Transformer Incipient Fault Detection Technique Based on Neural Network

Abstract

The paper presents an artificial intelligence approach to transformer incipient fault diagnosis. Gas concentration data for hydrogen gas (H2), methane (CH4), ethane (C2H6), ethylene (C2H4) and acetylene (C2H2) are obtained and processed by determining their relative gas concentrations. The relative gas concentrations are further transformed using a coding approach. The transformed relative gas concentrations are then fed into a multilayer perceptron neural network whose outputs give diagnosis for incipient faults. The proposed approach was tested on 145 data sets of gas concentrations. Results obtained show that the proposed approach has good prospects for quantitative diagnosis of incipient faults.