Condition monitoring and fault diagnosis of flyback switching power supply

Abstract This paper presents a fault diagnosis method for flyback switching power supply. The proposed method integrates input current and output voltage information to improve fault diagnosis accuracy. The flyback switching power supply's signal characteristics and fault separability are analyzed. Time domain features and frequency band wavelet packet dispersion entropy features are constructed to form multidimensional feature vectors that fuse time and frequency information, which enhances fault separability. Then, the MIV algorithm is used to screen the features to reduce the redundant information. Additionally, the diagnostic method of the NGO‐SVM model is proposed to optimize the multiclass SVM model by using NGO to improve the diagnosis model generalization performance. The experimental results show that the method proposed in this paper has good diagnostic effect for both single and multiple faults, and the diagnostic accuracy is up to 98.3% under ideal conditions, and up to 96.8% in the presence of noise interference.