Computers and Structures 205 (2018) 28–44

Contents lists available at ScienceDirect

Computers and Structures

journal homepage: www.elsevier.com/locate/compstruc

Parametric finite element analysis of RC wide beam-column connections

Hamdolah Behnam a,⇑, J.S. Kuang b, Bijan Samali a
a
Center for Infrastructure Engineering, Western Sydney University, Sydney, Australia
b
Department of Civil and Environmental Engineering, Hong Kong University of Science and Technology, Kowloon, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents the finite element analysis (FEA) of reinforced concrete wide beam-column connec-
Received 1 January 2018 tions using the theoretical context of the concrete damaged plasticity (CDP) model. The predictive capa-
Accepted 25 April 2018 bility of the model was verified by simulating the behaviour of four full-scale, exterior wide beam-
column connections that were tested under reversed cyclic loading conditions. A brief description of
the test program is reported. The finite element predictions showed excellent agreement with the test
Keywords: results, in particular, the lateral load-displacement response, reinforcement yielding, crack patterns
Concrete damaged plasticity
and failure modes. In addition, the sensitivity of the results to various model parameters, including the
Wide beam-column connection
Sensitivity analysis
viscosity parameter, mesh size, dilation angle, yield surface variables, damage parameter, fracture energy,
and the type of analysis was discussed. The key parameters of the model and their distinct characteristics
were identified. Finally, the parametric numerical study was conducted to explore the effects of several
parameters including the column axial load, column and beam dimensions, beam bar anchorage ratio,
and spandrel beam reinforcement, on the performance of exterior wide beam-column connections.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction world [11–13]. The exterior wide beam-column connections are

not only susceptible to joint shear failure but also they are highly
Shear failure of beam-column joints in reinforced concrete (RC) vulnerable to failure of the spandrel (transverse) beam in torsion.
moment-resisting frame (MRF) buildings may occur as a result of Torsion in a spandrel beam is produced by a wide beam’s longitu-
relatively large shear stresses in the joint, inadequate transverse dinal bars that anchored outside the column. The results of tests on
reinforcement, and inferior anchorage and bond condition. This both exterior and interior wide beam-column connections, without
type of failure is categorised as a brittle behaviour, and it has been reinforcement on transverse beams, showed that the transverse
investigated by many researchers in different studies, such as beams underwent severe torsion cracking, and the ductility and
experimental tests, analytical models, and finite element analyses ultimate energy dissipation capacities of the wide beam were
[1–3]. Numerous empirical equations have been proposed based found to be much reduced [11–16].
on test observations and analytical modelling result which provide The existing test database of wide beam-column connection is
the basis of development for the design provisions of RC beam- insufficient [10–28], and it cannot address all behavioural aspects
column joints in the current codes of practice [4–7]. According to of the connections. Unfortunately, the number of tested specimens
these codes, in the design of the RC beam-column joints against in previously published studies are very limited and various local
the seismic load, brittle shear and bond failure in the joints should reinforcement detailing and material properties are involved in
be prevented. In regions with a low to moderate level of seismici- the design and construction, which makes it difficult to extract a
ties, such as Hong Kong, Italy, Spain, and Australia, it has been a general design conclusions from the result. Recently, a comprehen-
widespread practice to utilise RCMRF with wide beams as the pri- sive experimental study has been conducted at the Hong Kong
mary structural system for resisting lateral seismic loads [8–10]. University of Science and Technology to investigate the seismic
The design and construction of such structural systems are effi- performance and failure mechanism of the wide beam-column
cient and profitable. However, the resistance of the beam-column connections. As part of this program, the effect of a number of vari-
connections in this structural system against lateral earthquake ables such as beam width, column width and depth, spandrel beam
loading is the major concern of structural engineers around the dimension and reinforcement ratio, joint shear reinforcement and
beam reinforcement ratio has been studied [29–32]. However due
to the cost and time limitations, it is not possible to test many
⇑ Corresponding author.
more specimens. Finite element analysis (FEA) is well-known for
E-mail address: [email protected] (H. Behnam).

https://doi.org/10.1016/j.compstruc.2018.04.004
0045-7949/Ó 2018 Elsevier Ltd. All rights reserved.
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 29

Nomenclature

e_ total strain rate h angle in degree

e_ el elastic strain rate n potential flow eccentricity
e_ pl plastic strain rate w the dilation angle
~epl t equivalent plastic strains in tension As total cross-sectional area of beam bars
~epl c equivalent plastic strains in compression bc column width
bw beam width
~ein inelastic strain
c Del initial (undamaged) elastic stiffness
e0 concrete strain at peak stress
d
0
scalar stiffness damage variable
rb max maximum eigenvalue of tensor r  dc damage variable in compression
r effective stress dt damage variable in tension
r c ð~eplc Þ the effective cohesion stress in compression dmax maximum aggregate size
r c ð~eplt Þ the effective cohesion stress in tension Ec initial modulus of elasticity of concrete
rt0 uniaxial tensile stress Es steel Young modulus of elasticity
rb0 biaxial compressive yield stress fc0 cylinder compressive strength of concrete
rc0 uniaxial compressive yield stress ft the tensile strength of concrete
r Cauchy stress fy the yield stress of steel
rc concrete compressive strength fu the ultimate strength of steel
a dimensionless constants in Eq. (4) Gf fracture energy of concrete
b dimensionless constants in Eq. (5) Gf0 base fracture energy of concrete
c dimensionless constants in Eq. (6) hb beam depth
gc material constant in Eq. (9b) hc column depth
kc material constant in Eq. (9c) Kc the shape factor of the yield surface
l relaxation time

its capability for accurate modelling of the mechanical properties, associated with stiffness degradation and inelastic deformations. In
crack formation and propagation, deflections, and possible failure order to capture these characteristics of concrete, the CDP model
mechanisms in RC elements. Moreover, the structural design of has been developed. The CDP model is a continuum, plasticity-
RCMRF with wide beam-column connections can be facilitated by based, damage model and based on the theory of plastic flow.
creating efficient material and element models in an available The model involves a combination of non-associated multi-
finite element software. However, only a few numerical studies hardening plasticity and scalar damaged elasticity to describe the
on wide beam-column connections have been reported to date damage in concrete. The model assumes two failure mechanisms,
[19–20,28–33]. including tensile cracking and compressive crushing of the con-
Accurate material modelling of concrete in tension and com- crete. The CDP model implemented in ABAQUS has proved to pro-
pression and reinforcing bars is the most challenging aspect of vide the stable regime with decent accuracy for modelling the
the finite element modelling of concrete structures. In the past dec- nonlinear and post-peak behaviour of concrete when compared
ades, different constitutive models of concrete have been devel- to the experimental results [33,47–51]. Further, CDP model can
oped and implemented in FEA software. These studies were be used in both ABAQUS/Standard and ABAQUS/Explicit, and it is
mainly based on elasticity theory [34], plasticity theory [35–37], suitable for modelling concrete structures subject to static and cyc-
concrete-damage theory [38,39], and coupled concrete damaged lic loading. The CDP is mesh dependent. Once the strain localisa-
and plasticity theory [40,41]. The attractiveness and popularity of tion occurs in few elements, the rest of the model starts to
each model depend on the degree of their accuracy, simplicity, unload and eventually, the analysis fails to converge.
and practicality. Genikomsou and Polak [48], and Wosatko et al. [49] studied the
ABAQUS [42] provides three crack models for simulating dam- behaviour of the RC slabs using CDP model in ABAQUS. In the cal-
age in RC, namely, concrete smeared cracking model, brittle crack- ibration of the models, they considered load-displacement curves,
ing model, and concrete damaged plasticity (CDP) model. The maximum displacement and cracking patterns. The results of both
brittle crack model is developed for the linear elastic materials studies showed that the FEA was sensitive to the concrete tensile
with tensile cracking. Thus, it cannot resemble the real behaviour behaviour, mesh sizes, dilation angle and the use of concrete dam-
of the concrete which has both tension and compression. In the age parameters. Genikomsou and Polak [48] emphasised that the
smeared crack concrete model, the crack is formed at any location mesh sensitivity analysis is indispensable for providing the most
that the concrete stresses extend beyond the failure surfaces either suitable element size due to mesh size-dependent model. Wosatko
in the biaxial tension region or in a combined tension compression et al. [49] pointed out that a regularised continuum description
region. The existence of cracks is taken into account by the way the should be incorporated in the CDP model to minimise the effect
cracks disturb the stress and material stiffness. The successful of mesh sensitivity, localised deformation, and numerical instabil-
application of the smeared crack formulation in RC elements can ities. A perfect bond between reinforcing bars and concrete was
be found in [43–45]. The main limitations of the model are the considered in both studies. The effects related to the concrete-
stress locking caused by stress allocation across broadly open reinforcement interface, such as bond-slip and dowel action was
cracks, mesh size dependency and instability at late stages of the roughly examined by using tension-stiffening in the concrete
loading [46]. model and by applying Hillerborg et al. [52] fracture energy crack-
For concrete material after loading and unloading, permanent ing criterion into the concrete model.
strains remains due to friction and sliding of the microcracks. Fail- This paper describes the modelling of RC exterior beam-column
ure mechanics of confined concrete in compression and tension is connections using the three-dimensional FEA software ABAQUS.
30 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

