PLASTICITY MODEL FOR CONCRETE UNDER TRIAXIAL COMPRESSION

By Iswandi Imran1 and S. J. Pantazopoulou,2 Member, ASCE

ABSTRACT: Using the experimental background of 130 triaxial tests conducted on cylindrical specimens, a
plasticity-based constitutive model of concrete behavior is developed. Parameters of the reference experimental
database include the water:cement ratio (i.e., f ⬘),
c degree of saturation at testing, and load path used in the tests.
In the model, damage is quantified by the volumetric expansion that builds up progressively in the material as
it approaches failure and is caused by propagation of microcracks. This behavioral index is calibrated with
reference to the available tests and subsequently used as the primary state variable in the model, determining
for any stress state the degree of stiffness and strength degradation and the ductility in the response. Because
failure is modeled as a damage-driven continuous process rather than a distinct event, the characteristic failure
envelope is expanding (hardening) or contracting (softening) as a function of a scalar measure of plastic defor-
mation. A nonassociated plastic flow rule calibrated against the experiments is used to describe the direction of
plastic deformation. The model was tested against published triaxial test series and empirical confinement models.
It was also used in the context of a finite-element formulation to study the mechanical behavior of reinforced-
concrete circular columns. This particular test problem was selected because it is a real-life example of the
experimental conditions used to derive the model.

INTRODUCTION (␰-axis) when plotted in ␳,␰-coordinates]. Yet, the degree of

saturation is seldom reported in experimental studies, with all
Early attempts at constitutive modeling of concrete under the attendant consequences such an omission may have on the
stress had been driven by theoretical argument and by limited shapes of the compressive meridian and the failure envelope
supporting experimental evidence. Pertinent frameworks of ap- derived by curve fitting of experimental data for concrete plas-
proach were established through these modeling efforts that ticity models.
succeeded in reproducing qualitatively the familiar features of Recently, a large volume of new triaxial test data has
concrete’s mechanical response (Chen 1982; Finite 1982). emerged in published literature (Jansen and Shah 1997, Mir-
However, despite the large volume of theoretical and experi- miran and Shahawy 1997, Ansari and Li 1998, Toutanji 1999).
mental research that has been conducted since, the quantitative An increased concern for the effect that the stiffness of the
accuracy of most available models as predictive tools of con- confining mechanism has on measured properties has recently
crete behavior is still limited in most cases (Li and Ansari led investigators to focus and report on the triaxial state of
1999). Reasons stem from the fact that, although many models deformation that accompanies the state of triaxial stress (Imran
reproduce successfully the qualitative trends of behavior under and Pantazopoulou 1996; Ansari and Li 1998). The broader
known load paths, the number of parameters that affect the range of concrete strengths and load paths for which data are
response in actual structures usually exceeds the dimensions now available, coupled with the information on deformation
of the experimental database that supports the mathematical paths and other important parameters routinely overlooked in
forms. Early published triaxial tests, used until recently for the past, encourage the anticipation that a better level of quan-
calibration of constitutive models for concrete, were obtained titative accuracy may be sought in constitutive modeling of
from concrete specimens of unconventional composition [e.g., concrete. In this paper this objective is explored through sys-
very small diameter aggregate or very large water:cement tematic formulation, calibration, and testing of a generalized
(w:c) ratio] (Richart et al. 1928; Launay and Gachon 1972; constitutive model that combines the framework of nonasso-
Smith et al. 1989). Furthermore, no single testing path suffices ciated plasticity with relevant measures of the state of damage
to reveal all aspects of the triaxial behavior of concrete. Many due to microcracking in the material structure. Derivation of
parameters are often overlooked in the tests and never reported the necessary functional forms is supported and guided by the
