Research on Estimation Method of Mechanical Fault Source Number Based on VbHMM

Abstract

The traditional source number estimation method must ensure that the signal sources are independent and noise-free interference. Based on the above deficiency in the traditional BSS method, combining variational Bayesian hidden Markov model (VbHMM) and autocorrelation determination (ARD), a estimation method of mechanical fault sources number based on VbHMM is proposed. In the proposed method, after the Bayesian networks are introduced, the Markov models (HMM) is used to capture the characteristics of a series of time-related time series information in the dynamic and nonlinear signals. The optimal number of hidden sources in the non-stationary signal is deduced by the unique model comparison function of Bayesian inference and autocorrelation determination (ARD). Simulation and experimental results verify the effectiveness of the proposed method.