The CDP model was adopted for the representation of concrete. total strain; epl is the plastic strain;: indicates Frobenius inner pro-
After a brief theoretical background on the CDP model, the mod- duct; d is the scalar stiffness damage variable, which can have val-
elling procedure was described. The model was validated by simu- ues in the range from zero (undamaged material) to one (fully
lating the behaviour of four large-scale exterior wide beam-column damaged material). According to the effective stress concept, the
connections, previously tested under reversed cyclic loading condi- plastic yield function is formulated in terms of effective stress, r .
tion at the structural lab of the Hong Kong University of Science
and Technology [10]. The numerical results were compared with 2.3. Hardening variables
the experimental test results. Moreover, the sensitivity of the FE
model to various model parameters is discussed, and the key Damaged states in tension and compression are considered by
parameters of the model identified. Finally, parametric numerical assigning two hardening variables, ~epl
t and ec , referring to equiva-
~pl
study was conducted to explore the effect of various design param- lent plastic strains in tension and compression, respectively.
eters on the lateral load transfer mechanism of wide beam-column Micro-cracking in the concrete is embodied by increasing values
connections. of the hardening variables.

2. Concrete damaged plasticity (CDP) model in ABAQUS 2.4. Yield criterion

Detailed descriptions of the CDP model and its parameters iden- The yield surface function, Fig. 1, is based on the model pro-
tification can be found in Refs. [33,40–42,48–51,53–55]. The model posed by Lubliner et al. [40] with the modifications by Lee and Fen-
is based on the following assumptions: ves [41]. The yield functions following the Kuhn-Tucker condition,
and takes the following form in effective stress space.
2.1. Additive strain rate decomposition 1
F¼  þ bð~epl Þhr
  3ap
ðq  max i  chr
 max iÞ  r
 c ð~epl
c Þ 6 0 ð3Þ
1a
An additive strain rate decomposition is assumed for the rate-
independent model, as given by Eq. (1). where p  is the equivalent von
 is the effective hydrostatic pressure; q
Mises stress; hxi = 0.5 (x + jxj) is the Macauley bracket; r  max is the
e_ ¼ e_ el þ e_ pl ð1Þ algebraically maximum eigenvalue of tensor r  c ; a, b, c are dimen-
sionless constants, which need to be defined in Eqs. (4)–(6),
where e_ is the total strain rate, e_ el is the elastic strain rate, and e_ pl is respectively.
the plastic strain rate.
ðrb0 =rc0 Þ  1
a¼ ; 0 6 a 6 0:5 ð4Þ
2.2. Stress-strain relations 2ðrb0 =rc0 Þ  1

In CDP model, the stiffness degradation is modelled by defining r c ð~eplc Þ

bð~epl Þ ¼ ð1  aÞ  ð1 þ aÞ ð5Þ
the relationship between effective stresses and stresses. Stress- r t ð~eplt Þ
strain model provided in CDP model is based on a simple model
which provides coupling between damage and plasticity using sca- 3ð1  K c Þ
lar damage and effective stress, as given by Eq. (2). c¼ ð6Þ
2K c  1
r ¼ ð1  dÞDel0 : ðe  epl Þ ¼ ð1  dÞr ð2Þ where ðrb0 =rc0 Þ is the ratio of biaxial compressive to uniaxial com-
pressive yield stress which influences the yield surface in a plane
where r is the Cauchy stress; Del0 is the initial (undamaged) elastic stress state; Typical experimental values of ðrb0 =rc0 Þ ratio is in
stiffness; Del ¼ ð1  dÞDel0 is the degraded elastic stiffness; e is the the range from 1.10 to 1.16, leading to values of a between 0.08

Fig. 1. Concrete damage plasticity model [42]: (a) yield surface in plane stress and (b) yield surface in the deviatoric plane.
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 31

and 0.12; r c Þ and r

 c ð~epl  t ð~epl
t Þ are the effective cohesion stresses in
of the beam in two opposite directions. The lateral load history
compression and tension, respectively. The coefficient c appears consisted of several sets of three cycles with different horizontal
only for triaxial compression stress states. This factor can be deter- displacements amplitudes. The amplitude of the cycles was con-
mined by comparing the yield conditions along the tensile and com- stant within each set, but it was increased in consecutive sets
pressive meridians. The parameter Kc is the coefficient ascertain the of cycles. Specimens were instrumented extensively by fixing
shape of the deviatoric cross-section, as shown in Fig. 1(b). strain gauges at critical locations on the reinforcement bars. Lin-
ear variable displacement transducers (LVDTs) were used to mea-
2.5. Flow rule sure the displacement at various locations on the specimen, as
shown in Fig. 2.
The concrete stress-strain relationships and the yield surface The main design parameters based on measured material prop-
are connected using the flow rule. The CDP model assumes non- erties are summarised in Table 1. The nominal flexural strengths
associated Drucker-Prager hyperbolic function for the flow poten- (Mn,c and Mn,b) and nominal torsional moment strength (Tn) were
tial function, G, as given by Eq. (7). calculated in accordance with ACI 318-14. As seen in Table 1, the
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ratios of Tu/Tn in specimens S2, S3 and S4 were 0.43, 0.87, and
G¼ ðnrt0 tan wÞ2 þ q 2  p tan w ð7Þ 1.29, respectively. According to ACI 352R-02, this ratio should be
smaller than one. The expected normalised joint shear stress ratios,
where n is the potential flow eccentricity; rt0 is the uniaxial tensile cd, were 0.74, 1.12, 1.63, and 2.03 in specimens S1, S2, S3, and S4,
stress, and w is the dilation angle. respectively. According to ACI 352-R02, the normalised joint shear
stress ratio should be lower than 1.25 for joints confined on three
2.6. Viscoplastic regularisation faces, or two opposite faces, and is set to 1 for other configurations.
Based on these design parameters beam flexural failure
Materials that exhibit softening behaviour and stiffness reduc- occurred in specimens S1 and S2, shear failure of the joint formed
tion may lead to convergence difficulties. To overcome some of in specimens S3 and shear failure of the joint core with the tor-
these convergence difficulties, viscoplastic regularisation of the sional failure of the spandrel beam occurred in specimen S4. Read-
constitutive material model was used. The CDP model can be reg- ers are encouraged to read the Ref. [10] for full experimental
ularised using viscoplastic regularisation according to the general- report.
isation of the Devaut-Lions approach. In the CDP model with a
viscous parameter, the plastic strain tensor is derived using an
additional viscosity parameter, which is known as the relaxation 4. Finite element simulations
time (l) and needs to be defined [39]. The default value of the vis-
cosity parameter in ABAQUS is zero, meaning that no viscoplastic 4.1. Methodology
regularisation is performed.
In summary, to use the CDP model in ABAQUS, two sets of uni- In the numerical study, the geometrical properties of the model
axial data, two sets of damage parameters and five additional were kept similar to the test specimens. Eight-node hexahedral
parameters should be assigned. Four data sets are the stress- (brick) elements (C3D8R) were used for concrete. Two-node linear
strain data in compression and tension and damage parameters truss elements (T3D2) were used to model reinforcements. The
in tension and compression. The additional five parameters are embedded method was adopted between the concrete and the
the values to complete the yield surface (rb0 =rc0 and Kc), the reinforcement which means perfect bond and displacement conti-
potential flow (w and n) and the viscoplastic regularisation (l). nuity between concrete and steel. It should be pointed out that
ABAQUS provides several different options for modelling interface
3. Test specimens of reinforcement and concrete. Interface modelling in RC elements
is important and in particular in modelling beam-column joints
Four large-scale wide beam-column connections were used to [56–61], because, an excessive bond slip and loss of anchorage of
verify the model. These specimens were designed, constructed, beam longitudinal bars are one of the common failure modes in
and tested under reversed quasi-static cyclic loading. A full report the connection region of RC structures. In this study, the perfect
of the experimental investigation can be found in [10]. The speci- bond between reinforcement and concrete is considered. Fig. 3
mens representing a portion of the framing system and are shows corresponding mesh and reinforcement.
obtained by terminating the beam at its mid-span and the columns The interaction between concrete and reinforcement after
at their mid-heights, where inflection points are likely to occur cracking, such as bond slip and dowel action was incorporated in
under the lateral loads. The columns had the same cross- a simplified way using the tension stiffening in the concrete model.
sectional dimension of 300  360 mm, with a height of 3100 mm, To apply the boundary conditions similar to the test setup, first, the
for all specimens. The beams had depths of 300 mm with lengths nodes at the surfaces of the column end’s attached to the reference
of 1500 mm in all sub-assemblages. Four different beam widths point using the coupling constraint. Then, the boundary conditions
were investigated. The width of the beam was 300 mm in the first were applied to these reference points. The reference point at the
connection (S1), and it increases to 450 mm, 600 mm, and 750 mm, right-hand side of the column was defined to be a pin, while the
in the second (S2), third (S3) and fourth (S4) specimens, respec- other end was set as roller support so the column end can move
tively. Therefore, specimens S1, S2, S3, and S4 had beam width in a horizontal direction when the axial force is acting on it. Anal-
ratio of 1, 1.5, 2 and 2.5. Wide beam specimens contained a shallow ogous to the test condition, a loading was determined in two steps.
depth spandrel beam with a cross-section of 360 mm  300 mm. First, the column axial load was applied, and then, a lateral dis-
The spandrel beam in these specimen consisted of 8D16 longitudi- placement was enforced in the beam end. The analysis was contin-
nal bars and having D10 closed stirrup at 70 mm center-to-center ued to the same displacement as the test. The mesh size of 30 mm
space. Fig. 2 illustrates the test setup, loading system and damage was used in all models. Static analysis in ABAQUS/Standard with
at 4% drift ratio. viscosity regularisation was performed. A computer with 12 GB
All the sub-assemblage were tested under a constant axial load memory (RAM) and Core i7 CPU of 2.67 GHz was used for the anal-
of 500 kN in specimens S1 and S2 and 480 kN in specimens S3 yses. For solving this model using ABAQUS/Standard, a full Newton
and S4. The lateral cyclic displacement was applied to the top solver with default matrix storage was used. An automatic
32 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