because of preconceived notions by the individual investiga- experimental results of a large series of triaxial tests conducted
tors as to their irrelevance or nonsignificance in the scheme of on cylindrical concrete specimens using a hydraulic-chamber
response. A characteristic example of this is the degree of sat- triaxial testing device (Imran and Pantazopoulou 1996). Pri-
uration at testing of specimens subjected to triaxial stress mary parameters of that study were the w:c ratio of concrete,
paths, which are routinely used in the derivation or calibration load path, level of confining pressure, and degree of saturation
of confinement models. Recent studies have shown that both at testing. An important finding was that the parameter best
the confined strength and the unconfined strength of normal organizing the available experimental database is the volu-
concrete differ by about 10 MPa from the dry to the saturated metric expansion of concrete resulting from internal micro-
state. Interestingly, a prevailing characteristic of test results cracking. Because experimental evidence conclusively identi-
that has not been corrected for the effect of saturation is non-
fies volumetric expansion as a meaningful measure of the
linearity in the compressive meridian at large confining
extent and intensity of damage in the material, this parameter
stresses [evidenced by its curving toward the hydrostatic axis
is used in the mathematical form as the primary state variable
1 driving the change of material stiffness and controlling fea-
Lect., Dept of Civ. Engrg., Technol. Inst. of Bandung, Bandung
40132, Indonesia. E-mail: iswan@bdg.centrin.net.id tures such as residual strength and ductility in the model.
2
Assoc. Prof., Dept. of Civ. Engrg., Demokritus Univ. of Thrace, Xan- The constitutive model is then used to correlate independent
thi 67100, Greece. E-mail: pantaz@cvil.duth.gr experimental summaries published in the literature. Combined
Note. Associate Editor: Arup Maji. Discussion open until August 1, with a pertinent stress-strain model for reinforcing steel, the
2001. To extend the closing date one month, a written request must be concrete model is used within the finite-element (FE) context
filed with the ASCE Manager of Journals. The manuscript for this paper
for calibration and sensitivity analysis of reinforced-concrete
was submitted for review and possible publication on January 3, 2000;
revised September 12, 2000. This paper is part of the Journal of Engi- circular-column tests. This particular example was selected be-
neering Mechanics, Vol. 127, No. 3, March, 2001. 䉷ASCE, ISSN 0733- cause it represents a classic case of 3D stress combination
9399/01/0003-0281–0290/$8.00 ⫹ $.50 per page. Paper No. 22206. found in civil engineering structures.
JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 281
CHARACTERISTICS OF CONCRETE’S BEHAVIOR The various aspects of the mathematical formulation of the
UNDER STRESS model are given in sequence in the following sections. The
database used for calibration of the various model parameters
Deviation from linearity owing to initiation and propagation consists of two triaxial test series, one conducted at the Uni-
of microcracking occurs in the uniaxial stress-strain response versity of Toronto (UT series) (Imran and Pantazopoulou
of concrete at a compressive stress approximately 30% of f ⬘c 1996) and one conducted at the University of Colorado (UC
(where f ⬘c is the uniaxial compressive strength). The corre- series) (Smith et al. 1989). The UT series consists of three
sponding axial strain is approximately equal to ⫺εcr /␯, where subgroups, namely, UTI representing high-strength concrete
εcr is the cracking strain of concrete under uniaxial tension with f ⬘c = 73.35 MPa, UTII representing normal-strength con-
(cracks develop parallel to the compressive load). Beyond this crete with f ⬘c = 47.40 MPa, and UTIII representing low-
limit, the ratio of lateral to axial strain increases continuously, strength concrete with f ⬘c = 28.62 MPa. Experimental values
highlighting the increasingly expansive tendency of the dam- used in calibration of the model corresponded to the dry state
aged material (apparent ␯ → 0.5). Due to the increasing rate of the material. Compressive strengths of identical specimens
of lateral expansion, the initial volumetric contraction of the in the UT series tested under saturated conditons were ap-
material is counteracted. From the reference triaxial test study, proximately 10 MPa lower than the dry values regardless of
