Detecting Arcing Faults in Switchgear by Using Deep Learning Techniques

Abstract

Switchgear and control gear are susceptible to arc problems that arise from slowly developing defects such as partial discharge, arcing, and heating due to faulty connections. These issues can now be detected and monitored using modern technology. This study aims to explore the effectiveness of deep learning techniques, specifically 1D-CNN model, LSTM model, and 1D-CNN-LSTM model, in detecting arcing problems in switchgear. The hybrid model 1D-CNN-LSTM was the preferred model for fault detection in switchgear because of its superior performance in both time and frequency domains, allowing for analysis of the generated sound wave during an arcing event. To investigate the effectiveness of the algorithms, experiments were conducted to locate arcing faults in switchgear, and the time and frequency domain analyses of performance were conducted. The 1D-CNN-LSTM model proved to be the most effective model for differentiating between arcing and non-arcing situations in the training, validation, and testing stages. Time domain analysis (TDA) showed high success rates of 99%, 100%, and 98.4% for 1D-CNN; 99%, 100%, and 98.4% for LSTM; and 100%, 100%, and 100% for 1D-CNN-LSTM in distinguishing between arcing and non-arcing cases in the respective training, validation, and testing phases. Furthermore, frequency domain analysis (FDA) also demonstrated high accuracy rates of 100%, 100%, and 95.8% for 1D-CNN; 100%, 100%, and 95.8% for LSTM; and 100%, 100%, and 100% for 1D-CNN-LSTM in the respective training, validation, and testing phases. Therefore, it can be concluded that the developed algorithms, particularly the 1D-CNN-LSTM model in both time and frequency domains, effectively recognize arcing faults in switchgear, providing an efficient and effective method for monitoring and detecting faults in switchgear and control gear systems.