A Rotating Machinery Fault Diagnosis Method Based on GVMD-PWVD and Improved Deep Transfer Learning

• Published in: 2025 2nd International Conference on Electrical Technology and Automation Engineering (ETAE) pp. 308 - 312
• Main Authors: Qin, Huabin, Zhang, Changhai, Mou, Qingzhi, Liu, Runyu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.05.2025
• Subjects: Adaptation models; Fault diagnosis; Feature extraction; Genetic algorithms; Measurement; Meta transfer; Metalearning; Metric learning; Rolling bearings; Time-frequency analysis; Transfer learning; Variational mode decomposition; Vibrations; Wigner-Ville distribution
• DOI: 10.1109/ETAE65337.2025.11089593

The fault diagnosis method that relies on deep learning necessitates a substantial amount of sample data and encounters challenges related to insufficient generalization capability when the operating conditions of the bearing change. This paper proposes a transfer learning method for diagnosing equipment faults using the time-domain and frequency-domain information representation of the vibration signals of electromechanical equipment to address the above challenges. Firstly, the variational mode decomposition (VMD) algorithm, optimized through a genetic algorithm (GA), is employed to extract the modal components of the original signal. In the feature extraction process, the correlation coefficient method is introduced to eliminate noise and other interfering information in the signals. Subsequently, the time-frequency representations of each decomposed vibration mode were obtained through the application of the pseudo Wigner-Ville distribution (WVD). Finally, a classification model based on meta transfer metric learning is used to achieve fault diagnosis of rotating bearings.