Sparse representation by novel cascaded dictionary for bearing fault diagnosis using bi-damped wavelet

• Published in: International journal of advanced manufacturing technology Vol. 124; no. 7-8; pp. 2365 - 2381
• Main Authors: Zhang, Long, Zhao, Lijuan, Wang, Chaobing
• Format: Journal Article
• Language: English
• Published: London Springer London 01.02.2023; Springer Nature B.V
• Subjects: Advanced manufacturing technologies; Algorithms; Bearings; CAE) and Design; Computer-Aided Engineering (CAD; Dictionaries; Engineering; Fault diagnosis; Feature extraction; Industrial and Production Engineering; Machinery condition monitoring; Manufacturing; Matched pursuit; Mathematical models; Mechanical Engineering; Media Management; Original Article; Representations; Rotating machinery; Singular value decomposition; Vibration monitoring; Wavelet transforms; Fault diagnosis; Bi-damped wavelet; Sparse representation; Feature extraction; Cascaded dictionary
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-022-10610-8

Vibration-based bearing condition monitoring of rotating machinery is of great importance for improving production efficiency and ensuring operational safety in the manufacturing industry. Sparse representation is able to effectively extract inherent impulse features from fault vibration signals corrupted by noise and harmonic interference, of which the performance is directly determined by dictionaries. In this study, the typical drawbacks of commonly used dictionaries are addressed using a novel cascaded dictionary. Period-assisted bi-damped wavelets with specific shapes are employed as the initial dictionary atoms to achieve overall matches with impulse features. Subsequently, the initial atoms are subjected to the K-singular value decomposition (K-SVD) for a secondary learning to obtain a cascaded dictionary that matches the real impulse features globally and locally. Finally, faulty vibration signals are recovered in segments using the cascaded dictionary and orthogonal matching pursuit (OMP). The results on the signals from the simulations, experiments, and real-world engineering confirm that the proposed cascaded dictionary consistently outperforms three other leading methods. Furthermore, the proposed cascaded dictionary is proved to be suitable for practical engineering diagnosis because of its outstanding anti-noise capabilities and self-adaptability.