Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Finite-Element Limit Analysis for Solid

Modeling of Reinforced Concrete


Mads Emil Møller Andersen 1; Peter Noe Poulsen 2; and John Forbes Olesen 3
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

Abstract: Complex triaxial stress states are present in many reinforced concrete structures. These structures are often analyzed using simple
hand calculations based on methods designed for plane structures. However, this can result in designs that are inefficient and excessive in
material usage. This paper introduces finite-element limit analysis (FELA) for modeling of reinforced concrete structures, with separate model-
ing of concrete and reinforcement in a so far unseen scale. The method provides results for the capacity as well as the stress state and failure
mechanism in the ultimate limit state. The FELA framework uses solid elements together with the modified Mohr-Coulomb and von Mises yield
criteria. The framework uses a computationally inexpensive tetrahedral FELA element, which makes it possible to model rebar details in three
dimensions (3D) with adequate discretization. The framework is demonstrated in two examples, one for verification and one showing the
practical use of the framework by analyzing a tension connection with overlapping U-bars. The numerical results are compared with the failure
mechanism and capacity from experiments. DOI: 10.1061/(ASCE)ST.1943-541X.0002979. © 2021 American Society of Civil Engineers.

Introduction by Lundgren (1949), and the strip method by Hillerborg (1974),


which are all lower-bound methods, and the yield-line method by
Reinforced concrete is a composite material consisting of reinforce- Johansen (1943, 1962), which is an upper-bound method. The de-
ment bars embedded in concrete. The design of reinforced concrete sign of reinforcement details is primarily based on the strut-and-tie
can, to a large degree, be simplified using assumptions of plane method and on failure mechanisms.
stress or plane strain. However, for massive concrete structures and Design and validation of reinforced concrete details can also be
rebar detailing in, for example, connections between beams and performed by using advanced nonlinear stiffness-based methods.
columns or slabs and walls, this assumption does not hold. To These methods use advanced constitutive elastoplastic models to
properly understand how these structures carry load and how they capture the behavior of structures subject to loading. The calculations
collapse, solid modeling with a triaxial stress states is required. are iterative, and the stepping is usually based on either load or de-
Currently, these structural members are designed based on codes formation. This provides detailed results about the behavior of the
and guidelines, but naturally, codes and guidelines cannot describe structures all the way to collapse as well as the limit load. Vecchio
detailing arising from all possible structural problems. This paper and Selby (1991) promoted this type of modeling for structures with
presents a novel approach for solid modeling of reinforced concrete triaxial stress states, and since then, research has been ongoing. Re-
in the ultimate limit state with complex triaxial stress states, where cent results by Gimeno (2017) showed modeling of reinforced
reinforcement bars are considered as discrete components. concrete with discrete modeling of reinforcement for an anchorage
Currently, limit analysis using hand calculations, based on block. Also, general-purpose finite-element method (FEM) pro-
idealized and simplified models, is the norm when validating rein- grams exist that can do this type of calculation, e.g., ATENA
forcement detailing. The methods are based on the lower- and (Červenka and Červenka 2017) and DIANA (Ferreira 2020).
upper-bound theorems, as first suggested by Gvozdev (1960) and Common for these methods and programs is that they are very
independently by Drucker et al. (1952). However, before the the- powerful; however, they require many additional physical param-
orems were formally written down, methods based on the ideas
eters, which are difficult to access. Finite-element limit analysis
had already been developed by researchers and engineers. These
(FELA) differs from these approaches by being based on a rigid-
methods were based on a thorough understanding of the structures
plastic material model, i.e., the model is based on stress equilibrium
rather than on a strict mathematical formulation. Examples of these
and not on a deformation/stiffness-based approach. This means that
methods include the truss model by Mörsch (1922) [now called
only information about the limit state is provided by the analysis;
the strut-and-tie method (Schlaich et al. 1987)], the stringer method
however, it also means that the material models simplify to the
1 modeling of yield surfaces, which are usually well-defined by only
Ph.D. Student, Dept. of Bridges International, COWI A/S, Parallelvej 2,
Kongens Lyngby 2800, Denmark (corresponding author). ORCID: https:// a few material parameters.
orcid.org/0000-0002-1501-2464. Email: mean@cowi.com; maema@byg FELA, which utilizes either the principle of the upper- or lower-
.dtu.dk bound theorem for rigid plasticity in a numerical framework, has
2
Associate Professor, Dept. of Civil Engineering, Technical Univ. of reached a mature state for plane-stress reinforced concrete struc-
Denmark, Brovej 118, Kongens Lyngby 2800, Denmark. tures. The method was first proposed by Anderheggen and Knöpfel
3
Associate Professor, Dept. of Civil Engineering, Technical Univ. of (1972) and has since developed into an active field of research.
Denmark, Brovej 118, Kongens Lyngby 2800, Denmark. ORCID: Plane structures subject to in-plane forces were treated by Poulsen
https://orcid.org/0000-0001-6695-7719
and Damkilde (2000), who presented a framework for modeling re-
Note. This manuscript was submitted on May 1, 2020; approved on
December 2, 2020; published online on March 4, 2021. Discussion period inforced concrete structures. The framework made use of a triangu-
open until August 4, 2021; separate discussions must be submitted for in- lar element in conjunction with a bar and a beam element and also
dividual papers. This paper is part of the Journal of Structural Engineer- facilitated material optimization. Further work by Larsen (2010),
ing, © ASCE, ISSN 0733-9445. and Herfelt (2017) has made the method commercially viable,

© ASCE 04021051-1 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


and the first commercial program for verification of reinforced con- on the extremum principals (Gvozdev 1960; Drucker et al. 1952),
crete structures subjected to in-plane forces was launched in 2019 which is also the case for hand calculations based on the rigid-
(Herfelt et al. 2019). plastic material model. The method can be based on either the
FELA, with solid modeling, using the Mohr-Coulomb yield cri- upper- or lower-bound theorem. If the analysis is based on the upper-
terion, was introduced by Martin and Makrodimopoulos (2008), bound theorem, the elements describe a kinematically admissible
who used semidefinite convex optimization to model the Mohr- state of collapse, and if the lower-bound theorem is used, the ele-
Coulomb criterion. Larsen et al. (2009) extended the approach ments describe a statically admissible stress field.
to include smeared reinforcement by using additional linear con- The FELA implementation described in this paper is based on
straints. Characteristic of the two studies was that the underlying the lower-bound theorem. The variables of the elements are stresses,
optimization algorithms and the available hardware were not suf- and equilibrium between elements is enforced through stresses on
ficient to provide detailed analysis. Recently, work by Vincent et al. the element surfaces. The equilibrium can be described through
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

