Research On Fault Diagnosis Method For Roller Element Bearing Based On Vibration Signal

Rolling bearing is one of the most common components, and it is widely used in rotating machinery. The working state of rolling bearing can directly affect the safety of the equipment. In order to ensure the normal operation of equipment, it is imperative for rolling bearing to carry out the condition monitoring and fault diagnosis.Due to the traditional fault diagnosis methods are difficult to fully characterize complex fault types, but it exists the nonlinear relationship between fault mode and fault feature vector of rolling bearing. This paper starts from the vibration signal processing, and Support Vector Machine(SVM) is used as pattern recognition method. Various statistical parameters which can be used as fault characteristics are analyzed and compared, and select the ones which have good and stable performance as the fault characteristics. Besids, combining wavelet packet decomposition with sample entropy is taken as another method of rolling bear fault feature extracting. Thus two feasible methods for rolling bearing fault diagnosis are proposed.Method one: The statistical parameters of vibration signal which are good and stable are extracted respectively in the time and frequency domain. 8 d feature vector is constructed in the form of multi featuren fusion, and the SVM method is employed to classify and forecast.Method two: Combining wavelet packet decomposition with sample entropy. First, the vibration signal is decomposed to 1,2,3 layer with wavelet packet method, and 2,4,8 frequency bands are obtained respectively.So there is a total of 15 frequency bands coupled with the original signal. Then, the sample entropy of each frequency division is calculated, and the trend of the sample entropy in each frequency division is used to characterize the fault types. Finally, the constructed 15 d characteristic vector is put into the SVM to classify and predict.Finally, in order to verify the feasibility and superiority of the presented methods, the two methods are tested with the bearing data of Western Reserve University. And the results are compared with the existing methods and it shows that the proposed methods have stable performance and better accuracy.