(a) Schematic view of the test setup (unit: mm).

(b) Damage at 4% drift ratio in specimen S2

Fig. 2. Test setup.

Table 1
Main design parameters and expected capacities.

Specimen bw mm Mn,c kNm Mn,b kNm Mra Vb,eb kN Tu kNm Tu/Tnc Vj,ed kN cde cACI
S1-BC1 300 142 92.6 3.07 70 0 0 483 0.74 1
S2-BC1.5 450 177 138.8 2.55 105 29 0.43 725 1.12 1.25
S3-BC2 600 189 195.0 1.94 148 61 0.87 1034 1.63 1.25
S4-BC2.5 750 189 243.8 1.55 185 91 1.29 1292 2.03 1.25
a P P
Mr ¼ M nc = M nb P 1:2.
b
V b;e ¼ Mnb =1:32.
c
T n ¼ 2A0 At f yt cot h=s, where Ao = 0.85Aoh, h = 45°, and s is the spacing of stirrup in spandrel beam.
d
V j;e ¼ 1:25As f yqffiffiffiffi
Vffi col , where As is the beam flexural reinforcement area and Vcol is the column shear force.
e
cd ¼ V j;e =ðbc hc f 0c Þ.

incremental with a small time step size and a large maximum concrete that represents the area under the tensile stress-crack dis-
number of increments were used to improve the convergence rate. placement curve. Adopting stress-crack width displacement based
on the Hillerborg et al. [52] fracture energy can help to prevent
4.2. Material parameters mesh-sensitivity and to enhance numerical convergence. The frac-
ture energy of concrete Gf (N/m) for normal weight concrete can be
The concrete behaviour in tension is characterised by a stress- obtained from Eq. (8) [62].
crack displacement response, as shown in Fig. 4, where ft is the  0 0:7
maximum tensile strength, and Gf denotes the fracture energy of Gf ¼ Gf 0 ðf c þ 8Þ=f cmo ð8Þ
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 33

(a) Geometric model (b) Reinforcement model

Fig. 3. FEM model of specimen S4-BC2.5.

Fig. 4. Uniaxial tensile stress-crack width relationship for concrete.

where fcmo = 10 MPa and Gf0 is the base fracture energy that Fig. 5. The uniaxial compressive stress-strain relationship for concrete.
depends on the maximum aggregate size, dmax. The value of the
base fracture energy is 0.03 N/mm for a maximum aggregate size
of 12 mm [62], as used in the construction of specimens [10]. rc;1 ¼ Ec ec ; ec 6 0:4f 0c =Ec ð9aÞ
The concrete stress-strain behaviour under compression was
modelled in three phases (Fig. 5) using the formulations given in  2
[53–54] with some modifications. The equations for the assumed gc ee0c  ee0c
rc;2 ¼ f ; 0:4f c =Ec 6 ec 6 0:0035
0 0
ð9bÞ
compressive stress-strain diagram are given in Eq. (9). Eq. (9a) rep- 1 þ ðgc  2Þ ee0c c
resents the linear-elastic branch in which ec is a variable changing
from zero to a 0.4fc0 /Ec, and Ec is the initial modulus of elasticity. !1
2 þ kc f c e0 kc e2c
0
The linear branch ends at the stress level of 0.4fc0 . Eq. (9b) describes
rc;3 ¼  kc e 0 þ ; 0:0035 6 ec 6 0:03 ð9cÞ
the second branch up to the strain level of 0.0035 in the decending 0
2f c 2e0
branch. The corresponding strain level at the peak stress is defined
as e0 = 2fc0 /Ec; gc is the material constant. The stress and strain The data required by ABAQUS are in terms of inelastic strain, ~ein
c ,
compatibility at the strain level of ec = 0.4 fc0 /Ec, for Eqs. (9a) and which is defined as the total strains minus the elastic strains corre-
(9b) gives the value of gc. Eq. (9c) shows the third and descending sponding to the undamaged material. ABAQUS automatically con-
branch; kc is the constant crushing energy as a material property verts the inelastic strains to plastic strain values using
c ¼ ec  rc =Ec , as shown in Fig. 5. An error message is issued
[54]. Using the stress and strain compatibility at the strain level ~ein
of ec = 0.0035, for Eqs. (9b) and (9c) enables the value of kc to be when the estimated plastic strain values are negative or decreasing
determined. The concrete ultimate strain eu was set to a large value with increasing inelastic strain, indicating that the input data are
of 0.035 to avoid any numerical difficulties. incorrect.
34 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

Table 3
Properties of reinforcing bars.

Reinforcement S1 and S2 S3 and S4

Bar diameter (mm) 10 16 10 16
Yield strength fy (Mpa) 485 522 511 558
Ultimate strength fu (Mpa) 622 599 620 642
Yield strain (%) 0.2425 0.261 0.255 0.279
Ultimate strain (%) 12 9 12 9

Tables 2 and 3 summarise the concrete and steel material

parameters used in the model. Fig. 8 shows the experimental curve
for the compressive strength of concrete from the tests together
with the theoretical curve from Eqs. (9a), (9b), (9c). Fig. 8 shows
the tensile test results on a bar of a diameter 16 mm and the pro-
posed bilinear model.
In addition to defining the material parameters, to complete the
Fig. 6. Tensile damage parameter-strain relationship for concrete. yield surface and the non-associated potential flow, the following
data were also provided.

 rb0/rc0 was set to 1.16;

 Kc was set to default value of 0.667;
 w was set to 40°;
 n was assumed with a default value of 0.1; and
 l was taken as 0.00001 s.

