Research on Generator Rotor Fault Diagnosis Based on CGAN and Features Fusion of Vibration Signals and Image
Yunjie Qi, Hong Qian, Jun Xu · 2023
This paper addresses the limitations of traditional vibration signals features in misalignment fault diagnosis, and proposes a method to fuse shaft trajectory image features with vibration signals features to compensate for the shortcomings in misalignment fault diagnosis. Firstly, the time domain, frequency domain and time-frequency domain features of the vibration signals are extracted, and Xgboost is used to select the features according to their importance. The identified axial trajectory map is fused with the selected vibration signals features to form the final feature vector. The unbalanced data set is then fed into a model integrated with multiple classifiers for training to obtain a diagnostic model that can accurately diagnose the fault. In the experiments, the original unbalanced small sample data was balanced by CGAN and the rotor fault diagnosis model achieved higher accuracy. The fault diagnosis model based on the features fusion of vibration signals and image proposed in this paper also achieves higher accuracy than the traditional rotor fault diagnosis method based on single extraction of vibration signals features and the traditional rotor fault diagnosis method based on single shaft trajectory diagram. It is demonstrated that CGAN can reduce the influence of unbalanced small samples on the diagnosis accuracy and the features fusion of vibration signals and image can compensate for the limitations of each diagnosis.