In No case are y+ values between 3 and 30 desirable.

Is this true even when using Wall Functions?

I am using the Automatic Hex-Dominant Mesh with layers combined with the k-omega SST model and wall functions. As I increase the Mesh Fineness, the model converges on the theoretical values for pressure loss in a pipe with a bend. But at this point, the y+ output says it is between 3-6. Is this a problem?

So what it is saying is if you are modelling as wall functions we want to start our layers at 30 < y+ < 200 as below 30 we model opposed to solving. Other wise we could solve the entire boundary layer, which we ideally want to solve from y+ < 3 (recommended to be even lower). The statement in question is kind of given from the other statements. So in short, my understanding is that by having a y+ of between 3 and 6 we risk placing the model in the wrong part of the boundary layer.

In my experience we can never get all surfaces to have the desired y+ value exactly, however, we can work towards getting the majority within the recommended bounds.

This is my understanding of it, please correct me if I’m wrong.
Kind regards,
Darren

So this what I wanted to understand better about the model used in SimScale. Implementation of the k-omega SST model may not always the same. Some models claim to be y+ independent. So is the k-omega SST model in SimScale y+ independent as well?

In my experience we can never get all surfaces to have the desired y+ value exactly, however, we can work towards getting the majority within the recommended bounds.

This is also a concern of mine as well because it would make meshing a complex geometry somewhat of a nightmare.

BTW, after meshing is there anyway to measure the first layer mesh distance y or any other mesh separation?

My understanding is that if it were independent we wouldn’t have the requirements that are laid out in the docs, but there was a very good discussion on turbulence. I don’t know if it is useful to you:

As for measuring y after meshing, we could look at evaluating the normals of the surface in paraview but from there I think it would become a lot of playing around to find the distance to the adjacent cells opposite face. But I reckon this is your best bet. It must be possible but might require some experimenting.

I tend you just rattle a simulation off for a minimal number of iterations and look at the results. In an ideal world, it would be great to have an adaptive layer meshing algorithm that could just keep adding layers until y+ is at desired value. Then we wouldn’t even be having this discussion

But check out that link, if nothing else you will see the community members that have much more knowledge about the turbulence models than me and would be a great place to ask any further questions and keep all the turbulence model related questions in the same place to increase it as a resource.

Thanks. I have seen that the discussion but did not read it in it’s entirety the first time although now I have. I have a few questions on a couple of the comments:

I think that the hardest to grasp aspect of turbulence modeling in SimScale is the use of automatic wall functions. I would like to focus on two turbulence models: the RANS k-omega SST and the DES of Spalart-Allmaras. In both it seems that automatic wall functions have been extensively used to promote an almost monotonic error decrease as the meshes are refined near the walls, bringing the adjacent to wall cell centers from y+ > 30 (inertial sublayer where the standard wall function would work) to the y+ < 1 range (laminar sublayer, where these turbulence models could be used without any wall function)

This would explain why the simulation I did with y+<6 seemed to converge and give a fairly accurate answer. Also, the following comment gives a y+<5 as a limit for k-omega which would confirms the results as well. It would appear that the k-Omega SST is capable of automatically determining where and if the wall function should be used.

The choice of y1+ and turbulence model must be made in unison. In general, a y1+ > 30 would be employed with k-epsilon models (with wall-functions calculating the flow for y1+ < 30) whereas y1+ < 5 would be employed with k-omega models.