Fault Diagnosis Of Rolling Bearing Based On Improved Wavelet Threshold De-noising Method And SVM

As the most common rotating component, the quality condition of rollingbearing can directly affect the performance of the whole mechanical equipmentand the safety of the job site. Therefore, continuing to carry out further researchon fault diagnosis of rolling bearing still has very important practicalsignificance.The essence of fault diagnosis of rolling bearing is state recognition, mainlyincluding signal acquisition, feature extraction and pattern recognition. In viewof that SVM can commendably solve the small sample learning problem, highdimensional pattern recognition problem and other difficult problem, the SVMwas introduced into the fault diagnosis of rolling bearing. And according to thecharacteristics of rolling bearing fault diagnosis, a set of method and specificprocess for parameters optimization of SVM was put forward, which wassuitable for bearing fault diagnosis. This method not only used average precisionof cross validation as the optimization goal, but also added control to the valueof penalty factor. Besides, in order to enhance the utilization value of rollingbearing samples and in view of the characteristic that support vector machines are sensitive to noises, a new method for rolling bearing running staterecognition based on wavelet threshold de-noising and SVM was proposed here.The samples from bearing vibration signals were de-noised with waveletthreshold de-noising method, and the corresponding de-noised samples were got.On this basis, the SVM model was established preliminarily by combining withparameter optimization of SVM. Then, the samples classified incorrectly werede-noised afresh and the SVM model were reconstructed until the penalty factorand the accuracy of cross validation met preconcerted requirements, therebyestablishing the most optimal SVM model and identifying rolling bearingrunning state.However, the disadvantages of the traditional soft-thresholding andhard-thresholding functions restrict the effects of signal de-noising and featureextraction. They cannot realize the adjustability of de-noising processing. Tosolve this problem, an improved wavelet thresholding function was presented,and the advantages of this function were verified by mathematical method andthe results of MATLAB simulation.Finally, a set of data including vibration acceleration signals of rollingbearing was analyzed respectively by method which was proposed in this paper,other SVM methods and BP neural network. The final results of diagnosisindicated that introduction of the improved wavelet threshold de-noising methodcan effectively enhance utilization rate of sample data, the generalizationcapability and anti-noise property of SVM and the reliability of the intelligent diagnosis of rolling bearing.