Hi,
I keep on getting non-convergence problems with my analyses, and cannot find a way of solving them. Can you please help?
Thank you.
How to solve non-convergence problems
-
seismosoft
- Posts: 1272
- Joined: 06 Jul 2007, 04:55
Re: How to solve non-convergence problems
Hi,
To start with, you should read very carefully the Help System, to make sure that you are modelling your structure correctly, paying particular attention to the numerous hints&tips (see 'Notes'), since these often relate to convergence difficulty issues.
Secondly, you should use the 'Search' facility of this Forum, to find past discussions on convergence issues.
Thirdly, you may consider the following suggestions:
- make sure the applied loads are not so large as to cause the structure to reach collapse (e.g. don't impose drifts larger than 3-5%)
- do not use an exaggerated number of load increments, since these may render the analyses unnecessarily slow and can also give rise to convergence difficulties (a load increment that is too small can be counterproductive). E.g. 50 steps in pushover analysis should suffice, as should a dt of 0.01s in dynamic analysis
- avoid the the introduction of diaphragm constraints when modelling of the slab is not really needed (e.g. relatively regular plan and well distributed loading).
- use an adequate number of fibres in elements with large sections (e.g. 500-1000 fibres in walls)
- use a smaller number of integration points (2-3) for short force-based elements (you can actually use displacement-based elements in such cases)
- increase number of force-based element internal iterations to 1000 and relax corresponding convergence criterion to 1e-4
In tandem with the above, consider the following standard model simplification procedure that may be of assistance in tracking the origin of your difficulties:
- backup the original model and then start simplifying it until the problem either gets obvious or disappears
- remove the masses
- change the force-based frame elements to the simpler more stable displacement-based ones
- remove the rigid diaphragms
- gradually decrease the model, removing columns and beams until only the core remains
- gradually remove the storeys
- eliminate element subdivision or discretisation
At every change save the model and try to run it. If suddenly the model runs, then the problem is caused by the parts just deleted. If the divergence problems occur even with the smaller model than it might be a problem of the settings (iterations, convergence criteria, etc).
Seismosoft Support
To start with, you should read very carefully the Help System, to make sure that you are modelling your structure correctly, paying particular attention to the numerous hints&tips (see 'Notes'), since these often relate to convergence difficulty issues.
Secondly, you should use the 'Search' facility of this Forum, to find past discussions on convergence issues.
Thirdly, you may consider the following suggestions:
- make sure the applied loads are not so large as to cause the structure to reach collapse (e.g. don't impose drifts larger than 3-5%)
- do not use an exaggerated number of load increments, since these may render the analyses unnecessarily slow and can also give rise to convergence difficulties (a load increment that is too small can be counterproductive). E.g. 50 steps in pushover analysis should suffice, as should a dt of 0.01s in dynamic analysis
- avoid the the introduction of diaphragm constraints when modelling of the slab is not really needed (e.g. relatively regular plan and well distributed loading).
- use an adequate number of fibres in elements with large sections (e.g. 500-1000 fibres in walls)
- use a smaller number of integration points (2-3) for short force-based elements (you can actually use displacement-based elements in such cases)
- increase number of force-based element internal iterations to 1000 and relax corresponding convergence criterion to 1e-4
In tandem with the above, consider the following standard model simplification procedure that may be of assistance in tracking the origin of your difficulties:
- backup the original model and then start simplifying it until the problem either gets obvious or disappears
- remove the masses
- change the force-based frame elements to the simpler more stable displacement-based ones
- remove the rigid diaphragms
- gradually decrease the model, removing columns and beams until only the core remains
- gradually remove the storeys
- eliminate element subdivision or discretisation
At every change save the model and try to run it. If suddenly the model runs, then the problem is caused by the parts just deleted. If the divergence problems occur even with the smaller model than it might be a problem of the settings (iterations, convergence criteria, etc).
Seismosoft Support