(2018) has shown examples of large-scale modeling of a pier head, a matrix vector product Hσ, where σ is a vector containing the stress
indicating a possible shift in this respect. Very few studies have variables of the elements, and H is the so-called equilibrium ma-
made attempts at modeling concrete and steel separately, with the trix. The equilibrium matrix has contributions from each individual
examples being simple. A typical example is the pullout of a element. The element equilibrium matrix, hel , is described sub-
reinforcement rod embedded in concrete (Herfelt 2017). sequently. The stresses of the elements must also obey the limita-
A fine-mesh discretization is required in order to model, in de- tions prescribed by the yield conditions of the materials. The
tail, the state of stress in reinforced concrete structures at the ulti- limitation of the material model is ensured in the stress points of the
mate limit state. The present work aims at giving a detailed account element by the inequality f i ðσi Þ ≤ 0 for i ¼ 1; : : : ; n where fi is
of the stress state in both concrete and reinforcing steel, which are the yield surface and σi is the stress state for the ith stress point,
modeled as separate parts. Therefore, it takes a large number of which is enforced in all n stress points. With equilibrium established
elements to model even small parts of a structure to study the local and the stress of the elements limited by a yield function, the loading
effects of reinforcement detailing. Thus, it is crucial to allow for the in the model can be optimized. The optimization problem may then
largest possible number of elements by applying a very simple solid be formulated
element, namely the four-node tetrahedral element with a constant
stress field. This is done although the element is known to perform maximize λ
badly in other circumstances, e.g., exhibiting volumetric locking subject to Hσ ¼ Rλ þ R0
when applied as a displacement-compatible element in cases of
zero, or near-zero, dilatation. fi ðσi Þ ≤ 0; i ¼ 1; : : : ; n ð1Þ
In the present FELA setting, the element is based on the mixed-
where the loading terms R and R0 are introduced for scalable and
formulation (Borges et al. 1996; Herfelt 2017), meaning that the
constant loads, respectively. Also the load factor, λ, is introduced,
requirement of stress equilibrium along element surfaces was re-
which is the optimization parameter. In this case, the optimization is
laxed, and instead, equilibrium was enforced in corner nodes. The
a maximization because the implementation is based on the lower-
mixed formulation was used to eliminate the problem of locking and
bound theorem. The optimal solution is the maximum lower-bound
has also been shown to enhance mesh convergence, as reported by
solution.
Krabbenhøft et al. (2007a, c) and Herfelt (2017). The element may
In the present work, the yield surfaces are either described by the
produce spurious hydrostatic stress states, which may corrupt the
modified Mohr-Coulomb yield criterion using semidefinite pro-
solution (Andersen and Koch 2018). However, in the present appli- gramming (Martin and Makrodimopoulos 2008; Krabbenhøft et al.
cation, reinforcing steel primarily experiences uniaxial stresses, and 2007a; Bisbos and Pardalos 2007; Krabbenhøft et al. 2007b), or
concrete hydrostatic stresses are limited by the low tensile capacity. von Mises yield criterion using second-order cone programming
Therefore, a limitation on the magnitude of the hydrostatic stress (Bisbos et al. 2005; Makrodimopoulos and Martin 2006). When
component for steel is introduced, which is demonstrated to elimi- implementing yield surfaces defined by either of these methods,
nate the spurious states of stress, and the element is demonstrated to additional linear equations and auxiliary variables α are needed.
produce reliable results for structural entities in so-far-unseen detail. These are described subsequently in this paper, together with the
This study sets out to investigate the usefulness of FELA to de- final form of the optimization problem.
termine the capacity of reinforced concrete with discrete modeling
of rebar details, the perspective being the enhanced ability of the
engineer to design complicated reinforcement details with adequate
capacity. As a case study, a connection detail between two precast
Constant Stress Tetrahedron
concrete slabs using overlapping U-bars is analyzed. The connec- Solid finite-element limit analysis can be a computationally de-
tion has been the subject of an experimental study by Joergensen manding task due to the nature of optimization and generally large
and Hoang (2013). The experimental series was designed to produce problem sizes for three-dimensional (3D) meshes, and especially
failure primarily in the concrete core. For this reason, the concrete for the purposes it is utilized in the paper. For these reasons, a sim-
effectiveness factor is important for the calibration of the model, ple and computationally inexpensive element with a constant triax-
and the experimental results are used to give an estimate of this ial stress state is used. The element is shown in Fig. 1, with the
parameter. Furthermore, the resulting failure mechanisms from the normal vector of Surface 1 shown, as well as the force components
numerical calculations are compared with the experimental findings. of Node 1. The following stress vector specifies the triaxial stress
state of the element:

Finite-Element Limit Analysis for Solids σTel ¼ ½ σxx σyy σzz σxy σxz σzy  ð2Þ

FELA is a combination of the domain discretization of the finite- The element, in the shape of a tetrahedron, is well-suited for mod-
element method and rigid-plastic limit analysis. The method was eling complex geometries. The element contains one stress point
first proposed by Anderheggen and Knöpfel (1972). FELA is based and thus describes a constant stress state. In the strict lower-bound

© ASCE 04021051-2 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


This is done by first computing the stresses on the element surface
and then distributing it equally into each of the three corner nodes.
The stresses ts for surface s are found from the relation ts ¼ σns ,
where σ is the Cauchy stress tensor and ns is the unit normal vector
of the surface. The transformation to the desired stress vector no-
tation presented in Eq. (2) can be performed by introducing the
stress equality
Fig. 1. Constant-stress tetrahedron.
σns ¼ Ps σel ;
2 3
ns;x 0 0 ns;y ns;z 0
60 0 0 ns;z 7
Ps ¼ 4 5 ð4Þ
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

