This is not the case for linear quads, or S4R elements, these give good results. My meshes are always linear quad meshes
Well, the shell elements are actually extruded into prismatic elements. So triangles to wedges and quadrilaterals to hexahedrons. No tetrahedrons are created.
1 Like
Just in case somebody isn’t aware of this: When using tie constraints for shells, calculix doesn’t create constraints for the slave nodes at corners, when perpendicular shells are attached (this would also have to be taken into account when making a comparison, even it doesn’t make much of a difference in this example). On the other hand, when using tied contact, these nodes will be taken into account.
1 Like
Isn’t that related more with the master and slave assignment and if node finds Surface where to project.?.
1 Like
The slave nodes cannot be projected then, no matter what settings you choose. This makes the structure softer/less rigid.
for sharp corner of shell element, CalculiX generate knot internally. It seems a previous version not accepted the condition when some region also constraint exist, but later have improvement by removing these conflicts.
Well, according to the documentation, tie constraints don’t work with shells at all 
Of course, it’s not true and should be corrected in the manual but some issues might happen. Tied contact is a good alternative in such cases even if it increases the computational cost.
That’s exactly the point I wanted to make. So far the comparison has only discussed the shell elements, but not the connection. Tie constraints were also used here; it would have been better to use tied contact. According to your answer, it must to be a Calculix bug.
As i wrote:
this post is addressed to users who are not aware of this.
Edit:
These pictures showing the different influence of selecting tie constraints or tied contact. The pattern is more symmetric just because of the correct connection (1. contact, 2. constraint).
Because calculix has no “real” shell elements (S4=C3D8I), it would be more interesting and meaningful to compare the calculix results with Nastran results where only one layer of 1st order solid elements are used for the mesh.
That’s why this whole exercise of comparing to Nastran on one layer of 1st order solid elements is a waste of time- we all know they will perform poorly due to the nature of the interpolation function and the associated problems with the integration point schemes. How bad will it be? Honestly, who cares at that point? Maybe some mathematical purist would like to highlight the differences, but a practicing CAE/FEA engineer would understand this and perform the appropriate fe model building.
In any case, I would not call “vast” or “big difference” to 0.16 inches over a 70-inch span - as pointed out earlier. Especially when understanding the expansion of the elements in ccx.
New users must do their due diligence and learn the software and its limitations before comparing apples to oranges.
2 Likes
But then you would compare the same element types and not apples to oranges (that would at least be a fair comparison i guess when the author wants to stay with this mesh…) I didn’t claim that i would expect accurate results.
Edit
Ops, i was wrong 
Its not the same…
As I wrote, linear elements cannot capture bending stresses. If local shell bending is not dominant, the results will be very similar.
1 Like
Yes i am aware of this. I just wanted to graphically highlight the point of interest. But these linear shells are not “that” bad as C3D8 elements.
1 Like
i have seen something unusual from these pictures. Below simplification of model, results shown not similar.
,
I changed the upper limit of the color bar to the max value of the linear results, if it’s that what you mean.
so, deflection result of 1st order quad element reported almost half of refined 2nd order quad element in your model? and notified as garbage. That’s unusual for me.
Man, that was ironic… related to:
That’s okay when someone is new and being migrated from another mainstream FE software with classical formulations, i.e Bernouli/Timoshenko Beams, Mindlin/Kirchhoff Plates. In case of linear triangular element (S3), the document also agreed with this statement.
However, become questioning when linear quad shell element being compared to quadratic (refines) and deflection result as almost half. If the condition is true., here i’m interested in shell knot as the probable causes.
btw, i can not reproduce here by simple model: 1st order quad is deflected less at almost half compared to 2nd order quad (refines).
are both model having the same load to make sure useful ones? or maybe, your model problems are in large deformation with plasticity and large strain which beyond the topics.
.