5. Comparison of nonlinear FEA predictions with experimental

results

Comparisons of the results obtained from nonlinear FEA with

the experimental results are presented. Fig. 9 shows the analysis
results together with the experimental hysteretic loops regarding
load-displacement. As illustrated in this figure, the lateral force-
displacement curve predicted by the FEA follows most of the
experimental curve closely. It should be mentioned that the full
Fig. 7. Compressive damage parameter-strain relationship for concrete. cyclic analysis was performed in ABAQUS, and it consumed a
tremendous amount of time. However, the hysteretic loops
obtained from the analyses did not exhibit a pinching effect. The
Damage parameters were introduced in the CPD model in ten- pinching of the hysteresis loops may occur due to different factors
sion and compression according to Figs. 6 and 7, respectively. It such as shear, bond deterioration, Bauschinger effects in reinforce-
was assumed that the concrete damage is occurred only in the soft- ment bars, etc. The complexity in the constitutive modelling of the
ening range in both tension and compression, as suggested in [48]. concrete and reinforcement and the adoption of the embedded
In compression, the damage values are presented after the peak method to simulate the bond between concrete and reinforcement
load corresponding to the strain level, eo. It was also assumed that caused this problem in the hysteretic simulations. Therefore, in
the damage parameter reaches a value of 0.8 when the strength this paper, only monotonic loading analysis is presented. Although
drops 80%, and the maximum value for the damage parameters a number of particular features of cyclic loading such as energy
was chosen as 0.95, which corresponded to a 95% reduction of absorption capacity were not measured by the model in the mono-
the stiffness. tonic loading, it permitted many more parameters to be examined.
The uniaxial stress-strain data of the reinforcement was mod- According to the CDP model, the concrete cracking is initiated
elled as elastic with Young’s modulus (Es) and Poisson’s ratio (v) when the maximum principal plastic strain is positive with the
of 200,000 MPa and 0.3, respectively. The plastic properties of direction of the vector normal to the crack plane, parallel to the
the reinforcement were determined based on the bilinear strain direction of the maximum principal plastic strain. Fig. 10 shows
hardening yield stress-plastic strain curve which fitted well with the cracking pattern on the tension side of the beam at the end
the experimental results of the tension test on reinforcement bars. of the test and comparison with the FEA results. In all specimens,
The local buckling of the reinforcing bar and Bauschinger effect the flexural cracking was formed across the full width of the beam
was not considered in steel material modelling. at early drift ratios, and up to around 1% drift, the crack widths

Table 2
Properties of concrete.

Specimen fc0 MPa Eca MPa e0 (%) eu (%) ftb MPa Gf N/mm Poisson’s ratio (v) Density tonne/mm3
S1, S2 36.1 28,240 0.255 3.5 3.05 0.085 0.2 2.4E009
S3, S4 34.7 27,685 0.250 3.5 2.80 0.083 0.2 2.4E009
qffiffiffiffiffi
a 0
Ec ¼ 4700 f c .
b
Measured using four-point loading tests on test day.
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 35

Fig. 8. Concrete and steel model.

Fig. 9. Load-displacement from experiment and numerical analysis.

36 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

(a) Specimen S1

(b) Specimen S2

(c) Specimen S3

(d) Specimen S4
Fig. 10. Cracking pattern on tension surface at 5% drift.

remained relatively narrow, with the specimens only having minor give ductile beam flexural failure as in the experiment, as shown
damage. After the 1% drift, the damage progressed at a higher rate. in Fig. 10(a) and (b). Specimen S3 had several torsional cracks in
In specimens S1 and S2 only hairline diagonal shear cracks were the spandrel beams starting at 0.75% drift, but these cracks
observed on the side face of the joint during testing and cracks remained tightly packed up to the end of the test. The failure mode
were developed mostly in the beam plastic hinge zone where con- of this specimen was a joint shear failure after beam yielding. Sim-
crete crushing was evident. These specimens failed in ductile mode ulation of the specimen shows joint shear failure after obtaining
by developing a complete a beam plastic hinge. The simulations maximum lateral load at a 3% (45 mm) drift ratio (Fig. 10(c)).
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 37

A higher torsional moment was applied to the spandrel beam of S3, and S4, respectively. The strain of the same reinforcement was
specimen S4. Hence the highest numbers of torsional cracks were monitored in the FE model, and it is plotted on Fig. 11.
developed in this connection. These cracks opened up on both sides The FE model-predicted values for the yield drift was 1.04%
of the spandrel beam. The failure mode of specimen S4 was beam (15.6 mm), 1.42% (21.3 mm), 1.74% (26.1 mm), and 2.78% (41.7
yielding followed by joint shear and spandrel beam torsional fail- mm), in specimens S1, S2, S3, and S4, respectively. Fig. 11 depicts
ure. The obtained FEA crack pattern of specimen S4 was similar that the FEA prediction for the yield drift was in close agreement
to the test cracks, concentrated on the beam-column interface, with the experimental results. During the tests, all the beam rein-
joint and spandrel beam. forcement yielded at a load of 66 kN, 105 kN, 142 kN and 180 kN,
In order to investigate the ability of the model in predicting the in specimens S1, S2, S3, and S4, respectively. In the FEA the beam
inner aspect of the specimens, the yield pattern of the wide beam reinforcement of the specimens S1, S2, S3, and S4, yielded at a load
longitudinal reinforcement was studied. During the construction of of 72 kN, 106 kN, 153 kN, and 177 kN, respectively.
time, a number of strain gauges were installed on the reinforce- The maximum lateral load and the drift in which all the beam
ment of the beam and column to measure the local strains level. bars yielded (Dy(all)%) predicted by the simulations are compared
Fig. 11 shows the drift ratio versus measured strain for the beam with the test results, and results of this comparison are presented
longitudinal bar that located at the corner of the beam in all spec- in Table 4. In Table 4, the error and mean (M) were used to
imens. The strain gauges were attached to the beam longitudinal describe the overall model accuracy, and associated average over-
bar at the 50 mm distance from the column face. The yield point estimate or underestimation of the model. Error (%) and M (%) are
is set at the first intersection of the measured strain curve with defined as:
the yield strain line (as listed in Table 3). From the measured drift  
FEA result  Test result 
versus strain plots, yielding occurred in drift ratios of 1% (15 mm), Error ð%Þ ¼    100
 ð10Þ
1.2% (18 mm), 1.5% (22.5), and 2.7% (40.5 mm) in specimens S1, S2, Test result

Fig. 11. Load-displacement from experiment and numerical analysis.

38 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

Table 4
The ratio of maximum strength to predicted strength and related drift.

Specimen DLa Vb,test (kN) Vb,FEA (kN) Dy(all) % test Dy(all) % FEA Strength prediction Drift prediction
Error % M% Error % M%
S1-BC1  73.6 74.6 1.00 1.04 1.36 101.36 4.0 104.0
+ 73.7 74.6 1.05 1.04 1.22 101.22 1.0 99.0
S2-BC1.5  106.5 107.6 1.20 1.42 1.03 101.03 18.3 118.3
+ 106.1 107.6 1.29 1.42 1.39 101.39 10.1 110.1
S3-BC2  156 157.1 1.50 1.74 0.71 100.71 16.0 116.0
+ 150 157.1 1.51 1.74 4.73 104.73 15.2 115.2
S4-BC2.5  185.8 178.1 2.55 2.78 4.14 95.86 9.0 109.0
+ 189 178.1 2.69 2.78 5.77 94.23 3.3 103.3
a
Loading direction.

FEA result agreement with the test results in terms of ultimate load and
M ð%Þ ¼  100 ð11Þ
Test result deflection. The results showed that assigning the small values of
It can be seen in Table 4 that in all cases, the model prediction of the viscosity parameter to the model helped to improve the rate
the maximum lateral force and the drift of yield, lead to an error of convergence, without affecting the load-displacement response.
below 6% and 19%, respectively, which once again shows that the The wall-clock computational time for the analyses using vis-
predictions of the FEA are in close agreement with the experimen- cosity parameters of 0.00001, 0.0001, 0.001, and 0.01 was 2144,
tal results. The simulated response of the test specimens was in 1100, 493 and 299 min respectively. Increasing the viscosity
good agreement with the results observed from the experiment. parameters allows for a larger stable time increment, and hence
The modelled responses verified the capability of the desig- it reduces the computational time, but it can affect the accuracy
nated material parameters and constitutive models to detect the of the model. For the viscosity parameter 0.01, the results overes-
behaviour of the connections accurately. However, regarding the timated the stiffness and strength of the specimen as shown in
complexity of the numerical modelling, it is crucial to investigate Fig. 12. The results indicated that the choice of the viscosity param-
the effect of the selected model parameters on the accuracy of eter value in practical computations using the CDP model should
the analysis. be made with great care. The results of the analysis showed that
assigning a proper viscosity parameter to the model help in reduc-
ing the time of analysis, however, viscous effect plays a minor role
6. Investigation of material parameters regarding strength and stiffness as expected, because, plastic
deformation of concrete elements is mainly caused by internal fric-
Specimen S3 was selected as a control specimen to discuss the tion. In all following analyses, the static analysis with the lowest
parameters of the CPD model. Fig. 12 presents a comparison analysed viscosity parameter (l = 0.00001) was used.
between the analyses and the experimental results regarding the A mesh-size effect on the results was studied. Three different
force-displacement response. The analysed cases had different val- mesh sizes (30 mm, 40 mm and 50 mm) were implemented in
ues of the viscosity parameter, l, from 0 through 0.0001, 0.001 the analysis of the specimen to investigate the level of mesh
until 0.01 s. When a value of the viscosity parameter was set 0 dependency of the model. In Fig. 13 the analyses are presented
(default value of ABAQUS), the analysis did not converge and with 30 mm, 40 mm and 50 mm mesh sizes. The mesh sizes were
stopped at an early stage. The value of the viscosity parameter is larger than the maximum aggregate size (12 mm). By having ele-
related to the time increment step, and in order to improve the ments with a mesh size of 30 mm, ten elements were considered
solution, it is suggested to assign a value around 15% of the time across the beam depth, while by having mesh sizes of 40 mm
increment step [42]. Regarding the high nonlinearity of the mod- and 50 mm, 8 and six elements were created over the beam depth.
elling of the connections, the time increment step was not fixed, The results of the force-displacement response for a mesh size
and it was set automatically. The results from the analyses, with of 30 and 40 mm were similar and in good agreement with exper-
viscosity parameters equal to 0.00001, 0.0001, 0.001, are in good imental test results. Providing reinforcement in most regions of the

