Hi
I have modeled a two dimensional braced frame.I have done static and adaptive pushover analysis without any problem.But, when I want to apply dynamic history analysis this message appears:
"Unable to apply the next load step"
I have fully read the forum and help system and applied their hint, however, I can not solve my problem.
I would appreciate if someone help me.
Covergence Problem
Re: Covergence Problem
Hi saman.
When does the message appear? Before the analysis even gets into one step of the time history? Or after several time steps have executed?
Perhaps the input ground motion is truly excessive for the structure capacity you have modeled. If you simply change the load factor on the input acceleration to a small number - say, one-tenth of the true value - and try to run again? If this runs, then you know the full input motion is too much for the structure as modeled. If you get the same message at the same time step, then you know something else is wrong.
Best of luck saman.
When does the message appear? Before the analysis even gets into one step of the time history? Or after several time steps have executed?
Perhaps the input ground motion is truly excessive for the structure capacity you have modeled. If you simply change the load factor on the input acceleration to a small number - say, one-tenth of the true value - and try to run again? If this runs, then you know the full input motion is too much for the structure as modeled. If you get the same message at the same time step, then you know something else is wrong.
Best of luck saman.
Tim Huff
Re: Covergence Problem
Hi Huffte
I have done what you suggested. However, the same message appears after passing several time steps. if I use dt=0.01 this message will not appear, but it took several hours just for few steps. It is worth mentioning that I have utilized nonlinear link to model buckling in braces.
I have done what you suggested. However, the same message appears after passing several time steps. if I use dt=0.01 this message will not appear, but it took several hours just for few steps. It is worth mentioning that I have utilized nonlinear link to model buckling in braces.
Re: Covergence Problem
Several hours for only a few steps does sound like there is a problem somewhere saman.
In the Constraints tab of the Project Settings, are you using Penalty Functions or Lagrange multipliers? I have had success using nonlinear links with Penalty Functions and, as noted in the Help, some careful tweaking of the Penalty Function settings may be required.
If the model is not too large and you wish to post it here, I'll take a glance at it as I get time saman.
In the Constraints tab of the Project Settings, are you using Penalty Functions or Lagrange multipliers? I have had success using nonlinear links with Penalty Functions and, as noted in the Help, some careful tweaking of the Penalty Function settings may be required.
If the model is not too large and you wish to post it here, I'll take a glance at it as I get time saman.
Tim Huff
Re: Covergence Problem
Hi Huffte
I would appreciate if you take a look at my model.
I have sent it to you by Email.
I would appreciate if you take a look at my model.
I have sent it to you by Email.
Re: Covergence Problem
Hi saman.
I notice a few things in your model very quickly.
First, there seems to be a problem with your input acceleration history. The time step is highly erratic, varying from a minimum of 0.0001 seconds to a maximum of 0.208 seconds, with a total duration of 12.113 seconds in only 219 specified lines. This is highly unusual, and I am not certain that SeismoStruct would even read this file in properly, much less provide an accurate solution with such an erratic and poorly-defined input.
I am not sure where you obtained your input time history, but I highly recommend downloading properly filtered and baseline adjusted accelerograms from ground motion database sites such as the European Strong Motion Database (http://www.isesd.hi.is/ESD_Local/frameset.htm), the Center for Engineering Strong Motion Data (http://www.strongmotioncenter.org/) or PEER (http://peer.berkeley.edu/peer_ground_motion_database). These records have been properly filtered and baseline adjusted to eliminate artificial drift in the displacement history, and these records have uniform time steps.
Secondly, the model is highly nonlinear. It is not uncommon for such structures to exhibit numerical instabilities when one attempts to solve them with the damping matrix set to zero. I notice that you were trying to solve the system as undamped. While this is usually good practice, I think you will often find that a small amount of damping will make the solution proceed in a much more stable fashion.
Finally, I notice that you are, in fact employing penalty function constraints but that you have the exponents set to the maximum possible value in the Project Settings.
