openimpala-cuda 4.4.4

Python bindings for OpenImpala — transport-property computation on 3-D voxel images

OpenImpala

OpenImpala is a high-performance framework for computing effective transport properties — tortuosity, effective diffusivity tensors, and effective conductivity — directly from 3D voxel images of porous media (X-ray CT, FIB-SEM, synthetic microstructures). It solves the governing PDEs on the Cartesian voxel grid using finite differences, bypassing mesh generation, and scales across MPI ranks and CUDA GPUs via AMReX and HYPRE.

Outputs parameterise continuum-scale models such as PyBaMM.

📖 Documentation: https://base-laboratory.github.io/OpenImpala/ 📓 Tutorials: tutorials/ (runnable on Google Colab)

Quick example

import numpy as np
import openimpala as oi

image = np.zeros((64, 64, 64), dtype=np.int32)
image[:, :, 16:48] = 1   # solid slab through the middle

with oi.Session():
    vf  = oi.volume_fraction(image, phase=0)
    tau = oi.tortuosity(image, phase=0, direction="z", solver="mlmg")
    print(f"Volume fraction: {vf.value:.4f}")
    print(f"Tortuosity:      {tau.value:.4f}")

Install

Python (recommended)

pip install openimpala            # CPU + optional CuPy GPU acceleration
pip install openimpala-cuda       # compiled HYPRE/AMReX CUDA wheel (Linux x86_64)

OpenImpala uses MPI for distributed parallelism — install an MPI runtime (libopenmpi-dev, openmpi, brew install open-mpi, or conda install -c conda-forge openmpi) before pip install. See Getting Started for full details.

Container (HPC)

Pre-built Apptainer/Singularity images are attached to each GitHub Release:

apptainer exec -B "$(pwd):/data" openimpala-vX.Y.Z.sif \
    /usr/local/bin/Diffusion /data/inputs

For batch SLURM scripts, see HPC Usage.

Build from source

See CONTRIBUTING.md for the native and containerised developer build, code style, and test workflow.

Features

• Steady-state diffusion / conduction on segmented 3D voxel images
• Tortuosity factor, full 3×3 effective diffusivity tensor, multi-phase transport
• Microstructural metrics: volume fraction, percolation, particle size, specific surface area
• TIFF / HDF5 / RAW / DAT image input; JSON output compatible with BPX / BattINFO
• Solvers: HYPRE (PCG, FlexGMRES, BiCGSTAB; SMG / PFMG preconditioners) and AMReX MLMG (matrix-free, GPU-native)
• MPI + OpenMP + CUDA parallelism — scales from a laptop to multi-node HPC

Citation

If you use OpenImpala in published work, please cite:

@article{LeHoux2021OpenImpala,
  title   = {{OpenImpala}: {OPEN} source {IMage} based {PArallisable} {Linear} {Algebra} solver},
  author  = {Le Houx, James and Kramer, Denis},
  year    = {2021},
  journal = {SoftwareX},
  volume  = {15},
  pages   = {100729},
  doi     = {10.1016/j.softx.2021.100729},
}

If you use the homogenisation-based effective diffusivity workflow, additionally cite Le Houx et al., Transport in Porous Media 150, 71–88 (2023), doi:10.1007/s11242-023-01993-7.

License

BSD 3-Clause. See LICENSE.

Acknowledgements

This work was financially supported by the EPSRC Centre for Doctoral Training in Energy Storage and its Applications [EP/R021295/1]; the Ada Lovelace Centre (STFC) project CANVAS-NXtomo; the EPSRC prosperity partnership with Imperial College, INFUSE [EP/V038044/1]; the Rutherford Appleton Laboratory; the Faraday Institution Emerging Leader Fellowship [FIELF001]; and Research England's Expanding Excellence in England grant at the University of Greenwich via the M34Impact programme. We acknowledge the use of the IRIDIS HPC facility, Diamond Light Source's Wilson cluster, STFC SCARF, and the University of Greenwich M34Impact cluster, and thank the developers of AMReX, HYPRE, libtiff, and HDF5.

Contact

Issues and feature requests: https://github.com/BASE-Laboratory/OpenImpala/issues. Questions: GitHub Discussions.