gamma-ReTheta turbulence model for predicting transitional flows
 

Hello, everybody,


since there seems to be a high demand for simulating boundary layer transition with OpenFOAM (see here), I am willing to publish the source code of my implementation of the gamma-ReTheta transition model originally proposed by LANGTRY and MENTER.

This implementation - though complete - is highly unvalidated and I found a couple of problems during my test simulations. I didn't want to release the code until I have fully tested the implementation but since people were asking for it and I don't have the time to continue testing the source code is attached to this post. I don't guarantee that the turbulence model works and leads to reasonable results! It worked well for me at simple zero-pressure gradient flat plate test cases but that's it.

However - there are a few things to account for when using the turbulence model. I won't give support on what boundary and initial conditions to use etc. For questions like that please refer to the literature, this model is well documented here:

R.B. LANGTRY: "A Correlation-Based Transition Model using Local Variables for Unstructured Parallelized CFD codes", 2006, Ph.D. thesis, University of Stuttgart


I created this thread so the community can discuss their results using this (experimental) model. Hopefully together we can validate or improve the model depending on how the turbulence model performs on different test cases. Also feel free to contact me via e-Mail if you have any questions. The address is found inside the header file of the code.



Greetings,
Felix.

Very good Felix!
I hope to have time to validate it soon, anyway it will be a great improvement for the OF community to have a modern transitional model in the code!

Ivan

I completely forgot to attach a sample case with the basic setup for running a simulation with the turbulence model. You can find it attached to this post.


Greetings,
Felix.

Thank you Felix!

Next Monday I will start to work with it and I will tell you how it is working for my case (thick airfoils).

Thank Felix!

I've been waiting for a transition model for a long time, but didn't have the time to implement it myself.

I'm running some airfoil tests at the moment. If the results for 0deg are ok, I will do a full polar tomorrow.

Best regards,
Alex

Alex, if you have any comparison with experiments of your polar (and if you can, of course!), please post here your results!
It could be a good start to validate this model in a collaborative way!

Ok, I have some first results using the default settings for the coefficients.

The case is a NACA63-618 airfoil at 0 degrees angle of attack. Re=6 million.
My grid has a maximum y+ of 0.48 and a stretching ratio of 1.1 close to the airfoil.
The turbulence intensity at the inlet is 0.1%

Experiments Fully turbulent Transition model
Cd 0.0055 0.0097 (+67%) 0.0061 (+11%)
Cl 0.4620 0.4870 (+5%) 0.5242 (+13%)
Cm -0.1094 -0.1180 (-8%) -0.1266 (-15%)

The experimental results are from Abbott & von Doenhoff

As you can see the drag prediction significantly improves. Unfortunately the lift coefficient gets worse. I have no experience with transition models, but perhaps somebody who is familiar with CFX can tell if this is also observed in CFX.

Hopefully, I have a full polar after the weekend.
Alex.

Hello, Alex,


your first results look promising. The reduction of draf actually is expected with any transition model so the much more interesting question would be, how well the current implementation predicts the location of transition onset?

As for the increased lift, I'm not quite sure about this. I suspect the shape of the boundary layers - a plot of pressure coefficients along the airfoil surface would be helpful to analyze the influence on the lift coefficient.


Greetings,
Felix.

Hi Felix,

I agree with you on the drag coefficient. Unfortunately I don't have an experimental skin friction distribution to compare with.

I was a bit to quick on the lift. It makes sense that it goes up, since the result is now more physical. The laminar part makes the lift go up. If you would have a fully laminar airfoil with a certain lift, the lift will be much lower if the same airfoil would be fully turbulent (rough or tripped). This is because of the effective camber of the airfoil.

The first test was with the coefficients as you have set them. Perhaps for airfoils, there is a better choice. For now, I'm reading Langtry's thesis and your source code to get a better idea of the model.

Best regards,
Alex.

Hello, Alex,


sounds like an explanation to me. So the conclusion is, the k-Omega-SST turbulence model overpredicts lift values using your current setup. For the fully turbulent case it should lie below experimental data. Could be a mesh dependency, couldn't it?


Greetings,
Felix.