formulation, interelement equilibrium needs to be ensured on ele- ns;y ns;x


ment surfaces. However, applying an element with a constant tri- 0 0 ns;z 0 ns;x ns;y
axial stress state, the resulting stress state would lock. The locking
happens because the surface stress equilibrium requires six equa- where Ps is the element stress to surface stress matrix. Because the
tions for the element. Each surface requires one equation for normal stress on each surface is constant due to the constant stress field, the
equilibrium and two for shear equilibrium; this amounts to a total of contribution to the equilibrium force in the corner nodes is a third of
12 equations for the element. However, because each surface in the the total force on the surface [Eq. (3)]. The total stress to nodal
interior is shared by two elements, the number of equations can be force matrix, P̂s , is then given by
divided by two, yielding an average close to six equations required
for surface stress equilibrium. With six stress variables per element As
qi;s ¼ P σ ¼ P̂s σel ð5Þ
this yields a determinate system of linear equations, i.e., there are 3 s el
no free parameters to optimize.
Instead of enforcing surface stress continuity directly between where qi;s is the force vector of Node i from Surface s; and
adjoining elements, force equilibrium is enforced in the corner no- As = surface area of Surface s. Each element surface contributes
des of the elements. The modified element has the same amount of to the nodal forces in its corner nodes. With the surfaces enumer-
equations per element (three equations in four nodes); however, on ated by the opposing node, the equilibrium for the entire element is
average, the equations are now shared by more than two elements. ensured by
This yields an indeterminate system of equation, which removes 2 3
the problem of locking. The modification is statically equivalent P̂2 þ P̂3 þ P̂4
6 7
to the original problem. However, equilibrium is now enforced 6 P̂1 þ P̂3 þ P̂4 7
in an average sense, which is a relaxation of the strict lower-bound qel ¼ 66
7σel ¼ hel σel
7 ð6Þ
requirements. With the relaxation, the element is based on a so- 4 P̂1 þ P̂2 þ P̂4 5
called mixed formulation. Elements of the mixed formulation type P̂1 þ P̂2 þ P̂3
interpolate stresses with shape functions. The contribution from
stresses on an element surface to the nodal forces are then found where the element equilibrium matrix hel is defined implicitly.
from the integration over the surface area As as follows:
Z
qs ¼ NTs ts dAs ð3Þ Yield Conditions of the Materials
As
In the FELA framework, the material models are defined by yield
where qs = contribution from surface s to the nodal forces; Ns is a surfaces. The modified Mohr-Coulomb and von Mises yield sur-
linear interpolation matrix on the surface; and ts is the surface stress face are used for concrete and steel, respectively. The modified
vector. Eq. (3) is the general formulation because the present for- Mohr-Coulomb yield criterion requires the use of semidefinite pro-
mulation this simplifies into an even distribution into the three cor- gramming because the yield criterion cannot be written using the
ner nodes. For elements of the mixed formulation, it is not known in general stress parameters in the triaxial stress state, but requires the
advance whether the failure load will converge from above or be- use of principal stresses. This is not the case for the von Mises cri-
low with increased mesh discretization. Examples show that for the terion, which is why the computationally cheaper method of second-
present formulation, the failure load converges from above. This order cone programming (Boyd and Vandenberghe 2004) can be
means that the load-carrying capacity is overestimated until the fail- used. The implementations of the yield surfaces are described in
ure load is converged. the following section.
The aforementioned relaxation has the unfortunate effect that
spurious modes of equilibrium are possible for material models
without hydrostatic stress limitations, which is the case for the von Modified Mohr-Coulomb Yield Criterion for Concrete
Mises yield criterion. The problem is overcome by a slight modi-
The modified Mohr-Coulomb yield criterion has a friction and a
fication of the implementation of the von Mises yield criterion,
separation yield criteria expressed by the following formulas:
which is described subsequently. The relaxation does not pose any
problem for the modified Mohr-Coulomb yield criterion due to the σ1 ≤ f t
limited tensile strength.
kσ1 − σ3 ≤ fc ð7Þ

Equilibrium Matrix where σ1 ≥ σ2 ≥ σ3 = principal stresses; f t = concrete tensile


strength; f c = concrete compressive strength; and k = frictional
In the present formulation, the equilibrium matrix transforms the parameter, which for concrete is usually taken as k ¼ 4 (Nielsen
stress state of the element into equivalent forces in the corner nodes. and Hoang 2011).

© ASCE 04021051-3 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


When a rigid-plastic material model is used for concrete, which 2 3
0 0 0 0 0 0 1 0 0 0 0 0 0
is an approximation of the real material behavior, the strength needs 6 1 1 7
6 − pffiffiffi pffiffiffi 0 0 0 0 0 1 0 0 0 0 07
to be reduced. The reduction is made by the so-called concrete 6 7
6 2 2 7
effectiveness factors in the following way (Nielsen and Hoang 6 7
6 1 1 7
2011): 60 − pffiffiffi pffiffiffi 0 0 0 0 0 1 0 0 0 07
6 2 2 7
6 7
6 1 1 7
f c;eff ¼ νfc ð8Þ 6 pffiffiffi 7
6 0 − pffiffiffi 0 0 0 0 0 0 1 0 0 07
6 2 2 7
6 pffiffiffi 7
f t;eff ¼ ν t f t ð9Þ 6 7
60 0 0 − 3 0 0 0 0 0 0 1 0 07
6 pffiffiffi 7
60 0 0 0 − 3 0 0 0 0 0 0 1 07
where ν and ν t = effectiveness factors on the compressive and ten- 4 5
pffiffiffi
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

sile strengths, respectively. These factors are case-specific and are 0 0 0 0 0 − 3 0 0 0 0 0 0 1


based on empirical results. 2 3
σxx
Larsen (2010) showed [based on work by Martin and Makro- 6 7
dimopoulos (2008), Krabbenhøft et al. (2007a), and Bisbos and 6 σyy 7
6 7
Pardalos (2007)] that the criterion in Eq. (7) can be cast as two 6σ 7
6 zz 7
semidefinite constraints and two inequalities by use of two aux- 6 7 2 3
6 σxy 7 fy
iliary variables α1 and α2 6 7
6 7 6 7
6 σxz 7 6 0 7
6 7 6 7
6 7 6 7
σ þ ðkα1 ÞI ⪯ 0 6 σyz 7 6 0 7
6 7 6 7
σ − α2 I ⪰ 0 6 7 6 7
× 6 α1 7 ¼ 6 0 7 ð14Þ
6 7 6 7
α2 ≤ ft;eff 6α 7 60 7
6 2 7 6 7
6 7 6 7
α1 þ α2 ≤ f c;eff =k ð10Þ 6α 7 60 7
6 3 7 4 5
6 7
6 α4 7 0
6 7
where I is the identity matrix of order 3; and σ is the Cauchy stress 6 7
6 α5 7
tensor. The operators ⪯ and ⪰ are symbols for linear matrix 6 7
6 7
inequalities implying negative and positive semidefiniteness, 4 α6 5
respectively. α7

The von Mises yield criterion for a triaxial stress state can then
Triaxial von Mises Yield Criterion for Steel
be expressed in the form of the second-order cone [Eq. (12)]
The von Mises yield criterion for a triaxial stress state can be
written in the following form: CvM ðαÞ ¼ α22 þ α23 þ α24 þ α25 þ α26 þ α27 − α21 ¼ 0 ð15Þ

