New publication: Artificial intelligence creates realistic electrode structures

The DigiCell team of Université de Picardie Jules Verne has recently published a new article entitled "From Top to Bottom: Manufacturing Process-Context Aware Resolution of Energy Device Electrodes Through a 3D Diffusion Generative Model". In this publication, they demonstrate how artificial intelligence can help researchers create realistic 3D models of electrodes.

The researchers developed a denoising diffusion probabilistic model (DDPM), a type of generative AI that learns the structure of electrodes from experimental and computer-generated data. Unlike approaches based on small cubic samples, the model uses data covering the full electrode thickness. This allows it to capture manufacturing-related gradients and heterogeneities, such as those caused by calendering and slurry drying. The approach was tested on lithium-ion battery NMC cathodes and catalyst layers used in proton-exchange membrane fuel cells. The generated structures reproduced important features of the original electrodes, including the volume fraction of each material phase, specific surface area, tortuosity, spatial correlations and, for the battery electrodes, the shape of the active-material particles. The model was also evaluated through multiphysics simulations. Finite-element and lattice-Boltzmann models showed that the AI-generated structures produced battery and fuel-cell performance comparable to the experimental or physics-based reference structures. This step was important because it helped identify structural artefacts that may not be visible when comparing images alone.

The method could therefore help researchers generate large numbers of realistic electrode models, study how manufacturing affects transport and device performance, investigate ageing and degradation, and digitally explore new designs before producing and testing them experimentally.

Read the full publication.