Hello,
which file i have to include in the make\options for compile it???
i have an error when i put wmake in the terminal:
salvatore@ubuntu:~/OpenFOAM/salvatore-1.7.1/applications/gammaReThetatSST$ wmake
options:7: warning: backslash-newline at end of file
SOURCE=gammaReThetatSST.C ; g++ -m64 -Dlinux64 -DWM_DP -Wall -Wextra -Wno-unused-parameter -Wold-style-cast -Wnon-virtual-dtor -O3 -DNoRepository -ftemplate-depth-40 -I/opt/openfoam171/src/turbulenceModels -I/opt/openfoam171/src/turbulenceModels/incompressible/RAS/RASModel -I/opt/openfoam171/src/transportModels -I/opt/openfoam171/src/finiteVolume/lnInclude -IlnInclude -I. -I/opt/openfoam171/src/OpenFOAM/lnInclude -I/opt/openfoam171/src/OSspecific/POSIX/lnInclude -fPIC -c $SOURCE -o Make/linux64GccDPOpt/gammaReThetatSST.o
g++ -m64 -Dlinux64 -DWM_DP -Wall -Wextra -Wno-unused-parameter -Wold-style-cast -Wnon-virtual-dtor -O3 -DNoRepository -ftemplate-depth-40 -I/opt/openfoam171/src/turbulenceModels -I/opt/openfoam171/src/turbulenceModels/incompressible/RAS/RASModel -I/opt/openfoam171/src/transportModels -I/opt/openfoam171/src/finiteVolume/lnInclude -IlnInclude -I. -I/opt/openfoam171/src/OpenFOAM/lnInclude -I/opt/openfoam171/src/OSspecific/POSIX/lnInclude -fPIC -Xlinker --add-needed Make/linux64GccDPOpt/gammaReThetatSST.o -L/opt/openfoam171/lib/linux64GccDPOpt \
-lincompressibleTurbulenceModel -lincompressibleRASModels -lincompressibleTransportModels -lfiniteVolume -lOpenFOAM -liberty -ldl -lm -o /home/salvatore/OpenFOAM/salvatore-1.7.1/applications/bin/linux64GccDPOpt/gammaReThetatSST
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 0 has invalid symbol index 11
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 1 has invalid symbol index 12
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 2 has invalid symbol index 2
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 3 has invalid symbol index 2
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 4 has invalid symbol index 11
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 5 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 6 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 7 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 8 has invalid symbol index 2
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 9 has invalid symbol index 2
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 10 has invalid symbol index 12
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 11 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 12 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 13 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 14 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 15 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 16 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 17 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 18 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 19 has invalid symbol index 13
/usr/bin/ld: /usr/lib/debug/usr/lib/crt1.o(.debug_info): relocation 20 has invalid symbol index 20
/usr/lib/gcc/x86_64-linux-gnu/4.4.5/../../../../lib/crt1.o: In function `_start':
(.text+0x20): undefined reference to `main'
collect2: ld returned 1 exit status
make: *** [/home/salvatore/OpenFOAM/salvatore-1.7.1/applications/bin/linux64GccDPOpt/gammaReThetatSST] Errore 1
salvatore@ubuntu:~/OpenFOAM/salvatore-1.7.1/applications/gammaReThetatSST$

could you help me please???

@Felix

Indeed you can say that kOmegaSST is overpredicting lift. Did you observe this in your simulations as well?

I have done a grid convergence study and I'm confident that this solution is not mesh dependent. I actually did the same case on a mesh (also checked grid convergence there) from a different mesh generator. So I'm pretty confident that the results are mesh independent. The far field is 50 chords away in all directions, so also boundary effects are not an issue. The mesh also complies with the recommendations of Langtry (page 43 of the thesis).

