Engineering Structures 32 (2010) 226–237

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

An improved design formula for a biaxially loaded slender RC column

Hyo-Gyoung Kwak ∗,1 , Ji-Hyun Kwak 2
Department of Civil and Environmental Engineering, Korea Advanced Institute of Science and Technology, Daejeon 305-701, Republic of Korea

article info abstract

Article history: Nonlinear analyses are conducted to evaluate the ultimate resisting capacity of slender reinforced
Received 20 November 2008 concrete (RC) columns subjected to an axial load with biaxial bending moments. Consideration is given
Received in revised form to the geometric nonlinearities caused by the P–∆ effect and the long-term behavior of concrete and to
1 September 2009
the material nonlinearities caused by the cracking of concrete and the yielding of steel. In addition, the
Accepted 3 September 2009
Available online 26 September 2009
biaxial stress state in an RC section is simulated on the basis of a fiber model. Because of the complexity
of Bresler’s load contour method, which was introduced in the ACI 318 code, this paper introduces a new
Keywords:
design approach to the construction of the failure surface of a slender RC column subjected to biaxial
Resisting capacity bending. Through a parametric study of slender RC columns, where consideration is given to the P–∆
Slender RC column effect and the time-dependent deformation of concrete, two regression formulas are proposed on the basis
Biaxial bending of the slenderness ratio and the creep deformation of concrete. Furthermore, the direct multiplication of
Long-term behavior the proposed formulas on the P–M interaction diagram for a short RC column subjected to axial force and
a uniaxial bending moment enables a P–M interaction diagram to be generated for a slender RC column
subjected to long-term axial force and biaxial bending moments. Correlation studies between analytical
and experimental results are conducted with the objective of establishing the validity of the introduced
numerical model. In addition, the ultimate resisting capacities calculated from the regression formula
are compared with those obtained from rigorous nonlinear analyses and from the ACI formula, with the
objective of establishing the relative efficiency of the proposed regression formula.
© 2009 Elsevier Ltd. All rights reserved.

1. Introduction buildings and the interior of lobbies. The high-rise building, ‘‘Burj
Al Arab’’, composed of interior columns with a slenderness ratio of
A reinforced concrete (RC) column is a structural member used about 60 is a typical example. Moreover, the use of higher strength
primarily to support compressive loads; it is usually subjected to steel and concrete has led to a corresponding increase in the use of
the axial force and bending moment which may be either due to slender members. In contrast to a short column, whose ultimate
the end restraint arising from the monolithic placement of floor load at a given eccentricity is governed only by the strength of
beams and columns or to the eccentricity of imperfect alignment. the materials and the dimensions of the cross-section, the design
Accordingly, the P–M interaction diagram, which represents a plot of slender RC columns requires the consideration of secondary
of the column axial load capacity against the developed moment, effects, such as the P–∆ effect and the creep deformation of
becomes the key concept for understanding the cross-sectional concrete, because these kinds of effects cause additional bending
behavior of RC columns and forms the basis for much of the design moments.
procedure for concrete columns. That is, the column section is In their research contribution to the behavior and design of
usually designed to ensure that the forces acting on a member fall slender RC columns, Bazant et al. [1] introduced an analytical way
inside the P–M interaction diagram. of calculating the resisting capacity of slender RC columns; in
particular, they used a sinusoidal function to describe the column
Recently, because of their architectural aesthetics and efficiency
deformation. Kim et al. [2] also conducted an experimental study
in the use of space, relatively slender columns have been widely
of RC short columns subjected to biaxial bending moments and
used in many building structures, either throughout the entire
numerically analyzed the specimens by considering the material
building or in some parts of the structure, such as the exterior of
nonlinearity on the basis of the later approach. Recently, Yalcin
and Saatcioglu [3] developed an analytical model which considers
the influence of anchorage slip and plastic hinge length on the
∗ Corresponding author. Tel.: +82 42 350 3621; fax: +82 42 350 4546. nonlinear behavior of RC columns. In addition, Kwak and Kim [4]
E-mail addresses: khg@kaist.ac.kr (H.-G. Kwak), alfis@kaist.ac.kr (J.-H. Kwak). proposed an improved design criterion that can be used to estimate
1 Professor. the load-carrying capacity of slender RC columns subjected to
2 Doctoral candidate. axial force and a uniaxial bending moment. They adopted a layer
0141-0296/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2009.09.009
 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237 227

