Analysis Of Vibration Characteristic And Intelligent Fault Diagnosis Of Rolling Bearings

Vibration fault diagnosis is an indirect diagnostic method based on the analysis of vibration signals.The vibration signal reaches the vibration sensor through the complex transmission path of the mechanical system,and the diagnosis information is unavoidably interfered in the transmission process.The uncertain fault diagnosis information transmitted from the complicated path is easily caused to the false diagnosis,which can lead to over-maintenance,erroneous repairs,or even significant safety accident.In order to solve the problem of rolling bearings',misdiagnosis caused by disturbance information in vibration transmission process,the research on rolling bearing fault modeling and transmission characteristics analysis,rolling bearing fault location,rolling bearing fault r ecognition and adaptability of rolling bearing fault diagnosis model are carried out.The research content of this paper are as follows:(1)Based on the coupling transfer process of rolling bearing fault vibration signal,the dynamic model of rotor-rolling bearing system considering the different slip ratio of bearing is established.The effect of dynamic combination of bearing stiffness on bearing vibration characteristics is analyzed.Based on Hertz contact theory,rolling bearings' fault models of outer-ring fault,inner-ring fault and roller fault are established.Based on analysis of dynamic characteristics and time-frequency characteristics of rolling bearing fault simulation signals on different measuring points,an expression of rolling bearing fault signal with amplitude modulation characteristics is proposed.The amplitude modulation information is taken as the research direction in the study of rolling bearing fault location method.(2)In order to solve the problem of rolling bearing fault location based on single-channel vibration signal,the fault location method of rolling bearing is proposed.The second-order cyclostationary of rolling bearing fault signal is verified based on cyclostationary theory.The discriminating indicators of the degree of cyclostationarity(DCS)is proposed for rolling bearing fault signals on different measuring points.Based on the prior knowledge of vibration coupling transfer mechanism,spectral kurtosis(SK)-DCS method and empirical mode decomposition(EMD)-DCS method are proposed.The DCS value of each frequency band decomposition signal is used as the characteristic parameter to determine the position relationship between the rolling bearing fault and the measuring point.A rolling bearing fault simulation test plateform is build to verify SK-DCS method and EMD-DCS method.(3)To solve the problem of identify the weak fault feature of rolling bearings,a long short-term memory recurrent neural networks(LSTM-RNN)model and the rolling bearing fault classifation method based on multi-label LSTM-RNN are proposed.Three feature vectors were tested,including square envelope signal spectrum,frequency spectrum of SK filtered signal and DCS of signal.The test results show that the square envelope signal spectrum has the best diagnosis effect.The validity of the multi-label LSTM-RNN fault recognition method is verified by the rolling bearing fault signals.The experimental results show that the multi-label recognition method is more correct than the single label recognition method by 30%.(4)To enhance the suitability of multi-label fault recognition model,a diagnosis method of fault signal based on transfer learning is proposed.A LSTM-RNN transfer learning model is trained by source domain samples and a small number of target domain samples.Based on this transfer learning model,the predicted auxiliary target domain samples is generated.The validity of the fault diagnosis method based on transfer learning under different working conditions and different structural parameters condition is verified.The diagnosis results under different target area sample size of transfer learning is studied.The results show that with increasing target sample size,the fault diagnosis result is improved,and the optimal effect of multi-label fault diagnosis model stop to be improved when the ratio of target sample to source sample reaches 20%.