I’m just showing the elements of interest. The whole model is a simple cube . Meshed with tets and all have the same dimension. I haven’t rate it but mesh quality is off the scale.
I hope the ccx community does not take this as malicious. My intention is to understand the weaknesses to get the best out of it…
Abaqus has show a similar response for the simplest case. But unaveraged result from Calc_em has light my bulb. Stress on the element with comond edge to the body edge is uniform. I have isolated some elments on top of the edge with that criteria and thinghs got much better. So that could be the workaround.
Don’t you think stress smoothing it’s masking the Stress Range.? Kind of like blurring the contour. Cheating oneself
wrt to stress smoothing. it would be nice if the results matched reality. in many of the pictures posted above there are like polka dot stress distributions, in cylinders with contact. i have seen that many times too. there is no way that is real. if you try increasing the mesh density it doesn’t really help. all of the results i have seen with ccx contact just don’t look right. in my case, i can ignore it, since those areas are not critical for me. but if you really needed an accurate stress value in a contact patch, using ccx, i’m not sure what you would do.
some of your other findings are interesting but i’m not sure what the issue is exactly. if you can test something without reentrant corners it may help debugging. i think if you have one node (reentrant) and the rest are not, it could be throwing off the element stress distribution. so it might be a type of issue, but not one you would encounter without reentrant corners. i don’t really know though.
my cube example supported by elastic spring using coupling type distributing, but for contact two part use rigid body. I try to switch between each other, still shown uneven distribution for bad mesh pattern. Extending to several elements and multiply the height will not change the condition, this can be mean not problem in distribution at face loads.
Yea, my explanation doesn’t make so much sense for contact since they’re already nonlinear springs and you’d think they could/would adapt to the load distribution. But maybe they don’t?
first i tried to replicate the problem by create similar mesh pattern in LISA, it seems to be failed in mesh conformity even solvable by CalculiX. Later i create from the base of CAD files then meshing using Gmsh,(Transfinity,Recombine:No) in PrePoMax, it has shown succeed. And the last i checked your mesh, did you trust the result and needed to go further when the mesh itself has a problem?
p.s i do some mesh conversion between Inp file formal with Msh and Vtk to be readable of both pre-processor.
I don’t see any problem with the mesh. Not with Mecway at least. ¿Could you explain a little more?
Inp file is a cube made of 8 x 8 x 8 cubes with 5 tets each one. 2560 Elements.
That’s great!!. I’m sure users will appreciate that unnoticed bug inside the meshing tools of “LISA” . Specially known that conformity (edge conformity) could be lost unnoticed after a refinement process.
My mistake and your findings will be very useful to others.
Once I have fixed that on my mesh:
-Things get slightly better but unfortunately don’t remove the wavy patterns.
This is not exactly right so I’m still in search of some procedure to detect nonconformal tet mesh (edge non-conformity) to avoid future problems.
I didn’t trust my results, that’s why I initiated this review process. Especially after you posted this:
If that’s true, which I don’t doubt, It is still not clear where those wavy patterns come from. Maybe your holes are fully structured (that’s not the common case) and all the nodes at the edge share the same number of elements. That’s only possible in a closed edge but that’s the case. If so , I would say the averaging would provide a uniform stress distribution to the eye , but maybe wrong value. I need to look.
indeed, my second test of contact two part by similar pattern but conformal, still shown result uneven in distribution.
so that’s puzzle, another setting can give true. First problem is in conformal mesh, and later in mesh pattern itself.
from fisrt screenshot of their global model, it has shown did not fit in curved geometry. Probably using linear tetrahedral element or midside node of quadratic did not set properly
LISA is an old, being used by many students here since years. Refinement and conformal mesh depend on the steps in creating. Previously, i’m using manual duplicate after one cube of 40 elements generate then it failed.
Thanks for this nice example! Haven’t seen this before.
I find it a little bit surprising that you mention TET10 only at the end, seemingly as an alternative. So, as a reminder for everyone new to CCX or FEM in general: you should never ever use TET4 but only TET10. Basically, it doesn’t matter how exactly TET4 behave in any scenario because they are useless. See CCX manual or ask anyone with experience.
You are right , the post started with TET10 until I focused more on the corners. Seems the source of those wavy patterns could be the different tet orientations converging on common corner nodes.
Well, I don’t agree completely. Everything has its use. You just have to find the right place for it.
. Specially known that conformity (edge conformity) could be lost unnoticed after a refinement process.