Fig. 12. Effect of different viscosity parameter. Fig. 13. Effect of different mesh size.
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 39

model and using Hillerborg et al. [52] fracture energy for defining
the concrete tension stress-crack width, helped to reduce the mesh
sensitivity. A mesh size of 50 mm seems to be a too coarse as it
could not give the strength reduction which occurred because of
joint shear failure. The wall-clock computational time for the mod-
els with mesh sizes of 30 mm, 40 mm, and 50 mm was 2144, 1123,
and 230 min, respectively. Increasing the mesh size can substan-
tially reduce the computational time. A mesh size of 30 mm was
selected in all following simulations not only based on the load-
displacement responses but also on comparisons with the
observed crack patterns.
Concrete is a brittle material, and under compressive loading, it
suffers significant plastic volume expansion. The volume change of
the concrete under compression is named dilatancy. In CDP model,
dilatancy can be modelled by assigning a value for the dilation
angle (w), and it is ranged from 30° to 42° [48,63–65]. The effect
of different dilation angles on the analysis accuracy was examined Fig. 15. Effect of different Kc.

using varying values of dilation angle from 31°, 34°, 37°, 40° and
42°.
Fig. 14 shows that the response of the model is highly depended
on the value of the dilation angle. The strength and ultimate dis-
placement increased with an increase in the magnitude of the dila-
tion angle. Fig. 14 also shows that the difference in response is
almost negligible for the value of dilation angle between 37° and
42°. Therefore, the dilation angle was preferred to be set as 40°
for all subsequent analyses. It can be concluded that the higher val-
ues of the dilation angle produce ductile behaviour while lower
values produce brittle behaviour.
Fig. 15 presents the results of analyses with the various values
for the parameter Kc. According to the CDP model [42], the param-
eter Kc is necessary to define the shape of the yield surface and is
ranged between 0.5 and 1, by the default value of 0.667. Four dif-
ferent values of Kc = 0.5, 0.667, 0.85 and 1, were given for investi-
gation. Fig. 15 shows that the effect of Kc on the response of the
model is significant. It can be seen that the default value of 0.667 Fig. 16. Effect of damage parameters.
gives the most accurate results in terms of strength and ductility.
Thus, in all analyses, the parameter Kc is set to 0.667.
In the CDP model, the degradation of the concrete stiffness after
the concrete plasticity model, presuming the plastic and inelastic
cracking is taken into account by assigning two separate damage
parameter data in tension and compression. Fig. 16 shows the strains are equal (~epl einc ) and the effective stress and the Cauchy
c ¼ ~

examination of the effect of the damage parameters on the stress (r) are equal. However, when considering damage parame-
response of the model. The results obtained from the analysis con- ters, the plastic strains are lower than the inelastic strains
sidering the damage parameters displayed that the higher strength (~epl
c < ~ einc ), and the effective stress (r  ) is larger than the Cauchy
compared to the analysis results with no damage parameters. The stress (r). In the case of no damage variable, many different
damage variables in CDP model are coupled with the plastic defor- parameters may affect, therefore the results of the modelling in
mations in the constitutive model formulations. In the absence of terms of strength and crack pattern can be substantially different
the damage parameters (dt = dc = 0), the CDP model is changed to from those by considering damage parameters. For example, in
S3 model with no damage parameter the cracks were localized in
the joint region of the specimen while by considering damage
parameters, cracks were distributed in both beam and joint region.
Fig. 16 shows that the existence of the damage parameter in the
model has no influence on the load-displacement response at the
early stage because there is no or little damage on the concrete
at this stage. The results slightly overestimate the ultimate loading
capacity of the connection by only considering tensile damage
parameter. Damage parameter in compression was found to have
a significant effect on the results. Overall, the results of the analysis
indicate that properly defined damage variables are vital for accu-
rate prediction of the experimental results in terms of strength and
crack pattern, as also emphasised in previous studies [48–50].
The fracture energy of concrete is related to the strength of con-
crete, as indicated in Eq. (8). For specimen S3, with the concrete
compressive strength of (fc0 ), 34.7 MPa, Eq. (8) gives fracture
energy of 0.083 N/mm. Fig. 17, demonstrates the impact of three
different values (0.06 N/mm, 0.083 N/mm and 0.1 N/mm) of
Fig. 14. Effect of different dilation angle. fracture energy on the connection response. Fig. 17 shows that
40 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

Fig. 17. Effect of different Gf. Fig. 18. Comparing different analysis types.

the difference in the response of the connections is not significant, ses provide closely matched curves compared to the experimental
indicating that the contribution of the tensile behaviour of the con- backbone curves. The difference between explicit and standard
crete is not critical. This result was expected as there was sufficient solution decreased as the loading rate decreased. This is because,
reinforcement in most of the connections. in reduced loading rate, the increments are small. Thus, more accu-
Overall, among the studied parameters, the dilation angle (w), rate results can be computed. In all analyses, ABAQUS/Standard
the shape factor of the yield surface (Kc) and the use of the damage was used because it could fairly predict the behaviour of the test
parameters (dt and dc) were found to be critical for accurate con- specimens in terms of strength, stiffness, reinforcement yielding
crete modelling when using CDP model. Using larger mesh size and crack pattern.
and viscosity parameter can significantly reduce the computational
time, but at the same time, they can affect the accuracy of the 8. Parametric analysis
results. It should be pointed out that issues related to uncertainty
in input parameters and their influence on the reliability of the Based on the demonstrated accuracy of the developed FEM, this
computational results can be investigated further by means of a section presents a parametric investigation to study the effect of
global sensitivity analysis and other available engineering tools different design variables on the connection performance. Speci-
[66–70]. mens S3 model was chosen as the basis for the parametric analysis.
Each model is briefly described in terms of variables that were
7. Investigation on solution procedure adjusted, and then the results are discussed. Key areas of compar-
ison focused on the load versus displacement backbone curve.
In every nonlinear analysis, the major contributing factor to the
computational time directly depends on the procedure of solution 8.1. Influence of column dimensions
of the nonlinear equations. In this paper, ABAQUS/Standard was
used for all the analysis. However, the behaviour of RC elements The effect of column width and depth on the seismic response
can be simulated using ABAQUS/Explicit. The explicit analysis is of wide beam-column connections has been investigated in previ-
usually used for dynamic problems, but it can also be used for sta- ous studies [12–20]. These studies showed that the wide beam-
tic solutions with a low rate of loading, such as quasi-static exper- column connections with rectangular columns indicated robust
imental studies [42]. ABAQUS/Standard solves the nonlinear hysteresis performance and adequate energy dissipation. The influ-
equilibrium equations based on Newton’s method, and the solution ence of column dimensions on the seismic behaviour of wide
is obtained as a series of increments. The iterations are required to beam-column connections was investigated by using two addi-
achieve equilibrium through each increment step. The computa- tional models. Table 5 presents the design parameters of the
tional efficiency depends on the choice of increment size as more models.
repetition is needed in the case of large increment. Also, very large In specimen S3, the column had a cross-sectional dimension
increment can cause a convergence issue. Newton’s method has its were 300 mm  360 mm and the joint shear stress ratios, cd, was
pros and cons. The primary advantage is its quadratic convergence 1.63, as shown in Table 1. ACI 318-14 requires that the joint shear
which provides an improvement to the solution. The method has stress ratio be lower than 1.25. The provided beam depth (hb = 300
two main disadvantages: algebraically difficulties in deriving the mm) to column bar diameter (dc = 16 mm) was 18.75 which was
form of the Jacobian matrix, and computational effort in solving 76% of that required by ACI 352R-02 (20fy/420 = 24.7). The modi-
such Jacobian matrix. The explicit analysis uses the forward Euler fied connections were designed in a way that they satisfied all
technique for integrating the equations of motion through time. the design requirements of the ACI 318-14 and ACI352R-02 codes.
The solution is determined by explicitly advancing the kinematic Models, S3-Col300500 has deep column of 300 mm  500 mm
state from the previous increment data that are known at the and model S3-Col420360 has wide column of 420 mm  360
beginning of a new increment. mm. The beam dimension and reinforcement detailing were kept
Specimen S3 model was analysed using both standard and similar to the reference specimens. As shown in Table 5, to keep
explicit solvers. In the static analysis in ABAQUS/Standard, dis- the column-to-beam moment strength ratio similar to S3 model,
placement was applied to the beam, while in the quasi-static anal- the diameter of the longitudinal column bars (dc) was reduced,
ysis in ABAQUS/Explicit, two different velocities were applied. and consequently, (hb/dc) ratio in these models were improved.
Fig. 18 presents a comparison between two types of analyses and The joint shear stress ratios in the new models were smaller than
the experimental results. The figure shows that the explicit analy- 1.25 as required by the ACI 318-14. Reference specimen S3 had a
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 41

Table 5
Effect of column dimensions.

