Fault diagnosis of synchronous motors by fusing shaft voltage and vibration features

• Published in: Dianji yu Kongzhi Xuebao = Electric Machines and Control Vol. 29; no. 7; p. 53
• Main Authors: Zhang, Hang, Guan, Xiangyu, Liao, Jingwen, Xu, Xinling, Chen, Xiaokun
• Format: Journal Article
• Language: Chinese; English
• Published: Harbin Harbin University of Science and Technology 01.01.2025
• Subjects: Algorithms; Artificial neural networks; Correlation analysis; Correlation coefficients; Defects; Electric potential; False alarms; Fault detection; Fault diagnosis; Generators; Machine learning; Physical simulation; Short circuits; Spectrograms; Synchronous machines; Vibration analysis; Voltage
• ISSN: 1007-449X
• DOI: 10.15938/j.emc.2025.07.006

During the operation of synchronous generators, various defects such as rotor eccentricity, turn-to-turn short circuits, and static charges may occur, jeopardizing the safe operation of the motor. A method for diagnosing defects in synchronous generators by integrating shaft voltage-vibration features with deep learning was proposed, based on a nonlinear correlation analysis of shaft voltage signals and mechanical vibration signals under different defects. Firstly, a physical simulation test platform for defects in a three-phase synchronous generator was established to obtain data on shaft voltage and mechanical vibration signals under various operating conditions and defects. The kernel canonical correlation analysis（KCCA） nonlinear correlation analysis algorithm was used to obtain the correlation coefficients between shaft voltage signals and vibration signals. Mel spectrograms were employed for preprocessing the spectrograms of shaft voltage and vibration signals. A parallel double-branch residual neural network（ResNet） was utilized to extract high-dimensional features from both the shaft voltage and vibration spectrograms. Furthermore, a bilinear pooling algorithm was applied to fuse high-dimensional features from different modalities, leading to the construction of a classification model for defects in synchronous generators based on the integration of shaft voltage and vibration features. The results indicates that the correlation between shaft voltage signals and the vibration signals of the synchronous motor exceeded 0.9 in both faulty and normal conditions. The proposed shaft voltage-vibration joint diagnosis model outperforms single shaft voltage and single vibration diagnosis algorithms in terms of accuracy, missed detection rate, and false alarm rate on the test dataset. This work aims to enable timely identification of potential faults and improve the reliability of generator operation by monitoring and analyzing their operational state.