The functional properties of materials, such as battery electrodes, are strongly influenced by their complex 3D microstructures. Unraveling structure-property relationships requires comprehensive databases that capture a broad range of microstructural scenarios, which is challenging when relying solely on experimental data due to cost and manufacturing constraints. Virtual materials testing approaches that combine stochastic microstructure modeling with numerical simulations offer promising alternatives. In particular, stochastic 3D models can be leveraged to generate virtual yet realistic microstructures, providing structural diversity for systematic analysis through numerical simulation. A key challenge is ensuring that these virtual structures accurately reflect real material microstructures. To achieve this, stochastic 3D models (e.g., random fields on the sphere or random tessellations) are calibrated using high-resolution microscopy data, which often requires machine-learning-assisted image preprocessing.

 

Convolutional neural networks can facilitate automated segmentation, whereas techniques such as generative adversarial networks can enhance image resolution or mitigate the limited availability of 3D image data by relying on 2D images instead. Calibrated stochastic 3D models capture multivariate distributions of geometric descriptors, enabling the generation of digital twins that statistically resemble real material microstructures. These virtual structures can then serve as input for numerical simulations of macroscopic functional properties, leading to a deeper understanding of structure-property relationships.

 

This talk highlights recent advancements in stochastic 3D modeling for battery microstructures, machine-learning-guided calibration and their application in virtual materials testing.

Dr. Orkun Furat is a Postdoctoral Researcher at the Institute of Stochastics at Ulm University. His research focuses on applied mathematics and statistics, specializing in spatial statistics, stochastic geometry and machine learning, with applications in materials science, energy research and process engineering. In particular, he develops data-driven stochastic 3D models for material microstructures that can be deployed to generate virtual, but realistic microstructures for virtual materials testing purposes.

He completed his doctorate studies at Ulm University with his dissertation on “Quantitative characterization of complex particle systems via machine learning and stochastic modeling”. Currently, he leads two projects funded by the German Federal Ministry of Education and Research (BMBF), where he applies methods from generative AI and stochastic geometry to model and design material microstructures in all-solid-state battery cathodes as well as in wall components of fusion reactors.