An Intelligent Classifier of Electrical Discharges in Oil Immersed Power Transformers

Electrical power systems are cost-efficiently structured such that the bulk power is transmitted by the means of high voltage levels. The stepping up/down of AC voltages relies principally on the presence of power transformers that link the various zones of the system together; thus, transformers are commonly described as the electrical counterpart of gearboxes. Transformer's lifetime is highly affected by the load cycling and the presence of discharges in its insulation system. Although cellulose paper and mineral oil represent the main materials of the transformer's insulation system, the presence of air voids and/or bubbles is inevitable. This paper aims at classifying the electrical discharges that may occur inside operating transformers such that corrective maintenance may be advised and/or their lifetime estimates are to be shortened. A point-to-plate discharge source was created either in air or in oil with the receiving UHF antenna being installed in either medium. To extract descriptive features, the captured signals were processed using two alternative techniques: the Discrete Wavelet Transform (DWT), or the Fast Fourier Transform (FFT). Finally, an intelligent classifier was capable of recognizing the medium of the discharge source using the feed forward with back propagation artificial neural networks (ANN). The ANN recognition rate was adopted for the comparison between the two signal processing techniques.