fðσxx ; σyy ; σzz ; σxy ; σxz ; σyz Þ where αT ¼ ½α1 ; α2 ; α3 ; α4 ; α5 ; α6 ; α7 . It is required that a given
stress state must be within or on the yield surface defined by the
1 basis second-order cone, i.e., CvM ≤ 0.
¼ ½ðσxx − σyy Þ2 þ ðσyy − σzz Þ2 þ ðσzz − σxx Þ2
2 Due to the relaxation of the stress equilibrium, as described pre-
þ 6ðσ2xy þ σ2xz þ σ2yz Þ − f2y ¼ 0 ð11Þ viously, the element is able to establish spurious stress states for the
von Mises yield criterion, i.e., equilibrium in the corner nodes is not
violated, and elements contributing to the equilibrium have very
where f y = material uniaxial yield stress. Considering the stress
large negative or positive hydrostatic stress states, respectively. This
deviations between the normal stresses as single components, it
shortcoming is solved by imposing an additional linear inequality on
can be seen that the yield criterion can be formulated as a second-
each individual stress parameter governed by the von Mises yield
order cone (Bisbos and Pardalos 2007)
criterion, in the following form:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Cðx1 ; x2 ; : : : ; xk Þ ¼ x22 þ x23 þ · · · þx2k ≤ x1 ð12Þ −χfy ≤ σ ≤ χf y ð16Þ

where χ ¼ 1 in the most simple case. For the examples studied in


The first six quadratic components need to be scaled to fit the this paper, this is a fair assumption.
formulation of Eq. (12). Seven auxiliary variables accounting for
this scaling are defined according to the number of unique stress
components and the material yield strength Interface between Material Models
No special considerations have been made in the modeling of the
σxx − σyy σyy − σzz σ −σ interface between rebars and the surrounding concrete. Therefore,
α1 ¼ f y ; α2 ¼ pffiffiffi ; α3 ¼ pffiffiffi ; α4 ¼ zzpffiffiffi xx
2 2 2 the model assumes a perfect bond at the interface. The perfect bond
pffiffiffi pffiffiffi pffiffiffi implies that the interface can only fail by a failure of either the modi-
α5 ¼ 3σxy ; α6 ¼ 3σxz ; α7 ¼ 3σyz ð13Þ
fied Mohr-Coulomb or von Mises yield criterion on either side of
the interface, which in practice would mean failure in the concrete.
The equations listed in Eq. (13) can be stated by the following This model would not be accurate for smooth rebars, which would
constraints matrix: have a low-strength bond to the surrounding concrete and would

© ASCE 04021051-4 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


experience a slip failure if not mechanically anchored. However, if
ribbed rebars are considered, some mechanical bond will be present
even for a straight bar, and it is reasonable to assume that a transfer
of forces is possible between the rebar and the surrounding concrete
due to actions by the ribs.

Detailed Optimization Problem


With the equilibrium matrix H and the additional linear equations
from the yield surfaces of the materials defined, the final assembly
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

of the optimization problem can be made. In the implementation


the optimization, variables can be separated into three different cat-
egories: the element stress variables σ, the load factor λ, and the
auxiliary variables α. The full system of linear equations is then
given by

maximize λ
2 32 σ 3 2 3 Fig. 2. Concrete cylinder in uniaxial compression with a steel tube
H −R 0 1 ¼ R0 providing confinement.
6 76 .. 7 6 7
6 76 . 77 6 7
6 766 7 6 7
6 76 7 6 7
6 76 σ n 7 6 7
6 76 7 6 7
6 7
Aα;eq 76 6 7
subject to 6 Aσ;eq 0 6 λ 7 7 ¼ 6 beq 7 Table 1. Material properties used in the compression test
6 76 7 6 7
6 76 α 7 6 7 Parameter Value
6 76 1 7 6 7
6 76 7 6 7
6 76 .. 7 6 7 fc (MPa) 40
4 54 . 5 4 5 ft (MPa) 0.004
Aσ;ieq 0 Aα;ieq ≤ bieq fy (MPa) 500
αn
fi ðσi ; αi Þ ≤ 0; i ¼ 1; : : : ; n ð17Þ

where the additional matrices Aσ;eq , Aα;eq , Aσ;ieq , and Aα;ieq and Validation Example: Concrete Cylinder with
vectors beq and bieq contain parameters defined from the imple- Surrounding Steel Tube
mentation of the yield surfaces.
An example of a concrete cylinder in pure compression with a sur-
The stress variables in general have no upper or lower bounds;
rounding steel tube of thickness t is considered. The diameter of the
however, as will be described subsequently, a lower and upper
cylinder is set to D ¼ 100 mm, the height of the cylinder to h ¼
bound is implemented for the von Mises yield criterion to mitigate
2D ¼ 200 mm, and the thickness is varied between t ¼ 0 and 5 mm
the formation of spurious stress states. The load factor λ has a lower
in steps of 1 mm. The example can be seen in Fig. 2. Typical con-
bound of zero and no upper bound. The auxiliary variables are
crete and steel parameters are used and are given in Table 1.
unbounded.
The analytical solution is found from a sliding failure of the
Besides the resulting load factor λ and stress variables σ, the
modified Mohr-Coulomb yield criterion. The steel tube will provide
solution to the convex optimization problem also yields a set of
confining radial stresses in the concrete core equivalent to its yield
parameters from the dual solution to the lower-bound optimization
capacity in uniaxial tension, 2fy t. Thus, the principal stresses in a
problem, which can be interpreted as a plastic deformation mode.
random material point inside the concrete core is given by
These parameters are linked to the equilibrium equations via H,
 
which ensure nodal equilibrium, and as such, provide a failure 2t 2t
mechanism of the structure (Poulsen and Damkilde 2000). σ ¼ fy ; fy ; p
D D

which, when put into Eq. (7), yields the analytical solution for the
Programs and Computational Requirements compression capacity of the confined concrete as follows:

A program based on the aforementioned theory has been imple- 2t


p ¼ f c þ 4fy
mented in MATLAB version R2019a to generate the convex opti- D
mization problem, which is then solved using the commercial
solver MOSEK version 9.0 (MOSEK ApS 2019). The element
Model and Mesh
meshes have been produced using GMSH version 4.4.1 (Geuzaine
and Remacle 2009), and visualization has been performed using The model is axisymmetric, and thus only a quarter of the cylinder
PARAVIEW version 5.7.0 (Ayachit 2019). was modeled. Besides the symmetry boundary conditions, a single
Computations of the tension connection described subsequently point in the loading direction was also supported to eliminate rigid-
in this paper have required computers with memory in the range of body translation. The meshing was unstructured and controlled by
30–100 GB for the mesh discretization considered. These calcula- a characteristic length lc , which is the target side length of the
tions have been performed on the high-performance computing element the meshing algorithm aims at providing. An example
(HPC) cluster at Technical University of Denmark. mesh with lc ¼ 5 mm can be seen in Fig. 3.

© ASCE 04021051-5 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Schematic of the experiment carried out at the University of


Fig. 3. Part of mesh for t ¼ 2 mm and lc ¼ 5 mm. Southern Denmark. (Reprinted from Engineering Structures, Vol. 52,
H. B. Joergensen and L. C. Hoang, “Tests and limit analysis of loop
connections between precast concrete elements loaded in tension,”
pp. 558-569, © 2013, with permission from Elsevier.)

spacings between overlapping U-bars, lengths of overlap, and


diameters of overlapping U-bars.

