inflation layer
 

Hi all.

I am modeling a free surface flow. In order to correct predict the boundary layer I have put some inflation layers in the bottom zone. In order to "predict" the minimum required space between the nodes in the bottom of my mesh I have used the following expression:

delta y= L (length of the domain)*y+*(80)^0.5 *Re^(-13/14). I have set the value of y+ to 30 and then computed the value of delta y. I am using k-epsilon turbulence model.Is this the correct procedure?

Any suggestion will be very helpful.Regards

The k-e model is best suited to wall functions, so you won't be modelling the full boundary layer. If your mesh is around y+=30 then this is probably a valid approach. If you want to model the full boundary layer you should consider some other turbulence models which integrate to the wall, SST is probably the best bet here. You will need to go to about y+=1 to get the full boundary layer this way, and that may be excessive for what you want to do. Up to you.

I seem to recall reading that modeling the boundary layer with the k-e model requires a resolution of y+ < 0.2, which is ridiculous.

Thanks a lot for your comments Mr.Glenn and Mr.Joshua.

If possible I would also like to clarify an additional aspect. As I have already said in order to predict the minimum required space between the nodes in the bottom I have defined y+ equal to 30 and then computed by hand the value of delta y. Then I have defined five inflation layers with this length (in my mesh).

My solution converges (imbalance=0.01 and residuals=10^-5). However in cfd post when I plot the yplus variable in the bottom wall it says that min value is 0 and the max value is 3.33. What can I conclude? Simply that I have an over refined mesh in the bottom zone correct?Typically it should be something like 20<y+<100 right?

Hi Antonio,

This would be a valid solution.

CFX will not have used the mesh close to the wall as it uses a wall function approach. You could try this same mesh with the Shear Stress Transport model which blends from k-e in the freestream to k-w at the walls.

You need to remember that the equation that you used t approximate Y+ is for flow along a flat plate and that it is just a starting point for your y+ study. As y+ varies with velocity you will likely get a range of Y+ on any surface.

As long as this is below 100 with k-e then you are ok, but if you want to predict separation points or recirculation then you need to be very close to 1.

Hope this helps a bit,

Scott

k-e cannot be integrated to the wall. That's why a y+>11 is the target for k-e.

Sorry, but this is not how it works. CFX does use the mesh close to the wall but it applies wall function boundary conditions to it even though they are not appropriate. That is because with a k-e model there is no appropriate model, so you are stuck with wall functions. So if you use y+=3 with a k-e model your sub-layer will be totally wrong and the drag predicted is probably wrong too.

If you want to integrate to the wall this is the approach to use. But the discussion here is about wall functions - the SST model has the advantage of automatic wall functions which automatically switch between integrating to the wall and wall functions depending on the local y+.

The choice of turbulence model is more important than the y+ for separations and recirculations. SST does a much better job than k-e in most cases, even with coarser meshes. SST also improves with mesh refinement which k-e does not do so well.

Hi Glenn,

I was trying to say that using a y+ of 3 for a k-e mesh is a bit of a waste as k-e will not resolve all of the way to the wall. Your comments are correct, of course it will calculate at the additional cells, but this may lead to erroneous solutions.

As far as my comments for using SST. I am implying that the current mesh, having a y+~1 would be better suited to SST. If the user does not want the additional cells, then they could remesh with a y+~30 and run with k-e. My suggestion was to not spend time remeshing, but to just switch to SST and rerun which is likely to be more accurate for wall bounded flows anyway.

My comments regarding a Y+ of 1 to capture separation were specific to someone already using the SST model. As you said, the solution will gain accuracy with increased mesh resolution and for accurate separation prediction it is bette to resolve the boundary layer to the wall than rely on wall functions.

Hope this clarifies things! We have the same position on this, you just got the point across better! :)

Cheers,

Scott