[Abhishek@123]

Hello all, I have done some simulations which involves turbulent air flow over a fin. I have used SST k-omega model for the simulation. I am getting a maximum Y+ value of 11 and an average Y+ value of 3.9. The simulations are performed in OpenFOAM 2112 software with wall functions. Can I rely on the result of this simulation?

[CFDfan]

It is a very good practical question I have as well and more specifically what percentage of the Y+ values have to below 1 (when employing the SST model) to have trustful results. I assume here that it can't always be 100% when the geometry is complex and the mesh size (i.e. RAM) is on the very high side, but still manageable.

Hopefully the CFD professionals in this forum will answer it.

[CFDfan]

It is a very good question I have as well and more specifically what percentage of the Y+ values have to below 1 (when employing the SST model) to have trustful results. I assume here that it can't always be 100% when the geometry is complex and the mesh size (i.e. RAM) is on the very high side, but still manageable.

Hopefully the CFD professionals in this forum will answer it.

[bigfootedrockmidget]

This is difficult to answer in general.


I presented some basic analysis on turbulence modeling with and without wall functions and the effect of y+ here:
Turbulence modeling with wall functions,

https://drive.google.com/file/d/16al...qaEu9QHDn/view
There is also a youtube video:

https://www.youtube.com/watch?v=x6PMfacKuK8


This work was done with the CFD code SU2, applied to SST and SA turbulence models, with and without wall functions.


You can see in the presentation that the prediction of basic quantities like skin friction deteriorates very quickly with increasing y+. y+ > 10 is basically garbage when you do not use wall functions. This is in the case that y+ > 10 everywhere. If it only happens in a few points it might not be so bad. It really depends where this occurs, but you should assume that errors generated in regions with large y+ are transported downstream with the flow.

[CFDfan]

Originally Posted by bigfootedrockmidget

This is difficult to answer in general.


I presented some basic analysis on turbulence modeling with and without wall functions and the effect of y+ here:
Turbulence modeling with wall functions,

https://drive.google.com/file/d/16al...qaEu9QHDn/view
There is also a youtube video:

https://www.youtube.com/watch?v=x6PMfacKuK8


This work was done with the CFD code SU2, applied to SST and SA turbulence models, with and without wall functions.


You can see in the presentation that the prediction of basic quantities like skin friction deteriorates very quickly with increasing y+. y+ > 10 is basically garbage when you do not use wall functions. This is in the case that y+ > 10 everywhere. If it only happens in a few points it might not be so bad. It really depends where this occurs, but you should assume that errors generated in regions with large y+ are transported downstream with the flow.

Thank you for your informative presentation. We established, that in real applications it is almost imposable to have all y+ values <1 unless one has an unbearably fine mesh. I noticed that I could "accommodate" say 85% of the y+ to be below 1 and the rest 15% would be >1 and spread out in an exponentially declining manner of up to say 10. The mesh size in such case would be reasonable and would require, say <200Gb or RAM.
In such cases it looks like the SST k-omg runs and converges OK, but I am never sure if the SST was the best approach, moreover, when using other turbulence models the results could be offset with 10-15%.
One could argue of course, that such fine meshes are not needed and we could employ raw mesh and use some of the well established, k-eps, wall function models dealing with y+ values of >30. The problem here is that complex geometries/flows very often fail to mesh successfully when raw meshes are employed. To overcome this issue one is simply forced to refine the mesh and would end up (naturally) with a large percentage of y+ values<1.
Another option is to use, DNS or LES models which are insensitive to the y+ values but the simulations take very long time to complete. In short, there is an uncertainty whatever you do.