Proposition and Validation of Computational Solutions for Fault Detection in Power Transformers Based on the Duval Pentagon

Abstract

Power transformers are used in substations of electrical power systems for transmission and distribution of electricity between generators and end consumers. Efficient insulation and cooling systems provided by insulating oil are essential for a transformer operates properly. During transformer operation, insulating oil can have changes in dissolved gas concentrations resulting from thermal or electrical discharges and the follow-up and monitoring of the proportion of these gases guarantee the transformer health. Among the methods used for this analysis and delimitation of faults, there is Duval Pentagon. The authors hope that the implementation of the algorithm, the aim of this work, will help in the interpretation of the Duval Pentagon, contributing to data analysis and preventive intervention in the event of a detected fault. A comparative experiment was carried out using Artificial Neural Networks. The proposed methods provides a direct identification of the fault, corresponding to the proportions of gases present in a sample of the insulating oil of a transformer under analysis.