Transformer Fault Location Classification Using FFT Based 1D-Convolutional Neural Network Model

Abstract

Vibration signals serve as indicators of an electrical device's condition, comprising multiple harmonics that elucidate its operational state. Analyzing the harmonic frequency and magnitude within the vibration signal enables the identification of fault locations in the machine or device. This work proposes the use of a fault location classification based on Fast Fourier Transform (FFT) for feature extraction and an 1D Convolutional neural network model to distinguish the difference between 3 types of deformation conditions that were collected from transformer surface. The vibration sensor data in the time-series domain, thus the primary reason for the development of 1D-CNNs. A reduction in the amount of computational work that the network needs to do is made possible by the proposed 1D-CNN model. The primary objective is to analyze vibration signals of the transformer core in order to diagnose fault location, with the data being measured in the time domain. With a classification accuracy of 96.6%, the CNN model that was created for the purpose of detecting faults in transformer cores displayed an amazing performance.