it has been shown that the volumetric strain, εv = dV/V, is confinement level. Thus, to use experimental values from tests
approximately equal to zero at peak compressive stress for any on saturated specimens, strength has to be corrected to account
level of confining pressure (Imran and Pantazopoulou 1996). for the reduction in stress in the three principal directions re-
In the postpeak stress range, εv is expansive, indicating un- sulting from the disjoining pressure of pore water (Imran and
controlled dilatancy. Thus, at initiation of postpeak softening Pantazopoulou 1996)
in the stress-strain response of concrete, the area strain of the
primary compressive strut is algebraically equal (and of op- ␴
¯ i = ␴i ⫺ p vc (1)
posite sign) to the axial compressive strain. (Area strain is the
sum of the two principal strains orthogonal to the principal in which vc = void ratio of capillary pores; ␴i and ␴¯ i = un-
compressive strain). The characteristics of this behavior im- corrected and corrected stress values, respectively; and p =
prove with increasing confinement pressure because of the kin- I1 /3 (note that the sign convention is tension positive). To per-
ematic restraint provided by the confining mechanism, marked form this correction, the degree of saturation and vc must be
by delayed expansion and reduced brittleness and strength known. When this information had not been reported in the
loss. Beyond some level of moderate confining pressure, the reference studies, it was assumed that specimens were tested
material experiences no strength degradation past the peak in a saturated condition, whereas the capillary void ratio was
load. This is the point of transition from brittle to ductile re- calculated from the following expression (Pantazopoulou and
sponse. Mills 1995):
Under uniaxial tension, concrete is very brittle, having a w0 ⫺ 0.3064␣⬁
tensile strength approximately equal to 10% of the correspond- vc = 3.22 (1 ⫺ Va) (2a)
1 ⫹ 3.22 w0
ing uniaxial compressive strength. Under this type of loading,
cracks form perpendicular to the direction of the load. Once where
cracks develop, the material experiences sharp strength deg-
radation, which is a symptom of very unstable response (Pe- 1.031 w0
␣⬁ = (2b)
terson 1981). 0.194 ⫹ w0
in which w0 = reported w:c ratio; and Va = volumetric fraction
PROPOSED CONSTITUTIVE MODEL of the aggregates (in concrete) used in preparing the mix.
To model the experimental trends discussed in the preceding When Va was not reported, it was taken as 0.65, as this is the
section, the theory of nonassociative, nonlinear strain hard- most common ratio used in practice. Estimated capillary void
ening plasticity is used as the basis for development of a gen- ratios for UTI (w0 = 0.4 and Va = 0.61), UTII (w0 = 0.55 and
eral constitutive model. The most important state variable in Va = 0.65), and UTIII (w0 = 0.75 and Va = 0.69) were 13.6,
the model is volumetric strain, taken as a measure of mechan- 14.3, and 16% (measured) and 10, 13, and 15% (theoretical)
ically induced internal damage. Therefore, the various com- values, respectively. Similarly, for the UC test series (w0 =
ponents of the plasticity model, outlined in the following, are 0.833 and Va = 0.68) vc was 16.5%.
based on the same characteristic deformation patterns as those
that were recorded in triaxial tests. Mathematical Form of Loading and Failure Surfaces
In the model, failure is a continuous process driven by the
The four-parameter Hsieh-Ting-Chen (HTC) criterion was
history of deformation rather than a marked distinct event.
used as the functional shape for the loading, failure, and de-
Contrary to the conventional use of a single yield or failure
grading surfaces of the model (Hsieh et al. 1988). This crite-
envelope, where attainment of failure is a binary event, in this
rion has a demonstrated ability to provide a good fit to ex-
model it is assumed that a yield envelope is reached at every
perimental results over a wide range of stress combinations
level of deformation regardless of its magnitude. Thus, there
(Hsieh et al. 1988; Imran 1994; Imran and Pantazopoulou
exists an infinite number of yield envelopes and the loading
1996) and conforms to earlier findings that all three stress
process is idealized as a process of transition from one yield
invariants (i.e., I1, J2, and J3) affect the failure pattern (Chen
envelope to the next, each such stepwise transition being ac-
1982). Its functional form has been given by (Hsieh et al.
