Identificação de fontes de descargas parciais por atributos extraídos da envoltória do sinal.

Abstract

The techniques for diagnosing partial discharges (PD) using the radiometric method require signal sampling in the order of nanoseconds. Therefore, a high sampling rate of the acquisition system is required to obtain the required accuracy of the method and a high data storage capacity of the acquisition devices. In order to reduce the hardware requirements of the acquisition system and increase the applicability of the radiometric method in PD monitoring in substations, the use ofa signal conditioning system between the antenna and the acquisition system was proposed. In this work, a system capable of acquiring radiometric PD signals, capturing the signal envelopes, extracting the most relevant attributes and separating and classifying the PDs was designed, developed, and evaluated, allowing a significant reduction in the sampling rate required for acquisition. Initially, signal smoothing was performed and evaluated using the Kernel density smoothing method. Subsequently, a signal conditioning system was developed and validated in measurements performed in the laboratory and in substations. The representative features of the signal envelopes were extracted, selected and applied to clustering algorithmssuch as K-means, Gaussian MixtureModel(GMM)andMean-shift, and to supervised machine learning models such as Support Vector Machine (SVM), Random Forest and Logistic Regression. The results demonstrate that the proposed system classifies and separates PD signals effectively, reducing the hardware requirements for the acquisition systems, expanding the potential of the radiometric method for use in substations.