Post-Peak behaviour

[chiaramgn]

Hello Seismo,
I'm trying to modelling a little small arch bridge with different kind of materials, reinforced concrete and masonry too, modifynig the constitutive law of concrete material of default. I'm working with static time-history analisys, but what happens in each model is that it works good in the elastic phase, but I can't obtain results for the post-elestic phase, because the program stops with problems of divergence (fbd_Ite or other kind of advices), also changing the maximum tolerance,tolerance values, number of steps and everithing possible. What can I do? I need to deduce a Load_displacement curve, but now I can obtain only the elastic part. Maybe I'm modeling something wrong, or I thought coul be the arch shape?
Thanks a lot for you help,
this is the link dropboxwhere I attached the model file, just for have a look if it's possible.
https://www.dropbox.com/sh/u7plj3cwod41 ... D0xva?dl=0
Chiara

[huffte]

Hi Chiara. I took a glance at your model over lunch today and the first thing I notice is the extremely fine resolution of your model. The element lengths are significantly shorter than the element widths and comparable to the element depths. A situation such as that might routinely be handled with shell elements, but that is beyond the intent of the excellent SeismoSoft software, I believe.

However, I don't see why a more relaxed resolution of the modeling strategy might not work just fine. It appears that you have used about 50 elements per span. I wonder if reducing that down to 10 or less might make the solution much quicker but still reasonably accurate? Also, you have 20 elements per pier. This is also extremely fine meshing. I wonder if you might reduce that to 4 or less and have an easier time debugging the model?

Best of luck, Chiara.

[chiaramgn]

Thanks Huffle for your suggestion. anyway, I did yet a model also with less element like 8 or 16 per arch but the problem is the same.

[huffte]

Would you post a link to the simplified model on the forum, Chiara? I am sure it runs much more quickly than the first one and should hopefully be easier to debug.

[chiaramgn]

https://www.dropbox.com/sh/u7plj3cwod41 ... D0xva?dl=0
I putin the folder also the model with 16FB elements. Thanks Huffte,
Chiara

[huffte]

Hi Chiara. I changed the time step to 0.1 seconds for both phases of your analysis and the model ran completely. I observe a good deal of inelastic behavior in the axial forces beginning at about 8 or 9 seconds and the full 50 seconds of results are available. So, my advice at this point is to do some research on the required time step for an accurate solution when static time history analysis is employed. While the 0.1 second step may be excessive for a dynamic time history analysis with an input accelerogram define at every 0.005 seconds, it could be that the solution for static time history analysis with a uniform ramp-up load as you have here may suffice.

Some other questions to ask yourself:

1. At what time does the analysis crash when you use the very small time step?
2. What are the stress states at the time when the analysis crashes?
3. Is there some way you could do a quick hand calculation to estimate the failure load? I wonder if it is close to the load at which your model crashes when using a small time step.
4. If you change the time step as I did, does the model run completely as it did on my computer? There could be some differences in computing power adding to the problem.

Best of luck and keep us posted on your progress, Chiara.

[chiaramgn]

Dear Huffte,
thanks a lot for your help.
I think it coul be a problem of my processor power because I tried a lot of attempts but it seems to give me always not good results.
Anyway, which steps you mean to change? The steps of the load curve or directly in the time history stages? I want to do the same model that you did, for understanding if it is a problem of my pc. If you can send me directly the model I will be really grateful to you. thanks, Chiara

[huffte]

Chiara, the only thing I changed was the number of steps in the 2 time history analysis stages. For the first stage I changed 500 to 70 and for the second stage I changed 1000 to 430. This produces a uniform time step of 0.10 seconds for both stages. You may want to identify the approximate time on the onset of inelastic response in the initial results and change the analysis stages parameters accordingly (use a larger time step up to the onset of nonlinearity and a smaller time step thereafter). I have e-mailed the revised model.

Best of luck, Chiara.