Abstract：During the operation of a synchronous generator, the shaft voltage signal on the shaft system can, to some extent, reflect the equipment’s operating condition. This paper proposes a defect identification method for synchronous generators using the shaft voltage signal as a feature signal, based on time-frequency transformation and a deep residual shrinkage network (DRSN). First, defect simulation experiments are conducted on a three-phase synchronous generator, where shaft voltage signals under normal conditions and three typical defects including rotor eccentricity, inter-turn short circuit, and electrostatic induction are acquired using an data acquisition card. Second, two time-frequency transforms, namely Mel spectrogram transform and synchrosqueezing transform, are respectively employed to construct datasets, and a deep residual shrinkage network is used to build a feature extractor. Finally, a full connected layer is adopted to classify the spectral features. The results show that the Mel-DRSN model achieves a classification accuracy of 99.05% for generator defects, outputing other compared models. The proposed method provides an effective aproach for defect identification in three-phase synchronous generators.
张新民, 常野, 朱建斌, 陈鸿昌. 基于轴电压信号时频变换和深度残差收缩网络的同步发电机缺陷识别[J]. 电气技术, 2026, 27(9): 21-27.
ZHANG Xinmin, CHANG Ye, ZHU Jianbin, CHEN Hongchang. Defect identification in synchronous generators based on time-frequency transformation of shaft voltage signals and deep residual shrinkage network. Electrical Engineering, 2026, 27(9): 21-27.
2026, Vol. 27 