I changed your model to use a realistic input acceleration history, reduced the penalty function exponent, and added Rayleigh damping to the effect of 2% and it ran in only 17 minutes, with the results looking reasonable in the sense that frame hysteretic curves were the proper shape. Link hysteretic curves, however, appeared to be somewhat erratic in nature. Of course, you will have to be the one to judge the adequacy and accuracy of the results for your model.
But I do recommend the following:
1. Use an input time history which has a uniform time step, of adequate resolution, which has been baseline adjusted and filtered (Note that you may also use SeismoSignal or SeismoSpect to do the filtering and adjusting yourself. However, I have never seen an input time history with such erratic time steps as yours and I would not even attempt to use the input currently specified in your model).
2. Perform a sensitivity analysis with various damping options. It is recognized by many researchers that tangent-stiffness-proportional damping is the more accurate option relative to initial-stiffness-proportional damping. And it is also a contention among many experts that values much smaller than the typical 5% should be used - see the book by Priestley and papers by Priestley and Grant for very convincing arguments of this, in my opinion. So don’t get carried away with damping just to make a model more numerically stable. See Project-Settings --> Damping in the Help System for further excellent discussion on this topic.
3. Read the Help System section on Project Settings --> Constraints for more guidance on penalty function exponents.
4. Examine closely the hysteretic curves for the links and frame elements you have modeled to ensure that they make sense to you. You are the one who knows what you are attempting to model.
Best of luck saman. It’s an interesting project you have there.
I notice a few things in your model very quickly.
First, there seems to be a problem with your input acceleration history. The time step is highly erratic, varying from a minimum of 0.0001 seconds to a maximum of 0.208 seconds, with a total duration of 12.113 seconds in only 219 specified lines. This is highly unusual, and I am not certain that SeismoStruct would even read this file in properly, much less provide an accurate solution with such an erratic and poorly-defined input.
I am not sure where you obtained your input time history, but I highly recommend downloading properly filtered and baseline adjusted accelerograms from ground motion database sites such as the European Strong Motion Database (http://www.isesd.hi.is/ESD_Local/frameset.htm), the Center for Engineering Strong Motion Data (http://www.strongmotioncenter.org/) or PEER (http://peer.berkeley.edu/peer_ground_motion_database). These records have been properly filtered and baseline adjusted to eliminate artificial drift in the displacement history, and these records have uniform time steps.
Secondly, the model is highly nonlinear. It is not uncommon for such structures to exhibit numerical instabilities when one attempts to solve them with the damping matrix set to zero. I notice that you were trying to solve the system as undamped. While this is usually good practice, I think you will often find that a small amount of damping will make the solution proceed in a much more stable fashion.
Finally, I notice that you are, in fact employing penalty function constraints but that you have the exponents set to the maximum possible value in the Project Settings.
I changed your model to use a realistic input acceleration history, reduced the penalty function exponent, and added Rayleigh damping to the effect of 2% and it ran in only 17 minutes, with the results looking reasonable in the sense that frame hysteretic curves were the proper shape. Link hysteretic curves, however, appeared to be somewhat erratic in nature. Of course, you will have to be the one to judge the adequacy and accuracy of the results for your model.
But I do recommend the following:
1. Use an input time history which has a uniform time step, of adequate resolution, which has been baseline adjusted and filtered (Note that you may also use SeismoSignal or SeismoSpect to do the filtering and adjusting yourself. However, I have never seen an input time history with such erratic time steps as yours and I would not even attempt to use the input currently specified in your model).
2. Perform a sensitivity analysis with various damping options. It is recognized by many researchers that tangent-stiffness-proportional damping is the more accurate option relative to initial-stiffness-proportional damping. And it is also a contention among many experts that values much smaller than the typical 5% should be used - see the book by Priestley and papers by Priestley and Grant for very convincing arguments of this, in my opinion. So don’t get carried away with damping just to make a model more numerically stable. See Project-Settings --> Damping in the Help System for further excellent discussion on this topic.
