Study On Fault Diagnosis Of Traction Motor Rolling Bearings Based On Mathematical Morphology

Rolling bearing is the core part of tranction system. Which the rolling bearing faulthappens, may have a serious effect on the stable operation of the whole system. For therolling bearing fault diagnosis a number of methods were put forward, mainly in view of thebearing vibration signal is analyzed. Because the traditional signal analysis methods havemany limitations, so continue to explore a new method of analysis to improve the diagnosticefficiency. Mathematical morphology as a new nonlinear non-stationary signal analysismethod, has been have been applied in many fields and is very suitable for the analysis of thebearing vibration signal. this thesis studied deeply on fault diagnosis methods based onmathematical morphology. The main contents are as follows:(1) To extract the fault characteristics of rolling bearings, whose fault signal is usuallymodulated to high frequency with lots of noise, this paper presents a method combining EMDand adaptive generalized morphological filtering based on LMS, it first uses EMD acquirehigh frequency signal and separates low frequency interference and noise. Then it uses theLMS morphological filtering and closed operation method to demodulate forms. At lastcharacteristics are extracted through frequency analysis. The experiments have proved thatthis method can effectively extract the fault characteristics of rolling bearings.(2) A method to diagnose the faults of rolling bearings accurately was presented，intowhich the1(1/2)-dimensional spectral entropy is introduced. First，the original fault signalgoes through EMD and gets some IMF. Second，the1(1/2)-dimensional spectral entropy valuewas calculated by the IMF we get. Then the value is input to the Elman Neural Network as anew eigenvector to characterize the fault type of the rolling bearing. At last，the fault statusand fault type of the rolling bearing were distinguished. Simulation analysis and experimentalstudy show that this method can effectively extract the fault features of rolling bearings.Compared with the Wavelet Packet Analysis-Neural Network fault diagnosis，this method ismore feasible and effective with a higher recognition rate．(3) In view of the limitations of mathematical morphology filter, this paper describes anew approach to extract rolling bearing fault features based on LMD and adaptivemorphological filter of multi-structure elements and multi-scale. Stimulation experimentsshow that this method can better avoid the influence of noise and extract impulses from theoriginal signals to effectively diagnose rolling bearing fault, comparing with three othermethods which use envelope demodulation, LMD-morphological close method andLMD-morphological difference filter. Then we can distinguish the fault status and type of therolling bearing by the Elman Neural Network.