Speaker
Description
We present a novel workflow for solving flow problems on multi-billion-voxel images using Direct Numerical Simulation (DNS) and High-Performance Computing (HPC). DNS is a powerful tool for investigating flow and transport in porous materials, but its application is typically limited by memory constraints, with images of approximately 500^3voxels often regarded as the practical upper limit. We demonstrate that this limitation primarily arises from the need for complex, conforming mesh generation.
To overcome this bottleneck, we developed a new workflow, implemented in our open-source, OpenFOAM-based simulator GeoChemFoam, that enables simulations directly on ultra-large micro-CT images comprising billions of voxels. A key aspect of the approach is the use of approximate immersed boundary methods (e.g. penalisation and volume-of-solid formulations), in which solid surfaces are represented by a volumetric indicator function rather than an explicitly resolved mesh. This allows the use of simple Cartesian meshes that can be generated efficiently and scalably in parallel.
We assess both weak and strong scaling using sub volume decomposition and show that, owing to the reasonable parallel efficiency at scale and the computational power of the UK national supercomputer ARCHER2, full-resolution CFD simulations can be performed without image coarsening or size reduction. In practical terms, flow simulations (permeability) on moderately sized images (e.g. 500^3voxels) can now be completed within minutes on a standard workstation, while simulations involving tens of billions of cells can be carried out within a few hours on ARCHER2. This work highlights the potential of modern HPC to enable detailed, full-scale simulations on high-resolution micro-CT data, opening new opportunities for scalable multiphase flow and reactive transport simulations in geological and engineering applications.
| Country | United Kingdom |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