model and considered the initial stress matrix for simulation of the 2.1. Concrete
P–∆ effect. For the biaxial behavior of the RC column, Bresler [5]
suggested a load contour method and Parme et al. [6] proposed Based on the principle of superposition, the total uniaxial
approximations to the exponents using a graphical interpretation. concrete strain, εc (t ), at any time, t, is assumed to be composed
Fafitis [7] analytically proposed the exact method using Green’s of the mechanical strain, εcm (t ), which is caused by short-term
theorem, and Rodriguez-Gutierrez et al. [8] presented an analytical service loads; and the nonmechanical strain, εcnm (t ), consists of
model for the prestressed case. Silva et al. [9] suggested a failure creep strain, εccr (t ), and shrinkage strain, εcsh (t ).
criterion based on the design optimization approach. However, on
account of the paucity of studies on extremely slender RC columns εc (t ) = εcm (t ) + εcn m(t ) = εcm (t ) + εcc r (t ) + εcs h(t ). (1)
subjected to biaxial bending [10,11], it is still difficult to directly
apply the obtained research results to design practice. Shrinkage strain can be evaluated directly by utilizing the
Many design codes [12,13], including the ACI method, use the shrinkage model proposed in design codes because it is defined
moment magnification method [13]. In this method, a column sec- as the volume change that occurs independently of any imposed
tion is designed to ensure that the ultimate load and the magnified stress; in contrast, creep is defined as an increase in strain under
ultimate moment exist inside the P–M interaction diagram of the sustained stress. In this paper, the first-order recursive algorithm
RC section. In spite of its simplicity in application, however, this proposed by Kabir and Scordelis [15], which is based on the
approach yields very conservative results for a column with a rela- expansion of creep compliance, was adopted because this model
tively high slenderness ratio. In addition, insufficient consideration can effectively simulate the stress history in spite of its simplicity
of the creep deformation of concrete accelerates the inaccuracy of of application.
estimating the ultimate resisting capacity of RC columns. Thus, an The increment of creep strain from time to time for the uniaxial
accurate estimation of the ultimate resisting capacity of slender RC stress state can be expressed as follows:
columns subjected to biaxial bending cannot be expected from the
m
additional consideration of the biaxial bending effect [13] upon the
Ain−1 (1 − e−λi 1tn ),
X
conventional design approach for slender RC columns. 1εnc =
n=i−1 (2)
This paper introduces a numerical model that can predict the
−λi 1tn
ultimate resisting capacity of slender RC columns subjected to Ain = Ain−1 e + ai (τ )1σn
biaxial bending moments. The geometric nonlinearity caused by
the P–∆ effect and the creep deformation of concrete are taken where 1εnc is the increment of creep strain; λi are the inverse
into account, and a fiber approach in which longitudinal fibers [14] retardation times; ai (τ ) are the constants depending on the age at
are subdivided into sections is used to simulate the different the loading, τ ; m is the number of time steps; Ai is a hidden state
material properties across the sectional depth. In addition, through variable that helps determine the effects of past time steps. The
many parametric studies that focus on changes in the major initial values are Ai1 ai (t1 ) · 1σ1 at n = 1.
design variables, such as the slenderness ratio, the eccentricity, Before the creep strain is calculated with Eq. (2), the parameters
and the ultimate creep coefficient, an improved design criterion m, ai (τ ) and λi must be determined. Because the use of the
is introduced for the design of RC columns subjected to biaxial compliance function (J (t , τ ) = 1/E (t ) + C (t , τ )) in the form of
bending moments. Because of the consistency of the current design a Dirichlet series induces some numerical difficulties if a separate
procedure, the introduced criterion should be based on the load term is not used to represent any instantaneous deformation, this
contour method of Bresler [5] and adopted in the ACI 318 code [13]. study uses the creep compliance C (t , τ ) directly, as shown in
In contrast to the conventional load contour method [6], in Eq. (3) for m = 4 and for assumed corresponding retardation times
which the exponential constant is determined from a graphical of 8.0, 80.0, 800.0 and 8000.0. Note also that the values of ai (τ ) are
interpretation, the proposed criterion can be used to directly determined by combining the method of least squares with Kabir’s
determine the exponential constant when the design variables Dirichlet series creep compliance [15]. Furthermore, the changes
are applied to the regression formula. Moreover, a more accurate in the material properties with time are considered on the basis of
estimation of the ultimate resisting capacity of slender RC columns the ACI model [16].
subjected to biaxial bending moments can also be expected
because the regression formula was proposed on the basis of m
ai (τ )[1 − e−λi (t −τ ) ].
X
the results obtained from rigorous numerical analyses. Finally, C (t , τ ) = (3)
the ultimate resisting capacities of slender RC columns subjected i =1

to biaxial bending moments are estimated with the aid of the For each fiber the concrete stress that corresponds to the me-
proposed criterion and compared with the corresponding values chanical strain can be calculated by using the stress–strain relation
calculated by rigorous nonlinear analyses for the purpose of of concrete. Because concrete is used mostly in compression, the
establishing the accuracy of the proposed design criterion. stress–strain relation in compression is of primary interest. Among
the numerous mathematical models currently used in the analy-
2. Material properties sis of RC structures, the monotonic envelope curve introduced by
Kent and Park [17] and later extended by Scott et al. [18] is used
The constitutive relations in a section of composite beam in this paper because of its simplicity and computational conve-
elements can be formulated on the basis of the following simplified nience. In this model, as shown in Fig. 1(a), the monotonic concrete
assumptions: (1) the element is divided into imaginary fibers stress–strain relation in compression is described by the following
to describe the different material properties; (2) plane sections three regions:
remain planes to represent the linearity in the strain distribution
of any section at any time; (3) a perfect bond between the concrete
"  2 #
εc εc
 
and reinforcing steel exists throughout all the loading stages; and σc = Kfc 0
2 − , εc ≤ εc0 (4)
(4) the constitutive materials only carry uniaxial stress. In addition, εc0 εc0
shear deformation is not taken into account in the formulation
σc = Kfc0 [1 − Zi (εc − εc0 )], εc0 ≤ εc ≤ εu (5)
because the shear effect is expected to be negligible in the slender
RC columns. σc = 0.2Kfc , 0
ε c ≥ εu (6)
228 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237

σc σc

f'c Kf'c
CONFINED CONCRETE ft
UNCONFINED CONCRETE

Z2
1
Z1
Et
Ec
0.2 Kf'c 1

εc εt
εc0=0.002 εu εcr ε0

(a) Compressive region. (b) Tensile region.

Fig. 1. Stress–strain relation of concrete.

where σs

ρs fyh
εc0 = 0.002K , K =1+ (7) Es2
fc0 fy
1
0.5
Zi = q , (i = 1, 2) (8)
3+0.0284fc0 h0
14.21fc0 −1000
+ 0.75ρs sh
− 0.002K Es1