companied by an attendant development of irrecoverable plas-
1988)
tic deformation. All yield envelopes have the same functional
form, but the shape is expanding (prepeak response) or con- J2
tracting (postpeak) depending on the state of damage. In this f (␴ij , f ⬘)
c = A ⫹ B兹J2 ⫹ C␴1 ⫹ DI1 ⫺ f ⬘c = 0 (3)
f c⬘
context as a matter of nomenclature, contracting yield enve-
lopes are referred to as degrading surfaces, whereas expanding where f c⬘ = uniaxial compressive strength of concrete; I1 = first
yield envelopes are referred to as loading surfaces. The en- invariant of the stress tensor; J2 = second invariant of the de-
velope corresponding to peak strength that separates the two viatoric stress tensor; and ␴1 = major principal stress. In (3),
regimes is referred to as failure surface. A, B, C, and D represent the effect of physical properties of
282 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001
concrete on its strength characteristics. The values of these curing room (hence they might have been subject to a moisture
constants may be calibrated from a set of four independent gradient). Being high strength and thus low porosity materials,
triaxial tests. Parameter values as obtained from calibration of and considering the short time interval between removal from
(3) with the UTI, UTII, and UTIII series (calculated R-square the curing room to testing, it was assumed that the condition
values were 0.92) are listed in Table 1. Fig. 1 plots the ex- of specimens at testing was nearly saturated. It was mentioned
perimentally derived compressive meridian of the HTC failure that the uncorrected strength characteristics of the wet concrete
criterion, indicating a successful representation of the relation- differ from those of the dry one. Note in Fig. 1 that the un-
ship between compressive strength and confining pressure. corrected data of Ansari and Li (1998) deviate from the pro-
Note that the UT tests used in calibrating the HTC failure posed failure envelope, curving down toward the hydrostatic
criterion corresponded to the dry state of the material. Addi- axis.
tional test data reported by Ansari and Li (1998) were also The same basic expression and coefficients are also used to
incorporated in Fig. 1 (Table 2). Ansari and Li’s specimens, describe the intermediate yield (loading) surfaces. The facility
with compressive strengths ranging from 47 to 107 MPa, re- to modify the size of these surfaces is described below.
portedly were tested 24 h after they were removed from the
Expansion of Yield (Loading) Surfaces up to Failure
TABLE 1. Summary of Parameters for Proposed Model Surface
Corrected for f ⬘c =/ 30 As loading in concrete progresses (i.e., in the prepeak re-
Value For f ⬘c = 30 MPa MPa gime), yield and failure surfaces are generated in the model
(1) (2) (3) using the isotropic hardening rule, thereby continuously evolv-
A 2.877 2.877 ing the size and actual shape of yield surfaces up to the failure
B 0.546 0.546 surface (peak point). Isotropic hardening has yielded satisfac-
C 9.315 9.315 tory results when used to model the hardening behavior of
D 0.274 0.274
E 0.41 0.41
concrete subjected to monotonic loading (Smith et al. 1989;
ko 0.37 0.37 Mizuno and Hatanaka 1989), and as such, it is adopted in the
G ⫺5.646 1.653 ( f ⬘c /30) ⫺7.3023 model. The degree of change of the loading surfaces is ex-
ε uv max ⫺0.00089 ⫺0.00089 pressed as a function of the state of damage in the material,
εs 0.05 0.05 quantified here by the amount of residual (or plastic) defor-
ε up max 0.0012 0.0012 mation.
