Aplicação de redes neurais convolucionais e correntes de sequência zero para identificação de motor com fuga a terra em sistemas de baixa tensão de neutro aterrrado com RAV

Abstract

This study proposes an approach to identify a ground fault motor in a low voltage three phase electrical system with a High Resistance Grounding (HRG) through the application of deep learning techniques. Electrical systems grounded by high value resistors have the advantage of limiting the leakage current to earth. However, the identification of the equipment that presented the fault is not simple, as such currents can be confused with eddy currents, which may eventually occur in the installation. The proposed method consists of using the zero sequence currents of the equipment in normal operating condition and in the ground fault condition, in order to classify these signals by their characteristics and create a model capable of separating these classes. This approach differs from existing techniques because it uses the fault current signature and not the RMS value. For the elaboration of classification models, raw signals were used for direct input into a one dimensional convolutional neural network. The best result (f1-score of 100.00%) demonstrates the feasibility of implementing the technique in industrial systems for detecting earth leakage equipment in this type of installation.