A common challenge from image-based modelling is the size of 3D tomography datasets, which can be of the order of several billion voxels. OpenImpala is a data-driven, fully parallelisable, image-based modelling framework designed specifically for these high computational cost problems.
3D datasets are used as the computational domain within a finite-differences-based model to directly solve physical equations on the image dataset, removing the need for additional meshing.
OpenImpala then calculates the equivalent homogenised transport coefficients for the given microstructure. These coefficients are written into parameterised files for direct compatibility with two popular continuum battery models: PyBamm and DandeLiion, facilitating the link between different computational battery modelling scales.
OpenImpala has been shown to scale well with an increasing number of computational cores on distributed memory architectures, making it applicable to large datasets typical of modern tomography.
The source code is available at https://github.com/kramergroup/openImpala.git .
cloud.sylabs.io holds a containerised build environment suitable for running OpenImpala and contains:
- Centos 7
- OpenMPI
- Hypre
- LibTiff
- AMReX
- HDF5
- h5cpp
- InfiniBand
- OpenImpala
To pull the singularity image from the hub, first, ensure you have Singularity installed on your system, then run the following command:
singularity pull library://jameslehoux/default/openimpala:latest
This command will pull the OpenImpala Singularity image to your current directory.
You can either operate the container interactively using the shell command, like so:
singularity shell openimpala_latest.sif
With this, you can compile a local version of OpenImpala, useful for development, or you can execute the OpenImpala install that's prebuilt within the container:
singularity exec openimpala_latest.sif /openImpala/build/tests/tTortuosity
to run the tortuosity test, or:
singularity exec openimpala_latest.sif /openImpala/build/apps/Diffusion ~/inputs
to run a Diffusion type problem, N.B., you would need an inputs file (https://github.com/kramergroup/openImpala/blob/master/build/apps/inputs) in your home directory for this to run
It is recommended to build OpenImpala using the provided Singularity image. With the image opened interactively, navigate to the main OpenImpala directory; here, run the following command:
make
This command will create a new folder, /build, populated with the .o files and executables of OpenImpala.
Once the make command is finished, navigate to the test directory to check the functionality of the created executables:
cd build/tests
./tTiffReader
Which will open a sample Tiff file and assert the dimensions are as expected, printing the output.
./tVolumeFraction
Which will open a sample 2-phase segmented tiff file and calculate the volume fraction of each phase.
./tTortuosity
Which will open a sample 2 phase segmented tiff file and calculate the effective diffusion and tortuosity in the x-direction.
The main programme of OpenImpala calculates concentration gradients across a steady-state diffusive problem for a two-phase dataset in the three cartesian directions.
To use Diffusion.cpp, first ensure your dataset is stored within the /data repository, or use the correct file path to navigate to your dataset, then navigate to the apps directory:
cd build/apps
./Diffusion inputs
It will now calculate steady-state diffusion in the x-direction and print the results, as well as volume fraction.
To run the same calculation but using more processors, try:
mpirun -np 2 ./Diffusion inputs
Compare the 'Total Run time' values between the two calculations to check MPI is running correctly.
Try modifying the inputs file to see if you can calculate in the y-direction.
OpenImpala is built on the AMReX software framework. Output plot files can be visualised using several open-source visualisation packages, e.g. Paraview, Visit, yt or AMRVis.
For further information on how to view the visualisation data, visit: https://amrex-codes.github.io/amrex/docs_html/Visualization.html
As an example of the visualisation, here is a concentration gradient for steady-state diffusive flow in the x-direction, for a 499 cubed voxel microstructure [1]:
[1]: Le Houx, J., Osenberg, M., Neumann, M., Binder, J.R., Schmidt, V., Manke, I., Carraro, T. and Kramer, D., 2020. Effect of Tomography Resolution on Calculation of Microstructural Properties for Lithium Ion Porous Electrodes. ECS Transactions, 97(7), p.255.
The code should run with no required input from the user.
You can now use OpenImpala in an HPC batch job, an example file:
module load singularity/3.2.1
export OMP_NUM_THREADS=1
cd /openimpala/build/apps/
mpirun -np 20 singularity exec openimpala_latest.sif ./Diffusion inputs
N.B., the OpenImpala singularity image needs to located in the same directory as the executable.
This repo is pull mirrored to a GitLab repository, https://gitlab.com/JLeHoux/openImpala, which runs a unit test suite to check functionality. The test suite can be found in .gitlab-ci.yml
OpenImpala Copyright (c) 2020, University of Southampton All rights reserved.
The license for OpenImpala can be found at LICENSE.