| As an important part of mechanical equipment,rolling bearings play an increasingly important role in modern industrial systems.In the process of industrial production,rolling bearings often fail due to various factors,such as performance degradation and component damage,which will cause serious harm to production enterprises and society.Therefore,it is necessary to propose a feasible fault diagnosis method for rolling bearings to identify the types of faults in time and reduce property losses caused by faults.Aiming at the fault diagnosis of rolling bearings,this paper establishes the corresponding fault diagnosis model from the two operating conditions of single working condition and variable working condition in real situation.The work done in this paper is summarized as follows:(1)Aiming at the problem that the traditional signal processing and machine learning methods can not fully extract the fault features and the model interpretation is poor,this paper first uses the Variational Mode Decomposition(VMD)method to decompose the vibration signal to obtain the Intrinsic Mode Functions(IMF);Secondly,the IMF component is subjected to Hilbert transform(HT)to obtain the envelope signal;then select the representative timefrequency domain statistical features in the envelope signal;Finally,a multi-scale convolutional neural network(MSCNN)is constructed to fuse shallow features to obtain deep feature information,so as to establish a single-condition fault diagnosis model and realize the fault diagnosis classification of rolling bearings,and the fault diagnosis and classification of rolling bearings is realized.The experimental simulation is carried out in the Case Western Reserve University(CWRU)dataset and the Paderborn University(PU)dataset.By comparing with other methods,the applicability and effectiveness of the research method in this paper for fault diagnosis problems with small sample size are proved.(2)Aiming at the problem of insufficient accuracy of traditional transfer learning methods in variable condition fault diagnosis,this paper uses the idea of extracting domain invariant features in domain adaptive methods to extract common features in source domain and target domain.The feature extraction module of traditional Domain-Adversarial Neural Networks(DANN)is improved,and a variable condition fault diagnosis model based on multi-channel CNN-LSTM-ECA network is proposed.Firstly,Convolutional Neural Networks(CNN)and Long Short-Term Memory(LSTM)are used to extract time series features.Secondly,an improved Efficient Channel Attention(ECA)module is added to increase the weight of effective feature channels.Finally,the frequency domain data of the source domain data after Fast Fourier Transform(DFT)is input into the network model for training.The simulation experiments on CWRU and PU data show that the proposed algorithm can effectively suppress the domain offset between the source domain and the target domain under variable operating conditions,and improve the fault diagnosis accuracy.(3)Aiming at the problem that the source domain information is not fully utilized in the single source domain variable condition fault diagnosis,which leads to the suboptimal solution of the fault diagnosis accuracy.This paper proposes a method based on multi-source domain adversarial learning strategy.Based on the single source domain variable condition fault diagnosis model,this method fuses the fault feature information in multiple source domains by designing multiple domain clas sifiers.Finally,the frequency domain data of multiple source domain data after fast Fourier transform are input into the network fault diagnosis model for training.The fault diagnosis experiment is carried out on the PU dataset,and the relevant indicators are used to prove that the proposed method can effectively extract the common features between multiple source domains and target domains under variable operating conditions,and improve the fault diagnosis accuracy. |
