Before and after the case¶
New PyFoam Features¶
or: Comparing solvers without any claim for correctness¶
Bernhard F.W. Gschaider¶
Wien, 2014-06-16¶
"The lesser the people know about how sausages and laws are made, the better they sleep in the night"
Otto v. Bismarck
Bismarck knew nothing about presentations.
This presentation is about how it was made
Who is this Bismarck anyway?
Overview¶
The aim of this presentation is to demonstrate two new features of PyFoam
- Semi-automatic case setup
- Working with IPython-notebooks
Structure of the presentation¶
- What is PyFoam
- Our "project"
- Setting up the case
- Getting the results
And who is PyFoam?¶
- PyFoam is a Python-library
- Aimed at working with OpenFOAM
- Manipulates case files
- Starts runs and analyzes their output
- There are utilities implemented with this library
- This is the part that people usually call "PyFoam"
- Most parts of PyFoam written by the speaker
- Which doesn't mean that he knows how to use it properly
Further information about PyFoam¶
- Webpage: http://openfoamwiki.net/index.php/Contrib/PyFoam
- Short overview
- Download of latest version
- Various presentations (mostly at workshops) linked from that page
Updates
Are usually announced:
- On the Wiki-page
- On the MessageBoard
- On Twitter @swakPyFoam
The case¶
p-U-coupled solver¶
- The next version of
foam-extendhas a p-U-coupled solverpUCoupledFoam- Sets up one big matrix with pressure and all three velocity components
- Solves them all at the same time
- I never worked with it before
- But people who did are quite excited
- Supposedly it converges faster and is more stable
The case¶
- There are tutorial-cases that come with
pUCoupledFoam- Never trust tutorials provided with the solver: Their purpose is to make the solver look good
- Instead a well-known case is used that also comes with (Open)FOAM
pitzDaily: a backward facing step for which measurment data by Pitz and Daily exists
In [6]:
Image(filename="pitzDailyOverview.png")
Out[6]:
Evaluations¶
This section shows some example evaluation. First we set up the environment (remember: most of this is done by pyIPythonNotebook.pyfor you)
In [14]:
%matplotlib inline
import matplotlib
matplotlib.rcParams['figure.figsize'] = (10.0, 8.0)
from PyFoam.Wrappers.Pandas import PyFoamDataFrame as DataFrame
from PyFoam.IPython.Case import Case
Now we get us the four cases
In [15]:
case=Case('/Volumes/Foam/Cases/Pfau2014/pitzDaily_coupled31')
case23=Case('/Volumes/Foam/Cases/Pfau2014/pitzDaily_coupled23')
caseSeg=Case('/Volumes/Foam/Cases/Pfau2014/pitzDaily_segregated31')
caseSeg23=Case('/Volumes/Foam/Cases/Pfau2014/pitzDaily_segregated23')
About a case¶
In [16]:
case.size(level=3)
Out[16]:
In [17]:
case.boundaryConditions(level=3)
Out[17]:
Reading the data¶
In [18]:
dataCouple=case.pickledPlots('PyFoamRunner.pUCoupledFoam.analyzed/pickledPlots')
In [19]:
dataCouple23=case23.pickledPlots('PyFoamRunner.simpleFoam.analyzed/pickledPlots')
In [20]:
dataSeg=caseSeg.pickledPlots('PyFoamRunner.simpleFoam.analyzed/pickledPlots')
In [21]:
dataSeg23=caseSeg23.pickledPlots('PyFoamRunner.simpleFoam.analyzed/pickledPlots')
Which datasets are there?
In [22]:
dataCouple.keys()
Out[22]:
Simple plotting
In [23]:
dataCouple["iterations"].plot()
Out[23]:
Comparing residuals¶
This looks worse than it is (renaming datasets for merging):
In [24]:
linCouple=DataFrame(dataCouple["linear"].select(lambda k:(k.find("_")<0 and k!="Up[2]"),1))
In [25]:
linCouple23=DataFrame(dataCouple23["linear"].select(lambda k:(k.find("_")<0 and k.find("Uz")<0),1).rename(columns=lambda n:n+"_cou23"))
In [26]:
linSeg=DataFrame(dataSeg["linear"].select(lambda k:(k.find("_")<0),1).rename(columns=lambda n:n+"_seg"))
In [27]:
linSeg23=DataFrame(dataSeg23["linear"].select(lambda k:(k.find("_")<0),1).rename(columns=lambda n:n+"_seg23"))
Comparing coupled and segregated¶
In [28]:
linSeg.addData(linCouple).plot(logy=True)
Out[28]:
Lets have a closer look:
In [29]:
linSeg.addData(linCouple).plot(logy=True,xlim=(0,400))
Out[29]:
Seems like coupled converges really faster
Comparing OpenFOAM-2.3 and foam-extend-3.1 (segregated)¶
In [30]:
linSeg.addData(linSeg23).plot(logy=True)
Out[30]:
I don't dare to say who is winning
Comparing coupled and segregated for 2.3¶
In [31]:
linSeg23.addData(linCouple23).plot(logy=True)
Out[31]:
There is a difference
Conclusion¶
- Two new utilities for PyFoam were demonstrated
- One to automatically set up parameterized cases
- The other to create notebooks for the evaluation of OpenFOAM-cases
- Use of IPython-notebooks as a platform for numeric evaluations
- Well suited to document the work
- Maybe not the best tool to write presentations
- But very nice to do commented evaluations
- The
pUCoupledFoamsolver was compared withsimpleFoam- Results are only preliminary
- Seems to converge much faster
- How this translates into CPU-time needs to be investigated further
Where to get it¶
- This presentation and the case files:
- Will be available on http://openfoamwiki.net/index.php/Contrib/PyFoam
- PyFoam-version with the new utilities
- Available on the USB-stick at the http://www.openfoamworkshop.org/ next week
- Through the regular channels some time later
- The
pUCoupledFoam-solver- In
foam-extend-3.1available at http://sourceforge.net/p/openfoam-extend/wiki/Home/ (get thenextRelease-branch in thefoam-extend-3.1-git - Testing is finished. Release at the Workshop
- In
Errata¶
- During the presentation it was suggested from the audience to use the inlet width $25.4 mm$ as $D$ and calculate the actual viscosity from the Reynolds-number provided
- Using that the viscosity is 50% higher than in the original tutorials
- The presentation has not been updated to reflect that
In []: