Convergence problem

04-Unexpected behaviour/errors
Post Reply
wangguojue
Posts: 6
Joined: 08 Jan 2018, 09:16

Convergence problem

Post by wangguojue » 11 Jul 2019, 10:38

Hi:

I just found that the convergence ability of Seismostruct 2018 is much worse than Seismostruct 2016.
The probelm is this, I made a static dynamic model in Seismostruct 2016, and the model was successfully calculated in all 41 seconds. However, when I use Seismostruct 2018 to run this model without any change, the calculation is terminated in the 28th second, then I have to increse the steps of the model. In seismostruct 2016, 41-seconds model have 820 steps and the calculation was successfully done, but, in seismostruct 2018, 41- seconds model have 8200 steps and the calculation still terminated at 38th second!

What is wrong with the newest Seismostruct 2018?

B.T.W I have sent the model file to support@seismosoft.com, I'm looking forward for the help.

Regards
Wang Guojue

User avatar
seismosoft
Posts: 592
Joined: 06 Jul 2007, 04:55

Re: Convergence problem

Post by seismosoft » 11 Jul 2019, 15:22

Wangguojue,

The fact that one model that ran in SeismoStruct v2016 has some convergence problems in v2018, especially when the structure is pushed beyond its limits deep in the inelastic range, does by no means indicate that 'the convergence ability of SeismoStruct 2018 is much worse than SeismoStruct 2016'!

Indeed, there are some changes in the convergence algorithms of SeismoStruct in v2018, but in general (99% of the cases) they lead to more stable analyses. In fact with a sample of hundreds (if not thousands) of models that we have tried last years, it is obvious that convergence is far better in v2018, with respect to v2016 (and with respect to similar nonlinear analysis programs).

In particular with your model, if you simply uncheck the 'Automatic Adaptation of the Convergence Norms' in the Program Settings, the model will run until the end of the analysis. So before asking about what is wrong with the newest Seismostruct 2018, it might be better to first check what might be wrong with the model. If you are not convinced however that the stability of SeismoStruct v2018 is very good, you could always shift to another package. There are plenty of nonlinear analysis programs worldwide that you could try and compare to SeismoStruct.

Seismosoft Support

wangguojue
Posts: 6
Joined: 08 Jan 2018, 09:16

Re: Convergence problem

Post by wangguojue » 11 Jul 2019, 15:26

Hi:

Thanks very much for your reply! I will check the software ASAP to see if your suggestions could help.

Regards
Wang Guojue

wangguojue
Posts: 6
Joined: 08 Jan 2018, 09:16

Re: Convergence problem

Post by wangguojue » 11 Jul 2019, 16:09

Hi:

I have followed your suggestions and checked the project settings in seismostruct v2018. The main difference of "convergence criteria" between v2018 and v2016 is that v2018 has one more option of "Automatic Adaptation of the Convergence Norms". The "Automatic Adaptation of the Convergence Norms" is defaultly used in v2018, after unclick "Automatic Adaptation of the Convergence Norms", the model works well in v2018.

Again, thanks very much for your explanation and help.
Regards
Wang Guojue

Post Reply