Convergence

[Jhno]

Hi,
I am conducting a displacement controlled analysis.
In a first attempt, I ran 30 000 steps for a 0.1 displacement total ( that make 0.1 / 30 000 displacement per steps).
Then it failed at around 28800.
I said fine, I'll just increase the number of step, to, say 35 000. There I thought I had it.
Well, now it fails at... 15 000 steps.
Have you ever had this problem? I changed absolutly nothing else in the model!
I thought decreasing the displacement increment per steps would help stabilize the system... never imagined it could ruins it? Any clue on what could be happening? Thanks!

 

[L_K]

A lot more things can affect achieving convergence than step size (nl contacts, material model, large deflections etc).
Although I too find it odd that decreasing step size has that effect on how far the analysis converges.
What software are you using and what kind of nonlinearities are in the model?

 

[Jhno]

I am using Opensees. I conduct a planstrain analysis of quad element with elastic-per.plastic behaviour (Druck.P).
When I take the same model with a little larger meshing, it's fine (but the answer is a little too high).
When I refine, with same steps number, I crash close to solution. When I increase the number of steps (which should help), I crash sooner... this I don't get.

 

[L_K]

Does the software indicate anyway what is the reason for diverging in the different models?

 

[Jhno]

No, not really. It just says that the solver failed... Then you gota figure it out by yourself.
I know it gotta have something to do with element distortion because when the mesh is large, I do not get this problems.

 

[scheah]

I'm not familiar with opensees. I assume it uses Newton-Raphson method. Do you have the option of arc-length method similar to in Ansys? [URL unfurl="true"]http://www.ansys.stuba.sk/html/guide_55/g-str/gstr8.htm[/url]


Best regards,
Sze Kwan (Jason) Cheah

 

[whitwas]

Look at the solution leading up to the failure. That can reveal what's happening. Something might be moving in an obviously wrong way.

I suspect buckling is happening. I've seen cases where refining the time steps caused convergence failure while coarser ones didn't. That doesn't mean the coarse step results were right. The fact that refining the mesh makes it worse is another sign in support of buckling since a finer mesh tends to be less stiff in bending. Try a linear buckling analysis for a rough idea to see if you're close to the critical load. It might not find it, but it might.

 

[L_K]

Buckling should also be seen from the nonlinear analysis results (deformation graph where the slope should get suddenly steeper).

 

[Jhno]

Maybe I am wrong but I do not think it could be buckling since am doing plane strain analysis with quad-element.
I am trying to obtain the ultimate bearing capacity of a foundation.
I am wondering if it could be some problem with the interpolation function of the element. Since am using quad-element with a stabilized single-point integration (just one gauss point at the element center) I thought maybe it had problem adapt to imposed deformation, hence the problem with finer mesh? I am trying with 9-noded element to see if it will fixe the problems.. any thought on this?
thanks for the help!

 

[scheah]

Looks like your difficulty may be due to hourglassing. Please let us know if the higher order element fixed your issue.


Best regards,
Sze Kwan (Jason) Cheah

 

[Jhno]

Well, I tried to use the 9-node element but it is designed for coupled fluid-structure (effective) stress analysis hence it's not doing a good job for what I need.
I was wondering if decreasing the size of the element could somehow solve this problems (using SSPquad element (SSP --> Stabilized Single Point, see here for "details" :

http://opensees.berkeley.edu/wiki/index.php/SSPquad_Element

).

 

[scheah]

Please let us know if a smaller element size works. Otherwise I assume your latest attempt was with Four Node Quad u-p Element. Perhaps give Quad element a try.


Best regards,
Sze Kwan (Jason) Cheah