Model
The experimental setup is symmetric about two axes transverse to
the loading direction; this is utilized to only model a quarter of the
test setup using symmetry boundary conditions. Furthermore, be-
cause the test setup is designed to fail in the joint concrete, only the
joint region itself is modeled.
The walls of the joint surface were greased before casting of the
Fig. 4. Cylinder capacity as a function of confinement for three dif-
joint. For this reason, it was assumed that no friction or cohesion
ferent characteristic lengths lc .
was present on these surfaces. Jørgensen and Hoang (2013) re-
ported that, typically, the first cracks that appeared were splitting
cracks along the two joint surfaces, and that they appeared at rel-
ative joint deformations in the order of 5%–12.5% of the ultimate
Results deformation capacity. This seems to indicate a significant separa-
Fig. 4 shows a graph showing the compressive stress capacity of the tion of the joint surfaces at the ultimate limit state. Because of this,
concrete cylinder as a function of the confinement stress. The graph the joint concrete surfaces were assumed to be free.
shows that even for the roughest mesh with a characteristic length The two diaphragms on either side of the joint were assumed to
of lc ¼ 10 mm, an accurate solution is obtained. This result vali- be strong and stiff enough to prevent transverse displacement of the
dates that the framework can find the optimal solution for a struc- longitudinal U-bars. For this reason, the surfaces of the longitudinal
ture consisting of both steel and concrete, modeled separately. rebars were fixed at the joint surfaces.
A figure of the model with loads and supports can be seen in
Fig. 6. An arrow in each corner of the surface indicates that the sur-
Application Example: Tension Connection Using face was supported in that direction. The load p was applied as a
Overlapping U-Bars surface unit load. The total capacity of the test specimen was then
given by N num ¼ 4Ap pλ, where Ap is the cross-sectional area of the
To show the ability of the presented framework for a practical ex- rebar on which the load was applied, and λ is the load factor.
ample, a tension connection using overlapping U-bars is consid-
ered. The connection has been the topic of an experimental study
by Joergensen and Hoang (2013), and the numerical results will be Mesh
compared with results from the experimental work. A schematic The mesh discretization was controlled by a characteristic length,
of the experimental setup can be seen in Fig. 5, with model param- denoted lc . The geometry of the curved U-bars is essential for the
eters given in Table 2. The experiment consists of 30 pairwise iden- correct transfer of stresses to the concrete. For this reason, the mesh
tical specimens with varying amounts of transverse reinforcement, was made finer in these regions. The refinement was achieved by

© ASCE 04021051-6 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


Table 2. Geometry, properties, and results of tested connections
b L a H ϕT ϕL fc f yT fyL N U;test A and B
Series ID Specimen ID (mm) (mm) (mm) (mm) (mm) (mm) (MPa) (MPa) (MPa) (kN)
Series 1 1A and 1B 210 540 100 170 6 12 38.4 650.4 583.3 214.6 and 191.1
2A and 2B 210 540 100 170 6 16 38.4 650.4 591.3 206.8 and 218.7
3A and 3B 210 540 100 170 6 20 38.4 650.4 560.9 206.9 and 205.3
Series 2 4A and 4B 210 540 100 170 8 12 38.4 553.4 583.3 243.8 and 230.1
5A and 5B 210 540 100 170 8 16 38.4 553.4 591.3 252.9 and 221.8
6A and 6B 210 540 100 170 8 20 38.4 553.4 560.9 238.8 and 269.8
Series 3 7A and 7B 210 540 100 170 10 12 38.4 632.1 583.3 273.1 and 248.5
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

8A and 8B 210 540 100 170 10 16 38.4 632.1 591.3 281.0 and 297.6
9A and 9B 210 540 100 170 10 20 38.4 632.1 560.9 281.4 and 285.3
Series 4 10A and 10B 210 460 80 170 10 20 38.4 632.1 560.9 387.1 and 391.4
11A and 11B 210 380 60 170 10 20 38.4 632.1 560.9 459.6 and 419.6
12A and 12B 210 300 40 170 10 20 38.4 632.1 560.9 509.4 and 595.3
Series 5 13A and 13B 265 540 100 225 10 20 38.4 632.1 560.9 479.5 and 470.5
14A and 14B 340 540 100 300 10 20 38.4 632.1 560.9 571.6 and 550.7
15A and 15B 490 540 100 450 10 20 38.4 632.1 560.9 597.5 and 648.4
Source: Reprinted from Engineering Structures, Vol. 52, H. B. Joergensen and L. C. Hoang, “Tests and limit analysis of loop connections between precast
concrete elements loaded in tension,” pp. 558-569, © 2013, with permission from Elsevier.

Fig. 7. Low- and high-resolution meshing of a circle.

Fig. 6. Supports and loads of the model. An arrow in either side of a


surface means the entire surface is supported.

prescribing a minimum number of points required when discretiz-


ing a circle. The minimum number of points on the circle denoted
ncirc was determined as a function of the characteristic length lc and
the transverse rebar diameter ϕT by ncirc ¼ 4ðϕT π=lc Þ. The results
were rounded up to the nearest integer. The factor 4 is simply a
scaling factor. For low values of lc , the meshing of especially the Fig. 8. Results of convergence analysis for Specimen 5.
transverse rebars was still quite rough. GMSH meshes curved sur-
faces by distributing points on the surface and generating elements
from these. This method produces a cross-sectional area of the
rebars that is too small. This is illustrated in the two-dimensional No effectiveness factor on the concrete compressive strength
(2D) case in Fig. 7. The problem was mitigated by iteratively was considered, and for the tensile strength, a value of f t;eff ¼
changing the rebar diameter until the cross-sectional area was cor- fc;eff =10,000 was used. This value was sufficiently small to im-
rect within a tolerance of 1 mm2 . prove numerical stability while not influencing the results. The
Specimen 5 was chosen as a representative specimen for a con- convergence plot can be seen in Fig. 8 and in tabulated form in
vergence analysis to determine the validity of the solutions and the Table 3. The solution for lc ¼ 8.2 mm was considered to be suffi-
required mesh discretization. The specimen model was subjected to ciently accurate. Therefore lc was chosen as lc ¼ 8.0 mm for
a convergence analysis with the characteristic length lc ranging further calculations. Fig. 9 shows an example of the surfaces of
from 20 to 5.25 mm, with a 25% decrease for each calculation. a mesh with this discretization.

