Radial Deformation Emplacement in Power Transformers Using Long Short-Term Memory Networks

TL;DR

This paper proposes using LSTM neural networks to analyze frequency response data for early detection and localization of radial deformation faults in power transformers, improving diagnosis accuracy.

Contribution

It introduces a novel application of LSTM networks for feature extraction in transformer fault diagnosis, enhancing early detection of radial deformation.

Findings

LSTM-based method accurately detects RD faults

The approach effectively locates deformation in transformers

Experimental results confirm high diagnosis precision

Abstract

A power transformer winding is usually subject to mechanical stress and tension because of improper transportation or operation. Radial deformation (RD) is an example of mechanical stress that can impact power transformer operation through short circuit faults and insulation damages. Frequency response analysis (FRA) is a well-known method to diagnose mechanical defects in transformers. Despite the precision of FRA, the interpretation of the calculated frequency response curves is not straightforward and requires complex calculations. In this paper, a deep learning algorithm called long short-term memory (LSTM) is used as a feature extraction technique to locate RD faults in their early stages. The experimental results verify the effectiveness of the proposed method in the diagnosis and locating of RD defects.