Fault Detection of the Solar Photovoltaic Modules Using YOLO Models

The growing adoption of solar panels, driven by climate change concerns, underscores the importance of ensuring the reliability of photovoltaic (PV) modules. However, outdoor PV modules deployment face a range of environmental challenges such as extreme temperatures, chemical exposure, and mechanical stress which can lead to aging, defects, and degradation. This research introduces a novel approach for identifying faults in solar photovoltaic (PV) modules. Leveraging deep learning techniques from the You Only Look Once (YOLO) family, specifically the recent YOLOv8 and YOLOv9 models, this paper aims to enhance the reliability and performance of PV systems by accurately detecting and classifying module defects to a thermal images database containing three photo-voltaic cell defects. By automating the fault detection process through computer vision, this work contributes to the ongoing efforts to optimize solar energy generation and maintenance. Further, YOLOv5, YOLOv6, and YOLOv7 are also trained, validated, and tested. The results showed that the novel technique of the GELAN architecture-based model outperformed all other models trained on the custom dataset of thermal images of solar PV modules, achieving a mean average precision (mAP) of 70.4%.