The Research on Blind Source Separation Method for Vibration Signals Base on the Second-Order Statistics

Abstract

A blind source separation (BSS) method base on second-order statistics was applied to separate the vibration mixtures based on the temporal structure of vibration signals. The single delay covariance matrix was used to compute the whiten matrix. A set of covariance matrices with different time delay was joint approximatively diagonalized to obtain their ‘average eigenstructure’. The separation of the real vibration mixtures from two motors illustrate that it is feasible to applied this method to the research of the health monitoring of the gearbox.