Stelios,
Thanks very much. I have sent both the original file I received from saman as well as the version which I modified in an attempt to fix problems with input motion, damping, and penalty function settings.
Regards,
huffte
Covergence Problem
Re: Covergence Problem
Tim Huff
-
Stelios_Antoniou
- Posts: 89
- Joined: 17 Jul 2011, 20:08
Re: Covergence Problem
Tim and Saman hi,
Thank you for the models, I had a look at them and found out the following:
1) there is no particular problem with the asymmetric curves, the same unexpected behaviour occurs with symmetric curves as well, when an adequately low yield value is input
2) When force equilibrium is sought for, it is highly recommended to employ a displacement+force convergence criterion. In the current model, since there are rigid links, a force-based criterion would lead to numerical difficulties, hence I would recommend more stringent displacement criteria (1/10 or the current ones). Preferably, you could simply remove the EqualDOF links that connect the floor nodes. Since the axial rigidity of the beams is very large with respect to the bending rigidity of the vertical components, the rigid links are not really necessary. However, that was not the main problem with the model and such change did not solve the problematic behaviour.
3) The differences in the nodal forces that you observed are caused by the introduction of Rayleigh damping, with period values that are different from the 1st and 2nd period value of the model. If one removes the damping (viscous damping is not really necessary, since the hysteretic damping is considerable and would dominate the structural behaviour) or inputs reasonable period values, the differences will disappear
4) In brief, what I would do to improve the model is: (i) remove damping, (ii) remove the rigid links, (iii) impose a loose force based criterion to control the forces (10 times larger than the existing values) and a more stringent displacement based criterion (1/10 of the existing values), (iv) since the yielding in the links causes serious instability problems, I would also apply a 5-10% hardening ratio both in the negative and positive direction. I think with all these the analysis will run effectively.
Hope this will help.
Thank you for the models, I had a look at them and found out the following:
1) there is no particular problem with the asymmetric curves, the same unexpected behaviour occurs with symmetric curves as well, when an adequately low yield value is input
2) When force equilibrium is sought for, it is highly recommended to employ a displacement+force convergence criterion. In the current model, since there are rigid links, a force-based criterion would lead to numerical difficulties, hence I would recommend more stringent displacement criteria (1/10 or the current ones). Preferably, you could simply remove the EqualDOF links that connect the floor nodes. Since the axial rigidity of the beams is very large with respect to the bending rigidity of the vertical components, the rigid links are not really necessary. However, that was not the main problem with the model and such change did not solve the problematic behaviour.
3) The differences in the nodal forces that you observed are caused by the introduction of Rayleigh damping, with period values that are different from the 1st and 2nd period value of the model. If one removes the damping (viscous damping is not really necessary, since the hysteretic damping is considerable and would dominate the structural behaviour) or inputs reasonable period values, the differences will disappear
4) In brief, what I would do to improve the model is: (i) remove damping, (ii) remove the rigid links, (iii) impose a loose force based criterion to control the forces (10 times larger than the existing values) and a more stringent displacement based criterion (1/10 of the existing values), (iv) since the yielding in the links causes serious instability problems, I would also apply a 5-10% hardening ratio both in the negative and positive direction. I think with all these the analysis will run effectively.
Hope this will help.
-
Stelios_Antoniou
- Posts: 89
- Joined: 17 Jul 2011, 20:08
Re: Covergence Problem
I carried out another sample analysis with the model and it turns out that a strain hardening parameter of 0.5-1.0% would be enough for stability. No need for larger values