The Research Of Rolling Bearing Vibration Signal Processing And Fault Recognition Method

In recent years, the study work on rolling bearing condition monitoring and faultdiagnosis get more and more attention, and the related theory research was rapidlydeveloped. Usually, the ideal means of the bearing condition monitoring and faultdiagnosis were based on the vibration signal processing way. The signal noise reductionmethod, the signal decomposition method and the application of neural network in faultdiagnosis were introduced detailed. Then, a bearing fault diagnosis method based onlocal mean decomposition and probabilistic neural network was proposed in this paper.To the four simulation modes of bearing signal acquired from test rig, themaximum correlated kurtosis deconvolution(MCKD) way was used to noise reduction.Then, the local mean decomposition was used to decompose the signal into a serialnumber of PFs (product function) for better signal feature extraction. then, with thecharacteristic vector which made of fault feature as the input samples of probabilisticneural network to complete identification. The main work and study conclusions are asfollows:(1) The application of traditional time-frequency analysis to non-stationary signalwas expounded, and the defects were pointed out. The rolling bearing fault mechanismwas analysised and a new noise reduction method for noise reduction to the four kindsof fault signals was introduced. According to the characteristics of rolling bearingvibration signals, a new adaptive signal processing method-Local Mean Decompositionwas cited to decompose bearing vibration signal，then，each component that reflects thebearing vibration was separated.(2) The local mean decomposition speed was improved, local mean decompositionalgorithm contains three important cycles, the three cycles enables signal decompositionat a slower speed. According to local mean decomposition slower speed problem,combined with engineering application, a corresponding improvement way was given,the signal decomposition speed was improved, the quantity engineering analysis wasreduced.(3) Combine the local mean decomposition with probabilistic neural network tofinish rolling bearing fault recognition. Extract the feature of each PF component andstructure characteristic vector as probabilistic neural network’s input data on faultdiagnosis, experiments proved the effectiveness and feasibility of the method.(4) Built a system for rolling bearing vibration signal time domain and frequencydomain analysis, time-frequency analysis, sample identification to four bearing models, completed system development based on MATLAB GUI. The system includes twosubsystems, subsystem1can make the signal processing and analyzing more direct andsimple, subsystem2can identify four kinds of faults to complete the bearing faultdiagnosis.Local mean decomposition theory in the bearing fault diagnosis of fever brought inbearing vibration signal processing worthy of further study.