Those strain softening constitutive models we like to use for concrete can lead to material nonlinearity that isolates, or localizes, in a single element or single integration point of your model. The resulting global response becomes non-objective, non-unique, or mesh-dependent.
The objective of this post is to show the localization problem then describe solutions that are available in OpenSees for force-based elements. I don’t want to get into details, I just want to show how different regularization techniques can improve localized response.
I will demonstrate localization with the reinforced concrete bridge pier shown below. Yes, it’s a well-worn example, but if I did anything more, this piece would be a journal article and not a blog post. The unconfined concrete strength is 32 MPa, and due to confinement, the core concrete strength is 39 MPa. The reinforcing steel strength is 510 MPa.
Before we get to force-based elements, let’s take a look at the response with displacement-based elements. As shown in a previous post, mesh refinement with displacement-based elements should lead to a convergent solution. However, that is not the case here. Although, as expected, the peak base shear decreases with more elements, the post-peak response is mesh dependent.
Now, turning to force-based elements, using more integration points should lead to a convergent solution. As shown below, the peak base shear is the same for all cases, but the post-peak response takes different paths as the number of Gauss-Lobatto integration points changes.
Some kind of regularization is required to obtain objective global response with force-based elements. I will show three regularization approaches.
The first approach proposed by Coleman and Spacone (2001) modifies the ultimate strain of the concrete material based on the number of integration points in order to maintain constant fracture energy release. As shown below, the results line up pretty well.
The second approach proposed by Scott and Fenves (2006) and similarly by Addessi and Ciampi (2007) modifies the integration weight at the end of the element. This allows the analyst to use an observed or empirically based plastic hinge length instead of whatever Gauss-Lobatto dictates. As shown below, the results match the first local approach.
The third regularization approach is a recent addition to OpenSees. This formulation by Sideris and Salehi (2016), implemented in the GradientInelasticBeamColumn element class, uses a set of gradient-based relations to ensure continuity of section deformations over the element length. The results match the two local approaches above.
Other formulations for force-based elements were developed by Valipour and Foster (2009) and Feng et al (2015). There is also a formulation for displacement-based elements by Kenawy et al (2020). These formulations are not (yet) available in OpenSees. If you’d like to implement them, let me know!
21 thoughts on “The Basics of Frame Element Localization”
Nice post! I like it.
Professor of Civil & Environmental Engineering
University of Washington
LikeLiked by 1 person
Thanks, Dawn 🙂
I think the gradientInelasticBeamColumn uses a lot of integration points along the element (high computational cost).
Yes, it can require a large number of integration points in some cases. For this example, I used 7 points and even got similar results with 5.
I would like to try Feng et al (2015).
Do you mean you want to try implementing the model? If so, before you start writing any code, I can ask the authors if they have implemented it in OpenSees.
Please let us know if they have implemented it in OpenSees.
Thanks Prof., let us know if they have implemented it in OpenSees. I read their paper. They have used in Abaqus.
Is there is any workaround for short force-based beam elements using “modified Gauss-Radau plastic hinge integration” when 4(lp+lj)>L?
To my understanding this condition results in negative integration weights of the two interior integration points. (Although I have seen people using Lp values to result in this condition).
There’s several integration options, one of which may be a workaround.
Click to access IntegrationTypes.pdf
You may want to look at “Regularized” hinge integration.
Thank you for the information. The Regularized Hinge Integration method discussed in  seems useful.
LikeLiked by 1 person
Did you use only one element of gradientInelasticBeamColumn with 7 points?
Yeah, Np=7 Simpson integration
Thank you for your post.
When we use Coleman method (first method) global behavior is good but local (M-curvature) is dependent on a number of integration points.Coleman provides post-process when LIP is not equal to Lplastic so M-curvature will be objective as well but when we ask for strain in the first section, it is calculated base on non-objective Moment-curvature ! so how can we trust the results ? (because I want to analyze section behaviour)
2- can I use the second method (scott) and use the regularization method both together to have objective global and local behavior?
Thanks in advanced
Yes, objective global and local behavior is the advantage of 2) over 1)
Thank you for your response.
One more question, about the boundary element in the shear walls, Do I have to consider all the boundary elements as confined concrete (In Regularization methods)?
You should consult the literature to see how people model shear walls in OpenSees.