-εu 1
where εc0 is the concrete strain at maximum stress, K is a εs
0 εu
factor which accounts for the strength in kgf/cm2 (1 kgf/cm2 =
0.098 MPA), fyh is the yield strength of the stirrups in kgf/cm2 ,
ρs is the ratio of the concrete core measured to the outside of
the stirrups, h0 is the width of the concrete core measured to the
-fy
outside of hoops or ties, and sh is the center-to-center spacing of
tie or hoop sets.
On the other hand, it is assumed that concrete is linearly elastic
Fig. 2. Stress–strain relation of steel.
in the tension region. Beyond the tensile strength, the tensile stress
decreases linearly as the principal tensile strain increases (see
Fig. 1(b)). Ultimate failure is assumed to take place as a result when the nonmechanical deformations of creep and shrinkage in
of cracking whenever the principal tensile strain, as shown in concrete are considered as a consequence of the restraint on the
Fig. 1(b), exceeds the value of ε0 = 2 · Gf ft0 · ln(3/b)/(3 − b), where b axial deformation by reinforcements embedded in the concrete
is the element length and Gf is the fracture energy that is dissipated matrix, additional consideration is required for the nonmechanical
in the formation of a crack of unit length per unit thickness strains.
and Gf is considered as a material property. The value of ε0 is
When the neutral axis is being determined in consideration of
derived from the fracture mechanics concept by equating the crack
the bending effects, the mechanical strains of concrete (εcm ) and
energy release with the fracture toughness of concrete, Gf [19]. An
experimental study of Welch and Haismen [20] indicates that, for steel (εsm = εst ) need to be partitioned into an axial strain (εca
m
=
normal strength concrete, the value of Gf /ft0 is in the range from εsa ) and a bending strain (εcb = εsb ) as shown in Fig. 3. Because
t m t

0.005 mm to 0.01 mm. If Gf and ft0 are known from measurement, the axial strain is constant across the section and the bending
ε0 can be determined. strain is zero at the neutral axis, the bending strains of concrete
and steel at any layer can be expressed by εcb m
= εct − εcnm − εcam

2.2. Steel and, εsb = εb = εs − εsa respectively. For a neutral axis, the
m t t t

stress corresponding to the total strain can be calculated from the

Reinforcing steel is modeled as a linear elastic and linear stress–strain curves of the constitutive materials, and iterations
strain hardening material with a yield stress, fy , as shown in with the bisection method can be repeated until an error for the
Fig. 2. The reasons for this approximation are as follows: (1) the axial force and bending moment calculated by the total strain and
computational convenience of the model; and (2) the behavior bending strain are within the given tolerance. The same iteration
of RC members is greatly affected by the yielding of reinforcing procedure is used for the other axis in the case of biaxial bending
steel when the structure is subjected to a monotonic bending behavior.
moment. In addition, thermal strain is the only nonmechanical
strain expected in steel. However, that factor is not considered in
this paper, which means that the mechanical strain can be directly 4. Construction of an equilibrium equation
calculated from the total strain of steel (εs (t ) = εsm (t )).
All the constitutive equations are derived on the basis of
3. Determination of a neutral axis the assumed displacement field formulation [21], and a spatial
beam element with five nodal displacement vectors in its local
Unlike a beam element subjected to a bending moment only, a coordinate system (d1 to d5 and d6 to d10 in Fig. 4) is adopted. If the
column element is subjected to axial force and bending moments; three-dimensional local coordinate system is expressed in terms of
hence, the neutral axis of a column section cannot be calculated the three independent axes of ξ , η and ζ , and their corresponding
solely on the basis of the normal force equilibrium. Note that lateral deformations are expressed as u, v and w , respectively, then
 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237 229

(a) RC column section. (b) Strain distribution in a section.

Fig. 3. Strain components at a section.

Fig. 4. Displacements and forces in a beam element.

the strain field can be defined by

2 !
∂u ∂ 2v ∂v ∂w
 2 
1
εξ ξ = −η 2 + +
∂ξ ∂η 2 ∂η ∂ζ (9)
εηη = εζ ζ = εξ ξ = εζ ξ = εηξ = 0.

In addition, the displacements u, v and w can also be expressed

in terms of the displacement shape functions of N eu , Nev and N
fw. Fig. 5. Imaginary fibers in a section.

eu · d̃ = (N1 , 0, 0, 0, 0, N6 , 0, 0, 0, 0)T · d̃ Z " T T

!
u=N (10) ∂ Neu ∂ 2 Nev ∂ 2 Nev ∂ Neu
K̃PM = − Eη +
v = Nev · d̃ = (0, N2 , 0, 0, N5 , 0, N7 , 0, 0, N10 )T · d̃ (11) V ∂ξ ∂ξ 2 ∂ξ 2 ∂ξ
T 2 T
!#
w=Nw · d̃ = (0, 0, N3 , N4 , 0, 0, 0, N8 , N9 , 0) · d̃
T
(12) ∂ Neu ∂ N f w ∂ 2N
f w ∂ Nu
e
+ Eζ
f
+ dV (15)
∂ξ ∂ξ 2 ∂ξ 2 ∂ξ
where N1 = 1 − p, N2 = 1 − 3p + 2p , N3 = N2 , N4 = l(−p +
2 3

2p2 − p3 ), N5 = −N4 , N6 = p, N7 = 3p2 − 2p3 , N8 = N7 , N9 = Z " T T 2

!#
∂ 2 Nev ∂ 2 N
fw ∂ 2Nfw ∂ Nv
e
l(p2 − p3 ), N10 = −N9 , and the non-dimensional parameter p K̃MM = E ηζ + dV (16)
denotes ξ /l, that is, the position along the axis of the beam element V ∂ξ 2 ∂ξ 2 ∂ξ 2 ∂ξ 2
and the nodal displacement vector d̃ = (d1 , d2 , . . . d9 , d10 )T . T T
!
∂ Nev ∂ Nev ∂N w ∂ Nw
Z l f f
With the application of the virtual work principle to a finite KG =
e + dV . (17)
element on the basis of the energy conservation law and neglecting 0 ∂ξ ∂ξ ∂ξ ∂ξ
a higher-order incremental term, the equilibrium equation, which Moreover, with the layer approach, where a typical section is
represents the relation between the nodal displacement vector, divided into imaginary layers (see Fig. 5), the sectional stiffness
d̃, and the nodal force vector f˜ = (f1 , f2 , . . . f9 , f10 )T , can be terms required in the calculation of stiffness and axial force
obtained [21] as components in Eqs. (13)–(17) can be evaluated by the summation
of all the layers
Pncin the two P orthogonal directionsPof ξ and η, that
(K̃ + P KeG · d̃) = f˜ (13) is, EA = Ec i=1 Aci + Es
ns
j=1 Asj , ESζ = −Ec i=1 ζci Aci − Es
nc

