Digital Electrode design for ASSB Cathode microstructures from 2D Data combining generative AI and stochastic 3D Modeling

All-Solid-State Batteries (ASSBs) are emerging as a key technology for next-generation energy storage, offering enhanced energy density and improved safety by eliminating flammable liquid electrolytes. The performance of ASSBs is heavily influenced by the 3D morphology of their electrode microstructures, e.g., transport path lengths within the solid electrolyte influence the effective diffusivity of ionic transport. This talk introduces a computational framework for generating digital twins for the 3D microstructure of ASSB cathode materials, using stochastic 3D models calibrated with 2D microscopic image data [1]. By enabling the model calibration from 2D image data, the method significantly reduces experimental imaging efforts while providing detailed insights into the 3D microstructure of ASSB cathode materials.

The framework combines the strengths of stochastic 3D modeling and machine learning techniques, such as generative adversarial networks (GANs). By integrating the parametric flexibility of stochastic models with the ability of GANs to capture intricate morphological details, this approach also enables the generation of a broad spectrum of differently structured 3D microstructures comprising the solid electrolyte, active material and pore phase. These structures can then be used as geometry input for numerical simulations to evaluate macroscopic effective properties. By enabling the systematic exploration of structural scenarios and their impact on macroscopic functional properties, this method supports the design of optimized materials, reduces reliance on costly trial-and-error processes, and contributes to the digitization of materials science. Integrated into GeoDict 2025, it provides a powerful platform for generating and analyzing digital twins of 3D morphologies of three-phased materials, accelerating the development of high-performance ASSBs and other functional materials.

In addition, this talk will highlight some further applications in which neural networks and methods from spatial stochastic modeling have been combined for the statistical reconstruction of various further 3D morphologies of material microstructures that have been measured with different 2D imaging techniques. This involves methods for the statistical reconstruction of (i) 3D grain architectures of polycrystalline battery cathode materials from 2D electron backscattered diffraction data by combining random tessellations and GANs [2], (ii) hetero-aggregate morphologies observed in projectional transmission electron microscopy images, by training convolutional neural networks with simulated training data [3] and (iii) the 3D microstructure in solid oxide fuel cell anodes imaged in 2D with scanning electron microscopy, by combining convolutional neural networks with an excursion set model [4].

[1] O. Furat, S. Weber, J. Schubert, R. Rekers, M. Luczak, E. Glatt, A. Wiegmann, J. Janek, A. Bielefeld and V. Schmidt, Generative adversarial framework to calibrate excursion set models for the 3D microstructure of all-solid-state battery cathodes. Working paper (under preparation).

[2] L. Fuchs, O. Furat, D.P. Finegan, J. Allen, F.L.E. Usseglio-Viretta, B. Ozdogru, P.J. Weddle, K. Smith and V. Schmidt, Generating multi-scale Li-ion cathode particles with radial grain architectures, using spatial stochastics and GANs. Communications Materials (in print).

[3] L. Fuchs, T.  Kirstein, C. Mahr, O. Furat, V. Baric, A. Rosenauer, L. Mädler and V. Schmidt, Using convolutional neural networks for stereological characterization of 3D hetero-aggregates based on synthetic STEM data. Machine Learning: Science and Technology 5 (2024), 025007.

[4] L. Schröder, S. Weber, L. Fuchs, V. Schmidt, B. Prifling, Predicting the 3D microstructure of SOFC anodes from 2D SEM images using stochastic microstructure modeling and CNNs. Working paper (under preparation).