© ASCE 04021051-7 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


Table 3. Convergence of solution for different mesh discretization for  
0.88 1
Specimen 5 ν ¼ pffiffiffiffiffi 1 þ pffiffiffiffi ð18Þ
fc H
lc (mm) nel N num
20.00 35,533 402.1 where H = height of the joint region. Eq. (18) gives values for the
16.00 58,348 378.1 effectiveness factor between 0.35 and 0.49. The effectiveness factor
12.80 81,723 372.0 is model-specific and should be calibrated from experimental re-
10.24 162,858 359.9 sults. For problems where shear is critical for the capacity, a low
8.19 367,982 345.2 effectiveness factor is expected.
6.55 663,226 335.8
Calculations are run for all 15 specimens, with a characteristic
5.24 1,271,293 328.1
length lc ¼ 8.0 mm and effectiveness factor ν ¼ 0.20, 0.25, 0.30,
Note: nel = number of elements. 0.35, 0.40, and 0.50. Comparison between the experimental results
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

N test and numerical results N num can be seen in Fig. 10, together
with the basic statistics of N num =N test . The values for N test are the
Calibration of the Effectiveness Factor
average value of the two experiments performed for each specimen.
When using rigid-plastic modeling of concrete structures, the The effectiveness factor that gives results closest to the experimen-
strength needs to be reduced by an effectiveness factor because tal data is 0.25, which has the mean value closest to unity and re-
concrete does not behave like a perfectly rigid-plastic material. sults that are generally distributed quite close to the theoretical line.
Joergensen and Hoang (2013) proposed an effectiveness factor This effectiveness factor is quite low even for shear prob-
for the connection based on beam shear problems lems, but, as previously mentioned, the effectiveness factor is

Fig. 9. Mesh of Specimen 5 with lc ¼ 8.0 mm.

Fig. 10. Comparison between experimental results, N test , and numerical results, N num , for different values of the concrete effectiveness factor. In the
plots, the lines corresponding to N num ¼ N test are shown together with basic statistics for N num =N test.

© ASCE 04021051-8 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

Fig. 11. Principal stress vectors and principal equivalent strain vectors for Specimen 5.

model-specific, and a FELA calculation can utilize the material in a in a band going from the loaded U-bar toward the U-bar at the sym-
much more optimized manner than hand calculations. Because of metry boundary condition, and almost no crushing takes place
this, it is reasonable that a lower effectiveness factor should be con- between the loaded U-bar and outer U-bar. The last plot shows
sidered when using FELA. For the rest of the results in this section, the largest principal strain, ε1 , which is an indicator of were sep-
an effectiveness factor of ν ¼ 0.25 has been applied. This is the ef- aration is occurring. Separation takes place perpendicular to the
fectiveness factor used in all elements of the model, whereas in crushing zone, as well as in two distinct zones on the left side
practice, only the critical areas of the model would have an effective- of the loaded U-bar. The separation, which is present perpendicular
ness factor this low. In reality, different effectiveness factors should to the concrete struts, arises from the tensile stresses, which are a
be employed in different regions of the model, corresponding to product of the bulging strut. The fact that relatively large strains are
the stress state experienced by the concrete in these regions. This present here is a good indicator of why such a low effectiveness
limitation has previously been tackled for plane problems by incor- factor is found because several authors have shown that transverse
poration the effectiveness factor into the yield surface (Andersen tensile strain lowers concrete compressive strength (Hoang et al.
et al. 2018), which has not been done for triaxial stress states. 2012). Some separation also takes place in the region just left of
the loaded U-bar; an explanation for this requires study of the fail-
ure mechanism.
Stress Distributions and Failure Mechanisms The failure mechanism reported by Jørgensen and Hoang (2013)
Specimen 5 is again chosen as a representative specimen and will in consisted of splitting cracks at the joint surfaces and lines of frac-
the following be studied in detail. First, the stress state will be stud- tures going from the bends on either side of adjacent U-bars. This
ied by analyzing principal stress vector plots together with prin- failure mechanism was reported to mainly consist of the pullout of
ciple equivalent strain vector plots. Because a rigid-plastic material the two central U-bars with their associated concrete core. The pull-
model is used, the magnitude of the strains are unknown. There- out was accompanied with a lateral motion of the outside section.
fore, the equivalent strains are derived from the failure mechanism The simplified mechanism they assumed in their upper-bound sol-
and are indicators of yielding. Lastly, the failure mechanism will be ution can be seen in Fig. 12. The figure includes numbering of the
studied and compared for the different specimens. different parts of the failure mechanism, which will be used in the
Fig. 11 shows a plot of the smallest and largest principal stress following text.
vectors, σ1 and σ3 , and the smallest and largest principal equivalent The deformations in the FELA calculations were similar to the
strain vectors, ε1 and ε3 . The concrete struts going from adjacent upper-bound solution for Specimens 1–8 and 10–14. A represen-
U-bars is seen when looking at the plot of σ3 . The struts originate tative example can be seen in Fig. 13. Fig. 13(a) shows the de-
from the relatively little surface area of the bend of the U-bars and formed shape of the specimen with strain vectors for the smallest
bulge out at the center of the strut. This bulging is possible due to and largest principles strains. A large part of the mechanism is sim-
the confinement provided by the transverse reinforcement because ilar to the simplified mechanism from Fig. 12. Parts 1 and 3 behave
the concrete itself is modeled with negligible tensile strength. At the almost exactly as stated. However, a large difference is apparent
bends of the U-bars and at the bends of the transverse reinforcement, with Part 2. Instead of moving laterally, it rotates around the sup-
large triaxial stresses are present due to a local state of confinement. ported part of the outside U-bar. The mechanism is further visual-
Tensile stresses are carried in the reinforcement, and as can be seen ized in Fig. 13(b), which shows the relative movement of the
on the σ1 plot, all of the available reinforcement is utilized. collapse as a vector field. The fact that Part 2 rotates, instead of
The smallest principle strain, ε3 , can be used as an indicator of rigid-body translation as the hand calculation prescribed, explains
where crushing takes place in the concrete, i.e., where the concrete the difference in strain distribution on either side of the loaded
is contracting in the failure mechanism. This contraction is present U-bar. From Fig. 13(a), it can be seen that the outer rebar bends

© ASCE 04021051-9 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


around a plastic hinge at the supporting surface of the outer rebar.
From this figure, it can also be seen that some degree of yielding is
also present in the loaded U-bar with one plastic hinge and in the
transverse U-bar with several plastic hinges.
Some differences are present in the failure mechanisms of Spec-
imens 9 and 15. Specimen 9 experienced yielding of the U-bar at
the symmetry boundary condition, meaning that Parts 1 and 3 move
together. The mechanism is visualized in Fig. 14. Parts 1 and 3
move almost entirely in the longitudinal direction, and some bend-
ing of the outside U-bar is still present, but here the transverse rebar
is stiff enough so that only a piece of Part 2 moves. This movement
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

is analogous to spalling of the corner concrete pushed aside by the


bending U-bar.
Fig. 12. Rigid-body motion assumed by Jørgensen and Hoang (2013) The primary failure of Specimen 15 can best be described as
in their upper-bound solution. The mechanism in this sketch show the spalling of the concrete on the outside of the joint. In fact, for Spec-
entire specimen. (Reprinted from Engineering Structures, Vol. 52, imens 13–15, a gradual change in the previously described mecha-
H. B. Joergensen and L. C. Hoang, “Tests and limit analysis of loop nism takes place. This can be seen in Fig. 15. The confinement
connections between precast concrete elements loaded in tension,”
stemming from the transverse rebar was not able to spread through-
pp. 558-569, © 2013, with permission from Elsevier.)
out the specimen when H was increased, and the transverse rebars

Fig. 13. Failure mechanism of Specimen 5 shown as (a) deformed shape with smallest and largest principal strain vectors; and (b) undeformed shape
with deformation vectors.

