SVD-based dictionary learning for bearing fault diagnosis

Bearing fault diagnosis is of great importance to maintain the high reliability and long-term safe operation of rotating machinery. The key factor of the processing result is the proper selection of basis, which is also called dictionary in sparse representation. In this paper, a data-driven method for designing dictionaries called singular value decomposition (SVD)-based dictionary learning method is proposed. Combining the K-SVD scheme and the idea of SVD based on hankel-matrix, the proposed method can extract the inherent components of signals, thus realizing the goal of training dictionary. The proposed method is applied to simulated signal and practical application in fault diagnosis of bearings. The processing result demonstrates that the proposed method outperforms the K-SVD method in learning dictionaries from vibration signal of rotating machine.