ζ A , ESη = −Ec ni=c 1 ηci Aci − Es nj=s 1 ηsj Asj , EIζ = Ec

Pns P P
where P = A E εξ ξ dA is the axial force applied at the centroid of a Pjn=c 1 s2j sj
R
i = 1 η ci A ci + E s η As , EIη = Ec i=1 ζci Aci + Es nj=s 1 ζs2j Asj ,
Pns 2 Pnc 2 P
section. In addition, the elastic stiffness, K̃ = K
e0 + K̃PM + K̃MM , and Pnc j=1 sj j
i = 1 η ci ζ ci A ci + E s j=1 ηcj ζsj Asj .
Pns
the geometric stiffness, KeG , can be represented by and EIηζ = Ec

Z " T T
∂ Neu ∂ Neu ∂ 2 Nev ∂ 2 Nev 5. Solution algorithm
e0 =
K E + E η2
V ∂ξ ∂ξ ∂ξ 2 ∂ξ 2
T
# Every nonlinear analysis algorithm consists of four basic steps:
∂ 2N
f w ∂ Nw
2f
the formation of a current stiffness matrix, the solution of the
+ Eζ 2 dV (14)
∂ξ 2 ∂ξ 2 equilibrium equations for the displacement increments, the state
230 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237

Loading Loading Table 1

point point Comparison of analytical (Pu,a ) and experimental (Pu,t ) data from Kim et al. [2].

Section Pu,a Pu,t Pu,a /Pu,t

d

d
RS0 206.92 205.98 1.00
e e
RS30 227.51 217.87 1.04
RS45 260.86 240.45 1.08
RS60 294.20 295.54 1.00
RS90 490.33 418.73 1.17
(a) SS series. (b) RS series. RS90 490.33 443.65 1.11
SS0 124.54 127.03 0.98
SS30 109.83 112.0 0.98
Fig. 6. Section shapes (e = 4 cm).
SS45 108.85 109.0 1.00

determination of all elements in the model, and a convergence

check. Because the global stiffness matrix of the structure depends to accurately predict the ultimate loads of hinged RC columns
on the displacement increments, the solution of the equilibrium subjected to biaxial bending moments.
equations is typically accompanied by an iterative method during An additional comparison is made with the numerical calcu-
the convergence check. The nonlinear solution scheme selected in lations introduced by Bazant et al. [23,1] to verify the proposed
this paper uses a tangent stiffness matrix at the beginning of each numerical model. The steel ratio and cross-sectional dimensions
load step and a constant stiffness matrix during the subsequent of selected columns with slenderness ratios of L/r = 10, 70 and
correction phase. This method is the incremental-iterative method. 100 are presented in Fig. 8. The same material properties of con-
In addition, the criteria for measuring the convergence of the crete and steel as those in the previous analytical study are used,
iterative solution are generally based on the accuracy with which and they have the following values: fc0 = 352 kgf/cm2 , Es =
the global equilibrium equations are satisfied or on the accuracy 2.04 × 106 kgf/cm2 and fy = 4220 kgf/cm2 .
with which the total displacements are determined. The accuracy For the design of slender RC columns, the ACI318 design
with which the global equilibrium is satisfied can be controlled code [13] recommends a simple approximate formula based on
by the magnitude of the unbalanced nodal forces. Hence, the the moment magnification factor. When a column is subjected to
convergence criteria for the unbalanced nodal forces are used in the ultimate loads of Pu and Mu , the load and moment used in the
this paper. These criteria can be expressed as design of the column section are assumed to be Pu and δ · Mu ,
where δ is the moment magnification factor and is calculated from
max
|Funbal .
. | ≤ Tol F ,
max
|Munbal .
. | ≤ Tol M (18) δ = Cm /(1 − Pu /φk · Pe ) ≥ 1; Pe = π 2 EI /(kl2u ) is the elastic
where | max
| and |
Funbal max
| are the absolute values of the maximum
Munbal buckling load where EI = (0.2Ec Ig + Es Ise )/(1 + βd ); and Cm is
. .
an equivalent uniform moment diagram factor. Note that Cm = 1
unbalanced axial force and bending moment, respectively, and
Tol .F and Tol .M are the specified tolerances that correspond to the
for the columns used in this example. Moreover, as mentioned
in a previous study [24], the stiffness reduction factor, φk , which
axial force and the bending moment. In this study, Tol .F = Tol .M =
was designed in consideration of the inevitable random variability
0.01 is used for the tolerance values. More details of the solution
of the materials, has not been incorporated for the purpose of
procedures, including a flow diagram, can be found elsewhere [22].
comparison with the numerical results (that is, φk = 1.0 is
assumed).
6. Verification of the analytical model As shown in Fig. 8, the results of the proposed numerical model
are in good agreement with those of Bazant et al. [1], which leads
6.1. Short-term loading to the conclusion that the proposed numerical model can be used
to accurately predict the ultimate loads of slender RC columns.
The experimental results from several hinged RC columns Fig. 8 also leads to the following conclusions: (1) as the slenderness
tested by Kim et al. [2] are used to investigate the validity of ratio increases, the difference between the ACI strength interaction
the proposed analytical model. As shown in Fig. 6, two groups curve and the proposed model gradually increases; (2) the ACI
of specimens are considered. The first group of specimens is method may underestimate the resisting capacity of slender RC
a series of square columns (the SS series in Table 1) with a columns; and (3) the ACI method does not achieve a uniform
width × depth of 100 mm × 100 mm; the second group of difference between the results predicted by the ACI method and
specimens is a series of rectangular columns (the RS series in the results calculated on the basis of a rigorous analysis, over the
Table 1) with a width × depth of 200 mm × 100 mm. All entire interaction diagram for columns with L/r = 70 and 100.
the columns maintained a primary eccentricity of e = 4 cm Hence the ACI method has limitations in the exact consideration of
from the centroid, regardless of the magnitude of the angle the P–∆ effect.
between the loading point and the horizontal axis; furthermore,
the numbers added subsequently to the series named in Table 1 6.2. Long-term loading
refer to the corresponding angles. The material properties and
sectional properties used in the experiment are as follows: fc0 = Correlation studies of the analytical and experimental results
247 kgf/cm2 , fy = 3940 kgf/cm2 , Es = 2.04 × 106 kgf/cm2 , l = of Drysdale et al. [26] were conducted to verify the accuracy
130 cm, L/r = 45, d = 23 cm and the steel ratio of ρs = 0.0285 of the proposed numerical model for long-term behavior. The
for the SS series and ρs = 0.0214 for the RS series. The detailed geometry and cross-sectional dimensions of the tested columns
geometry of the stirrups is equal to that of Kim et al. [2]. are presented in Fig. 9, and those columns have a slenderness
The ultimate loads and the corresponding load–deflection ratio of L/r = 107. The following material properties, which
curves of columns measured experimentally were compared with have the same values as those used in the experiments, are also
those obtained from the proposed numerical model in Table 1 and used: fc0 = 282 kgf/cm2 , Es = 2.04 × 106 kgf/cm2 and fy =
Fig. 7. Regardless of the combination of biaxial bending moments 3940 kgf/cm2 . When the time-dependent behavior of RC columns
induced from the magnitude of the eccentricity angle, good is traced, consideration is given to the relation between the
agreements were obtained for the individual columns, leading to creep and shrinkage of concrete and the aging effect of concrete;
the conclusion that the introduced numerical model can be used furthermore, an ultimate creep coefficient of cu = 3.0 and an
 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237 231

