Feature fusion using kernel joint approximate diagonalization of eigen-matrices for rolling bearing fault identification. (22nd December 2016)

• Record Type: Journal Article
• Authors: Liu, Yongbin; He, Bing; Liu, Fang; Lu, Siliang; Zhao, Yilei
• Is Part Of: Journal of sound and vibration. Volume 385(2016)
• Page Start: 389
• Page End: 401
• Publication Date: 2016-12-22
• Subjects: Sound -- Periodicals; Vibration -- Periodicals; Son -- Périodiques; Vibration -- Périodiques; Sound; Vibration; Periodicals; Electronic journals; 620.205
• DOI: 10.1016/j.jsv.2016.09.018
• Languages: English
• ISSNs: 0022-460X

Abstract

Abstract: Fault pattern identification is a crucial step for the intelligent fault diagnosis of real-time health conditions in monitoring a mechanical system. However, many challenges exist in extracting the effective feature from vibration signals for fault recognition. A new feature fusion method is proposed in this study to extract new features using kernel joint approximate diagonalization of eigen-matrices (KJADE). In the method, the input space that is composed of original features is mapped into a high-dimensional feature space by nonlinear mapping. Then, the new features can be estimated through the eigen-decomposition of the fourth-order cumulative kernel matrix obtained from the feature space. Therefore, the proposed method could be used to reduce data redundancy because it extracts the inherent pattern structure of different fault classes as it is nonlinear by nature. The integration evaluation factor of between-class and within-class scatters ( SS ) is employed to depict the clustering performance quantitatively, and the new feature subset extracted by the proposed method is fed into a multi-class support vector machine for fault pattern identification. Finally, the effectiveness of the proposed method is verified by experimental vibration signals with different bearing fault types and severities. Results of several cases show that the KJADE algorithm is efficient in feature fusion for bearing fault identification. Highlights: Rolling bearing fault diagnosis method via feature fusion analysis is proposed. Nonlinear dimensionality reduction using JADE with the kernel function. Obtain sensitive features by KJADE algorithm. Verify effectiveness and robustness by qualitative and quantitative analysis.