[SuperSonicCat]

Hello forum,


very often i read the following statement: If a wall function is used, the first wall adjacent node should be in range 30 <y+<300.
I understand that in that range the logarithmic law of the wall holds. But what if the wall functions (e.g. in Ansys, LS-Dyna ICFD) are implemented such that for y+ < 11.225 linear law of the wall (u+=y+) is applied? Am i allowed to place the first node in the viscous sublayer and will i get good results?
If its not recommended to do so, why is the switching between linear and log- layer is implemented in above mentioned solvers?


Best regeards : )

[FMDenaro]

Originally Posted by SuperSonicCat

Hello forum,


very often i read the following statement: If a wall function is used, the first wall adjacent node should be in range 30 <y+<300.
I understand that in that range the logarithmic law of the wall holds. But what if the wall functions (e.g. in Ansys, LS-Dyna ICFD) are implemented such that for y+ < 11.225 linear law of the wall (u+=y+) is applied? Am i allowed to place the first node in the viscous sublayer and will i get good results?
If its not recommended to do so, why is the switching between linear and log- layer is implemented in above mentioned solvers?


Best regeards : )

I think that the question can be reversed... If you can use a grid resolution fine until to resolve the viscous sub-layer, why do you want to use wall-modelled BCs?

[SuperSonicCat]

Originally Posted by FMDenaro

I think that the question can be reversed... If you can use a grid resolution fine until to resolve the viscous sub-layer, why do you want to use wall-modelled BCs?

Unfortunately, i have to use wall functions in LS-Dyna ICFD.

[sbaffini]

A wall function is always made of two pieces, the velocity (and other scalars) part and the turbulence model part. If a turbulence model can't be solved to low y+ then even having a linear log law won't help by itself.

But let's assume that the turbulence model can be solved to low y+ (Spalart Allmaras and k-omega models can). In this case you will have a single, approximate, near wall behaviour of the wall function to be matched with typically different behaviors of the different models. While you can adapt the constants of the law to each model, the fit will never be perfect (log linear is also very crude indeed). In this case you are probably safe to either work in full log or full linear, but don't expect great results in the buffer region.

There are exceptions. For example, the SA model allows for an exact wall function solution (which can be made the Musker profile with a simple modification to the model and also extended to other scalars).

So, in the end, it mostly depends from the model you are using, how is your software coded, what effort was put in it for certain cases, etc. But having a linear part, per se, doesn't solve that much.

[SuperSonicCat]

Thank you for the answer.
Can the k-omega-model always be used for low y+ ? In the book of Wilcox "Turbulence modelling for cfd" i have read that there exist two versions of that model: Low- and High- Re- models.

[sbaffini]

In practice, yes, it can be used both with and without that modification. But, as I said, the way a model is implemented (thus, the code) is key.