(a) SS0. (b) SS30. (c) SS45.

(d) RS0. (e) RS30. (f) RS45.

(g) RS60. (h) RS90.

Fig. 7. Comparison of load–deflection curves.

ultimate shrinkage strain of εsh ∞

= 600 × 10−6 are used on the values of Pou /Put than the results of the experimental data for
basis of the ACI model. the entire range of βd because the increase in concrete strength
Fig. 9 shows a plot of Pou /Put versus βd , where βd = PD /(PD + can accurately be considered in numerical analyses when the age
PL ) is the ratio of the dead load to the total load, Pou is the effect is implemented. In this respect, the two models proposed
short-term failure load, and Put is the long-term failure load. by the ACI [13] and Bazant et al. [25] yield conservative results
When an RC column is subjected to a small sustained load, and underestimate the resisting capacity of slender RC columns as
which means that βd has a small value, the points of failure βd increases. In addition, because the dead load represents 50% of
caused by the subsequent sudden overload generally indicate a the total design load in most RC columns (βd ≥ 0.5), any direct
substantial increase in strength. This phenomenon, however, is application of the ACI model or the Bazant model may result in a
almost nonexistent in the tests of Drysdale et al. because drying conservative design. On the other hand, the numerical results from
prevents hydration from increasing the strength of the concrete. the proposed analytical model agree well with the experimental
This effect means that the analytical results may represent lower results.
232 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237

Bazant et al. line in a general case, however, is difficult to derive because

14 This Study the shape varies with the section geometry, the strength of the
ACI materials, the arrangement and content of the steel, and the level of
12
the axial load. Thus, biaxial bending has been treated inadequately
or ignored by many designers in practice. Nonetheless, there are
several design approaches to biaxial bending in which the design
10 L/r = 10 effort is reduced by the use of simplified approximations, and the
constant load contour method, which is one of these approximate
P (X105kgf)

8 A methods, is popular in many design codes, including the ACI 318

code [13].
L/r = 70
6 B The load contour method, which was introduced by Bresler [5]
and extended by Parme et al. [6], can be represented as follows:
 α1  α2
4 Mux Muy
+ (19)
Mux0 Muy0
2
L/r = 100
where Mux = Pu ey , Muy = Pu ex , ex and ey are the eccentricities of Pu ,
and Mux0 and Muy0 are the uniaxial flexural strengths for the x and
0 y axes of the constant load Pu under consideration (see Fig. 10(b)).
0 2 4 6 8 10 12 14 16
The constants α1 and α2 depend on the column properties and are
determined experimentally.
Fig. 8. Failure envelopes for pin-ended RC columns.
Parme et al. [6] proposed the same values of α1 = α2 = logβ
2.0 0.5, where β is the parameter that dictates the shape of the
interaction line. In addition, the values of β are calculated
analytically and given in charts for a range of bar arrangements,
steel yield strength, reinforcement index and Pu /P0 values.
1 + βd However, the load contour method has some limitations in its
1.5 direct application to the design of slender RC columns subjected
to long-term biaxial bending moments, even though the method
is still widely used in many design codes. The limitations are
as follows: (1) the results of the load contour method are too
conservative for a long column with a large slenderness ratio;
Pou/Put

1.0 (2) besides the uniaxial bending behavior of slender RC columns,

1 + 1.3β 2d insufficient consideration is given to the fact that the creep
deformation of concrete is deepened when the biaxial bending
moments are applied; and (3) the calculation of the value of
0.5 φ β , which requires the use of charts constructed on the basis
of parametric studies, makes the design procedure complex and
inconvenient.

