Document Type : Research Article

Authors

• Adel J Yousif
• Fadhil Kadhem Zaidan
• Hassan K Al-Mahdawi

Diyala

Abstract

Solar energy is a sustainable and renewable resource that plays a vital role in mitigating climate change by reducing greenhouse gas emissions. However, its efficiency can be compromised by different operational faults such as dust accumulation, surface cracks, or electrical failures. Detecting these issues early is necessary to maintain optimum performance and avoid costly system failures. In this study, we propose an AI-based approach for automated fault detection in solar panels, built on a lightweight deep learning model adapted from the MobileNetV2 architecture. The model is trained and validated on a publicly available dataset with data balancing and augmentation to improve classification accuracy across different fault categories. To assess practical feasibility, we deployed the system on an embedded Jetson Nano platform. Extensive results and comparisons demonstrate the superior performance of the proposed method, achieving an accuracy of 93.14% and an F1-score of 93.12%, while maintaining a low model size (2.8M parameters) and an inference speed of 44.4 ms per image on the Jetson Nano, which is fast enough to meet real-time inspection requirements in embedded devices. Overall, the findings indicate that our solution provides an effective framework for on-site solar panel monitoring and maintenance without the need for cloud resources.

Keywords

• Solar Panel Defects
• Faulty Solar Panel
• Deep Learning
• Embedded Systems
• CNN