UAV-Based Autonomus Monitoring and Real-Time Thermal Anomaly Detection of Photovoltaic Modules

This study presents an end-to-end autonomous monitoring framework for photovoltaic (PV) modules using unmanned aerial vehicles (UAVs), integrating thermal anomaly detection, boundary segmentation, and adaptive coverage path planning. The proposed system utilizes deep learning-based YOLO classifiers to identify nine different thermal faults in PV arrays, ensuring accurate and efficient real-time fault detection during flight. YOLOv8l achieved the highest classification performance (83.02% Top-1 accuracy), and YOLOv8n demonstrated exceptional computational efficiency with an inference time of just 8 ms and a compact model size of 2.94 MB, making it ideal for resource-constrained UAVs. To guide the UAV flight inside a solar farm, YOLO segmentation models were trained to extract PV plant boundaries from aerial images. YOLOv8l, selected for its high segmentation accuracy (mAP@50 = 0.952), provided precise navigable zone delineation critical for the next planning stage. These boundaries were used to generate a grid-based representation of the inspection area, supporting a deterministic sweep coverage algorithm with real-time obstacle avoidance and fallback handling. The coverage path planning system dynamically adapts to the layout of each PV installation, ensuring all accessible areas are inspected while minimizing flight distance and energy consumption. This multi-component framework enhances UAV autonomy and inspection completeness, reduces manual effort, and supports scalable real-time maintenance strategies for large-scale solar farms.