Convergence Issue with Force Beam Element

[Dani_]

Dear STKO Team,
I'm trying to make comparisons between the DB and FB element using the same model and analysis while varying the mesh, excluding the regularization issue. With the same analysis settings, in this case Newton with Line Search and Norm Displacement Increment Test, FB seems to struggle more to achieve convergence, especially for the model with a higher mesh count and therefore a finer discretization. For example, with a discretization of 16 and 8 meshes in the case of FB, I do not achieve convergence in the analysis, even at the beginning of the curve (where it should still be in the elastic phase). I am attaching the model, in case there are any analysis settings that need to be modified.

Results_DB.png (33.33 KiB) Viewed 1335 times

Results_FB.png (21.42 KiB) Viewed 1335 times

To improve convergence in FB, I increased the analysis steps and changed the algorithm to Krylov-Newton, which allowed me to obtain more complete curves. However, the model with 16 meshes still fails to converge. Furthermore, due to the unrealistic oscillations in the curves, it appears that the analysis is not very robust.

Improvement_REsults FB.png (29.25 KiB) Viewed 1335 times

Since I've often noticed that the analysis never seems to have issues with the DB, but frequently encounters problems with the FB, and I need to improve the analysis settings as it often fails to converge, I wanted to ask if you have any suggestions for solving the problem. I'd like to have robust analyses for the FB similar to those for the DB.


Thank you

[STKO Team]

First I suggest you use more reliable materials.
Your concrete has some strange behavior (use the material tester and check it):

CFail.png (79.08 KiB) Viewed 1329 times

Also, your steel seems buggy, it does not provide any stress

SFail.png (83.1 KiB) Viewed 1329 times

Try something simpler like Concrete02 and Steel01 (or Steel02).
Then you may improve the convergence even more if you use more convergent materials like DamageTC1D (using the IMPLEX algorithm)


Finally

FB seems to struggle more to achieve convergence, especially for the model with a higher mesh count and therefore a finer discretization.

If you do not consider fracture energy regularization you will never solve this problem.
Fracturing is strictly related to the size of fracture zone (in your case the Mesh size).
For example,
you use 0.0035 as the concrete crushing strain.
First, it is very small (consider this is a limit for cover spalling, and is not even the strain at which the concrete loses all its bearing capacity).
And it is measured on specimens whose size is in the order of 100-200 mm.

Instead you have a mesh of 70 mm, with 5 intergration points (lobatto). With this integration scheme your end-integration-point is 14 mm. That's why the behavior becomes so brittle that you cannot achive convergence anymore if you do not properly consider fracture energy regularization