We used prepomax 2.0.0 with builtin calculix on a boxtruss made from shells,
compared to Nastran Calculix gives vastly different answer
Is the mesh the same? How about the boundary conditions? Are you aware of the expansion that Calculix does for the shell elements? Solver used in ccx (pardiso, spooles, pastix?)? We need more information to determine the root cause.
Pano,
A difference of just 0.16 inches on a structure that seems to span more than 70 inches is not a âvastâ difference.
Assuming that your analysisâ element type & thickness, contacts, materials, BCs, etc. are identical then at least tweak the results legend in PrePoMax to have the same upper and lower bands as in Nastran.
Then compare the displacements and isoplot pattern again. I have made many comparisons between PrePoMax and commercial software in the past and itâs almost always a user input difference.
1 Like
The mesh and loading, BCâs, material, glued contact and thicknesses is same in both cases. the load is 10psi in center as shown, using pardiso solver for calculix. I also verified the displacement with Ansys and with Nastran using quad4 elements. Since Im new to this the forum it doesnt let me attach the mesh file, but its not a big file, only 430kb for the inp mesh and 1.1mb for the prepomax pmx file.
Its the displacement pattern thats vastly different. Notice that the nastran correctly predicts the centered âbullseyeâ pattern for each loaded box section in the center of the truss, while calculix completely misses this. Why is the displacement pattern biased to one side of the truss in calculix when the load is centered?
Have you looked results calculix gives for large shell structures? In aerospace most structural work is with shells, even in automotive a car body is all shellsâŚwhat use is a calculix if it cant do shells right?
Check all contacts work properly with a frequency analysis.
No loads yet.
How do the firsts frequencies and mode shapes compare?
That is a good reference for comparison.
2 Likes
Could you please produce a displacement isoplot only in the Z direction (for both) and share it ?
By the way, you can upload the pmx file to Dropbox or Wetransfer and post the link here
1 Like
Yes, the results of shells are sometimes strange. So a user must be careful and informed when using them. However, a proficient CalculiX user can get accurate results.
2 Likes
CalculiX also has a true shell element. It could be interesting to compare the results using this element type: Support of shell user element (US3)
Thereâs no sign of any PrePoMax bug (rather an issue with CalculiX accuracy) so I moved it to General Questions.
1 Like
What I see weird in both results, is that for a simmetric model and bc, the displacement plot is not simmetric. Are those items true, model and bc are simmetric?
2 Likes
Yeah, I suspect that the top right leg is not fixed and perhaps allowed to slide. My guess anyway, but could be anything.
Would be good to get the pmx file.
1 Like
Here is the PMX fileâŚyes all the legs are fixedâŚhere is the pmx on google drive
I did the model with quads and Im not remeshing it with triasâŚmaybe next timeâŚ
If I didnt compare with Nastran I wouldnt know the results are wrong, intuition isnt reliable with this. So if one has to have Nastran, then Calculix is already redundant. probably what would be helpful is a disclaimerâŚâCalculix gives bad results with shellsâ and then people wouldnt fall into this trap
this is not the only model I have that Calculix did poorly on, I looked at many shell models and with all of them Calculix does poorly. It seems to be OK with small parts or solids, but with large shells it consistently gives bad results or doesnt converge at all but Nastran solves them all fast and with no problems.
Could you post the inp?.
the pmx is posted above, for some reason it wont let me post the inp
Results of your model submitted in Abaqus:
Btw. If you suspect issues with CalculiX, itâs better to post there: https://calculix.discourse.group/
Once confirmed, you can report CalculiX bugs here: Issues ¡ Dhondtguido/CalculiX ¡ GitHub
Oh yes, thank you for running it in Abaqus, it gets the displacement rightâŚit matches the Nastran pretty closelyâŚso its not my modeling thats the problem, its Calculix