Why AI: A Comparative Study for Detection Methods in DC Series Arc Fault
Yufei Mao, Sadaf Safa, Gianni Smith, Leon Wurth, Roland Weiß, Jens Hagemeyer
Abstract
With the advancement of direct current (DC) power grids, ensuring reliable fault detection for safety has become increasingly critical. DC series arc faults, in particular, remain challenging to detect accurately. Although artificial intelligence (AI)-based methods have shown promising performance in recent years, questions have arisen within the industry regarding the necessity of AI algorithms, especially when rule-based methods have also demonstrated strong results in previous studies. This study aims to identify a highly generalizable solution for industrial applications, especially for resource-constrained devices that operate with low sampling rates and limited computational power. Through a comparative analysis of AI-based and rule-based approaches, supported by a comprehensive review of the literature and experimental validation, this research provides statistical evidence highlighting the strengths and weaknesses of both methods. It is demonstrated that AI-based methods are more adaptable to varying conditions, and that integrating pre-feature extraction with domain knowledge can enhance general performance, particularly in such physical problems with considerable prior research.