Gear fault diagnosis method based on local mean decomposition and generalized morphological fractal dimensions

• Minkowski–Bouligand is sensitive to load and a little bit sensitive to speed and signal length.
• GMFDs are a little bit sensitive to signal length, load and speed.
• LMD can extract feature information.
• Data source can be selected based on mutual information theory.

Aiming at gear fault diagnosis, a fusion method of local mean decomposition (LMD) and generalized morphological fractal dimensions (GMFDs) is proposed. Firstly, a signal is decomposed by LMD into several product functions (PFs) which have physical meanings. Secondly, mutual information entropy value between each PF and original signal can be computed, and the PF corresponding to the maximum value is considered as containing the richest feature information of original signal, thus the PF is used as data source. Lastly, GMFDs are extracted from the data source, and some GMFDs which can quantitatively and comprehensively characterize nonlinear information of gear running states are adopted as feature vectors, hence gear faults can be diagnosed by kernel fuzzy c-means (KFCM). In order to demonstrate superiority of the proposed method, the GMFDs are extracted from signals of different lengths, ones sampled under three different working conditions of load and speed, ones without decomposition of LMD. The gear signals are tested and verified, and the result demonstrates that the proposed method is superior and can diagnose gear faults accurately.