[Shamoon Jamshed]

Dear CFD experts,

I wanted to know that in DES when the solver solves URANS in BL and LES outside, what sub-grid scale model it uses in case of LES? Please let me know.

Regards,

SJ

[MarcoC501]

Hello,
in DES you do not have a sub-grid scale model, since you only resolve with LES the core flow regions, in which the turbulent length scale exceeds the grid dimension.
If the turbulent length scale is less than the maximum grid dimension, your solver switches to RANS, so the small eddies that cannot be resolved with the grid are treated in the same way as what happens with RANS solution.
On the other hand, in "pure" LES you are modeling those eddies with a sub-grid model.
In the core flow regions, the LES is solved using a Smagorinsky-like empirical constant which slightly varies based on your RANS model.

[Shamoon Jamshed]

Oh i see, so you mean that instead of LES (sub grid) model we are employing DES to solve those sub-grid eddies, that was basic crux and I was not focusing on that point. Anyways, concept well-cleared now. Thanks,

[Shamoon Jamshed]

Originally Posted by MarcoC501

Hello,
in DES you do not have a sub-grid scale model, since you only resolve with LES the core flow regions, in which the turbulent length scale exceeds the grid dimension.
If the turbulent length scale is less than the maximum grid dimension, your solver switches to RANS, so the small eddies that cannot be resolved with the grid are treated in the same way as what happens with RANS solution.
On the other hand, in "pure" LES you are modeling those eddies with a sub-grid model.
In the core flow regions, the LES is solved using a Smagorinsky-like empirical constant which slightly varies based on your RANS model.

But My dear friend Marco,

It seems quite understandable but rather confusing also. Since LES without a sub-grid model is not picky concept. The LES region must have a sub-grid model to solve for turbulent stresses.

Regards,

SJ

[MarcoC501]

In Large Eddy Simulation, the motion is separated into small and large scales, often by spatially filtering the velocity field with a kernel, GΔ(x). The convolution kernel eliminates scales smaller than the prescribed length scale Δ (related to the filter width). The eliminated scales are the subgrid-scale (SGS) motions, which are calculated using various techniques.

DES is a hybrid approach, i.e. not a proper LES, since RANS ia applied to model the scales LES does not resolve. Also in this case the discriminant is related to the eddy length scale, but in a different way, as I explained in the previous answer.

I also suggest you to google some more infromation, or to consult the "Turbulent flows" textbook by Pope (best idea )

[Shamoon Jamshed]

Originally Posted by MarcoC501

In Large Eddy Simulation, the motion is separated into small and large scales, often by spatially filtering the velocity field with a kernel, GΔ(x). The convolution kernel eliminates scales smaller than the prescribed length scale Δ (related to the filter width). The eliminated scales are the subgrid-scale (SGS) motions, which are calculated using various techniques.

DES is a hybrid approach, i.e. not a proper LES, since RANS ia applied to model the scales LES does not resolve. Also in this case the discriminant is related to the eddy length scale, but in a different way, as I explained in the previous answer.

I also suggest you to google some more infromation, or to consult the "Turbulent flows" textbook by Pope (best idea )

Dear Marco,
I have read the book of Pope, there is no discussion of DES in it. Secondly, you have again said the same thing in the first para, that SGS motions are computed using various techniques. If large eddies are being resolved, then small eddies in LES must be modeled using sub-grid scale model. That is my argument.

[LuckyTran]

I feel the need to chime in here to prevent the further spread of blatant misinformation.

There are clearly subgrid models in LES, otherwise what even is LES? Filtered laminar navier stokes? LES is not a hybrid RANS/DES, DES is a hybrid RANS/LES. Otherwise what is DES? It should be rather apparent considering Smagorinsky model came first.

Now for what is common, DES and LES both use subgrid models in undamped regions.

But to answer the original question, DES uses a Cdes=0.65 which you can compare to the classical Smagorinsky constant or the same Wale constant of 0.325 for the core region. However, DES the way it is implemented in Fluent uses the largest x,y,z grid spacing and not the cube-root of the cell volume (hence why Cdes=0.65 is so vastly different than the ~0.1 Smagorinsky constant or 0.325 Wale constant). It's not important what you want to label these individually, they are all mixing length models with a tunable mixing length constant (i.e. linear strain models). Finally, the contribution of the subgrid scale model should be so negligible in the resolved areas that it is easy to overlook that DES even has a subgrid model since DES is designed for handling cases with walls, and wall errors are much more difficult and where most of the errors come from.

