Research On Fault Diagnosis Method Of Rolling Bearings Under Variable Working Conditions Based On Transfer Learning

A rolling bearing is the key component of rotating machinery,and its running state is influenced by many factors.In order to cope with the complex working environment,the working conditions of a rolling bearing are constantly changed,and the change of working conditions will directly lead to the change of the vibration information.Therefore,under variable working conditions,if the fault situation of the rolling bearing be diagnosed,it will be of great significance for ensuring the healthy running of mechanical systems.Two fault diagnosis methods for rolling bearings under variable working conditions are proposed in this research.(1)Based on variational mode decomposition(VMD)and structure of multi-feature combined with transfer learning,a fault diagnosis method for rolling bearings under variable working conditions is proposed.At first,the vibration signals of the rolling bearing are decomposed by VMD.The intrinsic mode function(IMF)components can be obtained and the matrix of IMF is decomposed by singular value decomposition to obtain the singular value and the singular value entropy,and multi-feature set is constructed by combining the time domain with frequency domain characteristics of the vibration signal.Then,the kernel function of the semi-supervised transfer component analysis(SSTCA)algorithm is improved by multi-kernel function,and multi-kernel SSTCA algorithm is proposed,which different feature samples are mapped into a shared reproducing kernel Hilbert space.At last,the maximum mean discrepancy embedding method is used to select the source domain feature samples which are more similar to the target domain feature samples,and the support vector machine(SVM)model be trained by the source domain feature samples,and which the mapped target domain feature samples be tested.The experimental results show that,the proposed multi-kernel SSTCA-SVM method has better results than other methods for fault diagnosis of rolling bearing under variable working conditions.In order to further characterize the vibration feature information of the rolling bearing,the features of vibration signals are adaptively extracted by deep learning,and another fault diagnosis method is proposed combining with the domain adaptation method of transfer learning.(2)Based on sparse denoising auto encoder(SDAE)and joint geometrical and statistical alignment(JGSA),a fault diagnosis method for the rolling bearing under variable working conditions is proposed.Firstly,the features of the frequency domain amplitude of the vibration signal are extracted using SDAE,then the domains shifting between the feature samples of two domains becomes smaller.Finally,the multi-state recognition of the rolling bearing under variable working conditions is realized by the K nearest neighbor(KNN)algorithm.The experimental results show that,the proposed deep transfer learning method compared with other based on SDAE methods,its sample feature visualization effect is the best after being transferred,and its fault diagnosis accuracy of rolling bearings under variable working conditions is better.Compared with method(1),the method is more suitable for occasion with higher real-time requirement,and method(1)is more suitable for occasion with higher fault diagnosis accuracy requirement.