Fault Diagnosis Of Rolling Bearing Based On Phase Space Reconstruction

Rolling bearing is a general machinery part, which is most widely used in varietyof rotating machinery. Whether it’s running properly tend to affect the performance ofthe whole machine. Therefore, the rolling bearing fault diagnosis has importantsignificance. Fault diagnosis of rolling bearing generally analyzes the signal representedby nonlinear time series, such as feature extraction, state recognition. The main contentsare as follows:(1) Rolling bearing often contains noise signal, in order to reduce the inlfuence ofnoise to the feature extraction,it is necessary prior to feature extraction for the noisereduction processing of signal.In this thesis, based on phase space reconstructiontechnique, the principal component analysis (PCA) noise reduction algorithm isproposed. The results show that the algorithm can effectively reduce the noise throughthe analysis of the simulation signal and the rolling bearing fault diagnosis dataexperiment.(2) The feature extraction of morphology filter is proposed in combination with thePCA based on the phase space reconstruction. After the signal is denoised by the PCAbased on the phase space reconstruction, firstly use the PCA bsed on phase spacereconstruction to conduct the signal noise reduction, then filter it to extract the faultfeature. The results show that these algorithms can effectively extract the fault featuresthrough the analysis of the simulation signal and the inner ring and outer ring faultdiagnosis data experiment of rolling bearing, proving the effectiveness of the proposedalgorithm.(3) The feature extraction of local mean decomposition (LMD) is proposed incombination with the PCA based on the phase space reconstruction. After the signal isdenoised by the PCA based on the phase space reconstruction, firstly use the PCA bsedon phase space reconstruction to conduct the signal noise reduction, then the de-noisedsignal is decomposed by LMD, the first component PF1which contains the highestpower is selected to conduct the envelope spectrum analysis and the fault features areexacted. The results show that the algorithm can effectively extract the fault featuresthrough the analysis of the simulation signal and the rolling bearing fault diagnosis dataexperiment. (4) Rolling bearing state recognition algorithm of multiscale and support vectormachine is studied finally. Appropriate scale is selected to construct feature vectors bycalculating the multiscale permutation of four state in every scale，then select a numberof feature vector samples and use support vector machine classifier to recognize therolling state，the results show that the method has a high recognition rate for the fourstates of normal，inner failure，outer failure and rolling body failure of rolling bearing.