Transformer fault diagnosis based on probabilistic neural networks combined with vibration and noise characteristics

Abstract

When the transformer is running, the vibration which is generated in the core and winding will spread outward through the medium of metal, oil, and air. The magnetic field of the core changes with the variation of the transformer excitation source and the state of the core, so the corresponding vibration and noise change. Therefore, the vibration and noise of the transformer contain a lot of information. If the information can be associated with the fault characteristics of the transformer, it is significant to evaluate the running state of the transformer through the vibration and noise signal, which improve the intelligence, safety, and stability of the transformer operation. Based on this, modeling and simulation of transformer multi-point grounding, DC bias, and short-circuit between silicon steel sheets fault are first carried out in this paper, and vibration and noise distribution of transformer under different faults are given. Second, a fault diagnosis method based on transformer vibration and noise characteristics is proposed. In the process of implementation, vibration and noise signals under multi-point grounding, DC bias, and short-circuit between silicon steel sheets are taken as the sample data, and the probabilistic neural network algorithm is used to effectively predict the transformer fault. Finally, the effectiveness of the proposed scheme is verified by identifying the simulation faults-the proposed fault diagnosis method based on PNN can be effectively applied to transformer.