Power transformer Fault Diagnosis using Image processing and Convolution Neural Networks

Abstract

Abstract This paper presents a novel approach for fault diagnosis in power transformers using Artificial Intelligence (AI) techniques, specifically Convolutional Neural Networks (CNN), in conjunction with Dissolved Gas Analysis (DGA). CNN is a powerful tool commonly employed in image processing, and its application in transformer fault diagnosis is explored here. Traditionally, the IEC standard includes two primary DGA methods: the ratio method and graphical representation. Surprisingly, these two methods often yield disparate results when applied to the same input data, posing a challenge in fault detection. To address this issue, this paper introduces a unique algorithm based on CNN.The CNN algorithm is employed to diagnose faults in power transformers based on gas ratios. Experimental data sourced from the Punjab State Transmission Corporation Ltd (PSTCL) laboratory serves as input data to showcase the implementation of this innovative CNN approach. During the training phase, gas ratio samples conforming to the IEC method are used. To facilitate this, the data is transformed into image format. Ultimately, the trained classifier successfully identifies six distinct fault types in transformers using gas ratios as inputs. The results obtained through the proposed CNN algorithm exhibit remarkable performance in distinguishing various transformer fault types. This analysis demonstrates that the CNN-based approach significantly enhances the precision and accuracy of transformer fault identification.