Artificial neural network based identification of deviation in frequency response of power transformer windings

Abstract

Deformations in windings can be diagnosed by a reliable and powerful method called sweep frequency response analysis (SFRA). In this work the deviation in the frequency response plots is derived in terms of statistical indicators. Nine statistical indicators have been used for the purpose. These indicators, then, complemented using artificial neural network approach, to derive a useful conclusion regarding the deviation based on the frequency responses. Winding deformation case data along with healthy transformer case data have been used to train a multilayer feed-forward neural network with the backpropagation algorithm. The trained neural network can help an expert to analyse statistical indicators to verify the level of deviation and in turn the level of deformation.