Skip to content

Latest commit

 

History

History
122 lines (93 loc) · 5.61 KB

tutorial.md

File metadata and controls

122 lines (93 loc) · 5.61 KB
Raw

Tutorial for the impatient

Download and compile

If you are working in AMCG/ESE then do these initial steps on valhalla.ese.ic.ac.uk as that is where the Tarantula mesh generator licensed.

mkdir ~/projects/
cd projects
git clone https://github.com/ggorman/poreflow.git
cd poreflow
cmake .
make
echo "export PATH=$HOME/projects/poreflow/bin:$PATH" >> .bashrc
export PATH=$HOME/projects/poreflow/bin:$PATH

Prepare data sample

THIS SECTION IS OUT OF DATE UNTIL WE FINISH UPLOADING NEW NDDR FILES All the data is available [here][http://www3.imperial.ac.uk/earthscienceandengineering/research/perm/porescalemodelling/micro-ct%20images%20and%20networks]. For this example click on [Berea Sandstone][http://www3.imperial.ac.uk/earthscienceandengineering/research/perm/porescalemodelling/micro-ct%20images%20and%20networks/berea%20sandstone]. You can see a full description of the data sample. Download the image by either clicking on [Download image (binary and ASCII)][http://www3.imperial.ac.uk/pls/portallive/docs/1/33505696.ZIP] or it may be easier to use wget:

wget http://www3.imperial.ac.uk/pls/portallive/docs/1/33505696.ZIP
unzip 33505696.ZIP
mv Image ~/projects/Berea

Next you have to convert the nhdr/raw format to something that the mesh generator can read (there is going to be a lot of converting so get used to it).

convert_microct -v -c vox -s 64 Berea.nhdr

If you just type the command by itself it will give you an example of its usage. The -v option enables verbose mode. As well as outputting messages to the stdout it may also output VTK files for diagnostic/visualisation purposes. -c vox specifies the output format as VOX and the optional argument -s specifies that only a sample of the data is selected.

These files can be huge so always compress them after generation or you will quickly run out of space and patience.

gzip Berea.vox

Mesh generation

Grab a template configuration file for Tarantula:

cp $HOME/projects/poreflow/data/Berea.conf .

Although obviously not required here, rename this to the rename this file to correspond to the dataset you are working on, and update the first 2 lines of the conf file so that the vox and mesh names are correct. I strongly recommend you use a different directory for each data sample or you will continuously clobber old results/data. Next, run the mesh generator:

/usr/local/tarantula/bin/linux/voxmesher Berea.conf

This can sometimes fail. When this happens try modifying the sample size with the -s option above. Powers of 2 are good. If this runs successfully you should see a Tarantula .spm file:

ls -ltr | tail -1 
-rw-r--r--  1 ggorman  esestaff   65685433 Sep  8 11:23 Berea.spm

This us a rather unique format that not many codes will recognise so we will convert this to GMSH format. For this we will use tarantula2gmsh. Tarantuala also drops the resolution parameter at the top of the VOX file. Therefore you should also specify the original nhdr file so that the meta-data can be re-read.

tarantula2gmsh -v -n Berea.nhdr Berea.spm

As well as converting the file format, this command does the following:

  • Images will typically have two materials (rock and void indicated by 1 and 0), you can toggle which material mesh it extracts using the -t flag.
  • Extracts only the active region - it throws away any connected region that is not connected to both sides of the domain along the X-axis.
  • Applies boundary labels: -x, +x, -y, +y, -z, +z, grain boundaries labelled as 1, 2, 3, 4, 5, 6, 7 respectively.
  • Add the -v option if you want verbose messaging and VTK files to admire your beautiful mesh!

Use paraview to take a look at the data. Does it look ok? Is it "fit for purpose"?

Running a simulation

We are going to use caloris.ese.ic.ac.uk because this has the master version of FEniCS and PETSc installed (complements of Patrick Farrell) which is required for the split field preconditioners used.

First set environment variables so you are using the right versions of everything:

source /data/pfarrell/src/local/install_fenics_opt.sh

Convert GMSH to Dolfin XML format so that dolfin can read it. Ideally we should convert to XDMF but I am having some trouble getting that working:

dolfin-convert Berea.msh Berea.xml
python ~/projects/poreflow/python/convert_dolfin_new_xml.py Berea.xml

Finally we are ready to run the Stokes flow solver in parallel using MPI:

mpiexec -n 24 python ~/projects/poreflow/python/stokes-dolfin-snes.py Berea.xml

 # the tail of the output will look something like this:
  211 KSP preconditioned resid norm 5.766382019060e-09 true resid norm 6.166533331081e-23 ||r(i)||/||b|| 6.843859574820e-14
  212 KSP preconditioned resid norm 5.523493457232e-09 true resid norm 1.886341229798e-23 ||r(i)||/||b|| 2.093535183190e-14
  213 KSP preconditioned resid norm 2.250052402294e-09 true resid norm 1.575254855321e-22 ||r(i)||/||b|| 1.748279372793e-13
  Linear solve converged due to CONVERGED_RTOL iterations 213
  1 SNES Function norm 1.575287828239e-22 
Nonlinear solve converged due to CONVERGED_FNORM_ABS iterations 1
[-1.0791655061523711e-16, 1.0791655010304546e-16, 0.0, 0.0, 0.0, 0.0, 0.0]
#################################
## In-flux = -1.07917e-16
## Out-flux = 1.07917e-16
## Permability = 3.20407e-13 m^2 (0.320407 D)
#################################

Note the permability above. Does it look the value you were expecting?

Finally have a look at the solution using paraview:

paraview --data=velocity.pvd 
paraview --data=pressure.pvd