Silicon solar cells: damage identification and classification based on electroluminescence measurements and AI tools

The reliability and performance of silicon solar cells are seriously affected by various degradation mechanisms and manufacturing defects. To ensure higher efficiency in photovoltaic plants, it is essential to effectively measure, identify, and possibly replace modules with seriously damaged cells. This thesis addresses these challenges by proposing an automated framework for the detection and classification of defects in silicon solar cells based on EL measurements and advanced Artificial Intelligence tools. The study design involves the development of a deep learning-based pipeline designed to process and analyze high-resolution EL images. Various image processing algorithms were implemented to enhance feature extraction, followed by the training of convolutional neural networks to perform multi-class defect classification. A test result demonstrates the precision of the proposed approach in accurately identifying common failure modes, such as cracks and dark areas, as well as in estimating solar cell parameters, including Voc, Isc, FF, Rs, Rsh. The integration of these AI tools significantly reduces diagnostic time while maintaining accuracy, providing an optimal solution for quality control in industrial and research environments.