Abstract:

A new method for performance degradation evaluation of rolling bearings based on multiple support vector machine and indicator selection was proposed in this paper. The degree of fault at different stage in rolling bearing full life is different. In the study, fault monitoring was divided into 4 parts including normal stage, incipient stage, mid-term stage and final stage. The advantages of spectral kurtosis which was sensitive to incipient fault was demonstrated by comparing with traditional indicators such as RMS, kurtosis value and peak to peak value etc. Then, the spectral kurtosis and other indicators were selected as inputs, and the multi-class support vector machine prediction model was constructed to predict the degradation stage of rolling bearing. The prediction model was finally tested by bearing full-life test data, which verified the effectiveness and feasibility of the method and had practical engineering significance.

Key words: spectral kurtosis, multi-class support vector machine, fault monitoring, rolling bearing, performance degradation assessment, machine learning