@salvatore

You have to compile a turbulence model as a library using: wmake libso

You can make a Make directory in the gammaThetatSST directory with two files: files and options.

files:
gammaReThetatSST.C
LIB = $(FOAM_USER_LIBBIN)/libGammaThetatSSTmodel

options:
EXE_INC = \
-I$(LIB_SRC)/turbulenceModels \
-I$(LIB_SRC)/transportModels \
-I$(LIB_SRC)/finiteVolume/lnInclude \
-I$(LIB_SRC)/meshTools/lnInclude \
-I$(LIB_SRC)/turbulenceModels/incompressible/RAS/lnInclude
LIB_LIBS =


To use it, you have to add to system/controlDict :
libs ("libGammaThetatSSTmodel.so");

This is well explained here:
http://openfoamwiki.net/index.php/Si...ry_/_Tutorials
or here:
http://www.tfd.chalmers.se/~hani/kurser/OS_CFD_2008/implementTurbulenceModel.pdf

I would put the model not under applications, but of course you're free to choose.

Good luck.
Alex.

@salvoblack: Can't really help you on this one... it compiles without complaining on my system. This is how I compile it:

- put the folder gammaReThetatSST inside the directory src/turbulenceModels/incompressible/RAS/
- in src/turbulenceModels/incompressible/RAS/Make/files I add the line gammaReThetatSST/gammaReThetatSST.C
- compile with wmake libso (working directory src/turbulenceModels/incompressible/RAS/)

@Alex: Thanks for doing a grid dependency study, I appreciate the effort you're putting into this.
I haven't done external aerodynamic simulations with kOmegaSST yet, it usually isn't my turbulence model of choice for airfoil calculations. So, no, no experience with this.

ok!! now it works thanks a lot!!

p.s. do you know if it is possible to do a run with a 3D mesh???...i would study the case of a wing...

Ok, I would suggest to try an airfoil first to get an idea of how this model works. And definitely read the thesis of Langtry. There are important mesh requirements on y+, stretching ratio and resolution on the surface.
Langtry shows a lot of 3D test cases in his thesis, so it is possible, yes.

Hello.
I have implemented the model of turbulence proposed by Felix and I have studied the case of a NACA 64312 airfoil.
These are the results
Re = 3 * 10 ^ 6 M = 0.16 Alpha = 4 °

Spalart-Allmaras model (fully turbulent)
Cd = 0.0325551
Cl = 0.662492
Cm = 0.0686287

K-omega model (fully turbulent)
Cd = 0.00995945
Cl = 0.701517
Cm = 0.0665175

Transition model
Cd = 0.0113176
Cl = 0.696949
Cm = 0.0652054

I also used xfoil and i heve these results with the same conditions cl=0.707 cd=0.008

As you can see the results of the model of transition are very good when compared to the model Spalart-allmaras (decrease of Cd and increase of Cl), while they are a little less when compared to the k-omega.
What do you think?

Finally could you clarify the discussion on the prerequisites that must have the mesh? where can I check that my mesh is ok??

Hello,


what k-Omega model were you using for your simulations? The "old" version of Wilcox' model (as implemented in OF) ist quite sensitive to freestream data.

For the grid requirements when using the gamma-ReTheta turbulence model please read the Thesis of LANGTRY. If I recall correctly he dedicated a whole subchapter of the appendix to this topic.


Greetings,
Felix

hello felix, thank you for your answer. i will study the thesis....
for the version of the k-omega model, i use the one that is present in the version 1.7. do you think that this version is not reliable???

Hello,


no, I wouldn't say that. It's just the model is quite sensitive to inlet conditions. You can try and vary omega at the inlet/freestream a bit and have a look if it changes the drag values.

You could also try another turbulence model - k-omega SST for example.


Greetings,
Felix.

felix sorry, but I have not explained very well.
the turbulence model that I use is just the k-omega SST model.

greetings.

Hi Felix,

I have a question about the correlation for ReThetac. You state that the CFX_v1.1 version is implemented according the AIAA paper of Langtry and Menter. In the source code I see that the correlation of Langtry and Menter is commented and you use a "corrected" correlation. Is this part of your own research, or is this new "corrected" correlation documented somewhere?

Furthermore, you have seem to miss something in the F1 blending function. The definition in the paper and thesis of Langtry is: F1=max(F1_orig,F3). The max is not in your implementation.

If I revert to the correlation of Langtry and Menter and include the max in the blending function, the results improve.
At 0deg AOA: Cd = 0.00579 instead of 0.0061 (experiments: 0.0055)
The simulations for higher AOA's are still running, but seem to improve even more. The lift of the original implementation is overpredicted a lot.

