LArcNet: Lightweight Neural Network for Real-Time Series AC Arc Fault Detection

Abstract

Detecting series ac arc faults in diverse residential loads is challenging due to variations in load characteristics and noise. While traditional artificial intelligence-based algorithms can be effective, they often involve high computational complexity, limiting their real-time implementation on resource-constrained edge devices. This article introduces lightweight arc fault detection network (LArcNet), a novel, lightweight, and rapid-response algorithm for series ac arc fault detection. LArcNet combines a teacher–student knowledge distillation approach with an efficient convolutional neural network architecture to achieve high accuracy with minimal computational demand. This streamlined yet robust design makes LArcNet ideally suited for resource-constrained embedded systems, achieving an arc fault detection accuracy of 99.31%. The model is optimized and converted into TensorFlow Lite format to reduce size and latency, enabling deployment on low-power embedded devices such as the Raspberry Pi and the STM32 microcontrollers. Test results demonstrate LArcNet's inference times of just 0.20 ms on the Raspberry Pi 4B and 3 ms on the STM32H743ZI2, surpassing other leading models in operational speed while maintaining competitive accuracy in arc fault detection.