BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//EURODYN//Schedule//DE
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:Europe/Berlin
X-LIC-LOCATION:Europe/Berlin
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:278@eurodyn2026.org
DTSTAMP:20261026T144037Z
DTSTART;TZID=Europe/Berlin:20260929T113000
DTEND;TZID=Europe/Berlin:20260929T115000
SUMMARY:Bridging Grain-Scale Fidelity and Continuum Efficiency on Ballast
 ed Tracks using Diffusion Accelerated Workflows
LOCATION:Room B305
DESCRIPTION:MS 13\n\nPresenter: Hassan\, Muhammad Moeeze\n\nAuthors: Hass
 an\, Muhammad Moeeze\; Cottereau\, Régis\; Gatti\, Filippo\; Dec\, Patry
 k\n\nChair: Ribeiro\, Diogo\n\nAbstract: Simulating granular media\, suc
 h as railway ballast with Discrete Element Method (DEM) remains a prohib
 itively expensive task\, especially during initialization phases\, which
  dominate total runtime through iterative force-displacement solutions a
 t sub-milli second time steps. Ballasted tracks are hence represented wi
 th overly simplified continuum models in traditional simulations to be c
 ompatible with conventional time domain solver pipelines\, often losing 
 important grain level characteristics.  We introduce the first end-to-en
 d generative pipeline based on diffusion models\, that sidesteps the len
 gthy initialization phase by learning the distributions of fully settled
  assemblies and directly sampling physically consistent final states. Th
 e three-stage pipeline is: i) Synthesis of 3D Representative Volume Elem
 ents (RVEs) using unconditional diffusion models ii) Stitching of RVEs u
 sing diffusion inpainting\, iii) Segmentation and exporting as DEM compa
 tible libraries. A speed up of 200x is observed on 1.2 meter of statisti
 cally validated ballasted tracks\, compared to conventional DEM initiali
 zation.  Leveraging this generation backbone\, we run a sleeper-track ca
 mpaign comprising of ~5000 DEM samples with varying loading conditions a
 nd homogenizing them using a custom toolkit to get a dataset of continuu
 m fields like macro and micropolar stresses/strains\, that capture heter
 ogenous details from DEM simulations. These fields become supervised tar
 gets for a neural operator that learns to output a rollout of these cont
 inuum fields given an initialized DEM sample\, without ever doing a comp
 lete DEM simulation.  Paired with fast granular sample generation\, this
  physically consistent rollout gives near realtime load-response estimat
 es for granular media applications\, bridging grain-scale fidelity with 
 continuum-scale computational efficiency.
URL:https://eurodyn2026.org/submission/278
END:VEVENT
END:VCALENDAR