Application of Artificial Intelligence for Selected Photovoltaic Faults Detection Based on Electroluminescence Images

Photovoltaics (PV), like other renewable sources, is characterized by a steadily growing public interest. However, the efficient generation of green energy from photovoltaic panels requires overcoming a number of barriers, including structural and technological barriers to achieve ever greater efficiency in the conversion of solar energy into electrical current. One of the known reasons leading to efficiency limitations in crystalline silicon photovoltaic cells is the different types of defects and damage, the causes of which can appear already in the production stage of the PV cells themselves. This article describes a brief overview of selected artificial intelligence-based solutions and their application to detect and identify defects in photovoltaic cells and panels using electroluminescence (EL) imaging, and is divided into two main subsections – the first is the comparison of selected solutions of using the AI in imaging the PV cells and panels using EL imaging and infrared (IR) imaging, as the most common ways to determine the presence of inhomogeneities or faults. The second part consists of the exemplary engineering application of the AI algorithms – the binary classification and multi-criteria analysis of the defects’ appearance on the photovoltaic cell EL images. The best results in case of binary classification was the use of Logistic Regression classifier with AUC of 0.9569 with the use of ResNet50 feature extraction. In the case of multicriteria analysis the best performance was achieved for the Faster R-CNN ResNet50 detector with [email protected] of 0.985. The performed analysis shows the usefulness of AI in the field of solar cell diagnostics.