[sampaio]

Hi All,
Is SFCD suitable for LESimulations?
I am getting oscillations with linear scheme, so would it be good idea to switch to SFCD (bounded, 2nd order)
Best regards,
Luiz

[henry]

No SFCD is not suitable for LES, it is far too diffusive.

[sampaio]

But than, which should I use?
(Incompressible LES, very fine mesh(y+=1 close to the wall), with linear scheme presents oscilations...)
Thanks again,
Luiz

[henry]

What time scheme are you using?
What max Co number?
What SGS model?
What coefficients?
What is the Re number?
Where are the oscillations?
Are they Temporal or spatial?

[sampaio]

time scheme: backwardDifferencing
max Co: 0.5
SGS: SmagDyn (after running some iter with Smag)
coefficients: default
Re: 213000
Where: everywhere close (one chord) to the airfoil. Even upstream...
They are spatial oscillations.

thanks luiz

[henry]

Do you get them with the standard Smagorinsky model?

[sampaio]

What do you have in mind?
I dont remember, I deleted the files because of diskspace.
I will run it again and tell you.
Bests,
Luiz

[henry]

Which dynamic Smagorinsky model are you using? The localised one? If so you might see some odd features due to the lack of averaging; it only uses the neighbour cell values. Better forms of dynamic models use some kind of lagrangian approach for the averaging but we haven't yet implemented any of these more recent models prefering the transport equation based models.

Also what tolerance are you using for p? How many interation does the p solution take?

[sampaio]

I was using dynSmagorinski
Yeah. I also like the transport models, but unless I use their dyn version, I think I will have problems with my delta (cubic root of the volume), because the grid varies a lot in xy plane, but the spanwise spacing (z direction) is fixed. I may get too much dissipation (or too litle) in some regions.

What do you think?

My tolerance for p is ICCG 1e-06 0.
In the initial stages, it requires 400 iter to converge (and 129 for the second iteration inside Piso).

By the way, I am using nCorrector = 2 and no nonOrtogonalCorrectors. Should I use nonOrthogonalCorrectors? I was trying to avoiding to keep computational time feasable...

[henry]

Why do you think the dynamic models will handle grid variations better than transport models? One option to help with the grid variation is to use a smoothed delta field which OpenFOAM supports.

You might be able to get away with a less-tight tolerance on the pressure and gain a bit in performance. I am not sure if you need non-orthogonal correction because I don't know what your grid is like. Run checkMesh to see what the maximum non-orthogonality is.

[eugene]

Hi Luiz,

There is something called deltaSmoothing implemented for LES that will gradually change the SGS length scale independent of the grid. So where you would normally have a big jump in delta because of a jump in cell size, you now have linear transitions. You should be able to select this option inside FoamX. It is called smoothDelta or deltaSmooth or somesuch. The coefficient determines how quickly the transition can occur. Generally set to around 1.15 which will allow a doubling of delta over roughly 5 cells. I reccommend you use either the oneEqEddy or Spalart-Allmaras transport models in conjunctions with this.

Another tip, lately I tend to use LES almost exclusively in compressible format. The reason: those 129 iterations of the pressure equation you are doing is partially because the speed of sound in an incompressible fluid is infinite. Thus the effect of every pressure fluctuation has to be transported to all parts of the domain (or until the wave completely diffuses). By using a compressible formulation, pressure oscilations only have to be transported at the speed of sound. Depending on your mesh and ambient conditions, this can be as low as a few cell deltas. The effect is to drastically cut down the number of pressure iteration (<10), with the added expense of solving the enthalpy equation. Overall your simulation cost should drop by at least a factor of two. Staggering will also lessen considerably. The code to do this is called coodles.

To find out if you need nonOrthogonal correctors, switch them on (2 or 3). If this causes significant additional cost, you probably need them.

Eugene

[sampaio]

Thanks Henry and Eugene,
What I was worried about was not fast variations in delta, but the following:

close to the wall, suppose my delta x, delta y and delta z are respectively 1, 5, and 100, giving a LESDelta = 8.

far from the wall, they are (for instance) 500, 1000 and 100 (because my spacing in z direction is constant). Thus LESDelta = 368.

Pehaps 8 is too big (too dissipative) close to the wall, and 368 is too low away from the wall.
(If this was a 2D case, one would expect LESDelta to be between 1-5 close to the wall and between 500-1000 far away from the wall, right?).

I heard that with dynamic models, you become less sensitive to this issue, since you are adjusting your constant with a dynamic proceedure. So if you get delta wrong, the dynamic adjustment will compensate for it...

Thanks,
Luiz

[henry]

But the dynamic model you have chosen averages over the whole domain, i.e. produces a single coefficient for the whole domain, how would that help with your delta variation?

[eugene]

Luiz

When using a transported SGS model, you will typically also employ some form of near-wall damping (explicit Van Driest for the oneEqEddy or DES for the Spalart-Allmaras). This manages the LESdelta value near the wall (more or less) independently from your cell size based on the distance to the wall.

Anyway, if your near-wall cell size is 1, 5, 100, then you have other problems, which a dynamic procedure definitely wont address.

The long and the short of it is really, the dynamic procedures havent seen much use and as a result much testing and development in foam. They may (unlikely) or may not be applicable to your problem, but I would discorouge their use purly on the grounds of unpredictability. If you are really set on using dynamic models, I would encourage you to thoroughly test them in canonical cases with good experimental data before applying them to more difficult problems.

As Henry mentioned, it is likely you will find they need some kind of lagrangian averaging to make them stable and generally applicable to complex domain flows. I'm sure several LES enthusiasts (including me) would like to see such a model implemented. I can give you a reference if you feel up for it.

Eugene

[sampaio]

Thanks for you very helpful comments.

Henry, I did not realized "average" in dynSmagorinski was a global average. Thank you for pointing this out.

Eugene,
I will be glad to have a look at your references if you send me. I have some references myself about lagrangian average (from ERCOFAT, I think). What I miss most is some more knowledge about foam. But it will come with time...

Thanks,
Luiz

[eugene]

There are a few references on this page:

http://pegasus.me.jhu.edu/~meneveau/pubs-numLES.html

You will be surprised how many hits google produces for "lagrangian dynamic LES".

I have a hunch that in the long run the best SGS model will be a lagrangian dynamic transport equation for the SGS Reynolds stresses. Some results show stress transport models like Deardoff's to be superior to one-equation and homogenous-space averaged dynamic Smagorinsky models. Given the high cost of LES calculations, I think spending a little extra on a good SGS model will be more than cost-effective.

Here is a paper that describes the implementation of a spatially averaged dynamic procedure to a SGS transport equation:
www.tfd.chalmers.se/~lada/postscript_files/ahmad_corsica.pdf

If you can combine this procedure with the lagrangian approach, extending it to the SGS Reynolds stress model in foam should not be too difficult.

[sampaio]

Could you also send me a list of publications using DES implemented in FOAM?
Thanks,
Luiz

[eugene]

Sorry, there are no such publications at present, except maybe for my thesis and that is still brewing.

If you are feeling brave, I can let you have the completed chapters.

Edit: Or are you referring to source publications? IF that is the case simply do a search on the Spalart-Allmaras DES model.

[sampaio]

Sure I am brave...
You can send your thesis chapters to
luizebs@mec.puc-rio.br or
luizebs@stanford.edu

Thanks a lot.
Luiz

[ivanwhlau]

Hi, Eugene,

I am trying to learn more about DES in FOAM. I know this thread was posted long time ago. Anyway, may I take a look with your thesis chapter too? my email address is ivanmailbox@yahoo.com

Regards,
iL