Fig. 14. Failure mechanism of Specimen 9 shown as (a) deformed shape with smallest and largest principal strain vectors; and (b) undeformed shape
with deformation vectors.

© ASCE 04021051-10 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

Fig. 15. Failure mechanism of Specimens 13–15 with smallest and largest principal strain vectors.

remained at the center of the joint. Therefore, the primary failure


became spalling.
The failure mechanism of the FELA calculations was generally
in line with what was reported by Jørgensen and Hoang (2013)
from their experimental work. Pictures of the specimens after re-
moval of the cover showed cracks corresponding to the areas of
large equivalent principal strains in Fig. 13. Furthermore, the ex-
perimental results from Specimens 9 and 15 also showed different
failures with yielding and spalling, respectively.

Comparison of Capacities
Fig. 16 shows the dependency of the ratio a=H on the capacity of Fig. 16. Comparison of results as a function of the ratio a=H with
the connection; the ratio is changed while keeping either a or H either U-bar spacing a or U-bar overlap H varying.
constant, yielding two different situations. It is seen that a lower
a=H ratio, that is either a larger overlap of the U-bars, H, or smaller
distance between U-bars, a, yields higher capacity. This result is in
line with the conclusions by Jørgensen and Hoang (2013), and the
numerical results generally show good agreement with the exper-
imental results. Interestingly the FELA calculations give the same
capacity in the two situations, which was not the conclusion from
the experiments. However, because the FELA calculation assumes
a negligible tensile capacity of the concrete, and therefore has no
way of transferring shear between the rebars and concrete except
for the shear enabled by confinement, the effect of a larger overlap
is not felt. For this model, with all other parameters remaining con-
stant, the slope of the compressive struts originating at the bend of
either U-bar is the parameter of importance.
Likewise, Jørgensen and Hoang (2013) also concluded that a
larger ratio of transverse reinforcement (here indicated by the diam- Fig. 17. Comparison of results as a function of transverse reinforce-
eter of the transverse U-bars, ϕT ) results in a larger capacity. This ment amount ϕT for varying longditudinal rebar diameters ϕL .
observation is also the case for the numerical results (Fig. 17).
Nevertheless, it seems like the FELA model overestimates this
effect based on the limited data available. Hoang (2013) based on the experimental data. The experimental
From Fig. 18, it can be concluded that an increase in the diameter data even suggest a slight decrease in capacity from ϕL ¼ 16 mm
of the longitudinal U-bars increases the capacity of the connec- to ϕL ¼ 20 mm. However, this could be due to the inherent
tion. This conclusion differs from the conclusion by Jørgensen and variability of experimental work, rather than an effect of the U-bar

© ASCE 04021051-11 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


COWI Foundation (Grant No. T-143.08). The first author would
like to thank and recognize the work done by Nikolaj Skafte Koch
in their joint master’s thesis, which has served as a basis for this
paper.

References
Anderheggen, E., and H. Knöpfel. 1972. “Finite element limit analysis
using linear programming.” Int. J. Solids Struct. 8 (12): 1413–1431.
https://doi.org/10.1016/0020-7683(72)90088-1.
Andersen, M. H., P. N. Poulsen, and L. C. Hoang. 2018. “Closed
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

Fig. 18. Comparison of results as a function of the longditudinal rebar form adaptive effectiveness factor for numerical models.” In Proc., Int.
Federation for Structural Concrete 5th Int. fib Congress. Melbourne,
diameter ϕL for varying transverse reinforcement amount ϕT .
Australia: fib.
Andersen, M. E. M., and N. S. Koch. 2018. “Three dimensional finite
element limit analysis of reinforced concrete structures.” Master’s
thesis, Dept. of Civil Engineering, Technical Univ. of Denmark.
diameter itself. Dependence of the longitudinal rebars is expected Ayachit, U. 2019. The ParaView guide. New York: Kitware.
when comparing the resulting capacities with the failure mechanism Bisbos, C. D., A. Makrodimopoulos, and P. M. Pardalos. 2005. “Second-
reported in the previous section. The capacity of the concrete strut, order cone programming approaches to static shakedown analysis in
going from the loaded U-bar toward the outer U-bar, is limited by steel plasticity.” Optim. Methods Software 20 (1): 25–52. https://doi
the lateral capacity of the system. In this case, the confinement .org/10.1080/1055678042000216003.
provided by the transverse U-bars and the bending capacity of Bisbos, C. D., and P. M. Pardalos. 2007. “Second-order cone and semide-
the U-bar itself. The dependency of N num on ϕL appears to be linear. finite representations of material failure criteria.” J. Optim. Theory Appl.
134 (2): 275–301. https://doi.org/10.1007/s10957-007-9243-8.
Borges, L. A., N. Zouain, and A. E. Huespe. 1996. “Nonlinear optimization
procedure for limit analysis.” Eur. J. Mech. A. Solids 15 (3): 487–512.
Conclusion
Boyd, S., and L. Vandenberghe. 2004. Convex optimization. 1st ed.
Cambridge, UK: Cambridge University Press.
A framework for the analysis of three-dimensional reinforced
Červenka, V., and J. Červenka. 2017. ATENA program documentation
concrete structures with triaxial stress states has been implemented. part 2-2 User’s manual for ATENA 3D. Prague, Czechia: Cervenka
The framework is based on FELA, and both the modified Mohr- Consulting.
Coulomb and von Mises yield criteria have been implemented for Drucker, D. C., W. Prager, and H. J. Greenberg. 1952. “Extended limit de-
modeling of concrete and steel, respectively. The FELA framework sign theorems for continuous media.” Q. Appl. Math. 9 (4): 381–389.
uses a computationally inexpensive tetrahedral element, which https://doi.org/10.1090/qam/45573.
has made it possible to model complex geometries using more than Ferreira, D. 2020. Diana user manual, release 10.4. Delft, Netherlands:
1 million elements. DIANA FEA.
Specimens from experimental work on tensile loop connections Geuzaine, C., and J. F. Remacle. 2009. “GMSH: A 3-D finite element mesh
were modeled. A comparison of model results versus numerical generator with built-in pre- and post-processing facilities.” Int. J. Numer.
Methods Eng. 79 (11): 1309–1331. https://doi.org/10.1002/nme.2579.
results was used to calibrate the concrete effectiveness factor ν,
Gimeno, C. M. 2017. “A finite element-based approach for the analysis
and the best fit was found with ν ¼ 0.25. This result was surpris- and design of 3D reinforced concrete elements and its application to
ingly low, even for a problem where shear is critical. However, it D-regions.” Ph.D. thesis, Departamento de Ingeniería de la Con-
must be expected to find lower effectiveness factors for FELA than strucción y de Proyectos de Ingeniería Civil’, Universitat Politècnica
hand calculations because results from FELA calculations are de València.
greatly optimized and use the material very efficiently. Gvozdev, A. 1960. “The determination of the value of the collapse load
The failure mechanism and stress distribution for a U-bar tension for statically indeterminate systems undergoing plastic deformation.”
connection with varying geometry and rebar sizes was analyzed. Int. J. Mech. Sci. 1 (4): 322–335. https://doi.org/10.1016/0020-7403
The failure mechanism and stress distribution were concluded to (60)90051-5.
be meaningful and in line with crack patterns reported for the experi- Herfelt, M. A. 2017. “Numerical limit analysis of precast concrete
structures—A framework for efficient design and analysis.”
ments. The capacity of the loop connection from the numerical cal-
Ph.D. thesis, Dept. of Civil Engineering, Technical Univ. of Denmark.
culation was compared with the experimental capacity. The capacity Herfelt, M. A., J. Krabbenhøft, and K. Krabbenhøft. 2019. “Practical de-
was studied as a function of different parameters, and good corre- sign and modelling of precast concrete structures.” Curr. Trends Civ.
spondence with test results was found. Struct. Eng. 3 (2): 8–11. https://doi.org/10.33552/CTCSE.2019.03
.000560.
Hillerborg, A. 1974. Strimlemetoden. Stockholm, Sweden: Almqvist &
Data Availability Statement Wiksell.
Hoang, L. C., H. J. Jacobsen, and B. Larsen. 2012. “Compressive strength
Some or all data, models, or code that support the findings of this of reinforced concrete disks with transverse tension.” Bygningsstatiske
Meddelelser 83 (2–3): 23–61.
study are available from the corresponding author upon reasonable
Joergensen, H. B., and L. C. Hoang. 2013. “Tests and limit analysis of
request. loop connections between precast concrete elements loaded in tension.”
Eng. Struct. 52 (Jul): 558–569. https://doi.org/10.1016/j.engstruct.2013
.03.015.
Acknowledgments Johansen, K. W. 1943. Brudlinieteorier. Lyngby, Denmark: Polyteknisk
Forening.
The work presented in this paper has been financially supported by Johansen, K. W. 1962. Yield-line theory. London: Cement and Concrete
the Innovation Fund Denmark (Grant No. 9065-00056B) and the Association.

