Fault diagnosis of rolling bearings under variable conditions based on VMD multi-domain features and MEDA

Abstract

The rolling bearing data under different working conditions show different distribution, resulting in the classifier trained cannot meet the task of multi-condition fault diagnosis. To solve this problem, a domain adaptation method based on multi-domain features of rolling bearings was proposed. Firstly, the original signal is decomposed by parametric optimization VMD, singular value and permutation entropy are extracted, and the time domain features of the original signal are combined to form multi-domain features. Secondly, multi-domain features are embedded into Grassman manifold space by GFK to achieve feature reduction and optimization and eliminate redundant information. Thirdly, the manifold characteristics of source domain data and target domain data are dynamically aligned. Finally, the crossdomain classifier is trained to realize cross-working condition fault diagnosis of rolling bearings. The results show that the proposed method can achieve better performance than the traditional intelligent fault diagnosis method and domain adaptive method in different working conditions.