Partial discharge type detection and identification based on its sources

Partial discharge (PD) is a symptom of initial damage to high voltage equipment insulators which if left for a long period will cause total damage to high voltage equipment. This study aims to detect and identify the type of PD based on the source of its discharge so that it can be useful in terms of monitoring and maintenance of high voltage equipment. In this research, the Hilbert fractal antenna sensor is used in the detection process of surface, cavity, and corona PD with different input voltage variables that successfully produce a total of 600 PD signal data on the oscilloscope. To reduce noise on the PD signal, the denoising process is done by utilizing the sym4 wavelet feature found in the MATLAB software. The denoising process generates new data so that the research data becomes 600 original PD signal data and 600 denoising PD signal data. With a statistical approach, all PD signal data is extracted successfully into the mean, skewness, kurtosis, and standard deviation parameters which are useful as input for the PD type identification process. From each of the PD signal statistical data, 450 data are used in the training data process and 150 data are used in the data testing process. The PD type identification process is performed using a back propagation neural network with a mean square error (MSE) level of 0.01. The identification results show that back propagation neural networks are able to identify PD types based on statistical input accurately. In addition, the denoising process also affects the accuracy of the identification results of the PD type that is 95.33% for the original discharge signal to 97.33% for the denoising signal.