7.2. The proposed formula for a uniaxial P–M interaction diagram

0.0
0 0.2 0.4 0.6 0.8 1 Before the biaxial bending problem is considered, however,
βd
there must be an exact evaluation of the P–M interaction diagrams
of RC columns subjected to the uniaxial bending moment because
Fig. 9. Long-term resistance of RC columns [13,26,25]. the application of the load contour method requires that the
flexural strengths of Mux0 and Muy0 on the uniaxial P–M interaction
7. The proposed formula for a biaxial bending design diagrams be determined for both independent axes, as shown in
Eq. (19). Thus, an incorrect calculation of the uniaxial flexural
7.1. Strength of an RC column strengths for the applied axial force does not allow an exact
evaluation of the biaxial moment resisting capacity. As reviewed
The strength of columns subjected to biaxial bending moments in Fig. 8, however, the ACI method gives very conservative
can be illustrated by interaction surfaces. The failure line of a P–M results when the slenderness ratio increases; hence, the results
interaction diagram of a column with uniaxial bending can be cannot be based on the calculation of the uniaxial flexural
determined on the basis of section analysis; a series of interaction strengths that correspond to the applied axial force, Pu . This
diagrams at various angles to the major axes of a section can also be conservativeness is exacerbated with the implementation of time-
obtained by varying the inclination of the neutral axis. A typical set dependent deformation of concrete.
of interaction diagrams for an arbitrary column section is shown in Accordingly, through parametric studies with changes in the
Fig. 10, and a complete set of diagrams for all angles describes the design variables, this paper introduces a regression formula
interaction surface, which means the failure surface. Each point of for the strength correction coefficient, C , so that the ultimate
this surface represents one particular set of axial load P values and resisting capacity of slender RC columns subjected to uniaxial
the moments Mx and My of the two major axes, which together bending moments can be determined directly. If the strength
produce the failure of a section. correction coefficient, C , is defined as the normalized ratio of the
If a horizontal section is taken through the interaction surface ultimate resisting capacities of a slender RC column as determined
of Fig. 10, the interaction line obtained gives the possible from rigorous nonlinear analysis to the corresponding values
combinations of Mux and Muy that would cause failure at a given as determined from the ACI method for the same eccentricity
axial load Pu (see Fig. 10(b)). This line is a constant load contour (C = OA/OB in Fig. 8), then the ultimate resisting capacity of a
of the interaction surface and is based in the design of the column slender RC column can be estimated easily and accurately by
with biaxial bending. An expression for the shape of the interaction (Pn · C , Mn · C ) without any additional rigorous numerical
 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237 233

(a) Failure surface at an angle θ . (b) Failure surface under constant P.

Fig. 10. A P–M interaction surface for an RC column with biaxial bending.

Table 2
Proposed formulas for the strength correction coefficient, C .
e emin eb
 2  2
(L/r ) (L/r ) (L/r )
cu = 0 1.20 100
− 0.65 100
+ 1.07 0.09 100
− 0.14 (100
L/r )
+ 1.03

C = a(L/r ) − b(L/r )(cu − 1)

 2  2
(L/r ) (L/r ) (L/r )
cu ≥ 1 a(L/r ) = 1.75 100
− 1.51 100
+ 1.23 a(L/r ) = 0.31 − 0.47 (100
L/r )
100
+ 1.11
2 
(L/r ) (L/r ) (L/r )
b(L/r ) = 0.27 100
− 0.06 b(L/r ) = −0.2 100 − 0.26 100 − 0.03

analyses. This estimation is possible because, in practice, the C , which corresponds to an arbitrary eccentricity, can be calculated
corresponding resisting capacity of (Pn , Mn ) can be directly through a linear interpolation of two boundary values for emin and
calculated in accordance with the ACI method. eb or eb and ∞. In addition, different regression formulas that
The calculated strength correction coefficient, C , however, take into account the creep deformation of concrete are proposed
does not maintain a constant value but changes according because the long-term loading shows a slightly different tendency
to the eccentricity and the slenderness ratio. Furthermore, an in the changes of the strength correction coefficients from that of
infinite number of RC column sections can theoretically be the short-term loading when cu = 0.
designed for the applied external forces. Hence, when the P–M Typical RC columns with different slenderness ratios and
interaction diagrams of RC columns are being determined, all the different ultimate creep coefficients of concrete are analyzed to
design variables need to be assumed on the basis of practical verify the effectiveness of the proposed regression formula. The
limitations and the design code requirements. The commonly used results are given in Fig. 11. In contrast to the ACI formula, which
compressive strength of concrete and yield stress of steel for a yields very conservative results for a column with a relatively
high slenderness ratio, the proposed formula effectively estimates
column design are fc0 = 352 kgf/cm2 and fy = 4220 kgf/cm2 ,
the ultimate resisting capacity of slender RC columns. Note, as
respectively. In addition, the steel ratio ranges from 0.02 to 0.08,
shown in Fig. 11, that the direct use of the ACI formula to
the slenderness ratio from 10 to 70, the ultimate creep coefficient
evaluate the resisting capacity of slender RC columns subjected to
from 0 to 3, the ratio of width to depth from 1.0 to 1.6, and the
biaxial bending moments accelerates the conservativeness of the
eccentricity from emin = 1.5 + 0.03 h(m) defined in the ACI 318
results.
code to infinity. These ranges of variables were selected for the
numerical analyses of this paper for the purpose of developing the 7.3. A regression formula for β̃
strength correction coefficient, C .
The regression formula for the strength correction coefficient, C , The construction of P–M interaction diagrams of RC columns
is developed conservatively on the basis of the lower limit values with the proposed strength correction coefficient, C , and the
so that the P–∆ effect and the creep deformation of concrete application of the ACI formula require the addition of the biaxial
can be included. This approach is taken because, for the sake of bending effect. As described in Eq. (19), however, the load contour
simplicity, the eccentricity, e, and the steel ratio, ρ , are not included method proposed by Parme et al. [6] still has some limitations in
in the regression equation for C . The regression results are listed terms of its direct application, even though it is popular in most
in Table 2. As shown in this table, the regression equations are design codes. To overcome those limitations, this paper draws on
defined for the two boundary eccentricities of emin and eb . If the parametric studies to propose a regression formula in which the
eccentricity is infinity, then the strength correction coefficient, C , improved parameter β̃ of α1 = α2 = logβ̃ 0.5 is used instead of β
must theoretically be 1. Hence, the strength correction coefficient, in Eq. (19).
234 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237

(a) cu = 0. (b) cu = 1.

(c) cu = 2. (d) cu = 3.

Fig. 11. P–M interaction diagrams of RC columns with a uniaxial bending moment.

(a) Equivalent parameter β̃ . (b) Transformed and normalized parameter β ∗ .

Fig. 12. Distribution of β̃ and its normalization.

