Rolling Bearing Faults Diagnosis Method Based on SVM-HMM

This paper presents a new scheme of bearing fault diagnosis based on SVM and HMM. Combining the classification ability of SVM and the ability of HMM to distinguish dynamic time series, by means of the sigmoid function and Gaussian model, we translate the information output of SVM into the form of posterior probability, and then introduce it into the observation probability estimation of hidden states in HMM model. Feature vectors used in diagnosis are established by AR parameters. The scheme was tested with experimental data extracted from the high frequency resonant vibration signal of bearing by wawelet analysis.