[maysmech]

Dear Foamers,
Happy new year.

Please tell me which wall function i should use to define wall B.C when my y+ is:

Patch 4 named wall y+ : min: 0.0516154 max: 2.20514 average: 0.344511

now, i use nutWallFunction for nut
kqRWallFunction for k
epsilonWallFunction for epsilon
but i think they dont give true answers.

Thanks in advance,

[stevenvanharen]

mmmm....I would use no wall functions, your yplus values are too low to be in the log layer:

http://www.cfd-online.com/Wiki/Wall_functions

[FelixL]

Hello, maysam,


I suppose you are using one of the k-epsilon based turbulence models? I can only guess but if you're not already using the Low-Reynolds corrected versions of the k-epsilon model you'll obtain poor results when integrating these turbulence models down to the viscous sublayer (which is clearly the case looking at your y+-values). See D.C. Wilcox's book "Turbulence Modeling for CFD" as a reference.

If your actually using the Low-Reynolds models then you shouldn't use wall functions, like Steven said in the post above. These models are supposed to work with more stringent boundary conditions (see http://www.cfd-online.com/Wiki/Low-Re_k-epsilon_models as a reference).


Greetings,
Felix

[maysmech]

Originally Posted by FelixL

Hello, maysam,


I suppose you are using one of the k-epsilon based turbulence models? I can only guess but if you're not already using the Low-Reynolds corrected versions of the k-epsilon model you'll obtain poor results when integrating these turbulence models down to the viscous sublayer (which is clearly the case looking at your y+-values). See D.C. Wilcox's book "Turbulence Modeling for CFD" as a reference.

If your actually using the Low-Reynolds models then you shouldn't use wall functions, like Steven said in the post above. These models are supposed to work with more stringent boundary conditions (see http://www.cfd-online.com/Wiki/Low-Re_k-epsilon_models as a reference).


Greetings,
Felix

Thanks for your answers.
the case (hydrocyclone) is not a low Re problem and i want run it with k-e, LRR and LES.
i think its low y+ is result of is small mesh at near wall.
now, what is your suggestion about Wall boundary field settings for k, epsilon and nut?

Best,
Maysam

[maysmech]

you mean i should set zseroGradient B.C for K, epsilon and nut for wall ?

wall
{
type zeroGradient;
}

Is it a suitable B.C?

[FelixL]

Hello, Maysam,


I guess the term "Low-Reynolds model" is a bit misleading. These models are actually suitable for any Reynolds number (as long as the flow is turbulent, of course). These Low-Reynolds k-epsilon-models are standard k-epsilon models with additional viscous damping functions to improve the models' results for near-wall regions. They're called "Low-Reynolds", because these models require an near wall y+ of the order of 1 (as in your case) and this is usually too costly for engineering applications with high reynolds numbers.

But if you insist in using the standard (High-Re) k-epsilon model for your case without using log-wallfunctions (i.e. y+>30), these would be suitable BCs, according to my knowledge:

nut: fixedValue 0; OR nutWallFunction; (this WF is continuous!)
k and epsilon: fixedValue 1e-10; (has to be nonzero to avoid division by zero errors!)

Let me repeat: The results will probably be inaccurate, especially the wall shear stress and heat transfer values. And this is not a BC problem, this is a problem of the k-epsilon model itself. Please refer to Wilcox' book regarding that issue.


Greetings,
Felix.

[maysmech]

Thanks,

What is your suggestion for high y+.

I changed my mesh to a coarse one and run it with k-e. y+ is near 70.

[vkrastev]

Originally Posted by maysmech

Thanks,

What is your suggestion for high y+.

I changed my mesh to a coarse one and run it with k-e. y+ is near 70.

This value of y+ at the near wall cells is suitable for the standard wall functions (kqR, epsilon and nutWallFunction), as the first nodes near the wall are supposed to be inside the log-layer. For what concerns the initial near-wall values of k, epsilon and nut, you can have a look at the pitzDaily RAS tutorial inside the Programmer's Guide and see how to set them properly (but, however, keep in mind that for practical appications with the standard k-epsilon model these initial values can be considered quite arbitrary)

Regards

V.

[subhkirti]

Hi,

I am trying to simulate the turbulent flat plate flow. I am using a yPlusRAS utility which gives me the following

Patch 3 named plate y+ : min: 63.4497 max: 90.5452 average: 73.1344

I am using kqRWallFunction for k, omegaWallFunction for omega, and nutWallFuction for nut. However, I am not getting the expected results, eg. the k value is not zero at the plate. I am not sure whether I am correctly putting up the BC. Can you post me your BC file, so that i can cross check if i am setting up the case correctly?

Originally Posted by maysmech

Dear Foamers,
Happy new year.

Please tell me which wall function i should use to define wall B.C when my y+ is:

Patch 4 named wall y+ : min: 0.0516154 max: 2.20514 average: 0.344511

now, i use nutWallFunction for nut
kqRWallFunction for k
epsilonWallFunction for epsilon
but i think they dont give true answers.