Specimen hb/dc Mn,b kNm Mr Tu kNm Tu/Tn Vj,e kN cd cACI Modea

S3-C300500 25 195.0 2.05 61 0.47 1034 1.17 1.25 B
S3-C420360 25 195.0 2.01 30 0.41 1034 1.16 1.25 B
a
B = beam flexural failure.

tional models were studied. In specimen, S3 beam depth was

300 mm. In the new models, the beam depth was increased to
360 and 420 mm. In the new models, the beam flexural bending
moment capacities were kept constant. Therefore, beam longitudi-
nal bars with smaller diameter were used in the models with
greater depth. Table 6 presents the design parameters of the mod-
els. Table 6 shows that increasing the beam depth and reducing the
beam flexural reinforcement leads to a reduction in the joint shear
stress ratio and torsional stress in the spandrel beam. The effect of
beam depth on the predicted lateral force is depicted in Fig. 20. As
can be seen, models with beam depths of 360 and 420 mm show
higher stiffness compared to the reference specimen.
The failure mode of the model was changed from joint shear in
the control specimen to beam flexural failure in the modified mod-
els. These results indicate that increasing the beam depth while
reducing the amount of beam longitudinal reinforcement has a sig-
Fig. 19. Influence of the column dimensions. nificant effect on improving the response of the wide beam-
column connections. The reason is that by increasing beam depth,
spandrel beam become stronger and reducing the beam bar size
joint shear failure after beam hinging. However, in the modified results in les torsion in spandrel beam and less joint shear stress.
connections, only beam flexural failure was expected. Fig. 19
shows the FE-predicted shear force capacity versus beam displace-
ment in the modified specimens. 8.3. Influence of column axial load
The predicted shear force capacity and stiffness of the connec-
tion with deep column increased compared to the control model. The exterior wide beam-column connection in the three-
As expected, no sign of the shear failure of the joint or torsional dimensional moment resistant frame is at the end of one direction
failure of spandrel beam was observed in the modified models. of the frame under seismic lateral loads and is therefore suscepti-
These results indicate that increasing the column dimensions and ble to high levels of column axial load variations due to dynamic
in particular column depth in wide beam-column connections, overturning effects. The effect of the column axial load on the per-
can significantly improve the seismic performance of the connec- formance of conventional and wide beam-column joints has been
tion. From the structural point of view, increasing the column investigated by many researchers. Haach et al. [43] evaluated the
depth in the wide beam-column connections will result in con- influence of the column axial load on the behaviour of monotoni-
tributing to the development of bigger compressive struts on the cally loaded exterior beam-column joints through numerical simu-
side face of the column, excessive anchorage length for the beam lations. They observed that there were significant values of strain
longitudinal bars, larger column-to-beam strength ratio and stron- in the stirrups inside the joint region in specimens with low col-
ger transverse (spandrel) beams. umn axial loads compared to specimens with high column axial
loads. Li and Kulkarni [20] investigated the effect of axial load on
seismic response of wide beam-column connection and found that
8.2. Influence of the beam depth the axial load improved the capacity of the joint up to a load of
0.25fc0 Ag, after which any further increases reduced the strength
The results of an experimental [71] and analytical [72] studies and stiffness. Masi et al. [25] tested two identical wide beam-
on conventional beam-column connections showed that for the column joint specimens under different axial load. Test results
same amount of beam flexural reinforcement, the shear strength showed that the axial load value is able to modify the damage evo-
of the joint decreased significantly as the beam depth increased. lution and, in turn, affect the deformation capacity of the joints.
The results of the analytical studies [30,31] showed that the wide The joint with higher axial load was subjected to delay vertical
beam width limitation has an inverse relationship with the beam cracking at the beam-column interface in compared to the joint
depth. By increasing the beam depth and keeping the same amount tested under low axial load. Davey et al. [28] studied the level of
of beam reinforcement, the beam width should be decreased. column axial load using two models: one with an axial load of
To investigate the effect of the beam depth on seismic perfor- 0.4fc0 Ag and one without axial load. No significant change in the
mance of the exterior wide beam-column connections, two addi- force versus deflection backbone curve was found with the

Table 6
Effect of beam depth.

Specimen hb As mm2 Mn,b kNm Mr Tu kNm Tu/Tn Vj,e kN ce cACI Modea

S3-B360 360 1232 196 1.92 61 0.68 777 1.22 1.25 B
S3-B420 420 1040 202 1.86 63 0.59 640 1.01 1.25 B
a
B = beam flexural failure.
42 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

8.4. Influence of beam longitudinal bar anchorage ratio

In addition to the geometrical parameters, reinforcement

detailing in a wide beam also plays a key role in effectively trans-
ferring the shear forces and bending moments from the wide beam
to the column. The beam bar anchorage ratio is the ratio of the
amount of wide beam longitudinal reinforcement which is
anchored inside the joint core to the total amount of longitudinal
bar. Through a set of experimental studies, Hatamoto et al. [16]
observed excessive pinching of the wide beam-column joints with
a smaller anchorage ratio. The revised ASCE 352-02 [5] provisions
recommend that at least one-third of flexural reinforcement in a
wide beam need to be anchored in or through the joint core to
ensure adequate anchorage. Kuang et al. [30] studied the behaviour
of previously tested specimens in literature and found out that for
wide beam-column connections with a square column section (hc/
Fig. 20. Influence of the beam depth. bc = 1), percentage of the beam reinforcement that passes through
the joint core should not be less than 50%, whereas for a rectangu-
lar column section with hc/bc > 3, a lower percentage of 30% may be
enough.
To study the influence of the beam anchorage ratio, the amount
of beam bar reinforcement anchored inside the column core was
varied. Table 7 presents the design parameters of the models. As
shown in Table 7, the direct effect of changing the anchorage ratio
is on the torsional demand on the spandrel beam. Anchoring more
reinforcement inside the column core reduces the torsional
demand on the spandrel beam.
Fig. 22 illustrates the force-displacement response of the con-
nections. Model with smaller anchorage ratio (S3-25%) had a tor-
sional failure of the spandrel beam and consequently a
significant reduction in strength and stiffness and ductility.

8.5. Influence of the beam length

As explained in [30,31], beam length has a critical role in the

Fig. 21. Influence of the axial load. seismic performance of the wide beam-column connections. It
was found that the beam width limitation can be relaxed when
using large span wide beams. However, the influence of beam
increase in axial load, and only a slight reduction in maximum drift
length on the performance of the beam-column connection has
was found for the model with no axial load.
In this study, the influence of column axial load on the seismic
behaviour of the wide beam-column connections was investigated
using the FEM analysis. The same loading histories as those used in
the previous analysis were applied, and the corresponding lateral
force versus horizontal displacements under different axial loading
levels are plotted in Fig. 21. To consider the effect of the axial force,
the applied column axial load varied in magnitude ranging from
0.0 fc0 Ag (full tension) to 0.5fc0 Ag (high axial compression force).
As observed in Fig. 21, the lateral force of the specimen increased
by around 4%, as the axial load was enhanced from 0.0fc0 Ag to 0.5fc-
0
Ag. When the axial load level was more than 0.3fc0 Ag, the exterior
wide beam-column joints experienced an optimum enhancement
in strength and stiffness.
The results of this study show that axial compression load on
the column enhanced the shear strength of the joint core through
increasing the width of the diagonal compression strut within the
core. A further increase in axial load, that is, beyond 0.3fc0 Ag had a
negligible effect on the behaviour of connection. Fig. 22. Influence of beam longitudinal bar anchorage ratio.

