Method of Gearbox Fault Diagnosis Based on Covariance Matrix Manifold Entropy Feature

摘要

For a complex vibration signal, the traditional sample entropy hardly quantifies itself inherent long-range correlation. The multivariate multi-scale entropy is a reflection of the nonlinear dynamic correlation as the promotion of multi-scale entropy in multivariable signal. Applying the improved multivariate multi-scale entropy, all variables can be embedded at the same time instead of the individual variable embedded mode in traditional multivariate multi-scale entropy. It not only solves the problem of memory overflow with the increasing channel number, but also is appropriate for the actual multivariable signal analysis. This paper analyzed the four-type signals, which were ones in normal operation, brokenperforation, mild wear and moderate wear of gear box. The experimental result showed that this method has a good performance to distinguish the correlation data and the fault diagnosis precision reached 98.4%.