Automatic Classification of PV Cell Defects From Electroluminescence Images

Abstract

Employing artificial intelligence, which enables the system to mimic the human brain by making sensible judgments based on prior experiences, technology has transformed the world by replacing manual systems with automatic ones. A photovoltaic (PV) cell is an energy harvesting device that converts solar energy into usable power through the photovoltaic effect. In this study, a computeraided method that can automatically classify PV cell faults from electroluminescence pictures is suggested. These faults need a great deal of physical labor and time to classify. Even for trained professionals, visually identifying damaged units is challenging. Many imperfections that lower the effectiveness of a PV module are not visible to the human eye, save from conspicuous cracks in the glass. The two modules that make up the proposed system are feature extraction and defective cell classification. Two alternative designs with CNN such as VGG-16 and resnet50, are different models used in this framework. Support vector machines and random forests are used as classifiers, and these architectures are used as fixed feature extractors. We test the proposed methodology using two open datasets, the Elvp dataset and the SDEL-EL dataset. First, the two datasets are split 80:20, with 80% used for training and 20% used for testing. We also used 10- fold cross-validation to gain insight on all the data later. In order to support vector machine and random forest classifiers, we employed pretrained VGG16 weights, trained datasets using these weights, and propagated these features. The findings demonstrate that, in comparison to other traditional forecasting models, the suggested technique yields better predicting outcomes.