Curvature on forcebeamcolumn elements

[apapaso]

I am using forcebeamcolumn elements to model the columns of a RC structure, for performing NL IDA analysis.
Subsequently, I use the curvature at the ends of the columns (sections 1 & N) to calculate a damage index.

However, I am greatly concerned about the variation of the curvature values calculated by opensses, regarding the number of the integration points:
- for small deformations (until the yield point), I get identical curvature values, regarless of the number o IPs.
- after yielding, curvature values diverge significantly, and the more the IPs, the grater the curvature values.

The largest values, obtained with 10 IPs (-integration Lobatto), are clearly erroneous (E.g. 1st floor column base curvature 0,44 for interstory drift 3%).

After reading some papers about forcebeamcolumn integration methods (like M. Scott & G. Fenves 2004), I am still confused.

How can I get reliable curvature values for calculating damage indices?

[selimgunay]

In the analyses with different number of integration points, are you using Lobatto? Also are you comparing the curvatures at the same integration points at the element ends?

[apapaso]

Indeed, I am using Lobatto. I always compare the curvatures at the first and the last section of the element.
(command: [lindex [eleResponse $i section 1 deformation] 1] -for the first section)
I understand that Radau is considered more accurate, but I 'd prefer a method which calculates internally the plastic hinge length Lp.

[EricsonEncinaZ]

If your elements have an almost EPP or softening behaviour then you may be having localisation issues, which are known to occur in force based elements http://ascelibrary.org/doi/abs/10.1061/ ... %281257%29. You can either use regularisation techniques for your materials and/or postprocessing your data to obtain objective results or you may use displacement based elements.

It is also important to note that a force based element assumes a linear moment distribution, unlike a displacement based element that assumes linear curvature distribution. This means that if you have a linear moment distribution (like columns and walls in buildings) you just need 1 force based element per level, regardless of the curvature distribution; but if you use DB elements then discretisation is quite important to accurately capture the NL curvature distribution.

I hope the paper above expands more the idea.

[apapaso]

Thank you for your reply, you have been most helpful.