[Bernhard]

"Assumption is the mother of all f* ups"
I think in your case it is easy to test with a periodic inlet domain, using the mapped boundary condition. Maybe it is worth giving it a try, to rule out this effect? Or to confirm your assumption

[Djub]

Periodic inlet domain?
In a very different case (atmospheric turbulent layer), I am mapping a plane on my inlet in order to developp my turbulence.
Here in the case of vortex shedding, I don't need to developp any turbulence because my flow is with no turbulence (or very low turbulence).
I can imagine that there is a problem with a too short inlet: upwind of an obstacle, there exists an over pressure "bulb" wwhich is able to repel the streamlines. Thus, the inlet must be longer than this "bulb". Limiting the inlet should add constrains on the shape of this bulb, and could affect slightly the frequency of vortex shedding.
I could try a longer inlet, just to check if the frequencies are getting better. But they are already not bad! I could also try to get a longer outlet, or a wider section (to get lower blockage effect), or better mesh (12, or 15 cell for the edge). Why do you thing the inlet length is the critical parameter?

[anishtain4]

I have not read all the posts but are you sure you should use LES? a 2-D case is not turbulent at all since it cannot model any vortex-stretch, by adding the LES model you are practically imposing a much more viscosity to your case, try running with laminar model

[Djub]

Hi anishtain4,
despite the figures, my case is fully 3D, with an extension in Z direction. I ran LES because it appeared to me a good choice: I have a high Reynolds, and need transient analysis. What could I have chosen?
Here -I think I've catched your idea- it seems not necessary to have such complicated LES, becasue in fact we have only one large eddie, not a whole spectrum turbulence. Nevertheless, my aim is to developp a high turbulence profile (atmospheric turbulent layer), and my idea was to use elementary turbulent wakes, making them interfering on each other to generate the whole spectrum. Thus, my first step is to have a correct vortex shedding, but in a simulation able to handle also smallest eddies.
What do you expect with a laminar case? I am quite happy with my results. How will they be improved?

[anishtain4]

In post #9 you mentioned a 2d simulation and given nor information is provided regarding 3rd dimension I assumed you have one cell with empty conditions on the z direction this is how I told you not to use LES.

in another post you said your maxCo is 10!! in LES you have spatial filter but no temporal one, so you should have small enough time so smallest resolved scale evolve in time, usually a maxCo of 0.2 is suggested.

with uniform inflow conditions you can't have a continuous spectrum as I see in your first picture. you should use some LES inflow features to have a broad range of eddies affecting each other. there was another topic in OpenFoam when a person was trying to make a good inlet conditions for LES in urban simulation, look it up, that maybe helpful.

[Djub]

No, my case is fully 3D. Actually I have about 50 cells along the Z direction; and my cells close to the beam have an aspect ratio of 1/2 (not good, I know).
Why do you want me to resolve small eddies? I was interested in the Vortex Shedding Frequency. Thus, the frequency of large eddies. And with no inlet turbulence. It works fine with an implicit algorithm (Pimple) and a maxCo of 10. Why to refine in time?
I know I am a a kind of iconoclast, not doing the way as other do. At the begining I wanted to have a correct vortex shedding frequency. Some people told me: refine in time; refine in space; change your SGS model. In fact, all this seems not so important! It works with a corse mesh, both in time and space. And SGS model doesnot change much the result. Why to chnage all this ?

[Bernhard]

"You throw dice and you get the Strouhal number" J. Ferziger

( http://www.google.nl/url?sa=t&rct=j&...42261806,d.d2k )

[Djub]

OK Bernhard.
I used the Strouhal Number as a proof of a good simulation. When my Strouhal number was incorrect, my case was incorrect! The author of your kind reference says that the "contrapositive" is not true, saying "a bad Strouhal Number is the proof of a bad case, but a good Strouhal Number is NOT a proof a good case".
Now I feel completely depressed. So much work, and faith, to have correct results in terms of unitary unsteady wake, and finally all this is wrong...
Cruel world...

[anishtain4]

Well then, if you have 3d grid then it's a 3d case
but about time refinement. You have to resolve small scales or you get wrong simulation for whole, having a converged run does not mean your case is solved right. One eq model you are using solves an additional equation for k (turbulent kinetic energy) and then uses that for approximating subgrid scale tensor. Even if you use any other LES model they all use smallest resolved scales to calculate subgrid tensor, now that you are using a big time step you are practically omitting your model.
It's not about being iconoclast it's about science, either you know theory of what you use, or you get unexpected results.

after what bernard mentioned if I were in your shoes I would have tried to validate my case with other factors.

[Djub]

Well my scientific background is much more experimental than numerical. And I know that vortex shedding, for simple polyedra, does not depend on the roughness, does not depend on Re, and depend only very few on the incident turbulence (for "reasonable" level of turbulence). Thus, I am expecting that the vortex shedding in the "numerical world" does not depend on the small eddies.


In Bernard's reference (Fröhlich and van Terzi, 2008), it is said that for a square cylinder:

• experiments gives St=0.143
• LES gives St=0.144
• RANS gives St=0.121

Conclusion: RANS is incorrect.
By the way, the Eurocode I (eurocode 1991-1-4 ) give St=0.12 for a square cylinder... And Lyn, Einav, Rodi and Park (1994 - 1995) concluded from their measurments that St=0.133 +/-0.004 (with Re=22k). Where is the truth ?

It doesnot appear evident to me that LES has to get a very high resolution: the SGS model is here to help us ! I think we have to deal with large eddies (here, saying the main vortex, sized about one diameter), and let the model deal with smaller things. Of course, we don't cut à 99% of D, but in my case about 1/10 sounds not idiot, does it?
In my case, Co=10 is about dt=0,02s. it seems to me already very short. My vortex shedding is about 5Hz, about 10 points per period. All have been chosen to discribe quite well the largest eddie (saying, the vortex shedding), and let the SGS model deal with smallest eddies. I am not interested in describing the small turbulence in the flow.

I Validaded my case changing Reynolds number, SGS model, time step, domain size, schemes, algorithms, mesh refinement. Maybe one important test, still lacking, is about mesh resolution.

Maybe I am throwing dices, not doing sciences. But now, my dices are very well cheated to give only the good results...

[Bernhard]

I am sincerely sorry for ruining your week. I hope you realize that numerical work is more than just running simulations. For each model, there is quite a lot of physics behind.

Some advised reading may be Eugene de Villiers PhD thesis: http://powerlab.fsb.hr/ped/kturbo/Op...ersPhD2006.pdf
Also, you may want to read the chapter in Pope (Turbulent Flows) about LES. This book is available almost everywhere and covers important issues.

[Djub]

Bernhard, you did not answer one question:

what is the strouhal number for a square cylinder ?

[Bernhard]

I have honestly no idea. This is not an easy question, but you can probably find (and did already) a lot in literature. Is it fundamentally different from vortex shedding around a circular cilinder?

[Djub]

You told me something about throwing dices.
I've read your reference: he is saying that RANS is bad because it founds St=0.121, instead of a measurement given as a reference saying 0.143. LES gives 0.144. He concludes that his LES calculation is good, at the opposite of RANS which results is bad (0.121), as bad as throwing dice.

Other sources in literature give 0.12 or 0.133 . I am not so sure that this RANS calculation is so bad, and this LES calculation so accurate!

Who is throwing dices ?

[anishtain4]

Does the cases match geometrically?