Research On Fault Diagnosis Of Metro Traction Motor Bearing Based On Data And Characteristicss

The change of speed and load causes the coupling and distortion of the bearing signal of metro traction motor to different degrees in the structure,which leads to the overlapping of the frequency bands of different types of faults and makes it difficult to judge the fault types.Aiming at the problem that it is difficult to accurately diagnose the fault types of metro traction motor under complex working conditions,this paper proposes a fault diagnosis method of metro traction motor which combines wavelet packet decomposition,information fusion and convolutional neural network,and designs a diagnosis scheme to verify it.Six kinds of fault bearings were prefabricated in the test.Based on the actual running speed of subway traction motor bearing at 2400 RPM and the actual load at7 k N,three levels of speed and load were set according to the conditions of the test bed,namely the speed at 800 RPM,1600RPM and 2400 RPM,and the loads at 5k N,7k N and 9k N.The whole test was divided into 54 sub-tests.And according to the test needs to adjust the control system and acquisition system,complete the signal acquisition.After the signal pretreatment,the difference between the fault bearing is determined by the analysis of time domain and frequency domain.The wavelet packet is used to decompose the original data,and then the root mean square value(RMS),kurtosis value(Kr)and margin value(C)are extracted as the features and the feature vectors are formed.At this time,the features contain time-domain and frequencydomain information.The bearing signal processing program was written in Python to achieve the above functions,and the eigenvalues were obtained by processing the original data and saved to a file.After comparing the advantages and disadvantages of data layer fusion,feature layer fusion and decision layer fusion,the feature layer fusion is selected as the fusion scheme of vibration signal and sound emission signal.By analyzing the format and physical meaning of three eigenvalues in different frequency bands,a scheme of eigenvalue fusion based on relativization and normalization is designed.The structural parameters of the convolutional neural network are built and set up.The results show that the load has the least influence on the diagnosis.Increasing the operating condition range of the training set is helpful to improve the accuracy of diagnosing complex operating conditions.The diagnostic accuracy based on information fusion is generally higher than that of single sensor.When the data of all working conditions are mixed,the diagnostic accuracy based on vibration signal and sound emission signal is 96.30% and 92.59% respectively,while the accuracy of the diagnosis scheme based on information fusion is 100%,which effectively improves the accuracy of the diagnsis system.