Automated analysis of performance losses in solar cells using explainable artificial intelligence

This study presents an automated approach for processing solar cell measurements and identifying key parameters that contribute to performance losses. The proposed method combines machine learning (ML) models to analyse the relationships between various parameters and cell performance, with explainable artificial intelligence (XAI) techniques to interpret the learned relationships. Two case studies demonstrate the application of this approach to numerical (one-dimensional) and image-based (two-dimensional) datasets. Results demonstrate that the developed ML models accurately captured parameter relationships consistent with established physics, despite having no prior knowledge of these relationships. Additionally, XAI techniques identified the features most significantly affecting solar cell performance. This fully automated approach is adaptable to diverse datasets, including in-situ measurements and environmental conditions during production. It has the potential to improve decision-making in production lines, supporting more efficient and reliable solar cell manufacturing.