Mesoscale Synthesis of 3D Multiphase Electrode Microstructures via AI Generative Diffusion Model with Manufacturing Aware Data Augmentation
ECS Meeting Abstracts · 07.07.2026 · MA2026-01 (6) · 674-674
DOI: 10.1149/ma2026-016674mtgabs ↗Kurz gefasst
Fachartikel in der Zeitschrift ECS Meeting Abstracts, erschienen am 07.07.2026. Verfasst von Victor Cid Ramirez und 2 weiteren Autor:innen.
Abstract (Originalsprache)
The ever-growing demand for energy storage and conversion devices for the future, such as Li-ion batteries for portable electronics and electric vehicles, motivates improving their performance today. Optimization of these devices can be achieved by studying their Process-Structure-Property (PSP) relationships. For the manufacturing of energy storage devices, these link the applied manufacturing parameters to the resulting microstructure of the electrode, affecting the properties that determine device performance at the macroscale level 1 . At the heart of these relationships lies the 3D electrode microstructure, however experimental techniques to resolve it are costly and time-consuming, which coupled with their vast manufacturing space, bottleneck the high throughput development of next-generation devices. Recent advancements in data driven methods based in Generative Artificial Intelligence attend to some of the shortcomings of traditional computer-based structure generation, resulting in compatible data for downstream modelling tasks in an accelerated manner 2 . This work explores the generative diffusion model for synthesizing high fidelity volumes of 3 phase electrode microstructures. Our presented model in this work leverages the DDPM version of the diffusion family to resolve new microstructures in 3D, statistically similar to the training dataset, handling various input sizes along the thickness of the material. Our training on non-isometric volumes improves data representativeness along a direction of interest, relevant to device operation, allowing the retention of structural heterogeneities that exist along the material thickness for model learning. These structural features originate from the manufacturing process of composite materials, often lost with common volume augmentation strategies in similar GenAI works, which train on isometric cubic samples. For Li-ion electrodes, for example, such heterogeneities stem from the calendering operation, affecting the tortuosity of the porous network along the main direction of ionic transport, directly impacting discharge rates at the cell level 3 . To test our model's ability for synthesizing microstructures, we trained the model on an X-ray resolved NMC532 Li-ion cathode volume, and an in-silico resolved catalyst layer for proton exchange membrane fuel cells. Qualitatively, the generated volumes exhibit strong resemblance to the ground truth data, with expected 3D phase morphologies and percolating networks. Quantitative analysis was also conducted by comparing statistical structure metrics such as the volume phase fraction, tortuosity factor along the thickness, and per phase n-point correlation function in 3D. Furthermore, operational performance simulations were performed, relevant to the corresponding energy device of each dataset. Specifically, galvanostatic discharge for the Li-ion cathode, and relative permeability measurements of the CL. Lastly, as both microstructures possess similar morphological features, we tested finetuning on the learned model weights from the NMC dataset with training on the CL data. This led to faster loss convergence along with improved structural metric accuracy versus training from scratch, showcasing the transfer learning potential from experimental to synthetic datasets across different materials, an emerging trend in microstructure optimization in the energy storage field 4 . References Troncoso, J. F., Zanotto, F. M., Galvez‐Aranda, D. E., Zapata Dominguez, D., Denisart, L., & Franco, A. A. (2025). The ARTISTIC battery manufacturing digitalization initiative: from fundamental research to industrialization. Batteries & Supercaps , 8 (1), e202400385. Niu, Z., Zhao, W., Deng, H., Tian, L., Pinfield, V. J., Ming, P., & Wang, Y. (2024). Generative artificial intelligence for designing multi-scale hydrogen fuel cell catalyst layer nanostructures. ACS nano , 18 (31), 20504-20517. Xu, J., Ngandjong, A. C., Liu, C., Zanotto, F. M., Arcelus, O., Demortiere, A., & Franco, A. A. (2023). Lithium ion battery electrode manufacturing model accounting for 3D realistic shapes of active material particles. Journal of Power Sources , 554 , 232294. Fernandez, F., Saravanan, S., Omongos, R. L., Troncoso, J. F., Galvez-Aranda, D. E., & Franco, A. A. (2025). Transfer learning assessment of small datasets relating manufacturing parameters with electrochemical energy cell component properties. npj Advanced Manufacturing , 2 (1), 14. Figure 1
Zitieren
APA
Ramirez, V. C., Franco, A. A., & Benammar, N. (2026). Mesoscale Synthesis of 3D Multiphase Electrode Microstructures via AI Generative Diffusion Model with Manufacturing Aware Data Augmentation. ECS Meeting Abstracts, MA2026-01(6), 674-674. https://doi.org/10.1149/ma2026-016674mtgabs
Vancouver
Ramirez VC, Franco AA, Benammar N. Mesoscale Synthesis of 3D Multiphase Electrode Microstructures via AI Generative Diffusion Model with Manufacturing Aware Data Augmentation. ECS Meeting Abstracts. 2026;MA2026-01(6):674-674. doi:10.1149/ma2026-016674mtgabs
BibTeX
@article{Ramirez2026,
author = {Ramirez, Victor Cid and Franco, Alejandro A. and Benammar, Nassima},
title = {Mesoscale Synthesis of 3D Multiphase Electrode Microstructures via AI Generative Diffusion Model with Manufacturing Aware Data Augmentation},
journal = {ECS Meeting Abstracts},
year = {2026},
volume = {MA2026-01},
number = {6},
pages = {674--674},
doi = {10.1149/ma2026-016674mtgabs}
}
EUVIMED