3. Read the Help System section on Project Settings --> Constraints for more guidance on penalty function exponents.
4. Examine closely the hysteretic curves for the links and frame elements you have modeled to ensure that they make sense to you. You are the one who knows what you are attempting to model.
Best of luck saman. It’s an interesting project you have there.
Tim Huff
Re: Covergence Problem
Hi Huffte
Thank you a lot for your guidance. Based on your suggestions the required time of my analysis decreases. But, I think the axial force of braces are not appropriate. I think the axial force of each brace and the links attached to it should be the same. However, their axial forces are different.
Thank you a lot for your guidance. Based on your suggestions the required time of my analysis decreases. But, I think the axial force of braces are not appropriate. I think the axial force of each brace and the links attached to it should be the same. However, their axial forces are different.
Re: Covergence Problem
Yes. I see your point saman. It appears to me that there is some erratic behavior in the asymmetric links.
If I change your asymmetric links to symmetric links, then it appears to me that the link force and brace force do, in fact, match.
If I plot the force in the asymmetric link versus time, it appears to match the brace for positive force values only, and some unexpected behavior occurs for negative force values in the link.
Perhaps it might be wise if you could employ symmetric links and determine whether or not you agree with my assessment. In the event that you get the same conclusion as I do, it would be a good idea submit this issue to SeismoSoft in the "04-Unexpected behavior/errors" section of the forum and pursue the issue with the SeismoSoft team further.
I will, in addition, that I have had better luck with bilinear kinetic links than with some of the link types that are in your current model.
Nonetheless, I would have expected behavior similar to that you had expected rather than that indicated by the model results. Either we are both missing something or there is a bit of a bug in the asymmetric link algorithm.
Best of luck saman.
If I change your asymmetric links to symmetric links, then it appears to me that the link force and brace force do, in fact, match.
If I plot the force in the asymmetric link versus time, it appears to match the brace for positive force values only, and some unexpected behavior occurs for negative force values in the link.
Perhaps it might be wise if you could employ symmetric links and determine whether or not you agree with my assessment. In the event that you get the same conclusion as I do, it would be a good idea submit this issue to SeismoSoft in the "04-Unexpected behavior/errors" section of the forum and pursue the issue with the SeismoSoft team further.
I will, in addition, that I have had better luck with bilinear kinetic links than with some of the link types that are in your current model.
Nonetheless, I would have expected behavior similar to that you had expected rather than that indicated by the model results. Either we are both missing something or there is a bit of a bug in the asymmetric link algorithm.
Best of luck saman.
Tim Huff
Re: Covergence Problem
Hi Huffte
Thanks a lot. I would like to mention that the links works correctly in static pushover analysis.But, the seems to have problem in Time history analysis.
Best regard,
Thanks a lot. I would like to mention that the links works correctly in static pushover analysis.But, the seems to have problem in Time history analysis.
Best regard,
-
Stelios_Antoniou
- Posts: 89
- Joined: 17 Jul 2011, 20:08
Re: Covergence Problem
Hi Saman and Huffte,
We have carried out some initial tests with the asymmetric curves (bl_asm & trl_asm) and things seem to be working correctly,
however this does not mean necessarily that there is not a problem somewhere (i.e. there might be a bug that we could not spot).
Hence, can one of you send me the file that gives the problems you described above?
Regarding the model itself, can it be that you are using a loose convergence criterion (i.e. displacement-based only with a relatively large convergence value)? In the highly inelastic range, as in your case, differences in the forces may occur.
Seismosoft Developer
We have carried out some initial tests with the asymmetric curves (bl_asm & trl_asm) and things seem to be working correctly,
however this does not mean necessarily that there is not a problem somewhere (i.e. there might be a bug that we could not spot).
Hence, can one of you send me the file that gives the problems you described above?
Regarding the model itself, can it be that you are using a loose convergence criterion (i.e. displacement-based only with a relatively large convergence value)? In the highly inelastic range, as in your case, differences in the forces may occur.
Seismosoft Developer