Development of a Handheld Hotspot Detecting Device Based on YOLOv5 Application on Thermal Images of Photovoltaic Panels

The booming of solar power systems has led to an increasing demand for monitoring and maintenance of solar panels. Hotspots on photovoltaic panels are a common dysfunction when operating solar power systems, reducing the generated power, system's efficiency, PV panel's lifespan and potentially causing serious hazard. Therefore, quick and accurate hotspots detection is important for maintenance, operation and safety. With the development of machine learning methods application in engineering, the combination of a machine learning model capable of detecting hotspots with thermal camera devices has high potential to improve the hotspot detection for better maintenance process. The main trend of research in the field is the application of UAV to collect thermal images of PV panels from aerial view to maximize monitoring area. However, UAV applications are facing many restrictions in several regions due to military capability, weather conditions and wide area requirements in operation. On the other hand, because of its complexity and flying control tasks, there is limited room for integration of machine learning models in UAV applications for real‐time hotspot detection. Currently on the market, most existing devices and manufacturers still haven't applied AI achievements into products. This research is about the designing, developing and testing of a handheld hotspot detecting device with YOLOv5 application to process captured thermal images to detect hotspot area appeared. The development of the device included case designing, electronic designing, user interface designing and firmware development. The developed device using common components of Raspberry Pi 4, thermal camera with application of pre‐trained YOLOv5 model with features of capturing and detecting hotspot areas on resulting thermal images and displaying result images on device's interface. Although the design is primitive with limited processing speed of 0.7 FPS when performing real time detections using YOLOv5 model, the device has capability of an Edge‐AI device to apply in PV panels hotspot detection. Furthermore, the research also shows the requirements and difficulties in work‐flow design when integrating machine learning models into practical applications and real time performance devices.