Diagnóstico de falhas em transformadores de potência pela análise de gases dissolvidos em óleo isolante com a utilização de redes neurais

Abstract

The main objective of this work is to develop a method using artificial neural networks, to perform the mapping of the gases generated in the insulating oil of the power transformers, coming from faults such as electric arc, corona effect, overloads and others. We will also describe the methods of detection of dissolved gases in oil currently used. The advantage of using neural networks over existing methods is because the method is fully numerical, well adapted to modern computational solutions practices, does not require specialized human interventions, and the analyzes are processed quickly. For the proposed system we used a multi-layer Perceptron neural network, trained by the Levenberg-Marquardt algorithm, because it is a nonlinear system, the results obtained were compared with the diagnostic criteria currently used for insulating oil, arriving at the conclusion that the systems Proposed have high accuracy in the diagnosis of faults, meeting all requirements and expectations. From a maintenance planning point of view, a well-defined maintenance schedule for these equipments means that costs are reduced, their operational efficiency and good quality of service rendered to society. It is worth remembering that the diagnoses used in the training and validation processes of the developed systems were obtained from the application of an international standard, which often does not fit the Brazilian reality. A future objective of this work is to employ the proposed diagnostic systems, taking as the desired outputs real diagnoses of lack, collected in the field.