CLASSIFICATION OF PARTIAL DISCHARGE (PD) SOURCES IN HIGH VOLTAGE EQUIPMENT USING ARTIFICIAL NEURAL NETWORK (ANN)

• Main Author: ROSSAL SUKMA - NIM: 23216049 , TAUFIK
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/31243
• Institution: Institut Teknologi Bandung

<p align="justify">Partial discharge (PD) is one of electrical phenomena which might occur in high voltage (HV) equipment and can be used for diagnosing the condition of the equipment. Artificial neural network (ANN) is then utilized to classify PD source in HV equipment. Waveform parameters generated from PD signal are used as input data for the ANN. Waveform parameters generated from each type of PD and noise sources differ each other. Therefore, they can be used to distinguish between PD (corona, void, surface discharge) and noise as well as for training and testing the ANN. <br /> <br /> PD measurements were conducted to generate waveform parameters in laboratory using four kinds of artificial PD sources, i.e., protrusion on conductor defect, protrusion on ground defect, void defect and surface discharge defect, by three kinds of sensors (transient earth voltage (TEV) sensor, surface current sensor (SCS) and high frequency current transformer (HFCT)). Moreover, noise signal from four kinds of noise sources were also measured using same sensors. Nine waveform parameters from one PD event, i.e., peak value, phase, rise time, fall time, pulse width, pulse area, event width, frequency giving maximum amplitude and frequency from method of moment were used for training and testing the ANN (ANN_WP). For further comparison, phase-resolved partial discharge (PRPD) pattern was also generated and used as input data for training and testing the other ANN (ANN_PR). Results reveal that the recognition rate of the ANN_WP achieves 94% while the ANN_PR achieves 96%. <br /> <br /> However, in order to evaluate more deeply regarding the ANNs performance, the trained ANNs are then tested using new PD data obtained from experiment laboratory that were not involved in the ANNs training process at all. Results reveal that the ANN_WP prediction result probability achieves 81% in average while the ANN_PR prediction result probability achieves 90%. Another test was conducted using new different artificial void defect. The results show that the ANN_WP predicted new PD data as void defect with 92% probability while the ANN_PR prediction probability was found 94%. <br /> <br /> Moreover, the trained ANNs then utilized to identify the PD source occurring in the cubicle-type gas insulated switchgear (C-GIS). As a result, the trained ANN_WP was found to predict the kind of PD source as void discharge with 99% probability while the trained ANN_PR was found to predict the kind of PD source as void discharge with 100% probability. These results indicate that the waveform parameters can be used as input data for the ANN as well as PRPD pattern to provide sufficient accuracy for identifying the PD source. The results suggest a possibility that developed ANNs can be used as a decision-support tool in HV equipment diagnosis by comparing PD data obtained in the field.<p align="justify"> <br />