[huangxianbei]

Hi,all:
As I read the help document, I find the LES near wall treatment is a simple method, which depends on the y+ value, that means, when y+ is small, u+=y+, when y+ is large, law-of -wall is used. Also, there is a blending when y+ is in the buffer region. This kind of approach is automatically used when doing LES in Fluent, however, I don't think it's appropriate since if the y+ is small, the wall should be fully resolved instead of using wall treatment. But things are out of my expectation that the results from Fluent is better than the case in which I turn off the wall function in OpenFoam near the wall. The y+ is aroud 1, I think the mesh is fairy enough. I don't know why the truly resolved solution can be worse than that in Fluent.
Any comment is appreciated. If my understanding on the LES near wall treatment of Fluent is wrong, please correct me without hesitation.

Xianbei

[sbaffini]

The Fluent formulation for the wall-treatment in LES is actually correct. Whenever the near wall resolution is within y+ = 5 it reverts to a classical no-slip boundary condition (that is, it uses interpolation from the linear profile, which is exactly equivalent).

Besides this, there is no reason to expect OF to perform better than Fluent.

[huangxianbei]

Originally Posted by sbaffini

The Fluent formulation for the wall-treatment in LES is actually correct. Whenever the near wall resolution is within y+ = 5 it reverts to a classical no-slip boundary condition (that is, it uses interpolation from the linear profile, which is exactly equivalent).

Besides this, there is no reason to expect OF to perform better than Fluent.

Hi,Paolo:
Thank you for your reply. I only find what I said before in the Fluent help manual, could you please point out where it is as you explained in the help manual? Or simply copy this part to the forum?
In my opinion, the wall treatment is needed only when the mesh resolution
is not enough, why do we need wall treatment when y+ is already so small?
I obtained similar results to OF's using CFX,in CFX, the wall function is automaic wall function without any choice. It's some kind of wall function to be adaptive to omega based model. Do you think in this way, Fluent is more accurate than the OF and CFX?

Xianbei

[sbaffini]

It is actually the same part cited by you. Indeed, using a wall function of the form:

u+=y+

actually means not using the wall function at all. Fluent uses this automatically when y+ is low (this is just a consequence of the Kader interpolation used by Fluent to blend the inner and outer zones).

I don't know how OF or CFX work but i don't expect them to be more accurate.

In conclusion, the part you read from the manual actually tells you that Fluent uses an automatic blending of the wall-function with the inner profile whenever it is required by the resolution.

To convince yourself, try plotting the wall function used by fluent.

[huangxianbei]

Originally Posted by sbaffini

It is actually the same part cited by you. Indeed, using a wall function of the form:

u+=y+

actually means not using the wall function at all. Fluent uses this automatically when y+ is low (this is just a consequence of the Kader interpolation used by Fluent to blend the inner and outer zones).

I don't know how OF or CFX work but i don't expect them to be more accurate.

In conclusion, the part you read from the manual actually tells you that Fluent uses an automatic blending of the wall-function with the inner profile whenever it is required by the resolution.

To convince yourself, try plotting the wall function used by fluent.

Thank you for your reply. I see, however, Isn't u+=y+ only a asymptotic relation in simple flows such as channel flow or shear flow? This is an artificial relation applied to the calculation, why not calculate just by the LES model?

[sbaffini]

Originally Posted by huangxianbei

Thank you for your reply. I see, however, Isn't u+=y+ only a asymptotic relation in simple flows such as channel flow or shear flow? This is an artificial relation applied to the calculation, why not calculate just by the LES model?

That asymptotic relation is the one you use to compute the stresses in well resolved simulations (at least for all the unstructured FV codes):

tau_wall = mu * (u_c0 - u_wall)/(y_c0-y_wall)

Besides this, there is however a huge conceptual flaw in mixing a model for steady turbulent flows (the classical wall function) with LES. In practice it works because the average stress predicted by the wall function is still much better than the one predictable in most cases by a not well resolved LES

[huangxianbei]

Originally Posted by sbaffini

That asymptotic relation is the one you use to compute the stresses in well resolved simulations (at least for all the unstructured FV codes):

tau_wall = mu * (u_c0 - u_wall)/(y_c0-y_wall)

Besides this, there is however a huge conceptual flaw in mixing a model for steady turbulent flows (the classical wall function) with LES. In practice it works because the average stress predicted by the wall function is still much better than the one predictable in most cases by a not well resolved LES

Thank you for providing so much information. Yes, in practice, it works well as in my case. I have a question, if I want to evaluate the LES models, this kind of artificial effect may eliminate the impacts come from the LES model itself. In this way, should I still apply the treatment?

[sbaffini]

The consequence of the discussed effect is that If your mesh is fine enough, it is like the wall function is not applied at all. Thus i don't expect its effect to cover those of the SGS model. Still, the SGS model in a low order code might still have a negligible effect independently from the wall function.

In practice, in Fluent you can't deactivate the wall function. So just be sure that the grid has y+ below 5 and you are ok.

[huangxianbei]

Originally Posted by sbaffini

The consequence of the discussed effect is that If your mesh is fine enough, it is like the wall function is not applied at all. Thus i don't expect its effect to cover those of the SGS model. Still, the SGS model in a low order code might still have a negligible effect independently from the wall function.

In practice, in Fluent you can't deactivate the wall function. So just be sure that the grid has y+ below 5 and you are ok.

The mesh is fairy good enough with a average y+<1 and also x+ and z+ in a relative low value. However, I'd prefer Openfoam to study the case for the easier way to modify or implement a new model. In Openfoam, usually second-order scheme is used so the precision can be guaranteed. While just as stated before, the velocity distribution obtained in the two codes varies near the wall, I can tell nothing except the near-wall treatment which causes this deviation. So that's why I'd like to discuss with you. I tried the schemes used in Fluent and didn't got similar result as in Fluent.

So in conclusion, it's the near-wall treatment that causes the deviation. I'm not sure which code is better in this simulation,for both codes contain a certain degree of deviation from the PIV data. The fact is that the velocity curve stands between the Fluent and OF results. The reason why I think the Fluent data near wall is more acceptable is that the same case calculated from another code is similar to the Fluent's(Published article).