SimScale CAE Forum

Very thin boundary layers and y+ on complex geometry


#123

Ah, now I understand your question. To my eye, I don’t see any cells in which the intersection of two cells actually defines a nodal point for the adjacent face. More simply, it doesn’t look like having two cells along a face actually splits the adjacent face, so all hexahedrons are still hexahedrons.

EDIT
And then I found a picture that invalidates the above, so I’m out. Regardless it still looks like shared faces = adjacency.


#124

Hey guys!

The only thing I can say about how the algorithm works is:

"Take an internal face, compute the volume of the cell on either side and divide the bigger by the smaller".

High volume ratio often correlates with a high aspect ratio. OpenFOAM will call everything below 100 “good” (checkMesh reports the inverse number, though, so it’ll tell you that everything > 0.01 is “good”).

Cheers!

Jousef


#125

Yes, but the lowest face of that ‘Cell of interest’ has 4 cells adjacent to it…

So, @jousefm, does that mean that lower face can have 4 volumeRatios (in this case that ratio is 0.125 or 8 (depending of which way you look at it) for those 4 volumeRatios)?


#126

In that pic, I do not think a lot of those dark outline faces are actually planar, I think the hashing shows planarity (if that is a word even :slight_smile: )


#127

I think that one cell of interest(COI) can have different number of volume ratio like- a cell on the outside will have different number of surrounding cells to a cell which lies inside. Also, the mesh contain different type of elements(hexahedral, polyhedral, prism cells) so this will also decide the number of volume ratios we get for the COI. I have calculated volume ratio for a particular cell and would like to share it with you guyzz.


Volume%20ratio%20calculation

Now, we have obtained 4 volume ratios for this cell and I don’t know what mathematical operation is further applied on these ratios(like they maybe taking an average of these values to calculate the overall volume ratio of our COI).

Thanks
Ani


#128

Now we are talking!

So, is that cell volumeRatio the average of those 4 values??? etc…

Also, you have not added the 3D ‘neighbouring’ cells yet, since it is pentahedron there should be 5 values but only if our interpretation of ‘adjacent’ is correct…

And when you add the 3D factor, you might not have highlighted any of the actual face to face adjacent cells to that COI (think about that one :wink: )


#129

But the upper face is not shared by any cell so only 4 values should be there.


#130

Not sure how you know that, this could be any one of its 5 faces that we see (and I think there is a chance that that face is not even planar to the screen (image plane) based on how I have seen some weird slicing depictions), what am I missing?


#131

This might help. I forgot to mention the shape of the adjacent cells.


#132

Yes, sorry, the COI face shown could be any one of 3 faces, not 5 of the pentahedron.

EDIT, sorry I misspoke, I think we have to keep our mind open that the COI face shown may even lie on the BMB surface. I think we can agree that it is very difficult to show/speculate on all faces of a group of volume mesh cells and faces in an image like that… :wink:

But in any case, unless those are right angle triangles, then none of your ||| cells are face adjacent to the COI. (even if right angle, only 1 could be face adjacent, and we then have to assume that face is planar to the image)

So, back to what does adjacent cells mean?

And back to the 7 permutations of pick 1,2 or 3 of common faces, edges or corner points?


#133

I think that you are forgetting that there can be skewness in hexahedral cells.
skewness
I am surely going to become the next picasso :rofl::joy:


#134

Yes, I think skewness could add the possibility that all 4 of your |||| cells could have face adjacent faces to the COI (but still possibly none of them could be) and in this case you have shown that we are seeing a pentahedron.

EDIT: I think the only way to say for sure that all 4 of your |||| cells are adjacent cells would be if the definition of adjacent cells includes faces with common edges…

To continue this discussion I think we need an answer on what does adjacent cells mean?

Which brings us back to which of the 7 permutations of pick 1,2 or 3 of common faces, edges or corner points, is the meaning of cell adjacency (especially WRT our volumeRatio)?

EDIT: If the adjacent cells definition includes all permutations, then an internal hexahedron cell (not touching the geometry or BMB surface) would have 26+ adjacent cells (I am not sure what the upper limit would be and I am not sure we need to know :wink: ). Sometimes I have trouble saying what my brain sees :wink:


#135

Okay @DaleKramer I think I’m ready to have you take a shot. See my Run 21 on CAC Deflected for the current best shot at this. This is the best coverage mesh I’ve been able to get without much in the way of illegal faces. I decreased sim time to 450s as that’s about when the previous best mesh converged, but it may take more.

Getting this simulation complete will give us a pretty good comparison with the 6MM cell mesh with “better” Y+ values discussed earlier in this thread.


#136

OK I have a couple sims running, just for a start I am using Dylan numerics on your Run 21 and a Run 1 with Dylan numerics on the 24M mesh in ‘For Dale’ sim. May not get different results yet but this is where I want to start :wink:

https://www.simscale.com/projects/DaleKramer/dale_copy_6-31pm_4-14-19_thrush_s2r_lateral_stability-_mk3/

EDIT: Dylan numerics did not get as far as Default in these cases, so I am at my limit of trying to get a Y+=1 Full res mesh to complete a simulation.

Please, can the sim expert @PowerUsers_CFD 's now have a look at why these 2 meshes that @jhartung has laboured over for so many hours, will not finish a simulation…

CAC Deflected\Run 22 11.6M mesh

For Dale\Run 1 25.6M mesh


#137

Hi everyone,

While you all are doing some fantastic work on snappy, I’ve proceeded to attempt the mesh generation on cfMesh on my PC. The problem is, I am unable to clean up the geometry sufficiently and cfMesh simply returns “too many errors”.

I’ll see what I can do but was wondering about how the geometry was created. Could you elaborate? @jhartung

Regards,
Barry


#138

I could never figure out what those errors were and the meshes that I saw were made with 0 illegal faces until we started relaxing some quality parameters for Full res mesh, hope you can figure them out…


#139

Hi both,

For this run, first attempt to disable potentialfoam, then if still unstable, try using smooth solvers for all conditions (U, P, k, w). There should be no reason why higher order schemes are not usable, but if the sim continues to be unstable, change the gradient scheme to gauss linear for grad(U).

Similar comments for this.

It is important to get the sim running first before attempting higher order schemes or faster solvers.

Cheers.

Regards,
Barry


#140

As of Post #139, the ball is back in your court @jhartung :wink:


#141

@DaleKramer I’m following along… are you suggesting I should try the above suggestions or you? Do you think any of them would reduce the memory requirements?

@Get_Barried CAD was built in Solidworks with special attention to minimizing small features and individual faces.


#142

Yes you, I would have to figure them out as well :wink:

Personally I do not think a # cores issue here, I have done larger than these with 32 cores, maybe @Get_Barried could answer that too…