Rolling Bearings Fault Diagnosis Method Under Variable Condition Based On Deep Transfer Learning And Domain Adaptation

With the continuous development of industrial society,the production of all walks of life is inseparable from electrical equipment.Among them,the rotating motor plays an important role,and its smooth operation is of great significance for industrial production.Rolling bearing is an important part of rotating motor.Once the fault occurs,it will affect the normal operation of the motor.The actual working condition of the motor is often complex and changeable due to the load and current changes.Therefore,it is of theoretical value and practical significance to carry out fault diagnosis under variable working conditions.This paper takes the rolling bearing of rotating motor as the research object,and introduces the deep learning method and domain adaptive algorithms into fault diagnosis.Aiming at the complex and changeable working conditions in actual industrial production,a fault diagnosis method based on transfer sparse coding and joint geometric and statistical alignment(TSCJGSA)is proposed.Firstly,the fast Fourier transform is used to extract frequency domain features from time domain signal,and then uses Transfer Sparse Coding(TSC)to extract deep features of frequency domain amplitude.Finally,Joint Geometric and Statistical Alignment(JGSA)algorithm is used for fault classification to realize fault diagnosis under variable conditions.In view of the problem of poor robustness of JGSA algorithm when the speed difference between the two conditions is large,two improved strategies of Laplacian graph matrix(LGM)and class conditional distribution weight(CCDW)are introduced,and three improved JGSA algorithms are proposed.Finally,aiming at the problem that the improved JGSA algorithm has a small increase in accuracy when classifying actual working condition data,the Local Preserving Joint Transfer(LPJT)algorithm is introduced,and the TSC-LPJT fault diagnosis model is proposed to effectively improve the fault diagnosis accuracy and robustness under variable working conditions.The effectiveness of the proposed classification method is verified by the experimental data of Case Western Reserve University and QPZZ-II rotating machinery.Experimental results show that TSC-JGSA method can achieve high classification accuracy of bearing fault diagnosis under variable working conditions.The improved JGSA algorithm can effectively reduce the distribution difference of data between domains,solve the problem of imbalanced data sets,and further improve the accuracy of bearing fault diagnosis.TSC-LPTJ method can solve the problem that the accuracy of improved JGSA algorithm is not greatly improved under actual working conditions and achieve better fault diagnosis effect.