Deep Learning Based Fault Detection and Diagnosis in Photovoltaic System Using Thermal Images Acquired by UAV

Abstract

Solar power is one of the largest renewable energy sources in the world. With photovoltaic systems, electrical energy can be generated wherever the sun is located. To prevent efficiency losses in photovoltaic systems, these systems should be tested at regular intervals. In this study, it is discussed to detect cell, module and panel faults in panels using thermal images obtained from solar panels. Within the scope of the study, a four-rotor unmanned aerial vehicle (drone) was designed and a thermal camera was placed on the vehicle. Thus, thermal images of the solar panels on the roof of Karabuk University buildings were taken. A thermal data set with cell fault, module fault and panel fault were created using the resulting thermal images. The YOLOv3 deep learning-based convolutional neural network was trained with the created dataset. This training was conducted on Nvidia Jetson TX2, an embedded AI (Artificial Intelligence) computing device. After the completion of the training of the YOLOv3 network, it was concluded that the faults mentioned in the tests were successfully detected.