Physics-Informed AI Models for Interpretable Data Analysis

Fraunhofer ISE has developed a digital twin of the solar cell that combines deep learning models and physics-based simulation methods. This digital twin enables real-time quality control and inline loss analysis of solar cells solely based on inline accessible measurement data and thus creates the basis for targeted and efficient process optimization. By combining simulation-based models and expert knowledge with machine learning, an interpretable and physically accurate representation of each individual solar cell is created. The physics-informed approach enables the fusion and compression of measurement data and thus efficient storage and analysis of the measurement data collected on the workpiece.

Fraunhofer ISE has developed a novel, physics-informed machine learning model for fast, precise and full-area determination of the layer thickness of transparent conductive oxides (TCO) in solar cells. Using multispectral images and synthetic training data, this model enables comprehensive quality control and process analysis during the production process. The technology enables the creation of complete thickness maps of the TCO layer in just one second and thus improves quality control in the production of heterojunction (HJT) solar cells and thus their efficiency and reliability.