I trans /fc 2.926 0.627 ( f ⬘c /30) ⫹ 2.3903
l
εp ult /εp max 8.632 8.1 ( f ⬘c /30)⫺0.95
From previous experiments, it is established that concrete
a u
0.58 0.58 experiences plastic deformation even under hydrostatic load-
␩ 0.34 0.34 ing (Launay and Gachon 1972; Lade 1988). This observation
imposes the requirement that the yield surfaces used for mod-
eling concrete plasticity should intersect the axis of the hydro-
static loading (Fig. 2). Empirically, this type of yield surface
is generated by introducing additional terms to the basic equa-
tion of the failure surface [(3)] so as to reduce the size of the
surface and bend it toward the hydrostatic axis. The proposed
functional form of the intermediate yield surfaces is
J2 1⫺k 2
A ⫹ B兹J2 ⫹ Ck␴1 ⫹ DkI1 ⫹ Ehtc I 1 ⫺ k f ⬘c = 0 (4)
k f ⬘c k f ⬘c
where the fifth term on the left-hand side of (4) controls the
intersection of the loading surfaces with the hydrostatic axis.
Variable k specifies the degree of hardening in the concrete
material during the course of loading; hence, it defines the size
of the loading surfaces. When k = 1, (4) reduces to (3). Its
value ranges from k = k0 for initial yield, to k = 1 when the
state of stress reaches the failure surface. All states of stress
described by (4) having the same value of k will lie on the
FIG. 1. Experimentally Derived Compressive Meridian HTC same loading surface.
Failure Envelope Values for the additional constants in the loading surface
function [i.e., the value of k at initial yield k0 and Ehtc in (4)]
TABLE 2. Properties of Analyzed Concrete Specimens

Degree of
saturation
Specimen f c⬘ vc assumed
ID (MPa) w0 Va (%) (%)
(1) (2) (3) (4) (5) (6)
(a) Ansari and Li (1998)
HS06 47.23 0.46 0.68 9.95 100
HS10 71.09 0.36 0.63 8.48 100
HS15 107.29 0.26 0.62 5.27 100
(b) Xie et al. (1995)
A 60.20 0.29 0.65 5.9 50
B 92.20 0.25 0.55 5.8 50
C 119.00 0.19 0.58 3.0 50
FIG. 2. Evolution of Loading Surfaces during Hardening

JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 283

are calibrated based on data from two independent stress states deformation pattern (into expansion) with ensuing failure (Im-
at the initiation of plastic deformation. Only a limited number ran 1994). The proposed relationship between εp max and εv max
of test data regarding these stress states are available in the is
literature; e.g., for uniaxial load cases, the initiation of plastic
deformation occurs when volumetric stress I1 reaches approx- εp max = G(εv max ⫺ ε uv max) ⫹ ε up max (8)
imately 0.3 ⭈ f c⬘ (Chen 1982), whereas for hydrostatic load where G = material constant. Parameters with superscript u in
cases, it takes place when I1 = 2 ⭈ f ⬘c (Launay and Gachon (8) are the corresponding strain data from uniaxial compres-
1972). Typical values of k0 and Ehtc evaluated from the ex- sive tests. Eq. (8) should be calibrated using strain data from
perimental data of the UT series are given in Table 1 (Imran a minimum of two different concrete tests (i.e., uniaxial com-
1994). pression and any other triaxial test). The estimated values of
Loading surfaces expand to represent the process of damage G for the UT series are given in Table 1 (R-square = 0.99 from
buildup in the material. For consistency with its intended phys- correlation of the entire test database). It may be noted from
ical significance, k should be expressed in terms of a perti- (8) that increasing the value of εv max while keeping the other
nent damage index. In this paper, a scalar measure of plastic variables (such as ε up max and ε uv max) constant causes a corre-
deformation εp is used as a measure of damage. The plastic sponding increase in the amount of accumulated plastic strain
strain εp is calculated based on the following statement of the εp max and therefore in the ductility of the material. Enhance-
equivalence of the plastic work increment, i.e.: ment of ductility because of prolonged volume contraction in

冕 冕
concrete (i.e., restraint to expansion) is an effect that charac-
␴ij d␧ pij
εp = dεp = (5) terizes the influence of confinement.