Alex.

Hello, Alex,


wow, yes you're right. I totally forgot to add the max function when returning F1. I owe you a beer, I would've never discovered this after doublechecking the code multiple times. This makes things more interesting, I think I will rerun my old test cases with this correction. I am a bit surprised the compiler doesn't complain about returning comma separated values to a volScalarField, though.

Yeah, the "corrected" correlation is based upon my personal research. I discovered that both OpenFOAM and FLUENT lead to too early transition onset locations (see attachement 2 in this post) using the original correlation of LANGTRY and MENTER, so I recalibrated the coefficients. In said attachement you can see the original correlation (FLUENT curve) and the improved correlation (OpenFOAM curve). LANGTRY states that the correlations may have to be corrected for each solver so I assumed this correction was neccessary for OpenFOAM.

But the missing max(...) function in F1 of course changes everything. I will simulate the test cases again when I find the time and present the differences in the results.

Thanks again for finding this flaw.


Greetings,
Felix.

Changelog:

You're welcome. That's what this forum is for.

I noticed that something was wrong when I saw the results for higher angles of attack, they were completely wrong. I think also your results will now improve a lot.

Alex.

Hello, everybody,


so I just did some of the flat plate simulations with the corrected version of the turbulence model I posted yesterday. Cf distributions along the plates for two different cases (T3A - moderate freestream turbulence intensity (FSTI), T3B - high FSTI) and as you can see the now correct F1 blending function only affects the skin friction at the transitional and turbulent regime.

In T3A the skin friction is reduced which might explain the improvement of Cd values for Alex' airfoil cases. Still, using LANGTRY and MENTER's correlation for ReThetac the transition onset location is way too early. So either there are still some errors somewhere or the correlation needs calibration.

In T3B the transition onset seems to be predicted quite well, the skin friction is much too high, though. This is a flaw of the model itself caused by the high turbulent viscosity (approx 100 times nu) in the freestream diffusing into the boundary layer. Still, the results in the paper of LANGTRY and MENTER look much better - and I don't know why.


Greetings,
Felix.

It's not even that early anymore but still I forgot the attachements. So here they are.

Hmm, perhaps calibrating the correlation will improve things. But indeed the difference, especialy for case T3A is quite big.

For the airfoil case, I would like to do some xfoil simulations to compare the transition location. Unfortunately, I don't have much time to spend on this the coming weeks.

I am running some simulations right now on a thick airfoil, as soon as I will have them I will show the results in here, this should be in 1 day or 2 maximum.

Hi Josè,
what's about your airfoil simulations? The new model worked well?

Hello!

Sorry for not answering. The results I have got until now for an AoA of 8 degrees are underpredicted 8% compared to the reference I am comparing to. Be aware that I am running a thick flatback airfoil, thus it is not an "easy" airfoil to simulate.

I have thought 2 things to have a look at:
1) Please have a look at the 2nd last message in the following thread: http://www.cfd-online.com/Forums/ope...tml#post298728
2) Use the correlation factos from another autor (i.e. Sørensen--> http://onlinelibrary.wiley.com/doi/10.1002/we.325/pdf)

I have no more ideas right now. I would really like to get some suggestions of these 2 things I have posted or from other aspects you consider.

Thanks for your attention and interest.

Regards,

José

Hello, José,


what physical quantity differs from literature by 8%? I assume it is the drag coefficient? If so, I think this is a pretty good start, especially for that angle of attack.

Regarding your two thoughts:

1) I'm not quite sure I understand correctly what you mean. Do you want to obtain the wall value OpenFOAM uses when applying the omegaWallFunction? It should be visible within paraView. What exactly do you want to do?

2) Yeah, of course different correlations can improve - or worsen - the results of a simulation with an empirical model. I can't access the paper but if you send me the equations/coefficients I'll be happy to implement that correlation.


Greetings,
Felix.

Dear Felix,

Sorry for not giving information enough. The difference of 8% I told you is for both the lift and drag coefficients, so I think this difference should be lower for the lift coefficient. However, I will plot the full polar and paste it in this thread if you want.