[Shamoon Jamshed]

Thank you LuckyTran, in fact, I was wondering that why are you still silent when a hot discussion is on the swing. Who said that LES is hybrid RANS/DES? Rest apart, that was my point and you answered what I was thinking that LES (in DES) must be using SGS. This also has been asked from Philipps Spalart on ResearchGate, who is the founder of DES, and lets see what he comes up with.

[MarcoC501]

My apologies! I have always thought wrong about the sub-grid scale model in DES.
Thanks a lot LuckyTran for bringing this up and correcting my statements.
Thanks a lot Shamoon to give me this opportunity to learn more

[Shamoon Jamshed]

I came to know with an important finding. It is from my lectures of Advance Turbulence Modeling from Dr. Dimitris Drakakis back in Cranfield in 2009. It is mentioned that
"The Spalart-Allmaras model functions as a sub-grid model when operating in LES mode and as a wall-layer closure when operating in RANS mode.

The form of the governing equations and the method of integration is the same as for unsteady RANS simulations. The turbulence model is modified so that it reverts to a sub-grid model (lower eddy viscosity) in regions where LES is employed and to a RANS model in regions where RANS is employed."

I hope this will clear many confusions.

Regards,

SJ

[LuckyTran]

More needs to be said on this topic rather than less. I see a general tendency of quoting things from sources without actually understanding what is going on.

When you do turbulence modeling, whether that be RANS or LES, it often comes down to finding an effective viscosity because linear strain models (i.e. eddy viscosity models) are the most popular form of of modeling. In RANS we call this the turbulent viscosity. In LES we call this the subgrid scale viscosity. It's just the label. In RANS the Reynolds stresses are proportional to a turbulent viscosity multiplied by the (Reynolds averaged) strain rate. In LES, the unresolved strain is obtained via the subgrid scale viscosity is multiplied by the (filtered) strain rate.

The governing equations for both RANS and commutative LES are the same. One can always modify (very little work!) a linear strain model to work in one framework and the other, because the governing equations are the same. All you need to add a simple if-else statement to the code to switch from a turbulent viscosity to a subgrid viscosity. One line of code.

But yes it can be very confusing if you have never derived and written down the Reynolds-averaged equations by hand yourself, I will give that allowance.

[Shamoon Jamshed]

LuckyTran,

So whats your point? Your reply does not offend the statement by Dr. Dimitris, i.e. S-A model is used as URANS and as SGS model in LES.

Regards,

SJ

[LuckyTran]

I don't offend it but I do not support it either.

S-A is not technically the sgs model. And you can easily appreciate this when you consider there is DES with other RANS models.

I can just as easily make statement that all CFD is just solving systems of linear equations. This would grossly oversimply what CFD is and still always be technically correct. Saying that one can modify an eddy viscosity model to work in DES is like saying someone can code. Subgrid models and eddy viscosity models are part of the same framework, all you do is put a hat on top of the variables RANS variables and it will look like DES/LES.

Say you have
Ax=b and at the same time WX=Y. Saying that I can apply the same linear solver to Ax=b as I do for WX=Y doesn't mean that Ax=b and WX=Y are the same thing. If I had a code I could certainly make this modification... But that doesn't mean that Ax=Y. Simply because I cannot disprove that water is wet doesn't mean I agree with you on any other issue.

Finally, just think about it. If S-A could be used as a SGS model then what is the difference between URANS with S-A and LES or DES? Do you think Smagorinsky was an idiot or something and people were running LES just to waste resources? Keep in mind that early weather data predictions were the most costly computations of their era (and they still are today) if we can disregard the calculations needed to make the atomic bomb. The answer of course is that certainly it can be used as a sgs model (because any eddy viscosity model can be used as a sgs model), but it would be a terrible one because it would give you the wrong result for the subgrid stress because it is not a subgrid stress model, it's an eddy viscosity model.