Thin Wall Cylinder Buckling

[kevinarden]

Elmer results compared to Abaqus using StressSolver to calculate buckling

StressSolver_Buckling.pdf

(326.38 KiB) Downloaded 57 times

https://github.com/mrkearden/Thin-Wall- ... r-Buckling

[raback]

Really nice case Kevin!

There is ~5% discrepancy in the eigenvalues compared to Abaqus. Do the results approach each other when the mesh density is increased?

I guess the tetrahedrons have quite a few more nodes (?) which can explain most of the difference in time spent. The hexahedral mesh seem rather similar to the eye.

-Peter

[kevinarden]

Tets have 19000 nodes
Hexa has 6000 nodes

I can try a refined mesh on the Elmer to see if the eigenvalue comes down

[kevinarden]

I took the Hex mesh to 19000 nodes two elements through thickness instead of one.
40 along the length instead of 20.

eigenvalues .76, .76, 1.02

run time 94 sec

[kevinarden]

I set the load to the theoretical buckling pressure so the eigenvalue should be 1.0
The equation involves picking a coefficient of a log-log chart so the theoretical is as good as my eye sight.

However NASTRAN was much closer to 1. However the mode shape is different.

chart.PNG

(13.53 KiB) Not downloaded yet

[raback]

Hi Kevin,

Very interesting! How did you obtain the quadratic elements in Elmer? What I'm wondering is whether they follow the cylindrical shape or is the mesh faceted so that each element has a straight face? That could have an effect.

Regarding the mode shape. Would this be similar in shape to the 3rd eigenmode of Elmer/Abaqus? They seem to be closer to one. I guess the 3rd one is the 1st antisymmetric buckling mode.

-Peter

[kevinarden]

The Elmer is faceted, I will try to change that.

[kevinarden]

The first antisymmetric mode in Elmer is mode 5 with an eigenvalue of 1.03.

[raback]

Hi Kevin,

I just added a consistency test case "CurvedBoundaryCylHquadratic" where you can see how to increase meshes from linear to quadratic + make the new nodes follow the cylinder surface. Only a few shapes are supported. There could be a more generic way but cylinder and spheres at least make it possible to test the theoretical convergence. The essential keywords are:

Simulation :: Increase Element Order = Logical True
Boundary Condition i :: Follow Cylinder Boundary = Logical True

You can also follow the shapes when introducing p-elements and using mesh levels. In these tests for capacitance computation one can see that the if the faceting is not addressed it does not much improve the results just throwing more elements / higher degree.

-Peter

[kevinarden]

With these settings
Simulation :: Increase Element Order = Logical True
Boundary Condition i :: Follow Cylinder Boundary = Logical True
and Element = "p:4"
the values came out to be 0.75, 0.75, .99
which is 2.7 % to ABAQUS, but the solution time went to 1478 sec