Using CNNs for Photovoltaic Panel Defect Detection via Infrared Thermography to Support Industry 4.0

Abstract

Background: This study demonstrates how convolutional neural networks (CNNs), supported by open-source software and guided by corporate social responsibility (CSR), can enhance photovoltaic (PV) panel maintenance. Connecting industrial informatics with sustainable practices underscores the potential for more efficient and responsible energy systems within Industry 4.0. The rapid expansion of solar power necessitates effective maintenance and inspection of PV panels to ensure optimal performance and longevity. CNNs have emerged as potent tools for detecting defects in PV panels through infrared thermography (IRT). Objectives: The review aims to evaluate CNNs' effectiveness in detecting PV panel defects, align their capabilities with the IEC TS 62446-3:2017 standard, and assess their economic benefits. Methods/Approach: A systematic review of literature focused on studies using CNNs and IRT for PV panel defect detection. The analysis compared performance metrics, economic benefits, and alignment with industry standards. Results: CNN models demonstrated high accuracy in defect detection, with most achieving above 90%. Integrating UAVs for image acquisition significantly reduced inspection times and costs. Conclusions: CNNs are highly effective in detecting PV panel defects, offering substantial economic benefits and potential for industry-wide standardisation. Further research is needed to enhance model robustness across diverse conditions and PV technologies.