CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM (http://www.cfd-online.com/Forums/openfoam/)
-   -   Vortex Shedding on LES (http://www.cfd-online.com/Forums/openfoam/106362-vortex-shedding-les.html)

 Djub August 27, 2012 06:14

Vortex Shedding on LES

Hi everybody !

(in fact, this is the continuation of another thread, but with new data. Thus, a new thread ?)

This WE was devoted to the calculation of the wake of a simple rectangular cylinder in an homogeneous flow. My purpose is to "tune" OpenFOAM in order to have a correct behaviour concerning the Von Karman Alley, or Vortex Shedding.

Here are some characteristics of my run:
• Mesh: the side is 3 times refined near the cylinder, up to about a cell of 1 mm
http://www.cfd-online.com/Forums/mem...r-than-1mm.png
• The SGS model is oneEqEddy with cubeRootVol delta.
• Schemes:
• Time: CrankNicholson 1/2,
• Grad et div: Gauss linear (everywhere but grad(U): cellLimited Gauss linear 1, div(phi,U) Gauss linearUpwindV grad(U),
• Laplacian: Gauss Linear Corrected
• Interp: linear
• Solution: "classical" PCG/DIC tol 1e-6 for p, PBiCG/DILU tol 1e-5 for others
• pisoFoam with 2 correctors and 1 nonOrthoCorrector.
• decompound on 8 cores using scotch
• Inlet: uniform u=3.3, k=0.1 and nuSGS=0.1
• Boundary Conditions: kqRWallFunction with default parameters
After some time, the solution seems quasi periodic.
A snap shot (arbitrary phase):
http://www.cfd-online.com/Forums/mem...o-periodic.png
And zoom near the cylinder:
http://www.cfd-online.com/Forums/mem...esentation.png

Any advice or comment concerning this run ?

The solution seems OK, but the Strouhal number is completely wrong: I found about 0,18, whereas the Eurocode says 0,06 (three time less!!!). What is that ? What did I make wrong to have so huge error?

 niaz August 27, 2012 08:12

Can I ask you what kind of mesh do you use to create these mesh?

 Djub August 27, 2012 10:47

blockMesh with very small expanding , and then snappyHexMesh. Does it look strange in your opinion?

 niaz August 28, 2012 14:19

the triangle cells may affect your solution

 Djub August 29, 2012 03:50

I agree. And most of all, some of these triangulars seem in the incorrect way. How to prevent snappyHexMesh from doing this during refinning?

Do you think a bad mesh could explain the error in the Strouhal number? Don't I mysunderstand strongly something else? The wake is 2~3 times faster that it should be! There is a strong bias !

 Bernhard August 29, 2012 04:09

It is not a snappyHexMesh issue but a Paraview issue. Deselect 'decompose polyhedra' when loading your foamfile.

 Djub August 29, 2012 08:45

Thanks Bernhard. So, these triangular have nothing to do with the solution, and thus nothing to do with the tremendous bias on the Strouhal number ...

Nobody has allready modelled correctly the Von Karman Alley with OpenFOAM ?

 fs82 August 29, 2012 11:03

I never dealed with a square cylinder but first think come into my mind: You should think about your Reynoldsnumber and the regime. May be you have to adjust the model constants of our oneEqnEddy SGS model. It could also be that you are using the wrong SGS model.
I also would refine the mesh in the wake of your square cylinder, not only at the surface. Have you considered to get rid of the wallfunction and resolving the viscous sublayer (y+ < 1)?

Fabian

 Djub August 29, 2012 11:52

Fabian,
My Reynolds number is about 22 000. I didn't know the SGS model was dependent on this Re ? Do you have any suggestion instead of "regular" OneEqEddy (saying with default parameters) ?

Concerning my mesh, my moderate computing power limits me: refining either the wake or the cylinder's surface would imply a 2D study. Do you think it should be better?

 fs82 August 29, 2012 12:13

Most of the SGS models are developed for a specific case. As far as I know there exists no SGS model which could be applied to universal cases. Often the coefficients vary if you deal with a low tubulent flow (cylinder at Re=3900) or a high turbulent flow (cylinder 180000) or so. The model is tricky I guess. I would check other papers to see which model the authors use. Probably a ordinary smagorinsky (SM) is not to bad. I would try more than one model and see how the Strouhal number behaves. But especially with a SM model be carfull with the coefficients.
You have a very high Re number. I assume the regimes are comparable to circular cylinder flows. For my opinion, without a refinement in the wake you will never get a right Strouhal number. What are your boundary conditions for the inflow and outflow? You should also ensure that your domain is large enough, to ensure you have no effects from the downstream BC influencing your solution. I have no clue right now how far, but you definitly should check the various papers on cylinder flows for the dimensions.
Ok final and most important guess: Take a paper for a circular cylinder, e.g.

