Mechanical Fault Diagnosis Based on Variational Bayesian Hidden Markov Model

Abstract

In order to overcome the deficiency in the traditional static independent component analysis(ICA) source separation, i.e. the traditional static independent component analysis blind separation is incapable to separate the dynamic time series signals, through combining the advantage of Variational Bayesian Hidden Markov model(VbHMM), a blind source separation method of mechanical faults based on VbHMM is proposed. The characteristics of the proposed method are: the dynamics variation of the hidden states in the data generation process can be obtained by the HMM; and when the signal is dynamic and nonlinear, a series of time related information in the signal can be captured to improve the accuracy of blind separation. Finally, the effectiveness of the proposed method is verified by a bearing fault blind separation experiment.