Research On Vibration Signal Feature Extraction And Fault Diagnosis Of Rolling Bearing

Rolling bearings become the most widely used component in rotating machinery due to the small friction,simple lubrication,easy replacement and various other advantages.The normal operation state of the rolling bearings is very important to the device.Therefore,it's very necessary to study the fault diagnosis of the rolling bearings.Some existing vibration signal analysis methods,such as Fourier transform,wavelet analysis,blind source separation,Hilbert-Huang Transform and higher-order statistics analysis have very good results for non-stationary signals,but now there are also some defects and shortcomings that is difficult to overcome.To analyze vibration signals better,in addition to improving the existing signal analysis method,some other good methods applied well in other fields can also be introduced.In this paper,the probability density in statistics is firstly applied to fault diagnosis of rolling bearings.After finding that the probability density estimation method of the existing software is not effective to extract the feature information of the rolling bearings,the probability density is calculated accurately according to the original definition of the probability density against the deficiencies of probability density calculation method in existing software.And the feature information of the rolling bearings is further extracted.Secondly,after calculating the probability density,the multi-fractal spectrum is calculated and the multi-fractal parameter is extracted as the feature vector of the rolling bearings.After using the two methods to extract the feature of rolling bearings,Support Vector Machine is used to realize the classification and recognition of rolling bearings' various working conditions.However,when the fault conditions of rolling bearings increase,compared with the probability density method,the classification of multi-fractal algorithm would be slightly worse.Therefore,after confirming the advantage of probability density algorithm for fault diagnosis of rolling bearings,in order to improve the classification accuracy,on the basis of obtained probability density of the signal,the more specific fault feature information of rolling bearings is extracted and fault severity of rolling bearings are recognized well through calculating the value of correlation statistics,and an effective fault diagnosis method of rolling bearings is presented.