Thanks in advance,

[vkrastev]

Originally Posted by subhkirti

Hi,

I am trying to simulate the turbulent flat plate flow. I am using a yPlusRAS utility which gives me the following

Patch 3 named plate y+ : min: 63.4497 max: 90.5452 average: 73.1344

I am using kqRWallFunction for k, omegaWallFunction for omega, and nutWallFuction for nut. However, I am not getting the expected results, eg. the k value is not zero at the plate. I am not sure whether I am correctly putting up the BC. Can you post me your BC file, so that i can cross check if i am setting up the case correctly?

Your y+ is good for the kind of wall functions you are using. Don't be surprised about the wrong k value at the wall: the wall functions in question are designed to give the correct value of the turbulent quantities at the first node away from the wall, not at the wall itself (for instance, the wall function for k simply applies a zeroGradient boundary condition, which means that the value at the wall should be equal to the value at the firs near-wall node thus, in general, not equal to zero)

Hope this helps

V.

[subhkirti]

Hi,
Thanks for your reply. I compared my results to the different posts in the forum, and they are coming out quite close. The problem that now I am having is that when i decrease the y+ by increasing the cell grading, my solution does not converge. In my case, as y+ approaches 30, the residuals blow up. Is there something I'm missing in my solution.

Originally Posted by vkrastev

Your y+ is good for the kind of wall functions you are using. Don't be surprised about the wrong k value at the wall: the wall functions in question are designed to give the correct value of the turbulent quantities at the first node away from the wall, not at the wall itself (for instance, the wall function for k simply applies a zeroGradient boundary condition, which means that the value at the wall should be equal to the value at the firs near-wall node thus, in general, not equal to zero)

Hope this helps

V.

[vkrastev]

Originally Posted by subhkirti

Hi,
Thanks for your reply. I compared my results to the different posts in the forum, and they are coming out quite close. The problem that now I am having is that when i decrease the y+ by increasing the cell grading, my solution does not converge. In my case, as y+ approaches 30, the residuals blow up. Is there something I'm missing in my solution.

About the near-wall modeling, as your y+ goes below 30 you are entering into the buffer layer, so I would expect less accurate results (the standard wall functions do not take into account for the buffer region), but not significant numerical troubles...I need some more details about your numerical setup (BC's, fvSchemes and fvSolution dictionaries, etc.)

Regards

V.

[subhkirti]

Hello V,

Thanks yet again. I am attaching the three files for your review (blockMesh, system files, and BC). These are the files for the case where I get the non-converged solutions. yPlus utility gives me the avg. value of y+ as 43.62 and the min y+ as 30.643.

Originally Posted by vkrastev

About the near-wall modeling, as your y+ goes below 30 you are entering into the buffer layer, so I would expect less accurate results (the standard wall functions do not take into account for the buffer region), but not significant numerical troubles...I need some more details about your numerical setup (BC's, fvSchemes and fvSolution dictionaries, etc.)

Regards

V.

[vkrastev]

Originally Posted by subhkirti

Hello V,

Thanks yet again. I am attaching the three files for your review (blockMesh, system files, and BC). These are the files for the case where I get the non-converged solutions. yPlus utility gives me the avg. value of y+ as 43.62 and the min y+ as 30.643.

1) Set the tolerances in the fvSolution from 1e-05/1e-06 to 1e-11
2) Assign an initial value for k and omega at the plate (as you have done with nut, but this time using the same values of the internal field initialization)
3) Underrelax omega and put the underrelaxation factors for k and omega to 0.5 (instead of 0.7)

I'm quite sure that the main problems are in point 1 and 3. Hope this helps

V.

[romant]

Originally Posted by FelixL

Hello, Maysam,


I guess the term "Low-Reynolds model" is a bit misleading. These models are actually suitable for any Reynolds number (as long as the flow is turbulent, of course). These Low-Reynolds k-epsilon-models are standard k-epsilon models with additional viscous damping functions to improve the models' results for near-wall regions. They're called "Low-Reynolds", because these models require an near wall y+ of the order of 1 (as in your case) and this is usually too costly for engineering applications with high reynolds numbers.

But if you insist in using the standard (High-Re) k-epsilon model for your case without using log-wallfunctions (i.e. y+>30), these would be suitable BCs, according to my knowledge:

nut: fixedValue 0; OR nutWallFunction; (this WF is continuous!)
k and epsilon: fixedValue 1e-10; (has to be nonzero to avoid division by zero errors!)

Let me repeat: The results will probably be inaccurate, especially the wall shear stress and heat transfer values. And this is not a BC problem, this is a problem of the k-epsilon model itself. Please refer to Wilcox' book regarding that issue.


Greetings,
Felix.

So in order to use a low-Re k-epsilon model, one could use those boundary conditions, if I understand this right?