A Novel Approach for Incipient Fault Diagnosis in Power Transformers by Artificial Neural Networks

A good functioning of power transformers must be assured, since they play a vital role in distribution and transmission systems, an early diagnosis of incipient faults in such equipment enables their predictive maintenance. Traditional dissolved gas analysis interpretation methods are the most popular approach for fault diagnosis in power transformers, however, their application suffers from several drawbacks; some studies have applied machine learning techniques towards overcoming them. This article introduces a novel methodology for incipient fault diagnosis in power transformers that uses artificial neural networks (ANN). It considers feature selection for multilayer perceptron (MLP) networks in a cascade structure for classifying faults. Its performance was compared with those of two other ANNs, i.e., a traditional MLP and another cascade structure with no feature selection. The methodology is of simple implementation, highly accurate, and capable of correctly classifying over 85% of the test samples.