Solution converges before max iterations are reached

mvass

Hello.

 

I am working on the NL buckling of a composite panel (geometric NL implicit) and I successfully setup the model and ran the analysis within radioss after automatic conversion from the optistruct input. The problem is that although I specify I need 50 increments (NINC) with PW convergence criteria, the solution ALWAYS converges at 25 increments and of course the total buckling is not captured. Below is my NLPARM card from the .fem file. I am not specifying anything else in the NLPARM card or in any other cards. Can you please help?

 

NLPARM         6      50     0.0               6      50      PW
+               0.01    0.01                          20                
+                    1.0

 

Thank you  

mvass

OK, here is an update, after solving my .fem file on two different versions (thanks to my friends):

 

HW version 13.0 completes the solution in 50 increments as requested, and the analysis is complete and correct.

HW version 2017.2 (student version) completes the (same) solution in 25 increments the analysis says that it is completed but the results are not comparable to the previous ones.

 

Any ideas why this happens? 

Rahul Rajan_21763

A lot has been changed since OptiStruct version 13.0 was released. NLGEOM case should be solved as NLSTAT with PARAM,LGDISP,1(large displacement analysis)

mvass

Thanks for your reply and sorry for my late feedback. I cannot remember what I did with the latest version, however I'll give it a go with NLSTAT and large displacement option and contact you again. However I still cannot understand why when you specify that you want a NL analysis with fixed number of increments (constant DT), the problem converges with less increments completed (and the result doesn't capture the phenomenon as it should). 

Rahul Rajan_21763

If implicit problem converges properly then there should not be any difference in output whether you use NINC as 50 or 25.For better convergence don't use PW. Simply run with NINC as 50 or 25.However if you still want to have 50 increment then set control card: PARAM: EXPERTNL: NO

mvass

Altair Forum User said:

If implicit problem converges properly then there should not be any difference in output whether you use NINC as 50 or 25

I agree with you. That's why I was surprised to see different results when NINC was 50. Maybe I couldn't see where the mistake was, however there was no error shown in both runs (NINC=50, NINC=25).

 

Thanks for the advice about the param option.

mvass

OK, here is my update to the problem. Unfortunately I cannot share the model with you, however, please find attached the .out file and the solution scheme from the .fem (input file).

 

As mentioned before, the solution converges normally but the results are not correct. It is a NL buckling problem (compression of a panel) and I am supposed to have a max displacement (magnitude) on the panel of 20mm. This doesn't happen and the max. displ. magnitude I get is 10mm. My solution was a NL geometric (impl) that I set up in Optistruct and automatically translated to Radioss (version 2017.2).  

 

I tried to run the problem as quasi-static NL with large displacement activated in optistruct but it didn't converge.

 

Thank you in advance for your assistance. 

Unable to find an attachment - read this blog

mvass

Second update:

 

I managed to solve the afore-mentioned problem, by specifying a value of 1.0 (instead of the default 1.1) to the DTSCI entry of the NLPARMX card (see the attached picture). However the question still remains: Why the solution is different when the default values are used? How come there is no warning/error in the previous situation? 

Unable to find an attachment - read this blog

mvass

Altair Forum User said:

Why the solution is different when the default values are used?

 

Hi, any views on my updates about the problem?

 

Thank you in advance

Rahul Rajan_21763

So this is the nonlinear buckling case solved with Riks or Arc length method.You can refer nonlinear DTRAN for post buckling analysis which currently has generalized alpha and backward eulerian time stepping schemes. Currently we don't support Riks/Arc-length for nonlinear DTRAN .

OptiStruct Nonlinear Learning Center

mvass

Thank you for your reply. In my study, I am not looking for post buckling and my solution scheme does not include a velocity BC like the example shown. Besides the obvious constraints, the load is exerted on the model via an enforced displacement. In addition, I am not looking on a transient NL analysis. But I guess it wouldn't hurt if I'll try to solve the problem with NLPARM and TSTEP card, provided that it is possible. Should I get something worth to write about, I'll post it here.