CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   y+=1 vs Wall Function (http://www.cfd-online.com/Forums/fluent/109374-y-1-vs-wall-function.html)

paduchev November 16, 2012 03:34

y+=1 vs Wall Function
 
Does anybody knows how fluent chooses when to use wall function and when to model boundary layer directly when I'm using turbulenxe model K-W-SST?

Shamoon Jamshed November 16, 2012 06:38

There is no separate option to use wall f/n in kw SST case. Just keep near wall cells fine enough to get y+=1 in FLUENT.

LuckyTran November 16, 2012 13:57

Quote:

Originally Posted by paduchev (Post 392446)
Does anybody knows how fluent chooses when to use wall function and when to model boundary layer directly when I'm using turbulenxe model K-W-SST?

Are you asking about any equation in particular? Wall functions for velocity, the k-equation or w-equation? The velocity boundary condition uses the identical enhance wall treatment found on the realizable-k-epislon model if you are familiar with that. The k-equation uses the identical wall boundary condition from the k-epsilon model. There is a blended-two-layer wall function for the w equation as well (similar blending to get the enhanced wall treatment).

Also, a bit of a technicality, but there are always wall functions used to model the wall boundary condition. Hence, it's not good practice to say wall function and modelling the boundary layer as you are saying.

Quote:

Originally Posted by Shamoon Jamshed (Post 392483)
There is no separate option to use wall f/n in kw SST case. Just keep near wall cells fine enough to get y+=1 in FLUENT.

There is no option but it is there in the background. In other words, you don't have a choice but the wall functions are always there. You can use the kw SST model on a coarse grid and it will automatically "switch" to the log-law. There actually is no switch, the blended wall function is just a smoothed version of linear and log laws.

paduchev November 18, 2012 01:19

Quote:

Originally Posted by LuckyTran (Post 392591)
Also, a bit of a technicality, but there are always wall functions used to model the wall boundary condition. Hence, it's not good practice to say wall function and modelling the boundary layer as you are saying.

There is no option but it is there in the background. In other words, you don't have a choice but the wall functions are always there. You can use the kw SST model on a coarse grid and it will automatically "switch" to the log-law. There actually is no switch, the blended wall function is just a smoothed version of linear and log laws.

Thanks very much. Have I understood correctly - fall function is always using, when I'm use k-w SST, but if i have fine near-wall mesh, boundary layer will be modeled by logarithmic law and if i will have coarse mesh the smoothed version of logarithmic law (wall function) will be using? But why in fluents tutorial posted: (13.2.1.8.*Grid Resolution RANS Models) "However, for simulations with high accuracy demands on the wall boundary layer (especially for heat transfer predictions) near wall meshes with y+*~1 are recommended. When wall functions are used, it is essential to avoid meshes with * values lower than y+ ~30 as the wall shear stress and the wall heat transfer can and will seriously deteriorate under such conditions."
Does it means that when i'm using k-w-SST model i must avoid meshes with y+ from 2 to 30?
And the last. I have modeled experimental data - flow around conical nozzle and the "wall function mesh" y+ near 60 showed better match with test data, than mesh with y+ = 1

Far November 18, 2012 11:57

In older versions there was switch which changes the formulation from integration to wall (ITW) approach to wall function (WF) at Y+ (Rather Y*) = 11.06. Below 11.06 ITW was used and above 11.06 WF was used irrespective the fact that WF were not valid in the buffer zone (from 5-30) and similarly ITW was used irrespective the fact the viscous sub layer requires Y+ = 1 or at-least Y +< 5.

Now Fluent is using the hybrid wall functions (similar to the apporach in reference give below and technology comes from CFX due to F. Menter) which is combination of ITW and WF. Upto Y+ = 6, it uses the ITW apporach and above Y+ > 30 it uses the Wall function (yes standard wall functions) and in the region between Y + 6-30 it uses the combination of both.

Important points to remember:

1. Keep Y+ below 10 for viscous dominate flows with separation.
2. Use Y+ < 1 for the transition prediction and heat transfer calculations.
3. Use Y+ < 10 for SST model (omega based models in general)
4. Use K-Epsilon model with scalable wall functions (default option in new version of Fluent). It will keep the Y+ above 11.06 so that it does not suffer the limitation of old standard wall functions. Moreover now you can carry the successive refinment of mesh, which was not possible with standard WFs.

http://num.math.uni-goettingen.de/ba...ings/knopp.pdf

Far November 18, 2012 12:01

Matching the Experimental data is not the goal !
 
Quote:

And the last. I have modeled experimental data - flow around conical nozzle and the "wall function mesh" y+ near 60 showed better match with test data, than mesh with y+ = 1
First make the mesh refinement study. once you choose the fine enough mesh, then half the first cell height (in other words you are reducing Y+) and check the results. Do they change? or double the first cell height and compare the results. Again do they change?


Moreover CFD is not done to match the experimental data, it can be done in MS Excel :D. You must ensure the best practices of CFD and there are many source of errors e.g. turbulence modelling, grid, boundary conditions, flow scheme and it is also possible that some source of error is cancelling other factor and you are getting good results.

paduchev November 18, 2012 22:30

Quote:

Originally Posted by Far (Post 392798)
3. Use Y+ < 10 for SST model (omega based models in general)

It's turbulence models requirements or Fluent's?
Why I could not use k-omega SST model with wall-function mesh with y+>30?

LuckyTran November 18, 2012 22:32

Quote:

Originally Posted by paduchev (Post 392843)
It's turbulence models requirements or Fluent's?
Why I could not use k-omega SST model with wall-function mesh with y+>30?