Table 7
Design parameters in models for studying the effect of anchorage ratio.

Specimen An Mn,b kNm Mr Tu kNm Tu/Tn Vj,e kN ce cACI Modea

S3-25% 25% 195 1.94 91 1.25 1034 1.63 1.25 BJS
S3-75% 75% 195 1.94 30 0.42 1034 1.63 1.25 BJ
a
B = beam flexural failure; J = joint shear failure; S = spandrel beam failure.
 H. Behnam et al. / Computers and Structures 205 (2018) 28–44 43

produce ductile behaviour while lower values produce brittle

behaviour.
 It was found that providing damage variables are necessary
when predicting the experimental results regarding crack pat-
tern. Damage parameter in compression was found to have a
significant effect on the results.
 Using larger mesh size and greater viscosity parameter can sig-
nificantly reduce the computational time, but they can affect
the accuracy of the results.
 The results indicated that the choice of the viscosity parameter
value in practical computations using the CDP model should be
made with great care and too large values of the relaxation
parameter should be avoided.
 Increasing a mesh size is led to the coarse mesh which causes an
error in prediction of the strength and failure mode. Results
confirmed that both Standard and Explicit analyses could be
Fig. 23. Influence of the beam length. used for modelling wide beam-column connections.
 Increasing the column width and depth have a favourable effect
on the seismic performance of the wide beam-column
never been studied before. To investigate the influence of beam connections.
length on the performance of exterior wide beam-column connec-  Increasing the beam depth while reducing the amount of beam
tion, two additional numerical models were considered. longitudinal reinforcement also enhance the response of the
The numerical models were same as the control specimens, but wide beam-column connections. The reason is that by increas-
the beam length was increased from 1500 mm to 2500 mm and ing beam depth, spandrel beam become stronger and reducing
3500 mm. In Fig. 23, the comparison was made with the flexural the beam bar size results in les torsion in spandrel beam and
strength of the beam. The beam flexural strength was constant less joint shear stress.
and was obtained by multiplying the peak beam load by the beam  Increasing beam bar anchorage ratio inside the column core was
length. The FEA showed that in the same lateral drift, the models found to have a beneficial effect on the seismic performance of
with larger beam length had a smaller stiffness and lower joint the wide beam-column connections.
shear cracks as most of the cracks, were developed along the beam  The FEA showed that in the same lateral drift, the models with
length. larger beam length had a smaller stiffness and lower joint shear
cracks as most of the cracks, were developed along the beam
9. Conclusions length.

This paper presents the FEA to predict the behaviour of RC exte-

Acknowledgments
rior wide beam-column connections. Four different beam-column
connections with various beam width ratios and different failure
The support of the Hong Kong Research Grants Council under
modes were simulated and analysed. The concrete damaged plas-
grant number 16209115 is gratefully acknowledged.
ticity model (CDP) in the finite element software ABAQUS was
adopted. The FEA predictions were compared with the experimen-
tal results. A sensitivity analysis was performed to investigate the References
effect of different material parameters on the accuracy of the [1] Kim J, LaFave JM. Key influence parameters for the joint shear behaviour of
results. ABAQUS/Standard was used because it could fairly predict reinforced concrete (RC) beam-column connections. Eng Struct
the behaviour of the test specimens regarding strength, stiffness, 2007;29:2523–39.
[2] Kotsovou G, Mouzakis H. Exterior RC beam-column joints: new design
reinforcement yielding and crack pattern. Moreover, the paramet-
approach. Eng Struct 2012;41:307–19.
ric investigation was carried out to explore the influence of the [3] Wong HF, Kuang JS. Predicting shear strength of RC interior beam-column
design parameters on the seismic performance of the wide joints by modified rotating-angle softened-truss model. Comput Struct
beam-column connections. Based on the evaluation of the beha- 2014;133:12–7.
[4] ACI Committee, American Concrete Institute, International Organization for
viour of the test specimens from the tests and the FEA, the follow- Standardization. Building code requirements for structural concrete (ACI 318-
ing conclusions can be drawn. 14) and commentary. ACI 318-14, Farmington Hills, MI; 2014.
[5] ACI Committee 352. Recommendations for design of beam-column joints in
monolithic reinforced concrete structures. ACI 352R-02, Farmington Hills, MI;
 In all cases of modelling, the ratio of the predicted maximum 2002.
lateral force and the drift in which all the beam longitudinal [6] NZS 3101. The design of concrete structures. Wellington, New Zealand;
bar yielded to that of the test results were within errors of 6% Standards Association of New Zealand; 2006.
[7] European Standard, EN 1998-1. Euro code 8-Design of structures for
and 19%, respectively. earthquake resistance-part 1: general rules, seismic actions and rules for
 The finite element analysis results, confirm the ability of the buildings. Brussels; 2004.
proposed model for predicting the behaviour of the wide [8] Benavent-Climent A. Seismic behavior of RC wide beam-column connections
under dynamic loading. J Earthquake Eng 2007;11:493–511.
beam-column connections. [9] Davey M, Abdouka K, Al-Mahaidi R. Post-tensioned band beams as moment
 Among the studied parameters, the dilation angle (w), the shape resisting frames under earthquake loading: a state-of-the-art review. Aust J
factor of the yield surface (Kc) and the use of the damage param- Struct Eng 2013;14:193–205.
[10] Behnam H, Kuang JS, Huang RYC. Exterior wide beam-column connections:
eters (dt and dc) were shown to be critical for accurate concrete
effect of beam width ratio on seismic behaviour. Eng Struct. http://doi.org/10.
modelling using CDP model. 1016/j.engstruct.2017.05.044.
 A default value of Kc = 0.667 gives the most accurate results [11] Gentry TR, Wight JK. Wide beam-column connections under earthquake-type
regarding ultimate strength and failure mode. loading. Earthq Spectra 1994;10:675–703.
[12] LaFave JM, Wight JK. Reinforced concrete exterior wide beam-column-slab
 The dilation angle (w) range between 38° and 42° for use in CDP connections subjected to lateral earthquake loading. ACI Struct J 1999;96
model is recommended. The higher values of the dilation angle (4):577–85.
44 H. Behnam et al. / Computers and Structures 205 (2018) 28–44

