Partial discharges (PD) are characterized by a natural phenomenon that occurs in cavities or inclusions that, in a partial way, breaks the dielectric strength due to the action of a high-intensity electric field. It is necessary to monitor these discharges, since they precede future failures in the isolation of high voltage machines, allowing a better planning and avoiding unplanned shutdowns of generators, which would lead to great socioeconomic losses. The diverse and different causes of PD sources generate specific patterns of partial discharges classified according to the nature of their origin. Each PD pattern can be plotted on a graph called the Phase Resolved Partial Diagrams (PRPD), better known as statistical maps. Due to this particularity, this work aims to use artificial intelligence (ANN - Artificial Neural Network) to classify partial discharges in stator bars of hydro generators. The network Multilayer Perceptron was used for the classification. The PD database is made up of measurements made at the hydroelectric plant of the Northern Region of Brazil, UHE-Tucuruí, in the state of Pará. The processing of the database used in ANN is in the methodology of the work and, in the results, It is possible to observe the performance of each configuration through analysis through confusion matrices and performance graphs.
|
