While synchronous generators are in operation, faults such as rotor eccentricity, inter-turn short circuits, and static charge defects can compromise the motor’s safe functioning. To reduce the effects of these issues on equipment and system performance, it is essential to identify the fault types precisely and swiftly, followed by implementing appropriate actions. This project aims to achieve defect diagnosis of the motor by visualizing the bispectrum transformation of shaft voltage and employing a Deep Residual Shrinkage Network (DRSN).Firstly, physical simulation experiments on the motor were conducted to obtain shaft voltage data under different artificial defect conditions. The shaft voltage data were preprocessed through bispectrum transformation for feature extraction and visualization. A dataset comprising four operating conditions was constructed. A deep learning neural network incorporating Convolutional Neural Networks (CNN) and the DRSN was designed for defect classification. The model’s performance was evaluated on the test dataset, and the results showed an accuracy rate of 99.3%. Therefore, the intelligent defect detection model for synchronous generators based on the Bi-spectrum-DRSN transformation provides a reliable support and value for the safe and stable operation of the motor.