Roller bearing fault diagnosis based on LMD and multi-scale symbolic dynamic information entropy

摘要

This paper presents a new fault feature extraction method based on the combination of local mean decomposition (LMD) and multi-scale symbolic dynamic information entropy (MSDE). The LMD method decomposes the multi-component signal into a finite number of product functions (PFs) to extract the characteristic information of the original signal. In this paper, the comparison of MSDE and the multi-scale sample entropy shows that the symbolic dynamic information entropy (SDE) has simple calculation and has better stability than sample entropy at multiscale. Entropy values of PFs that combine LMD and MSDE are extracted as the feature sets, which are inputted into the affinity propagation clustering model to identify the fault types and fault degree of roller bearings. This paper also discusses the influence of different filtering algorithms on the noise removal of bearing signal. Application results demonstrate the effectiveness of the proposed fault diagnosis method.