Research And Application Of HSMM In The Fault Diagnosis For The Rolling Bearing

Rolling bearing is the basic part of engineering machinery and other machineries;many mechanical faults are related with bearing condition. According to statistics,thirty percent of faults in rotating machinery are caused by the bearing. Whether thebearings are normal or not affects working state of the whole mechanical device.Therefore it is of practical value to study fault diagnosis technology for rollingbearing.The purpose of the research is aiming to solve the problem of fault prognostics ofrolling bearings. The mechanisms of the faults and the trends of failure process areanalyzed systematically. Hence a fault diagnosis technology of rolling bearings basedon IMF energy moment and Hidden Semi-Markov Models (HSMM) is studied in thisthesis. The main contents of the research are as follows:（1）Extraction method of the eigenvector of roller bearing IMF energy momentIn this paper, the self-adaptive method of EMD has applied in the eigenvalueextraction of the rolling bearings fault, and put forward eigenvector extraction methodbased on IMF energy moment. This paper has proved that the engenvector extractionmethod based on IMF energy moment will more remarkably display the differencesbetween non-stationary signals, compared to the wavelet packet frequency bandenergy method. In this paper the wavelet transform is adopted to denoise the vibrationsignal, so the interference composition of vibration signal is effectively eliminated,which improves signal-to-noise ratio.（2）Fault diagnosis of rolling bearings based on HSMMGiven that traditional method of HSMM parameter initialization for faultdiagnosis (random method, K-means) can lead to convergence in local optimization ofmodel parameters rather than global optimization problems, A modified algorithm ofK-means is proposed and improves the fault diagnosis rates which comparing withtraditional methods. In order to solve the problem such as the uncertainty of parametersetting, avoiding underflow due to the multi-samples training, a modified algorithm ofHSMM is studied in detail. Based on the above researches, a fault prognostic methodof rolling bearings based on HSMM is studied through taking the IMF energy momentas the prognostic feature information.（3）Verification for fault diagnosis of rolling bearing based on HSMM Finally, an experimental plan of rolling bearings has been designed and provedthat, in the process of fault diagnosis of rolling bearings, the combination of IMFenergy moment and HSMM will remarkable increase the precision and the real-timecharacteristic compared to the traditional HSMM.At the end of this thesis, the summarizations of the research and expectation ofthe related technology development are presented.