인공지능 기반 태양광 모듈의 열화상 이미지 결함 진단

Abstract

Photovoltaic power systems are prone to faults in photovoltaic modules due to damages that can occur in poor outdoor environments, potentially reducing power generation efficiency. Traditional fault diagnostic methods are time-inefficient due to manual inspection. The advancement in artificial intelligence (AI) has significantly contributed to improving the speed and accuracy of fault diagnostics process in photovoltaic modules. Therefore, we have developed an AI-based thermal imaging fault diagnostic model for photovoltaic modules to enable efficient fault identification. However, a scarcity of fault data for these modules poses challenges for diagnostics. This paper particularly addresses the critical task of diagnosing faults in photovoltaic modules under conditions of limited data availability. The primary contributions of our study are threefold. Firstly, we introduce a novel method for automatically extracting photovoltaic modules from images of photovoltaic power systems captured using unmanned aerial vehicles equipped with thermal cameras. This method significantly reduces the time and cost involved in building a database for fault diagnostics, thereby addressing the issues arising from limited data availability. Secondly, our study conducts a quantitative comparison and analysis of the classification performance among models with and without the application of transfer learning, using a meticulously preprocessed dataset comprising 10,682 thermal images of photovoltaic modules. Thirdly, the practicality and field applicability of our system have been validated through tests in real-world environments. Our findings demonstrate that our research not only simplifies human intervention in efficient photovoltaic module fault diagnostics but also contributes to alleviating the data scarcity issue in the photovoltaic module diagnostic field through automation in database construction.