Real-Time AC ARC Detection Method Based on AI Ensemble

A real-time AC arc detection method was developed to predict potential fires and failures by detecting small arcs at an early stage, even on low-specification systems. To overcome the existing challenges related to limited data, time-series data closely resembling arc faults in real systems were acquired by introducing small arcs into the system. Data were collected from sensors installed inside switchgears and combined with simulation results to facilitate AI training, analysis, and validation. An AI framework for analyzing and predicting switchgear conditions was used to select appropriate AI algorithms and optimize input parameters. Additionally, an anomaly prediction device was developed to integrate precise time-series digital data acquisition and analysis for current and voltage, specifically for integrated AI switchgear. The developed AI-based real-time arc detection system successfully identified small arc phenomena through various filter chains and ensemble filters, demonstrating accurate detection performance.