Research On Fault Diagnosis Methods Of Electric Traction Rectifier Based On CEEMD-DNN

High-speed railway has a significant purpose in China’s transportation industry.Its rapid development promotes the development process of China’s economy and society.As an important unit of high-speed train(HST),the role of electric traction system is to provide power.As one of the main components of the traction system,the normal operation of rectifier is also the basis to promise the secure and stable running of the HST.The odious working environment and changeable working conditions lead to frequent device failure of rectifier.The complex topology makes it very hard to locate the faulty components.To settle those problems,based on Complete Ensemble Empirical Mode Decomposition – Deep Neural Network(CEEMD-DNN),an opencircuit fault diagnosis method is proposed.The main studies are as follows:1.Simulation of open-circuit fault of rectifier based on Matlab/Simulink.The opencircuit fault of Insulate-Gate Bipolar Transistor(IGBT)of rectifier is simulated in Matlab/Simulink.The current signal of the grid side is collected for analysis.Time frequency analysis methods,such as Fast Fourier transform(FFT)、Short-Time Fourier Transform(STFT),are used to analysis and process the fault current.2.Effective eigenvector selection based on EMD.In view of the fact that the original signals of IGBT open-circuit fault happen in the same conduction interval are very similar,the adaptive decomposition algorithms,such as Empirical Mode Decomposition(EMD)、 Ensemble Empirical Mode Decomposition(EEMD)and Complete Ensemble Empirical Mode Decomposition(CEEMD),are proposed to decompose the fault currents into a series of IMFs.Then the correlation coefficient method is used to calculate the correlation between each IMF and the original signal.Next the Intrinsic Mode Function(IMF)which has the greater correlation with the original signal are chosen as the input eigenvectors of the follow-up fault diagnosis model.3.Establishment the open-circuit fault diagnosis models,DNN,Dense Net,CRNN.Aiming at the division of the feature of the original signals are unconspicuous,this paper proposes deep neural networks such as Deep Neural Network(DNN),Densely Connected Network(Dense Net),Convolutional Recurrent Neural Network(CRNN)to realize feature extraction and fault classification.Taking the original signals as the input feature vectors of the above three networks,the multi-scale features in those signals are extracted through those models and the IGBT open-circuit faults of the rectifier are located.4.Establishment the open-circuit fault diagnosis model based on CEEMD and DNN.Considering the advantage of CEEMD in signal decomposition and DNN in deep feature extraction,a composite fault diagnosis model based on CEEMD and DNN is proposed.Using the correlation coefficient method to select the IMFs.The decomposition results of EMD,EEMD and CEEMD,which are highly correlated with the original signals are chosen as the eigenvectors to input into the DNN.Then the DNN is used for deeper feature extraction to lacated the fault IGBT of rectifier.