Abstract: Aiming at the problem of difficulty in identifying the fault category caused by the nonlinear and non-smooth characteristics of rolling bearing fault signal, a fault identification method based on the combination of Local Mean Decomposition (LMD), multi-scale sample entropy and GG clustering algorithm is proposed. Initially, the fault signals of rolling bearings are decomposed using LMD to obtain multiple product function (PF) components, enabling the preliminary extraction of rolling bearing state features. Then, the optimal PF components are selected by correlation analysis, and the sample entropy value are computed at multiple scales. Finally, the principal component analysis is employed to visualize and reduce the dimensionality of the high-dimensional feature vectors. The new feature vectors are visualized by principal component analysis, which are then input into the GG clustering algorithm to realize the identification of rolling bearing fault categories. The results show that the proposed method has the advantage of better clustering effect compared with other pattern combination methods.

Key words: local mean decomposition, multiscale sample entropy, correlation analysis, GG clustering