Fault diagnosis in high voltage transformers with Artificial Neural Networks Ricardo Luiz Nacarato, Dorotéa Vila nova Garcia, Maurício Conceição Mário, João Inácio da Silva Filho

Abstract

In the present work, in its introduction, some aspects of the high power transformers, fundamental parts in the electric power distribution systems, are presented, theoretically grounding the power transformers, the power transformers diagnostic methods and the artificial neural networks (RNA ). The main objective of this work is to describe the use of an artificial neural network (RNA) of the multilayer type and its feasibility as an auxiliary tool in the diagnosis of faults in high power transformers, and specific, the analysis of the dissolved gases in the oil, propose a solution for the automatic analysis of the tests performed on the transformer oil samples and to evaluate the possibility of providing tools for the rapid and accurate emission of reports and opinions. The materials and methods chapter details the neural network that was implemented, using the toolbox, in the analysis of 224 power transformers installed in the power substations of an electric sector operator. From the results it was verified that the architecture that obtained the best rates of identification of the diagnoses of the transformers were the networks with individually identified conditions. The focus of this project is the prevention of problems in power transformers through the technique of dissolved gases in the soda oil, concluding that the neural networks can aid in the learning of laboratory technicians in this process of analysis.