RE: convergence error.

[ryancwl95]

Hi,

I am currently conducting a transient analysis of a pile subjected to cyclic loading. The analysis was successful and provided reasonable results. However, I encountered a problem when I significantly increased the number of nodes and elements, which resulted in a convergence error.

Screenshot 2024-11-24 223003.png (11.85 KiB) Viewed 2913 times

I am currently using these analysis object options for my analysis steps (gravity, pile removal, pile installation and pile loading):

# Analysis objects
constraints Penalty 1.0e+16 1.0e+16
test NormDispIncr 1e-3 100 1
algorithm KrylovNewton
numberer RCM
system UmfPack
integrator Newmark 1.50 1.0
analysis Transient
InitialStateAnalysis on
analyze 5 5e6

Could you advise on the error and how to around it?
Thank you in advance and I look forward to your reply!

Sincerely,
Ryan

[STKO Team]

That error says the system is singular.
Probably there is a modeling error.
Unless you reached millions of DOFs and a sequential solver like UmfPack cannot handle it

[ryancwl95]

Hi,

Thank you for your reply. I am using a 20-8 brickUP element in my analysis, and I'm utilizing STKO as my post-processor.I create the mesh model using a script that defines the initial nodes in three dimensions. The allocation of nodes for elements and fixities is handled automatically by the script.

It seems that the issue arises when I increase the number of nodes (and, consequently, elements) beyond a certain limit. When I reduce the number of nodes, the model runs successfully.

How would you suggest I approach this problem? Should I consider using a different solver? What would you recommend?

Sincerely,
Ryan

[STKO Team]

How many nodes are you talking about?
If it's too much you may need to go parallel (OpenSeesSP or OpenSeesMP)

[ryancwl95]

I have approximately 52436 nodes, each with 4 degrees of freedom (u-p), and 3,048 nodes with 3 degrees of freedom (u). Would this type of analysis require parallel computing (SP or MP)? Additionally, since I am only using STKO as a post-processor, could you advise me on what I need to do to get started with parallel analyses? Do I need to download MUMPS, and how do I install it?

sincerely,
Ryan Chia

[STKO Team]

You can follow the instructions here to build the parallel versions of OpenSees:
https://opensees.github.io/OpenSeesDocu ... build.html

Or if you are on windows you can use the pre-compiled un-official version from prof. Tarque:
https://www.researchgate.net/publicatio ... ltext=true

We have a lot of webinars where we use OpenSeesMP within STKO. But if you are not using the STKO pre processor probably it's better to use OpenSeesSP (though some material or elements might not work with it)