1) I am not using wall functions for omega (i.e. I am not setting any omegaWallFunction value in the BCs files), neither for any other variable. This was just a misunderstanding of Gerard (from the other thread I pasted in here before).
What I was saying is that I run the same computation with ANSYS CFX and it gave me a different value for all the values of omega at the wall, while in OpenFOAM what I do is that I fix it to a value (based on the equation of w_wall of the following link --> http://turbmodels.larc.nasa.gov/sst.html). Thus, if I could set it to a "calculated" value somehow maybe I would get better results. Another thing that I am thinking is that maybe ANSYS CFX gives as values at the wall, values from the 1st cell....

2) I cannot attach the paper of Niels Sørensen because of its size, so tell me your e-mail address and I will send it to you by e-mail (if you don´t want to do so in here, just send me an e-mail to the address I have sent to you in a private message). I was thinking on implementing it my self but since I am not experienced with C++ it can take a while to implement it. But if you do it and upload it I will test it on this airfoils and tell you the results, of course.

What do you think about everything I have told you?


Thank you again for all your help.

Regards,

José

Hello, José,


sure, go ahead and post the polar. It would be interesting, too, to compare the polar with results obtained with the k-Omega SST model, if you have that data available.

So you use fixedValue for omega at the wall using the equation proposed by Menter? This is correct, yes, but only if your first cell spacing over the airfoil surface is constant! I recommend the usage of omegaWallFunction - this BC automatically calculates the correct value and you don't have to bother with manual calculations. OmegaWallFunction is valid for the whole y+ regime. You can find some more infos about near wall values of omega here:
http://www.cfd-online.com/Forums/ope...tml#post295498

I don't know about CFX, maybe the software uses different omega values. Maybe you can find information about that in the manual?


Greetings,
Felix.

I added the correlation of Sorensen (2009). If you want to use it, please uncomment it in the source code.

Changelog:The results of my testcases didn't really improve (actually they got worse) when using this correlation (see attached PNG), but maybe you're luckier!


Greetings,
Felix.

FelixL,

Have you seen the reference of Tue in a older post?
http://www.cfd-online.com/Forums/ope...tml#post233057

He mentioned a paper that has some corrections to Langtry-Menter model. It may be another option for you.

By the way, just a final question, do you know how to fix the transition onset and end points in OpenFOAM? I am simulating a rough cylinder case and there is no correlation or prediction model for that. However, I can still use the intermittency concept (\gamma).

In a simple algebraic transition model, gamma is multiplied by the turbulent viscosity. Where gamma is zero, the flow is laminar. Where it goes to unity, the flows becomes turbulent. There are several models of intermittency function such as Narasimha, Reynolds/Kays/Kline or Abu-Ghannam & Shaw. This is not a problem. However, I got stuck in calculating the distance or Reynolds along a surface. Is this possible in OpenFoam?

Regards,

Guilherme da Silva

gammaReTheta model skin friction and yplus problem
 

hi Felix,
i am testing your gammaReTheta model in O.F. 1.7.1 with T3A test case and i am not able to get the skin friction magnitude/vector neither the yplus. I always get zero values for every time step. I wonder their is something wrong, may be you can help me. Looking forward to your reply.

regards,
taxalian.

Dear all,

I am doing a master thesis using OpenFOAM to compute the flow around thick flatback airfoils. I have run some simulations in steady state until now.
Some time ago I promised I would upload my results obtained using the transition model Felix L. implemented in OpenFOAM.

This is what I am going to do in the next messages. You are very welcome to ask questions to me if necessary.

Regards,

José

Results for a NACA0012 with a blunt trailing edge (4% of the chord is cut off). The reason why I do this is to have a mesh topology closer to the one I am using later on the thesis (which is about thick-flatback airfoils).

Comparison for the cases with and without transition for the boundary layer with OpenFOAM.

TWS=data from the book "Theory of Wing Sections"

Attachment 7465

Attachment 7466

Attachment 7467

Comparison for the cases run with OpenFOAM and ANSYS CFX. Exactly the same mesh and BCs.
Attachment 7468

Attachment 7469