Fast 3D diffusion for scalable discrete granular media synthesis
Generating dense, realistic packings of discrete granular media is a critical prerequisite for Discrete Element Method (DEM) simulations of materials like railway ballast and soil mechanics. Traditional gravitational settling methods scale poorly and often require prohibitive computational time for large-scale domains. In this work, we propose a scalable 3D generative diffusion approach utilizing unconditional generation and boundary-conditioned inpainting to synthesize granular assemblies in seconds. The generated samples faithfully reproduce target particle size distributions, void ratios, and fabric tensor orientations, accelerating multiscale homogenization and dynamic contact simulations by over two orders of magnitude.
@article{hassan2026fast,
title = {Fast 3D diffusion for scalable discrete granular media synthesis},
author = {Hassan, Muhammad Moeeze and Cottereau, R{\'e}gis and Gatti, Filippo and Dec, Patryk},
journal = {Computers and Geotechnics},
volume = {198},
pages = {108336},
year = {2026},
publisher = {Elsevier},
doi = {10.1016/j.compgeo.2026.108336}
}