© ASCE 04021051-12 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051


Krabbenhøft, K., A. V. Lyamin, and S. W. Sloan. 2007a. “Formulation and Martin, C. M., and A. Makrodimopoulos. 2008. “Finite-element limit
solution of some plasticity problems as conic programs.” Int. J. Solids analysis of Mohr-Coulomb materials in 3D using semidefinite program-
Struct. 44 (5): 1533–1549. https://doi.org/10.1016/j.ijsolstr.2006.06.036. ming.” J. Eng. Mech. 134 (4): 339–347. https://doi.org/10.1061/(ASCE)
Krabbenhøft, K., A. V. Lyamin, and S. W. Sloan. 2007b. “Three- 0733-9399(2008)134:4(339).
dimensional Mohr-Coulomb limit analysis using semidefinite program- Mörsch, E. 1922. Der Eisenbetonbau, seine Theorie und Anwendung.
ming.” Commun. Numer. Methods Eng. 24 (11): 1107–1119. https://doi 5 ed. Stuttgart, Germany: Konrad Wittwer.
.org/10.1002/cnm.1018. MOSEK ApS. 2019. “MOSEK optimization toolbox for MATLAB
Krabbenhøft, K., A. V. Lyamin, S. W. Sloan, and P. Wriggers. 2007c. “An manual.” Accessed December 1, 2019. https://www.mosek.com/.
interior-point algorithm for elastoplasticity.” Int. J. Numer. Methods Nielsen, M. P., and L. C. Hoang. 2011. Limit analysis and concrete
Eng. 69 (3): 592–626. https://doi.org/10.1002/nme.1771. plasticity. 3rd ed. London: CRC Press.
Larsen, K. P. 2010. “Numerical limit analysis of reinforced concrete Poulsen, P. N., and L. Damkilde. 2000. “Limit state analysis of rein-
structures: Computational modeling with finite elements for lower forced concrete plates subjected to in-plane forces.” Int. J. Solids
Downloaded from ascelibrary.org by Indian Institute of Technology Bombay on 04/05/22. Copyright ASCE. For personal use only; all rights reserved.

bound limit analysis of reinforced concrete structures.” Ph.D. thesis, Struct. 37 (42): 6011–6029. https://doi.org/10.1016/S0020-7683(99)
Dept. of Civil engineering, Technical Univ. of Denmark. 00254-1.
Larsen, K. P., P. N. Poulsen, and L. Otto Nielsen. 2009. “Limit analysis of Schlaich, J., K. Schäfer, and M. Jennewein. 1987. “Toward a consistent
solid reinforced concrete structures.” In Proc., Int. Conf. on Computa- design of structural concrete.” PCI J. 32 (3): 74–150. https://doi.org/10
tional Technologies in Concrete Structures. Jeju, Korea: Computational .15554/pcij.05011987.74.150.
Technologies in Concrete Structures. Vecchio, F. J., and R. G. Selby. 1991. “Toward compression-field analysis of
Lundgren, H. 1949. Cylindrical shells. Lyngby, Denmark: Danish Techni- reinforced concrete solids.” J. Struct. Eng. 117 (6): 1740–1758. https://
cal Press. doi.org/10.1061/(ASCE)0733-9445(1991)117:6(1740).
Makrodimopoulos, A., and C. M. Martin. 2006. “Lower bound limit analy- Vincent, H., M. Arquier, J. Bleyer, and P. de Buhan. 2018. “Yield design-
sis of cohesive-frictional materials using second-order cone program- based numerical analysis of three-dimensional reinforced concrete struc-
ming.” Int. J. Numer. Methods Eng. 66 (4): 604–634. https://doi.org/10 tures.” Int. J. Numer. Anal. Methods Geomech. 42 (18): 2177–2192.
.1002/nme.1567. https://doi.org/10.1002/nag.2850.

© ASCE 04021051-13 J. Struct. Eng.

J. Struct. Eng., 2021, 147(5): 04021051

You might also like