Análise de medidas ultrassônicas de descargas parciais em sistema ponta-plano por meio da tranformada hilbert-huang e de redes neurais de Regressão generalizada

Abstract

The electric power system comprises generation, transmission and distribution systems. To ensure the operation of this complex system, it is crutial to verify the operation condition of the involved components. Thus, the power equipment insulation is subject to degradation due to the possibility by partial discharges. The ultrasonic measurement is a non-invasive inspection technique of power equipment, which allows checking the insulation degradation through the analysis of acoustic emission data. This work proposes a methodology for partial discharge detection and classifying through the Hilbert-Huang Transform and generalized regression neural networks. The Hilbert-Huang Transform is used to verify behavioral patterns in partial discharges within time frequency domain. In this sense, variables that presented relevant patterns to describe this phenomenon were extracted. A statistical analysis is applied on patterns and analyzed by artificial neural networks in order to classify the database thereby, determinate the existence of partial discharges according to severity by the acoustic signal. As a result, it is noted that the acoustic pulses of partial discharges are more evident with the increase in voltage. Furthermore, the categorization of three behavior patterns was allowed through the investigation of frequency spectrum of the acoustic signals, which were recognized as Background Noise, Moderate Discharge and Several Discharge.