Wissink, J. & Rodi, W. Numerical study of the near wake of a circular cylinder International Journal of Heat and Fluid Flow, 2008, 29, 1060 - 1070

and try to set up the same case with OF and if you get the right Strouhal number and velocity profiles in the wake. Tune OF as long as you do not fit their results. Than switch back with the same setup to your squared cylinder and see what happens.

Fabian

P.S: 2D makes no sense in my opinion. This Re number is to high for assuming a mainly 2d flow. If it would be laminar, you can do 2d but not in this highly turbulent case. Please correct me anybody of I am wrong.

 Djub August 29, 2012 12:47

OK, thanks Fabian. I agree with you for everything now (I just don't want to make the same case as others, because I am over confident... and I naively think in this way, I will save time!!!).
'see you in two years when I will have done all what you are saying :D

Sincerely,

 fs82 August 29, 2012 13:55

I do not think it will take you 2 years. But it would be the right way. The basic setup you have to figure out anyhow. If you do it with your squared cylinder there is much guessing and less knowing. It will coast you very much time to find the right setup. If you calculate a circular cylinder frist and compare it to known results, is less guessing and you get a good setup very quick. Than the only thing you have to change is the geometry (basic mesh should remain the same) and change the inlet velocity. The postprocessing is also established already, because you did need it for the validation and your done. Think about it again :-D

Fabian

 cosimobianchini September 3, 2012 04:14

First of all: how did you calculate the Strouhal number?
i.e. Which parameter did you base it on? How did you get the principal frequency?

Regarding your solution it seems to coherent for being a properly resolved LES. I guess this comes from a too coarse mesh and also setting nuSgs at the inlet to 1e-1 makes no sense (if evolving fluid is air or something with equivalent laminar viscosity).
Regards,
Cosimo

 Djub September 3, 2012 05:08

Good Morning Cosimo ! And thanks for your attention.

My Strouhal number is evaluated with the width of the cylinder (dimension perpendicular to the flow, here the shortest side, 0.1 m), the upsteram homogeneous velocity (here 3.3m/s).

I put a probe in the wake, and my frequency is evaluated thanks to the Ux velocity time series (see post 378281).

'sure my small eddies are not resolved here. But I don't mind! I want the large ones, specially the vortex shedding. I thought the principle of LES was exactly this: resolve large eddies and model small ones. Am I wrong?
In fact, the effective BC concerns k, not nuSGS (calculated from k). And my inlet-k is 0.001. Calculated nuSGS is about 5e-5.

Do you think I have to refine my mesh near the cylinder or in the wake?

Sincerely

 cosimobianchini September 3, 2012 05:46

Quote:
 Originally Posted by Djub (Post 379968) My Strouhal number is evaluated with the width of the cylinder (dimension perpendicular to the flow, here the shortest side, 0.1 m), the upsteram homogeneous velocity (here 3.3m/s).
This is not influent just be consistent with the definition for this case.

Quote:
 Originally Posted by Djub (Post 379968) I put a probe in the wake, and my frequency is evaluated thanks to the Ux velocity time series (see post 378281).
As someone already suggested to you in the other thread, I would use an integral parameter such as lift coefficient to compute the Strouhal number. Since the flow is symmetric, the stream-wise velocity signal should have principal frequency about the double compared to the vortex detachment one.

Quote:
 Originally Posted by Djub (Post 379968) 'sure my small eddies are not resolved here. But I don't mind! I want the large ones, specially the vortex shedding. I thought the principle of LES was exactly this: resolve large eddies and model small ones. Am I wrong?
Yes that's the LES concept, but the definition of large and small is tricky. In order to obtain reliable results, you should verify to correctly resolve all the turbulence scales not modeled by the sgs model. Since sgs models are developed to model the behavior of the dissipation scales you should have a filter width falling at least in the inertial subrange.
You can read this paper by Prof.Davidson to have an overview on methods employed to verify the reliability of your LES. http://www.tfd.chalmers.se/~lada/pos...r_qles2009.pdf

Quote:
 Originally Posted by Djub (Post 379968) In fact, the effective BC concerns k, not nuSGS (calculated from k). And my inlet-k is 0.001. Calculated nuSGS is about 5e-5. Do you think I have to refine my mesh near the cylinder or in the wake?
Probably both. Indeed using an almost uniform grid in the entire domain is not a very efficient strategy. You should refine more where you aspect to have most of the smallest turbulent structures.

