[cfd_newbie]

I am new to DES so my question may seem to be obvious.
I was reading a CFX document where it is mentioned that -
"For DES two-dimensional and axisymmetric simulations are not feasible, as turbulent structures are always three-dimensional."

My doubt is if were are trying to validate a 2D airfoil how do we do DES ?
Do we have to extrude our airfoil to make it 3D ?, if yes then by how much ? and in that case will it be OK to compared a 3D CFD result with a 2D experimental results.

Dear fellow members/visitors please enlighten me on this.

Regards,
Raashid

[Josh]

I wouldn't say it's not feasible, but a 3D simulation will allow you to take advantage of the mostly physical LES resolution in the separated regions. If you're only running 2D, you might as well use URANS.

Read some DES literature. Basically, you want the airfoil as wide as the width of the larger turbulent structures. Do a sensitivity check - run at a few different spans, e.g., 0.1c, 0.2c, and 0.3c, until the solution is span-independent. Your span requirements will depend on the simulation - we can't tell you what will work.

[Martin Hegedus]

I don't consider myself a turbulence expert, so this is just my two cents. Also consider all this my own opinion based on my experience.

Be careful, many experimental results which are advertised as 2D, are not. Once separation occurs on the airfoil, wall effects can become important. In general, once separation occurs, you should model the wind tunnel. This can be seen with the S809 airfoil that was discussed on an earlier thread. There are people who have compared turbulence models to the "2D" S809 stalled values. Unfortunately, from my CFD calculations where I model the tunnel, their comparisons of turbulence models for stall conditions are inappropriate.

Yes, true turbulence is 3D. In a sense this is similar to the vortex with occurs on the WT wall when the airfoil separates. The vorticity vector when created is perpendicular to the flow vector but wants to rotate so it points downstream. A RANS, or URANS, because of the higher eddy viscosity, will prevent this from happening. So URANS stays 2D.

I assume the lower your Reynolds number, the wider your airfoil needs to be. You'll need to do trial and error to find the proper width.

"Numerical Study of Wind-TunnelWalls Effects on Transonic Airfoil Flow" (http://agarbaruk.professorjournal.ru...=DLFE-6665.pdf) discusses interpreting 2D transonic WT data. Probably transonic data is not explicitly what you are interested in. But I could not find a paper that discuses subsonic 2D data. However, it gives you a heads up on the difficulty in comparing to 2D WT data. The interesting part is that this paper is from 2003 and earlier papers don't discuss this topic much, or even at all. In fact, one doesn't see, even in more current papers, much quantitative comparisons about 3D WT affects. Usually it's constrained to one sentence hand waving.

[julien.decharentenay]

The rational behind the CFX statement is that DES is a turbulence model that is supposed to "transition" between a RANS turbulence model and a LES turbulence model. The transition is based on a number of parameters, the dominant ones being the mesh resolution and the turbulent length scale. You/the user has no a-priori control over where the transition occurs.

When DES operates in LES mode, it behaves a smagorinski model. To capture the appropriate turbulence decay one has to do a 3D simulation. Running a DES model in 2D results in an unphysical solution. If you wish to capture the large coherent structures using a 2D simulation, you should put your efforts in URANS as pointed out by Josh.

In addition to the above comments, I would advise using periodicity (or cyclic depending on what the software is calling them) for the spanwise boundary conditions rather than symmetry.

Note on the DES transition between "RANS" and "LES":
1) The transition zone is difficult to define. My best guess is that one can a posteriori assess areas where turbulence kinetic energy was modeled (this would be the RANS areas) and resolved (LES ares);
2) The user do have some control on the transition through the mesh;
3) The following statement is my personal opinion and should be taken as such. The transition between the 2 modes of operation is not clear-cut and I would personally refer to it as a buffer zone (and a potentially unphysical one). This buffer zone is the reason why DES model works well for some cases (massively detached) and not for others (channel flow for example).

Good luck.
Julien

[Josh]

Originally Posted by julien.decharentenay

3) The following statement is my personal opinion and should be taken as such. The transition between the 2 modes of operation is not clear-cut and I would personally refer to it as a buffer zone (and a potentially unphysical one). This buffer zone is the reason why DES model works well for some cases (massively detached) and not for others (channel flow for example).

Agreed. It's also the reason why there are such large inconsistencies between "identical" hybrid RANS-LES simulations.

On that note, yes, the user has some control over the transition region by refinishing spatial (mesh) and temporal (timestep) scales. However, the buffer region seems somewhat ambiguous. It is not strictly based on mesh/timestep. The simulation works ideally when URANS is used to model the attached flow and LES is used to partially resolve the separated regions. Controlling how the model does this is difficult.

Also, just because your DES seems to capture qualitative small-scale, 3D structures does not indicate the solution is correct. DES can fortuitously produce better results than URANS for the wrong reasons.

Sounds like you have some reading to do!

[cfd_newbie]

Dear Joshua, Julien, Martin,
Thanks a lot for putting light on this niche topic.
Surely I need to read a lot of things.
Will get back once I am finished and have more doubts.
Regards,
Raashid

[Martin Hegedus]

Since the original question was about 2D airfoils, I've got a question. Does anyone know of a good experiment to compare to which has a large separated or turbulent region?