i’m not yet to go further since your model seems to have several problems shown by graphics and logs. Rebuild or replicating the model independently by following problem sketch in geometry, loads and boundary conditions could give more insight for myself.
Well… this is the problem of the original file, which was uploaded by the author… not my mesh. Why should i waste my time? Rebuilt it yourself and share your results.
the original poster is asking for something unexpected. However, at the end you do modify by refining to quadratic with dense mesh and took some conclusions without correction of surface alignment.
Man, learning something is not wasting a time and i’m done by simple model. I understand the logs and graphics shown in your model to have something serious consequence.
Here are the geometry files for this, both midsurfaces and solids. PrePoMax should be able to mesh these structures. This type of box truss is very common in aerospace, architecture, marine, automotive, its the basic engineering structure. In aerospace one deals with more complex curved shells, and I have several examples of those, but this geometry keeps things simple…
Anyone else here could have noticed that. I was at least working with the model instead of just expressing assumptions and theories that didn’t prove anything concrete. Never mind, next time I’ll hold back.
Results with improved 2nd order quad (S8 elements only) mesh:
Note that the units are SI(mm). I’m not used to imperial units.
Can you share please…the model
Sure
table improved mesh.pmx (1.0 MB)
I was rerunning gunnar’s platte.pmx above and got some very odd behaviour from Pastix with Pastix_mixed_precision=1. I’d be interested if someone could repeat it.
Did you set it in Tools → Settings → Calculix → Environment variables ? What happened ? I haven’t noticed anything strange.
The Nastran predicts 0.65" and Calculix 0.535…the difference between 0.535 and 0.65 is 11.5%…Do the basic math…In world of linear structural analysis 11% error means something is wrong, its not an “OK” result…you would expect nothing more than 1 or 2% difference…
here is the geometry file, you can remesh it from scratch if you want. I meshed it with 1st order quad mesh for Nastran which gives accurate enough results with that, Calculix will require a 2nd order fine mesh to do the same thing.
Yes in the settings environment. I’ve no interest in the “correct” answer for this particular problem.
Nastran has 0.65…so your 2nd order analysis is 7% off from that, but its getting alot better than the original 1st order analysis …lesson learned is I would say never use 1st order elements in Calculix for static structural analysis…
Did you use 1st order mesh in Nastran too ? Maybe it has some inaccuracy because of that. It would be interesting to try with the same (or similar) improved mesh in Nastran.
And what was the issue ? I’ve tried it and it looked normal.
Oh yeah, Nastran does fine with 1st order elements, even if they are a coarse mesh…The issue is to raise awareness that Calculix doesnt have good accuracy with large shells even with a 2nd order mesh its still going to be 7% (shown above) from the Nastran benchmarks. So I guess that if you were to use Calculix doing analsis for a paying customer you would have to run Nastran anyway, or else you’ll be giving your customer part designs that are not optimized and how would you know otherwise? So Calculix doesnt replace Nastran, it needs alot more testing and refinement. Yes, it is opensource free software…but No, you cant count on it to replace Nastran (or Ansys etc…)
No thanks. I know how i can get good results, even with linear elements.