Parametric studies were undertaken to determine the ultimate interaction diagrams regardless of the changes in the eccentricity
resisting capacity of RC columns in relation to changes in the angles. The marks in Fig. 12(a) show some of the typical results
eccentricity angles after the angle had been uniformly divided from from a square RC column with a steel ratio of ρ = 0.03. As shown in
0◦ to 90◦ into six parts. In addition, equivalent parameters of β̃ , this figure, the changes in the ultimate creep coefficient of concrete
which were to be used in Eq. (19), were determined in relation also affect the values of the equivalent parameter β̃ as well as the
to the value corresponding to the lower boundary of the P–M values of the strength correction coefficient, C , which are shown
 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237 235

respect to the common point of β̃ = 0.565 when L/r = 10. Even

though the equivalent parameter β̃ can be graphically determined,
the use of a mathematical approach in constructing a general
equation for a direct calculation of β̃ is more computationally
convenient. However, if there is a disagreement over the common
point of β̃ = 0.565 with the origin, the β̃ values must be
transformed to construct the rotational rule, and this construction
can be achieved by normalization. As shown in Fig. 12(b), all the
values of the equivalent parameter β̃ are transformed by β ∗ =
(β̃ − 0.565) × 10 with respect to the normalized slenderness ratio
of (L/r )∗ = (L/r − 10)/60; after that, the rotational rule is applied.
Because the regression formula that corresponds to the upper
boundary of cu = 0 was determined as β ∗ = a(L/r )∗ 2 − b(L/r )∗ =
0.538(L/r )∗ 2 −0.072(L/r )∗ , the transformation can be defined for
an arbitrary point (p, q) as follows:
δ cos θ sin θ
    
p
Fig. 13. Flowchart for biaxial parameter α . =
β∗ − sin θ cos θ q
in Table 2. This outcome means that the conventional approach in
cos θ sin θ
  
p
Eq. (19), which excludes the changes in the ultimate creep = (20)
− sin θ cos θ ap2 − bp
coefficient of concrete, usually yields excessively conservative
results for RC columns subjected to biaxial bending moments. where δ is an arbitrary normalized slenderness ratio and β ∗ is the
The distribution of the equivalent parameter β̃ has a uniformly corresponding normalized parameter of β̃ . Accordingly, upon the
decreasing trend when the ultimate creep coefficient, cu , increases. determination of the given slenderness ratio and the creep coeffi-
As shown in Fig. 12(a), an increase of 1 in the value of the ultimate cient, δ can be determined by δ = (L/r − 10)/60 (see Fig. 12(b)),
creep coefficient requires a clockwise rotation of 13.6◦ with and the first row of Eq. (20) then enables the values of p and q

(a) cu = 0. (b) cu = 1.

(c) cu = 2. (d) cu = 3.

Fig. 14. P–M interaction diagrams of RC columns with a biaxial bending moment (θ = 45◦ ).
236 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237

Table 3
Comparison of the resisting capacity of RC Columns.
Slenderness ratio Creep coefficient Normalized area ratio of the P–M interaction diagram
θ =0 θ = 15 θ = 30 θ = 45
Prop.a ACI Prop. ACI Prop. ACI Prop. ACI

10 0 1.041 1.014 0.965 0.975 0.946 0.976 0.964 0.994

10 1 1.016 1.001 0.965 0.961 0.946 0.961 0.940 0.980
10 2 1.017 1.002 0.968 0.964 0.946 0.961 0.945 0.985
10 3 1.019 1.004 0.965 0.961 0.953 0.968 0.951 0.991
30 0 0.993 1.015 0.925 0.974 0.915 0.980 0.939 1.006
30 1 1.010 0.872 0.913 0.833 0.903 0.841 0.947 0.867
30 2 0.980 0.888 0.922 0.849 0.902 0.851 0.908 0.885
30 3 0.992 0.944 0.967 0.902 0.939 0.900 0.908 0.946
50 0 0.981 0.969 0.934 0.927 0.947 0.945 0.960 0.959
50 1 0.953 0.641 0.943 0.607 0.941 0.615 0.915 0.650
50 2 0.962 0.727 1.076 0.705 1.057 0.709 0.929 0.758
50 3 0.969 0.828 1.207 0.805 1.192 0.820 0.914 0.866
70 0 0.973 0.869 0.967 0.844 0.970 0.841 0.995 0.871
70 1 0.999 0.499 1.076 0.499 1.075 0.515 1.022 0.539
70 2 1.043 0.607 1.268 0.604 1.274 0.634 1.040 0.668
70 3 1.023 0.702 1.446 0.707 1.440 0.744 1.014 0.786
a
Prop. The proposed method.