 Djub September 3, 2012 06:28

Thanks a lot. I will read this paper. I am still young in LES problems and I don't "feel" if a LES simulation is OK or not.
Nevertheless, I disagree with you about one point:
Quote:
 Since the flow is symmetric, the stream-wise velocity signal should have principal frequency about the double compared to the vortex detachment one
This is true only in the symetry plane, and wrong every where else. Of course I didnot place my probe in the symmetry plane.

So my huge error in frequency (x3) does not come from this estimation error (x2). Do you think my bad mesh is responsible of multiplying by 3 the vortex shedding frequency ?

 newOFuser February 6, 2013 12:51

Hi Julien

Any updates on this? Were you able to get the correct St #?

Many thanks,
Amit

 Djub February 11, 2013 07:54

Hi everybody!
Excuse me for my long silence: in my language, we say "No news? Good news!". Generally speaking, I write my posts when I have problems. So no post = no problem!
Nevertheless, I feel bad thinking of all this kind community dealing only with problems... Thus, I would like to share those good results finally I had with FOAM, concerning the vortex shedding.
Basically, I wanted to simulate the vortex-shedding frequency of a rectangular beam, depending on its aspect ratio. This was also to train myself with Foam, LES, meshing,...
I worked a lot, with a lot of different parameters: I would not be able to detailed all the steps. But changing things and things, finally I had the correct frequency on the 1:1 case, and then changed the aspect ratio, or certain parameters. Thus, I can say what changes do not work. I cannot explain what to do to have the correct situation, but I know from the correct situation what to do to have a bad case!;)

Basically, I wanted to reproduce this picture from the Eurocode I (NF EN 1991-1-4 Annex E):
http://www.cfd-online.com/Forums/mem...pect-ratio.png

My case is tuned like this:
• Uniform mesh (with no refinement). There is only 10 cell in the edge of the cylinder
• Turbulence model: smagorinsky or oneEqEddy, no major change (even in calculation speed)
• Algorithm: pimplefoam, with a max Co of 10
• Scheme: cranknicholson 1/2 and Gauss linearUpWindV gradU
• Solver: GAMG with PBiCG DILU for U (and k)
The frequency is estimating by a probe in the wake, at the top level of the beam (see http://www.cfd-online.com/Forums/ope...tml#post378281). The main problem is to wait enough time to let all the transcients die. Here are the results:
http://www.cfd-online.com/Forums/mem...pect-ratio.png
In fact, the situation is not so clear, because is several cases, there are several "modes" in the wake. Is those cases, there are several peaks in the Fourrier spectrum, with no relations (such as harmonics: octave, fifth,...). Here you can see those additional modes:
http://www.cfd-online.com/Forums/mem...requencies.png
One of my mystakes has been not to wait enough time: during the transient part, other modes can be wider than the stable mode!
I cannot explain these several modes. Nevertheless, I cannot explain neither the high complexity in the Eurocode curve... My guess is that there is actually competition between several modes, at each aspect ratio, but one mode is accentuated: the Eurocode curve should be parts of different curves. But this is a long other story...
Now, more basic considerations: by changing some parameters, the calculation fails. Here are those results:
http://www.cfd-online.com/Forums/mem...-frequency.png

• at d/b=2, refining the mesh close to the beam makes totally wrong frequency. I concluded that the mesh has to be uniform (or slightly refined, but with an expansion factor of ~5%, no more)
• LUST or bakward sheme both fail.
All this is not very detailed; I would be pleased to discuss this with you, if this work presents an interest for you.

 Bernhard February 11, 2013 08:38

Thanks for sharing!

You are using a uniform inlet velocity (as far as I read here). Did you study the effect of this? How long is your inlet section?

 Djub February 11, 2013 09:09

Yes, I use a uniform inlet. I think it is maybe a bit short, but it is the same as a my reference data (AGARD AR 345 , CMP20 by Lyn, Einav, Rody and Park). Nevertheless, I don't think it affects the Vortex Shedding frequency.

All times are GMT -4. The time now is 12:15.