fault diagnosis and remaining life prediction is of great significance to ensure the reliability and safety of mechanical equipment.deep learning algorithms,with superior feature extraction capabilities and the ability to process complex nonlinear relationships in time series data,provide powerful technical support for bearing fault diagnosis and remaining life prediction.In this paper,the research on bearing fault diagnosis and remaining life prediction is carried out by using deep learning algorithm,and the main research contents are as follows:(1)To address the problem of insufficient granularity in bearing fault type recognition and the inability to maintain fault diagnosis accuracy in noisy environments of existing fault diagnosis methods,An end-to-end fault diagnosis method is designed that combines structure optimized convolutional neural networks(SOCNN)with fast batch nuclear-norm maximization(FBNM).During training,SOCNN is responsible for extracting fault features from the raw vibration signals of the bearings,while the FBNM algorithm is used to increase the discriminability and diversity of the network’s batch output matrix,thereby improving the network’s fault diagnosis performance.The SOCNN-FBNM method is validated on the CWRU bearing dataset,and the experimental results demonstrate that the model can accurately identify the fault location and degree of damage in bearings under various noise environments.(2)To address the problem of difficulty in obtaining bearing fault data with fault label,A bearing fault diagnosis method based on unsupervised migration learning is designed by combining the SOCNN-FBNM method with the local maximum mean discrepancy(LMMD)algorithm.During training,the LMMD algorithm is used to improve the similarity between the source domains and target domains,thereby improving the network’s fault diagnosis performance on the target domain without fault labels.The SOCNN-LD-FBNM method is validated on the PU bearing dataset and the CWRU bearing dataset,and the experimental results demonstrate that the method significantly improves the fault diagnosis performance of bearings under variable operating conditions,and also exhibits good noise robustness and generalization ability.(3)To address the problem of high network model complexity and computational cost in deep learning algorithms,this paper combine variational mode decomposition(VMD)with kernel principal component analysis(KPCA)algorithms to form the VMD-KPCA data processing method,which can achieve data decomposition,dimensionality reduction,and noise reduction.The processed data is then input into a gate recurrent unit(GRU)model for network training and remaining life prediction.The VMD-KPCA-GRU model achieves excellent prediction performance in predicting the remaining life of bearings using the full life operation dataset of bearings from our university.The experimental results demonstrate that the VMD-KPCA-GRU model can predict the remaining life of bearings more efficiently and accurately.