Research On Bearing Fault Monitoring Based On Transfer Learning

As a key part of mechanical equipment,bearings usually work in relatively harsh environments,and failures will cause safety accidents and economic losses.Therefore,it is necessary to monitor the working state of the bearing.However,in practical applications,due to the constant changes in the working conditions of the bearing,the diagnostic effect of the established model is not good under different working conditions.This paper takes the vibration signal of the bearing when the mechanical equipment is working as the research object.Based on transfer learning and generative confrontation network,transfer learning research is carried out for bearing fault diagnosis under different working conditions,and domain difference research is carried out for cases where the transfer effect is not good,so as to quickly complete the bearing fault diagnosis task through a small amount of vibration signals,without additional labor costs.The main work of this paper is as follows:Aiming at the problem that the bearing fault diagnosis algorithm based on deep learning has poor diagnosis performance when the fault samples are lack of labels in different working conditions and real environments,an unsupervised domain adaptive bearing fault diagnosis method is proposed to realize the unsupervised fault diagnosis of bearings under different working conditions.First,the bearing fault sample data is preprocessed by Fast Fourier Transform(FTT)and the features of bearing faults samples are extracted using Convolutional Neural Network(CNN).Then,the feature distributions output of the source domain and the target domain are converged by the method of reversing labels in the Generative Adversarial Network(GAN).Finally,the classifier of the source domain is exploited to complete the bearing fault diagnosis task under different working conditions.In order to verify the effectiveness of the proposed method,relevant comprehensive experiments are carried out on the bearing dataset of Case Western Reserve University(CWRU)and the bearing dataset of the University of Paderborn(PU)in Germany.The experimental results demonstrate that the proposed method can use the unlabeled target domain data to complete the transfer task,and it shows a good transfer performance on the two datasets and achieves a high diagnostic accuracy.This thesis aims at the problem that the data difference between the source domain and the target domain is not obvious when the bearing fault data is migrated,and the migration effect is not good.An unsupervised domain adaptive bearing fault diagnosis method based on maximum domain discrepancy is proposed.This algorithm inputs the target domain data into the source domain feature extractor and the target domain feature extractor respectively and ensures the forward optimization of the target domain feature extractor by maximizing the distance between the two domains.The algorithm is experimentally verified on the bearing dataset of Case Western Reserve University.The results show that the algorithm is more stable during training and has a higher accuracy rate.The algorithm in this paper has achieved good results,which verifies the feasibility of transfer learning in cross-condition bearing fault diagnosis.