Manufacturing-conditioned hierarchical generation of cathode active-material microstructures with global electrode context
Abstract
Three-dimensional cathode microstructures are important inputs to electrochemical simulation, yet a manufacturing descriptor does not determine a unique voxel field and experimentally labelled electrodes are scarce. We introduce Manufacturing-Conditioned Hierarchical Electrode Generation with Diffusion and Global Context (MC-HEDGE), a two-scale conditional generator that first predicts an electrode-level low-resolution active-material field and then samples high-resolution patches conditioned on local coarse context, patch coordinates and a global encoding of the parent electrode. Parent-electrode splitting prevents leakage among correlated patches. On 16 ETHZ binary NMC cathode electrodes (12/2/2 train/validation/test), the model is evaluated after coordinate-aware assembly using conditional volume-fraction Wasserstein distance, directional two-point-correlation distance and a standardized morphology-feature distance, together with a seam diagnostic. MC-HEDGE obtains the lowest ETHZ two-point-correlation distance (0.0135) and seam error (0.2882), while a CVAE gives lower volume-fraction and multi-feature errors. True-context denoising is substantially more accurate than generated-context denoising, quantifying error propagation from the electrode-level stage. The results establish a reproducible computational route to manufacturing-conditioned candidate structures, while the binary representation, small dataset and absence of independent experimental validation delimit the claims.
Related
- Fiber-Scale Discrete Element Modeling Approach for Microstructural Evolution in Carbon Felt Electrode Manufacturing
- Manufacturing process modeling of bilayer cathode for graded thick electrode design
- Manufacturing of Hierarchical (Nano-)Composite Structures: Numerical and Experimental Perspectives
- High Fidelity Stochastic Electrode Microstructure Generation for Low Temperature Electrochemical Systems
- Hybrid Laser–Etching Fabrication of High-Density Antireflective Microstructures on DLC-Coated Silicon Using Burst-Mode Bessel Beam Processing
BibTeX
@misc{xing2026manufacturingconditioned,
title = {{Manufacturing-conditioned hierarchical generation of cathode active-material microstructures with global electrode context}},
author = {Mengda Xing and Jean-Marie Lagniez and Alejandro A Franco},
howpublished = {HAL},
year = {2026},
month = sep,
url = {https://hal.science/hal-05747338v1},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Manufacturing-conditioned hierarchical generation of cathode active-material microstructures with global electrode context
AU - Mengda Xing
AU - Jean-Marie Lagniez
AU - Alejandro A Franco
PY - 2026
DA - 2026/09/11
PB - HAL
UR - https://hal.science/hal-05747338v1
AB - Three-dimensional cathode microstructures are important inputs to electrochemical simulation, yet a manufacturing descriptor does not determine a unique voxel field and experimentally labelled electrodes are scarce. We introduce Manufacturing-Conditioned Hierarchical Electrode Generation with Diffusion and Global Context (MC-HEDGE), a two-scale conditional generator that first predicts an electrode-level low-resolution active-material field and then samples high-resolution patches conditioned on local coarse context, patch coordinates and a global encoding of the parent electrode. Parent-electrode splitting prevents leakage among correlated patches. On 16 ETHZ binary NMC cathode electrodes (12/2/2 train/validation/test), the model is evaluated after coordinate-aware assembly using conditional volume-fraction Wasserstein distance, directional two-point-correlation distance and a standardized morphology-feature distance, together with a seam diagnostic. MC-HEDGE obtains the lowest ETHZ two-point-correlation distance (0.0135) and seam error (0.2882), while a CVAE gives lower volume-fraction and multi-feature errors. True-context denoising is substantially more accurate than generated-context denoising, quantifying error propagation from the electrode-level stage. The results establish a reproducible computational route to manufacturing-conditioned candidate structures, while the binary representation, small dataset and absence of independent experimental validation delimit the claims.
ER -