WG1 - Alternative model-based methods

For start, the aim is to apply a few existing ML-based surrogate modelling developments for rotating drum data

• Of course one could also opt for other methods such as Reduced Order Modelling, for example.

Links to the available rotating drum datasets

• Yade - https://surfdrive.surf.nl/files/index.php/s/z4AMRCAOAFIZ4qG (Tri-axial data)

• MercuryDPM - https://doi.org/10.5281/zenodo.15281830

• Or feel free to bring your own rotating drum data

List of techniques to pick from

• Graph Neural Simulator (GNN-based)

  • https://www.geoelements.org/publications/gns3d/

  • https://github.com/geoelements/gns

  • https://github.com/GrainLearning/graphs

• Universal physics transformer

  • https://github.com/ml-jku/UPT

• NeuralDEM

  • https://arxiv.org/abs/2411.09678

  • https://nx-ai.github.io/NeuralDEM/

• Simpler RNN

  • https://github.com/GrainLearning/grainLearning/tree/main/grainlearning/rnn

• Simple surrogate model script

  • https://bitbucket.org/mercurydpm/mercurydpm/src/2c3ad7d1fb19b55a06230cf9451f39058e70a840/Drivers/Calibration/CalibrationTools/surrogate_models.py?at=feature%2FAWSCalibration

As a sandbox, it would be great to have all our developments under one roof. For more details checkout https://github.com/deepaktunuguntla/ON-DEM-WG1-AlternativeMethods