[13] Elsouri A, Harajli M. Seismic response of exterior RC wide beam-narrow [44] Belletti B, Esposito R, Walraven J. Shear capacity of normal, lightweight, and
column joints: earthquake-resistant versus as-built joints. Eng Struct high-strength concrete beams according to Model Code 2010. II: experimental
2013;57:394–405. results versus nonlinear finite element program results. J Struct Eng 2012;139
[14] Siah W, Stehle J, Mendis P, Goldsworthy H. Interior wide beam connections (9):1600–7.
subjected to lateral earthquake loading. Eng Struct 2003;25(3):281–91. [45] Rabczuk T, Akkermann J, Eibl J. A numerical model for reinforced concrete
[15] Stehle JS, Goldsworthy H, Mendis P. Reinforced concrete interior wide-band structures. Int J Solids Struct 2005;31 42(5):1327–54.
beam-column connections subjected to lateral earthquake loading. ACI Struct J [46] Jirasek M, Zimmermann T. Rotating crack model with transition to scalar
2001;98(3):270–9. damage. J Eng Mech 1998;124(3):277–84.
[16] Hatamoto H, Bessho S, Matsuzaki Y. Reinforced concrete wide-beam-to- [47] Mohamed AR, Shoukry MS, Saeed JM. Prediction of the behavior of reinforced
column subassemblages subjected to lateral load. Design of beam-column concrete deep beams with web openings using the finite element method.
joints for seismic resistance. Detroit, Michigan, USA: ACI Publications SP-123; Alexandria Eng J 2014;53:329–39.
1991. p. 291–316. [48] Genikomsou AS, Polak MA. Finite element analysis of punching shear of
[17] LaFave JM, Wight JK. Reinforced concrete wide-beam construction vs. concrete slabs using damaged plasticity model in ABAQUS. Eng Struct
conventional construction: resistance to lateral earthquake loads. Earthq 2015;98:38–48.
Spectra 2001;17(3):479–505. [49] Wosatko A, Pamin J, Polak MA. Application of damage-plasticity models in
[18] Quintero-Febres CG, Wight JK. Experimental study of reinforced concrete finite element analysis of punching shear. Comput Struct 2015;151:73–85.
interior wide beam-column connections subjected to lateral loading. ACI [50] Abdelatif AO, Owen JS, Hussein MF. Modelling the prestress transfer in pre-
Struct J 2001;98(4):572–82. tensioned concrete elements. Finite Elem Anal Des 2015;28(94):47–63.
[19] Kulkarni SA, Li B. Seismic behavior of reinforced concrete interior wide-beam [51] Nzabonimpa JD, Hong WK, Kim J. Nonlinear finite element model for the novel
column joints. J Earthquake Eng 2008;13:80–99. mechanical beam-column joints of precast concrete-based frames. Comput
[20] Li B, Kulkarni SA. Seismic behavior of reinforced concrete exterior wide beam- Struct 2017;30(189):31–48.
column joints. J Struct Eng 2009;136:26–36. [52] Hillerborg A, Modeer M, Petersson PE. Analysis of crack formation and crack
[21] Benavent-Climent A, Cahis X, Zahran R. Exterior wide beam-column growth in concrete by means of fracture mechanics and finite elements. Cem
connections in existing RC frames subjected to lateral earthquake loads. Eng Concr Res 1976;6:773–82.
Struct 2009;31(7):1414–24. [53] Jankowiak T, Lodygowski T. Identification of parameters of concrete damage
[22] Benavent-Climent A, Cahis X, Vico J. Interior wide beam-column connections plasticity constitutive model. Found Civ Environ Eng 2005;6:53–69.
in existing RC frames subjected to lateral earthquake loading. Bull Earthq Eng [54] Birtel V, Mark P. Parameterised finite element modelling of RC beam shear
2010;8(2):401–20. failure. In: ABAQUS users’ conference; 2006. p. 95–108.
[23] Goldsworthy HM, Abdouka K. Displacement-based assessment of non-ductile [55] Grassl P, Jirasek M. Damage-plastic model for concrete failure. Int J Solids
exterior wide band beam-column connections. J Earthquake Eng 2012;16 Struct 2006;43:7166–96.
(1):61–82. [56] Rabczuk T, Zi G, Bordas S, Nguyen-Xuan H. A geometrically non-linear three-
[24] Fadwa I, Ali TA, Nazih E, Sara M. Reinforced concrete wide and conventional dimensional cohesive crack method for reinforced concrete structures. Eng
beam–column connections subjected to lateral load. Eng Struct Fract Mech 2008;75(16):4740–58.
2014;76:34–48. [57] Rabczuk T, Belytschko T. Application of particle methods to static fracture of
[25] Masi A, Santarsiero G, Mossucca A, Nigro D. Influence of axial load on the reinforced concrete structures. Int J Fract 2006;137(1–4):19–49.
seismic behavior of RC beam-column joints with wide beam. Applied Mech [58] Celik O. Probabilistic assessment of non-ductile reinforced concrete frames
Mater 2014;508:208–14. Trans Tech Publications. susceptible to Mid-America ground motions PhD thesis. Atlanta, GA: Georgia
[26] Elsouri A, Harajli M. Interior RC wide beam-narrow column joints: potential Institute of Technology; 2007.
for improving seismic resistance. Eng Struct 2015;99:42–55. [59] Rabczuk T, Zi G, Gerstenberger A, Wall WA. A new crack tip element for the
[27] Mirzabagheri S, Tasnimi AA, Mohammadi MS. Behavior of interior RC wide and phantom-node method with arbitrary cohesive cracks. Int J Numer Methods
conventional beam-column roof joints under cyclic load. Eng Struct 2016;15 Eng 2008;75(5):577–99.
(111):333–44. [60] Rabczuk T, Zi G, Bordas S, Nguyen-Xuan H. A simple and robust three-
[28] Davey MJ, Abdouka K, Al-Mahaidi R. Exterior post-tensioned band beam to dimensional cracking-particle method without enrichment. Comp Methods
column connections under earthquake loading. Aust J Struct Eng 2016;17 Appl Mech Eng 2010;199(37):2437–55.
(1):14–27. [61] Deaton JB. Nonlinear finite element analysis of reinforced concrete exterior
[29] Behnam H, Kuang JS. Exterior wide beam-column connections: effect of beam-column joints with non-seismic detailing PhD thesis. Atlanta,
spandrel beam on seismic behaviour. J Struct Eng 2018. ASCE. STENG-5981. GA: Georgia Institute of Technology; 2013.
[30] Kuang JS, Behnam H, Huang Q. Effective beam width of wide beam-column [62] CEB-FIP-Model Code 1990: design code. London: Thomas Telford; 1993.
connections. J ICE-Struct Build 2017:1–18. https://doi.org/10.1680/ [63] Wu JY, Li J, Faria R. An energy release rate-based plastic-damage model for
jstbu.15.00124. concrete. Int J Solids Struct 2006;43(3):583–612.
[31] Behnam H, Kuang JS, Wong HF, Huang Q, Al-Mahaidi R. Analysis of laterally [64] Voyiadjis GZ, Taqieddin ZN. Elastic plastic and damage model for concrete
loaded exterior wide beam-column connections. Mag Concr Res 2017;20:1–2. materials: part I-theoretical formulation. Int J Struct Changes Solid 2009;1
[32] Behnam H, Kuang JS, Abdouka K. Effect of post-tensioned spandrel beam on (1):31–59.
wide beam-column connections. Mag Concr Res 2018;1(70):28–41. [65] Malm R. Predicting shear type crack initiation and growth in concrete with
[33] Luk SH, Kuang JS. Seismic performance and force transfer of wide beam- non-linear finite element method, 2009 Doctoral dissertation. Stockholm,
column joints in concrete buildings. Proc ICE Eng Comput Mech 2017;170 Sweden: Department of Civil and Architectural Engineering. Royal Institute of
(2):71–88. Technology (KTH); 2009.
[34] Richard Ralph M, Abbott Barry J. Versatile elastic-plastic stress-strain formula. [66] Stefanou G. The stochastic finite element method: past, present and future.
J Eng Mech Div 1975;101:511–5. Comput Methods Appl Mech Eng 2009;198(9):1031–51. 15.
[35] Grassl P, Lundgren K, Gylltoft K. Concrete in compression: a plasticity theory [67] Vu-Bac N, Rafiee R, Zhuang X, Lahmer T, Rabczuk T. Uncertainty quantification
with a novel hardening law. Int J Solids Struct 2002;39:5205–23. for multiscale modeling of polymer nanocomposites with correlated
[36] Papanikolaou VK, Kappos AJ. Confinement-sensitive plasticity constitutive parameters. Compos B Eng 2015;68:446–64.
model for concrete in triaxial compression. Int J Solids Struct 2007;44:7021–48. [68] Vu-Bac N, Silani M, Lahmer T, Zhuang X, Rabczuk T. A unified framework for
[37] Yu T, Teng J, Wong Y, Dong S. Finite element modeling of confined concrete-II: stochastic predictions of mechanical properties of polymeric nanocomposites.
plastic-damage model. Eng Struct 2010;32:680–91. Comput Mater Sci 2015;96:520–35.
[38] Loland K. Continuous damage model for load-response estimation of concrete. [69] Hamdia KM, Msekh MA, Silani M, Vu-Bac N, Zhuang X, Nguyen-Thoi T, et al.
Cem Concr Res 1980;10:395–402. Uncertainty quantification of the fracture properties of polymeric
[39] Lubarda V, Krajcinovic D, Mastilovic S. Damage model for brittle elastic solids nanocomposites based on phase field modeling. Comp Struct
with unequal tensile and compressive strengths. Eng Fract Mech 2015;133:1177–90.
1994;49:681–97. [70] Engen M, Hendriks MA, Köhler J, Overli JA, Aldstedt E. A quantification of the
[40] Lubliner J, Oliver J, Oller S, Onate E. A plastic-damage model for concrete. Int J modelling uncertainty of non-linear finite element analyses of large concrete
Solids Struct 1989;25:299–326. structures. Struct Saf 2017;31:64:1–8.
[41] Lee J, Fenves GL. Plastic-damage model for cyclic loading of concrete [71] Wong HF, Kuang JS. Effects of beam-column depth ratio on joint seismic
structures. J Eng Mech 1998;124:892–900. behaviour. J ICE-Struct Build 2008;161(2):91–101.
[42] ABAQUS. Analysis user’s manual 6.10. Providence, IR: Dassault Systems [72] Park S, Mosalam KM. Analytical model for predicting shear strength of
Simulia Corp. SIMULIA; 2013. unreinforced exterior beam-column joints. ACI Struct J 2012;109(2):149–60.
[43] Haach VG, El AL, El Debs MK. Evaluation of the influence of the column axial
load on the behavior of monotonically loaded RC exterior beam-column joints
through numerical simulations. Eng Struct 2008;30(4):965–75.