Clustering Group-Sparse Mode Decomposition and Its Application in Rolling Bearing Fault Diagnosis
Bin Pang, Bo-Cheng Wang, Yuzhi Hu, Tianshi Cheng, Zhenli Xu
Abstract
Group-sparse mode decomposition (GSMD) is a novel signal decomposition algorithm based on the concept of group-sparse. It is fast, robust against noise and has good anti-mode aliasing performance, and these characteristics show that it has great potential for bearing fault feature extraction. However, GSMD estimates the intrinsic mode functions (IMFs) by a set of ideal filters designed based on energy detection over short windows. It is suggested that GSMD is susceptible to mode-splitting of the fault impact component that has wide bandwidth when processing bearing fault signals. Therefore, in this paper, an improved version of GSMD, called clustering group-sparse mode decomposition (CGSMD), is put forward through the introduction of the mean shift clustering to overcome the mode-splitting of GSMD for fault impacts. The proposed method aims to facilitate fault diagnosis for those rolling bearings running at constant speeds. Firstly, GSMD is employed to decompose the vibration signal and acquire the preliminary IMFs. Then, these IMFs are grouped based on mean shift clustering algorithm. Finally, the ultimate decomposition components are obtained by combining the IMFs in the same cluster. The analysis of simulated and experimental bearing fault data indicates that CGSMD is capable of overcoming the mode-splitting associated with GSMD and results in improved fault feature extraction performance. Moreover, it outperforms the clustering variants of state-of-the-art signal decomposition methods such as variational mode decomposition (VMD) and empirical Fourier decomposition (EFD).