Bearing Fault Diagnosis Using Multiclass Self-Adaptive Support Vector Classifiers Based on CEEMD–SVD

Bearing fault diagnosis under variable conditions has become a research hotspot recently. To solve this problem, this paper presents a new classifier: multiclass self-adaptive support vector classifier (MSa-SVC). Firstly, self-adaptive SVC is created by combination of SVC and information geometry. Then, several binary Sa-SVCs are constructed as a multiclass classifier for fault diagnosis. The proposed MSa-SVC, in conjunction with complementary ensemble empirical mode decomposition (CEEMD) and singular value decomposition (SVD) is utilized for bearing fault diagnosis: (1) each signal is processed into singular features by CEEMD–SVD. (2) MSa-SVC is used for fault clustering under variable conditions. Finally, the proposed method was applied on bearing fault diagnosis in practice. The results show that this method provides an efficient approach for bearing fault diagnosis under variable conditions.