Multivariate comb mode decomposition and its application in mechanical compound fault diagnosis

Abstract

The mechanical system exhibits a complex structure with characteristics such as multi-source vibrations and strong coupling effects. Existing multivariate signal decomposition methods fail to adequately account for the underlying fault mechanisms, making it difficult to achieve decoupling and feature extraction for compound faults. To address this, this article proposes a novel multivariate signal decomposition method called multivariate comb mode decomposition (MCMD). The MCMD method consists of three key steps: Firstly, based on the frequency characteristics of mechanical faults, MCMD innovatively represents fault features using a frequency comb structure, decomposing the signal into mode components with comb-shaped frequency patterns. Secondly, MCMD introduces a joint spectrum segmentation algorithm to determine the frequency bands of the mode components, significantly reducing noise interference while achieving cross-channel modal alignment and information fusion. Finally, MCMD employs an iterative optimization process to search for the optimal frequency comb structure across all frequency bands, enabling precise separation of strongly coupled fault features. As a novel multivariate signal decomposition method, MCMD is particularly effective in decomposing compound mechanical faults, even in cases of spectral overlap. It integrates multidimensional vibration information to achieve the decoupling and feature extraction of complex faults. Its working mechanism is fundamentally different from existing multivariate signal decomposition methods. Experimental results applying MCMD to mechanical fault diagnosis validate its effectiveness in decoupling composite faults and extracting their features, offering a fresh perspective for mechanical fault diagnosis.