to be determined in advance. Moreover, the corresponding nor- the changes in design variables; (2) the ACI method yields very
malized coefficient β ∗ can be calculated from the second row of conservative results when the slenderness ratio and the ultimate
Eq. (20), and, finally, the equivalent parameter β̃ can be uniquely creep coefficient of concrete increase; and (3) the results of the
determined and applied to the relation α1 = α2 = logβ̃ 0.5, in proposed formula are in good agreement with the results of a
Eq. (19). A flowchart to obtain the biaxial parameter α correspond- rigorous analysis when a consistent difference is maintained over
ing to determined λ, cu values is given in Fig. 13. the entire eccentricity for all slender RC columns.
The effectiveness of the introduced relation is reviewed on the Although rigorous numerical methods that consider material
basis of the normalized area of the P–M interaction diagram with and geometric nonlinearities play an increasingly important role
respect to that obtained by a rigorous nonlinear analysis. As shown and will become the standard for final design checks, the formula
in Table 3, the ACI method gives more conservative results, and proposed in this paper can be effectively used to determine the
the conservativeness is proportionally increased with an increase initial section of a slender RC column. Moreover, a more rational
of the slenderness ratio and a decrease of the symmetricity in approach can be developed by conducting extensive studies,
the applied axial load. However, the proposed method gives more including experimental studies, on reliability assessment.
consistent results rather than those by the ACI method, regardless
of the slenderness ratio and the eccentricity of the applied load. Acknowledgments
Table 3 and Fig. 14, which show some of the obtained results,
confirm that the ACI method can be effectively used in the design This study has been a part of a research project supported
of RC columns with a relatively small slenderness ratio regardless by Korea Ministry of Education, Science and Technology (MEST)
of the creep coefficient and the eccentricity angle, though the via the research group for control of crack in concrete and also
results are too conservative for an increasing slenderness ratio. supported by a grant (07High Tech A01) from High tech Urban
In addition, the creep deformation of concrete may accelerate Development Program funded by Ministry of Land, Transportation
the conservativeness of the ACI method in the evaluation of the and Maritime Affairs of Korean government. The authors wish to
ultimate resisting capacity. On the other hand, as shown in Table 3, express their gratitude for the financial support that made this
the proposed method may slightly overestimate the resisting study possible.
capacity in the case of very slender RC columns subjected to an
axial load with a large eccentricity. The overestimation, however, References
is less than 20% (see Table 3) and limited to the range of e > eb (see
[1] Bazant Z, Xiang Y. Inelastic buckling of concrete column in braced frame.
Fig. 14). Therefore, the proposed method can be effectively used in J Struct Eng 1999;123(5):634.
the preliminary design in which the initial section of a slender RC [2] Kim J, Lee S. The behavior of reinforced concrete columns subjected to axial
column is assumed. force and biaxial bending. Eng Struct 2000;22(11):1518–28.
[3] Yalcin C, Saatcioglu M. Inelastic analysis of reinforced concrete columns.
Comput & Structures 2000;77(5):539–55.
8. Conclusion and recommendations [4] Kwak H, Kim J. Ultimate resisting capacity of slender RC columns. Comput &
Structures 2004;82(11–12):901–15.
[5] Bresler B. Design criteria for reinforced columns under axial load and biaxial
A numerical model that simulates the material and geometric bending. ACI J Proceedings 1960;57(11):481–90.
nonlinearities of RC columns, including the time-dependent [6] Parme A, Nieves J, Gouwens A. Capacity of reinforced rectangular columns
deformation of concrete, is presented in this paper. The proposed subject to biaxial bending. ACI J Proc 1966;63(9):911–24.
[7] Fafitis A. Interaction surfaces of reinforced-concrete sections in biaxial
model is verified by comparison of the model’s results with the bending. J Struct Eng 2001;127(7):840–6.
results of previous experimental and analytical studies. Moreover, [8] Rodriguez-Gutierrez J, Aristizabal-Ochoa J. M–P–φ diagrams for reinforced,
through section failure and P–∆ analyses of slender RC columns, a partially, and fully prestressed concrete sections under biaxial bending and
axial load. J Struct Eng 2001;127:763.
simple but effective regression formula is proposed for the design [9] Silva M, Swan C, Arora J, Brasil R. Failure criterion for RC members under biaxial
of slender RC columns subjected to uniaxial or biaxial bending bending and axial load. J Struct Eng 2001;127(8):922–9.
moments. The following conclusions can be drawn from the results [10] Bonet J, Romero M, Fernandez M, Miguel P. Design method for slender columns
subjected to biaxial bending based on second-order eccentricity. Mag Concr
of this limited investigation: (1) the results of the ACI method are Res 2007;59(1):3–19.
in good agreement with the results of a rigorous analysis of RC [11] Furlong R, Hsu C, Mirza S. Analysis and design of concrete columns for biaxial
columns with a relatively small slenderness ratio, regardless of bending-overview. ACI Struct J 2004;101(3):413–22.
 H.-G. Kwak, J.-H. Kwak / Engineering Structures 32 (2010) 226–237 237

[12] Comite Euro-International du Beton-Federation Internationale. CEB-FIP Model [19] Kwak H, Filippou F. Dept. of Civil Engineering, B. U. of California. Finite element
code for concrete structures. London: Thomas Telford; 1990. analysis of reinforced concrete structures under monotonic loads. Dept. of Civil
[13] ACI Committee 318. Building code requirements for structural concrete (ACI Engineering, University of California; 1990.
318-08). Farmington Hills (MI): American Concrete Institute; 2002. [20] Welch G, Haisman B. U. of New South Wales School of Civil Engineering.
[14] Taucer F, Spacone E, Filippou F. N. S. F. (US, California, E. E. R. Center, B. U. Fracture toughness measurements of concrete. School of Civil Engineering,
of California, D. of Transportation). A fiber beam column element for seismic University of New South Wales; 1969.
response analysis of reinforced concrete structures. Earthquake Engineering [21] Hughes T. The finite element method: Linear static and dynamic finite element
Research Center, College Engineering, University of California; 1991. analysis. Englewood Cliffs (NJ): Prentice-Hall; 1987.
[15] Kabir A. Nonlinear analysis of reinforced concrete panels, slabs and shells for [22] Kwak H, Seo Y. Numerical analysis of time-dependent behavior of pre-cast pre-
time-dependent effects. 1977. stressed concrete girder bridges. Constr Build Mater 2002;16(1):49–63.
[16] ACI Committee 209. Prediction of creep, shrinkage, & temperature effects on [23] Bazant Z, Cedolin L. Stability of structures: Elastic, inelastic, fracture, and
concrete structures (ACI 209R-92). Farmington Hills (MI): American Concrete damage theories. Dover Publications; 2003.
Institute; 2002. [24] Mirza S, Lee P, Morgan D. ACI stability resistance factor for RC columns. J Struct
[17] Kent D, Park R. Flexural members with confined concrete. J Struct Div 1971; Eng 1987;113(9):1963–76.
97(7):1969–90. [25] Bazant Z, Tsubaki T. Nonlinear creep buckling of reinforced concrete columns.
[18] Scott B, Park R, Priestley M. Stress–strain behavior of concrete confined J Struct Div 1980;106(11):2235–57.
by overlapping Hoops at low and high strain rates. ACI J Proc 1982;79(1): [26] Drysdale R, Huggins M. Sustained biaxial load on slender concrete columns.
13–27. J Struct Div 1971;97(5):1423–43.