You can use a larger mesh with y+ > 30 even with k-w SST.

Far November 19, 2012 02:57

Quote:

Originally Posted by paduchev (Post 392843)
It's turbulence models requirements or Fluent's?
Why I could not use k-omega SST model with wall-function mesh with y+>30?

strictly speaking, SST model is low Reynolds number model. Which means you need the Y+ ~ 1. But in commercial codes (as they know that it is not possible to get the Y+ 1 for every problem by every one), therefore they have implemented WF. So you can use the WF and it is automatically turned on when you have Y+ > 30 (Which is not always possible to ensure due to presence of separation, stagnation zone and your yplus can go to zero!) .

You must also understand that in the log region, you are solving the transformed K-Epsilon model and it is good idea to use the original K-epsilon (which is designed for WFs) in combination with scalable wall functions for the reasons mentioned above and in previous post.

In my limited experience K-epsilon model with scalable wall function is better choice than the SST model with hybrid functions for the Y+> 30.

paduchev November 19, 2012 03:15

Quote:

Originally Posted by Far (Post 392871)
strictly speaking, SST model is low Reynolds number model. Which means you need the Y+ ~ 1. But in commercial codes (as they know that it is not possible to get the Y+ 1 for every problem by every one), therefore they have implemented WF. So you can use the WF and it is automatically turned on when you have Y+ > 30 (Which is not always possible to ensure due to presence of separation, stagnation zone and your yplus can go to zero!) .

You must also understand that in the log region, you are solving the transformed K-Epsilon model and it is good idea to use the original K-epsilon (which is designed for WFs) in combination with scalable wall functions for the reasons mentioned above and in previous post.

In my limited experience K-epsilon model with scalable wall function is better choice than the SST model with hybrid functions for the Y+> 30.


Thanks a lot, I have writed you a privat message, could you answer on it?)

LuckyTran November 19, 2012 03:19

Quote:

Originally Posted by Far (Post 392871)
strictly speaking, SST model is low Reynolds number model. Which means you need the Y+ ~ 1.

I think this is mis-leading. Strictly speaking, none of the commonly available turbulence models are restricted to any Re. The SST formulation is already a hybrid blending of a the k-epsilon and k-omega models and is not restricted to low or high Reynolds number grids. What is restricted is the wall modelling approach for the wall boundary conditions which can be high or low Re and still be consistent with the turbulence model being used.

If one desired (and it seems paduchev definitely wants to) use a grid with y+ of wall adjacent cell to be >30, one can certainly do so. Will a low Re number grid do better? Probably. But I don't think any of the previous explanations were of any help.

Far December 1, 2012 10:08

1 Attachment(s)
Please look at the document by Lars for details of SST model http://www.os-cfd.ru/cfd_docs/chalme...5ebe3f5385c13b

And don't confuse the hybrid wall treatment with the hybrid turbulence model like SST model which is combination of K-omega in inner and buffer layer. And K-epsilon model in outer and fully turbulent region.

The hybrid wall treatment is used to relax the Y+ requirements on the K-omega model in SST model. Which requires Y+<2

Moreover SST model was made by mentor in1994 and work on enhanced wall treatment (automatic wall treatment) was published in 2002-2003. Which means SST model was there and without hybrid wall functions. Now question arises what type of wall treatment was being used before the intorudction of hybrid wall functions. The answer is it is same as required by K-omega model. This is also clearly mentioned in CFX theory guide where this model was implemented first by mentor.

http://www.kxcad.net/ansys/ansys_cfx.../i1346068.html

Quote:

Automatic Near-Wall Treatment for Omega-Based Models

Automatic near-wall treatment automatically switches from wall-functions to a low-Re near wall formulation as the mesh is refined.

One of the well known deficiencies of the model is its inability to handle low turbulent Reynolds number computations. Complex damping functions can be added to the model, as well as the requirement of highly refined near-wall grid resolution (y+ < 0.2) in an attempt to model low turbulent Reynolds number flows. This approach often leads to numerical instability. Some of these difficulties may be avoided by using the model, making it more appropriate than the model for flows requiring high near-wall resolution (e.g., high wall heat transfer, transition). However, a strict low-Reynolds number implementation of the model would also require a near wall grid resolution of at least y+ < 2. This condition cannot be guaranteed in most applications at all walls. For this reason, a new near wall treatment was developed by ANSYS CFX for the based models that allows for a smooth shift from a low-Reynolds number form to a wall function formulation. This near wall boundary condition, named automatic near wall treatment in ANSYS CFX, is used as the default in all models based on the -equation (standard , Baseline , SST, -Reynolds Stress).
I am also attaching the some notes from Wilcox book on turbulence modelling.

LuckyTran December 2, 2012 05:34

Quote:

Originally Posted by Far (Post 395158)
Please look at the document by Lars for details of SST model http://www.os-cfd.ru/cfd_docs/chalme...5ebe3f5385c13b

And don't confuse the hybrid wall treatment with the hybrid turbulence model like SST model which is combination of K-omega in inner and buffer layer. And K-epsilon model in outer and fully turbulent region.

Exactly. The SST approach for high-Re grid is equivalent to a standard k-epsilon model. Which means that the requirements should be the same as the standard k-epsilon model and not the k-omega model (for a high-Re grid). In the low-Re grid case I agree, the requirements for k-w SST should be treated as same as k-omega.

macfly December 2, 2012 12:05

Here's the complete and possibly fresher version of Lars Davidson's document that Far is referring to in post #12, very nice : http://www.tfd.chalmers.se/~lada/pos...-modelling.pdf


All times are GMT -4. The time now is 20:27.