Recognition of Transformer Fault Based on Multi-modal Vibration Signal

Abstract

Listen

Translate article

The transformer is the core component of the distribution system, and its safe operation performance is related to the reliability of the whole distribution network. The accurate identification of transformer fault modes is an essential guarantee for active safety management. However, the complex mechanical structure and changeable working environment make many faults occur at the same time, which makes it difficult for the physical model based on single modal information to accurately identify the fault modes. Transformer vibration signal contains rich fault information, which is an important parameter to realize transformer fault pattern recognition. In this paper, a convolutional neural network model based on multi-modal vibration information is constructed to extract fault features from transformer vibration information and accurately identify transformer fault modes under complex working conditions. Firstly, the wavelet transform method is used to obtain the time-frequency domain modal information of the vibration signal. Secondly, a multi-channel convolutional neural network is constructed to extract the fault features of multi-modal vibration information, and the features of the last convolutional layer and the pooling layer of the previous layer are fused to retain the global and local features of the vibration signal, and the fused features are input to the fault mode of the transformer output in the classification layer. Experimental results show the effectiveness of the proposed method, and the proposed method has higher diagnostic accuracy than the shallow network structure.