Multiple Shell Sections in one Model not working?

Hello,

At the moment I am testing PrePoMax 2.1.0 for the buckling analysis of a wind turbine tower, the tower is a shell with multiple different sections (different wall thicknesses


).

I start with a “Buckle” Analysis and it seems to me that PrePoMax can only handle 2 different sections, If I use only 2 Sections I get about the expected result (buckling factor = 3.4), If I try more than 2 different sections I get (buckling factor = 1).

“Update: During writing of the post I tried the same analysis with quadelement and second order and it works ! in the first try I used only linear tri elements. Still strange error though”

I would avoid them in CalculiX - one such example: Calculix vs Nastran big difference in results for shells - #22 by synt

Hello, I am encountering similar problems. When I try to solve the model using S4/S4R elements it works correctly for a cylinder of constant thickness, but when I include several thicknesses it does not give adequate results. Can this be attributed to the computational capacity of my PC (8Gb RAM), or what is the reason?

Did you try refining the mesh or using different element types ? Is it a linear buckling analysis ?

I am modelling the cylinder with shell elements, so I where using S4/S4R. I tried different meshes finer and coarser, but similar results. Yes, I have specially noticed this in LBAs, but it is true that I tried in other computer (with less RAM) and it has similar problems for LA, that’s why I asking about computational capacity.

I solved a similar problem (looks like the same) some time ago based on a well documented result of a FEM contest.

Mesh was dense + S8R. See picture.
In my experience from diferent problems solved along the time, S4R is not reliable for Buckling analisys.
Pay special attention to boundary conditions on the base. You have to constrain rotations on a curved surface. That’s not trivial to achieve.

Deviation from reference value was minimum.

What matrix solver are you using ? There might be some issues in those analyses with Spooles and PaStiX. Pardiso is the most reliable.

Thanks for the comment. I have been working a lot with shell buckling (it is true that it is in ABAQUS), and I have obtained very good results using S4R. In fact, the second part of the paper you have shared is the reference solution and they proposed the use of S4R. So, my question here is, are the problems with S4R limitation of Calculix?

Yes, I am using Pardiso. It was the fastest one too

Check this benchmark study: Buckling of composite panels - #34 by lucas_bueno - CalculiX (official versions are on www.calculix.de, the official GitHub repository is at https://github.com/Dhondtguido/CalculiX).

That whole thread can be interesting too.

In CalculiX, S4 and S4R four-node shell elements are expanded into three-dimensional C3D8I and C3D8R elements, respectively.

By other hand a node leads to a knot if

• the direction of the local normals in the elements participating in the node

differ beyond a given amount. Notice that this also applies to neighbor-

ing elements having the inverse normal. Care should be taken that the

elements in plates and similar structures are oriented in a consistent way

to avoid the generation of knots and the induced nonlinearity.

• the thickness is not the same in all participating elements.<------

• the offset is not the same in all participating elements.

Thank, it is a very interesting post and quite helpful

Ok, thank you. So, now I understand why this problem occurs when considering different thicknesses. So, how do you solve this by considering S8R since in this case it is not possible to avoid the generation of such knots because of the variable thickness?

Sometimes it makes sense to try with corresponding solid elements in the case of such issues with expanded shells in CalculiX. PrePoMax has a tool for that - Thicken Shell Mesh.

I have translated my old file with S8R from Mecway to most recent Prepomax version and it has perform excellent too.

S4 and S4R results are empty. It reports Buckling factor =1 but mode shapes are almost zero.

Yes, that’s exactly what I am obtaining for S4R but I am not able either to obtain the results for S8R. This is how they are looking

is it possible to know how you made the model??

You are very close.
Try to increase the accuracy in the Buckling step window. That should improve your mode shape.
Regarding your buckling factor (slightly low) , read the paper carefully because you are making a mistake that is described in it.

Thank you so much. I think that more or less I got it.

The only question I still have is why with the S8R elements it works but with the S4R didn’t? the knots are also appearing