Reseach On Fault Diagnosis Methods For Rolling Bearing Based On Relevence Vector Machine

Rolling bearing is a kind of precision component and also is a key quick-wearing par ofrotating machinery. Because of different reasons, it is more vulnerable than the other. So, thenegative effect and pecuniary loss should not be overlooked when it is widely used. To takeeffective diagnostic methods and timely exclude operational failures has become one of theactual and key focus on studying rolling bearing.This article summarizes vibration principle and characteristics of the different rollingbearing faults, makes researches on the basic theory of Relevance Vector Machine, discussesthe function of kernel function, introduces a new intelligent diagnosis method based onRVM for four condition of rolling bearing: the normal, inner fault, outer fault, rolling bodyfault. First, failure data is decomposed by harmonic wavelet packet to obtain feature vectorbased on wavelet decomposition coefficients at any frequency. Second, the modified kernelfunction which optimize the traditional Gaussian keeps certain attenuation when the distanceis infinitely far from the test points, combines with “K Nearest Neighborhood” algorithmimprove recognition efficiency. Third, Decision Tree-DT RVM model is constructed byRVM and "Decision Tree" algorithm for fault identification problems, which put easyidentification pattern on front to improve accuracy. Finally a new algorithm is proposed baseon One Against One, which combines with RVM to build diagnosis model for judging faults.The actual performance of the different diagnostic methods is verified with the vibration dataof rolling bearing in the case western university. The results show the proposed methodpossesses higher accuracy and efficiency when diagnosing roller bearing faults compared withSupport Vector Machine method, and the number of relevance vector is fewer. Differentproposed methods based on RVM all reduce fault recognition error rate and the needed timeon the different level.