Machine Learning in Solar Plants Inspection Automation

Abstract

The emergence of large photovoltaic farms poses a new challenge for quick and economic diagnostics of such installations. This article presents this issue starting from a quantitative analysis of the impact of panel defects, faulty installation, and lack of farm maintenance on electricity production. We propose a low-cost and efficient method for photovoltaic (PV) plant quality surveillance that combines technologies such as an unmanned aerial vehicle (UAV), thermal imaging, and machine learning so that systematic inspection of a PV farm can be performed frequently. Most emphasis is placed on using deep neural networks to analyze thermographic images. We show how the use of the YOLO network makes it possible to develop a tool that performs the analysis of the image material already during the flyby.