[Ali Blues]

Hi,
I'm trying to mesh a cylindrical tank equipped with baffles and a Rushton impeller using snappyHexMesh. Since I'm going to apply a turbulence model with wall functions, I cannot refine the mesh to arbitrary fine levels, i.e. the center of the first cell next to boundary needs to be placed at a certain y+. For my case, setting y_{p} = 6 mm will result in y+~ 30, and it corresponds to 2 levels of refinement relative to the background mesh from blockMesh (level 0) . However, if I try to limit the refinement to ~ y_{p}, I get a completely distorted discrete representation of the impeller as I've indicated in the attached image.


I would be able to properly capture the complete impeller but then I have to refine the mesh to unacceptable levels, even with the SurfaceFeatureExtact utility (the white edges in the image are those extracted from the utility). I've tried playing around with some of the parameters such as tolerance, nSolveIter, nRelaxIter, but the outcome is the same.


I've seen a lot of sHM examples and in all, these parameters don't seem to vary that much from case to case. Specially when I'm not using any inflation layers. So I would greatly appreciate suggestions on how to resolve this issue. I've attached the case here: ( http://we.tl/x7iw3jXQhJ )



Thanks,
Ali

[Tobi]

Dear Ali,

I still have some time before my flight start (: . I want to give some hints and experience:

• If you will rotate the propeller you will have y+ different at each face due to the complex flow, so in my opinion it is a hard topic to get y+ to ~ 30 everywhere; imagine a flow around a cylinder - it is the same;

• I am not a specialist in turbulence modeling and wall refinement but as I understand it correctly, you have to choose y+ in a way that either you have y+ ~ 1 and a lot of cells that you resolve the viscouse sub-layer or you choose y+ that the first cell is in the log-law; therefore y+ has to be > 30. If you calculate the transition between log-law and viscouse sub-layer we get something like y+ ~ 26 (I can not remember but in Fertziger & Peric you will find the value). Important is that the first cell center is within the log-law. Hence, it does not matter if 30, 40, 50 if you do not have too much interaction near the wall (like heat-transfere, or mass transfere).

• For meshing it is clear that you will end up with such a mesh if you will not refine the mesh enough. Starting from a pure cube background mesh, think about the position of the parts that have to be meshed. The parts are not aligned with the mesh and therefore some cells have to be deformed in a way that the mesh quality settings can not be fullfilled. At last refinement has to be done, otherwise you should use some tet-mesh (salome etc.), or you have to generate a complex backgroundmesh (e.g. different blocks that have some alignment like the single parts; this is not easy to handle)

• Additionally, if you will change the speed of the impeller, y+ would change. At last you will not get a mesh with perfect y+ everywhere.

Finally, remember that I am not sooo familiar with turbulence stuff, maybe I am wrong (:
Good luck,
Tobi

[nimasam]

Dear Ali

if you are going to use wall function you should be worried about y+ , however it seems you can resolve all boundary layers using the fine grids , so i suggest use y+ around ~1 and not use wallfunction

[Ali Blues]

Thanks a lot guys for the input.
I still have some questions which I would greatly appreciate it if you would help me sort out.

@ Tobias (well I hope to hear from you once you're back from your trip):
Indeed, given the complexity of the velocity field, we cannot expect only a certain value of y+ for the same wall normal distance from the wall. Following the discussion in
http://www.dicat.unige.it/guerrero/o...turbulence.pdf

, I was planning to run the simulation for a few iterations to get an estimate of the actual y+ value at various patches, and then make modifications to the grid near the interface accordingly.

Q1) In my initial post, I didn’t intend to mean that I want to strict my y+ to only certain value but just y+ more or less equal or greater than 30 so that using the wall functions make sense. In the presentation the refineWallLayer utility is suggested. I was wondering for using this utility the current mesh already needs to have inflation layer or not. Also this utility logically seems to work when aiming for lowering values of y+ compared to the current mesh. But does it also allow for “un-refinement” in areas where y+ is too low.

Q2) I think it would be best to discuss this on a separate post, but already curious to know if it is possible to do adaptive mesh refinement/unrefinement based on y+ which seems to be a different mentality than the refineWallLayer perhaps.

Q3) Same thing here regarding use of salome. There is actually a nice webinar on generating a hybrid mesh of a similar case using pointwise software. So then the question is whether Salome is capable of meshing in a similar manner? perhaps yes, but how is the question, as there are barely any step-by step tutorial using salome for meshing complex geometries using a hybrid mesh.

@ Nima:
Q1)So indeed one approach would be to just use a mesh with y+ ~ 1 and then not use wall functions. But then if I'm not wrong, I have to use a low-Re turbulence model. I don't have experience using low-Re models and but from what I saw here http://www.dicat.unige.it/guerrero/oftraining/8_1OF_advanced_modeling_turbulence.pdf , sufficient number of inflation layers are required to capture the viscous sublayer. So it seems to be a relatively expensive approach compared to the high-Re?

Q2) In pg 52 of the presentation, it is mentioned that if 1<y+<300 then it seems I can get away with high-Re number turbulence model, and still be able to get a good estimate of some "wall related" parameters such as the required power input which is sum of the pressure and viscous moments on the impeller. Though I'm limited to using k-omega SST model. What do you think of this approach?

Q3) Having said that, based on your experience, which of the low-Re turbulence models do you suggest, they're just too many of them


Thanks,
Ali