kf ⬘c
Degradation Surfaces in Postpeak Regime
where ␴ij and d␧ pij = stress- and plastic-strain vectors, respec-
tively. By means of (5), the relationship between the stress- The hardening function derived above regulates the expan-
and plastic-strain vectors in the model are replaced by their sion of loading surfaces for concrete in the prepeak regime.
equivalent scalar measures. Typical loading and failure sur- Past the peak point, response of unconfined or lightly confined
faces as defined by (4) are illustrated in Fig. 2 for values of k concrete is characterized by strength degradation with increas-
ranging between k0 and 1. The surfaces are plotted in the ␳,␰- ing deformation. The strength degradation is caused by the
coordinate system, where ␳ and ␰ are defined as the deviatoric formation of macrocracks in the material structure. This results
and hydrostatic components of the stress tensor, respectively. in debonding of the material constituents, an action that effec-
tively reduces cohesion and decreases the material resistance
Hardening Function ␬(␧p) (Van Mier 1986).
With increasing deformation, interparticle cohesion contin-
It was mentioned that k determines the size of the yield ues to decrease until it is entirely eliminated. Without this
(loading) surfaces and is a function of accumulated plastic mechanism of resistance, concrete behaves like a cohesionless,
strain εp. Boundary conditions to this evolution of size are the frictional material. This type of material derives its strength
following: (1) at zero total strain, the loading surface collapses from the friction between material constituents only. For un-
to a point (state of zero stress); and (2) at total strain ε0 cor- iaxially loaded concrete, the frictional action cannot be mo-
responding to peak stress, the loading surface converges onto bilized, as there are no applied lateral forces present to keep
the failure surface (k = 1). The evolution of k subject to these the debonded concrete constituents together after cohesion
end conditions is described using a Hognestad-type parabola vanishes. The strength generated from frictional action, re-

冉 冉 冊冊 ferred to as residual strength, becomes apparent when concrete

2
2ε ε
k = ␬(ε) = ⫺ (6) is confined laterally (e.g., by introduction of some degree of
ε0 ε0 lateral kinematic restraint). Frictional strength increases with
in which ε = total current strain (scalar quantity) computed increasing confining stress (the magnitude of frictional resis-
from superposition of elastic and plastic components. The elas- tance is equal to the product, coefficient of friction ⫻ normal
tic component is computed using a Hookean-type stress-strain pressure). At some critical level of confining stress, the resid-
law, whereas the plastic component is calculated from (5). ual strength may eventually equal the peak stress of the ma-
Based on this superposition, (6) is rewritten in terms of εp terial. Beyond this transition point, concrete experiences no
strength degradation but rather exhibits a ductile response that
(2兹εp εp max ⫺ εp) resembles plastic flow (Smith et al. 1989).
␬(εp) = (1 ⫺ k0) ⫹ k0 (7)
εp max Strength degradation of concrete during postpeak strain soft-
ening is modeled here by contraction of the failure surface.
where k0 = value of ␬(εp) at εp = 0 (i.e., at the initiation of This model has been used successfully before to simulate the
plastic flow). The plastic modulus H p that defines the rate of softening response of frictional materials such as concrete
expansion of the yield surface (or the rate of hardening) is (Ohtani and Chen 1989). The extent of contraction is given as
obtained by differentiating (7) with respect to εp. The variable a continuous function of the state of damage in the material.
εp max in (7) represents the amount of accumulated plastic de- As in the prepeak regime, damage is quantified by the amount
formation when the state of stress reaches the failure surface. of accumulated equivalent plastic deformation; i.e., during the
This strain measure characterizes the ductility of the material, postpeak response, an increase in the equivalent plastic defor-
because it controls the amount of plastic strain, and therefore mation causes the failure surface to contract. Degradation sur-
the total strain, at which the material reaches maximum stress. faces that represent the strength reduction are assumed to have
Ductility increases with the amount of this plastic-strain mea- the same functional form as the failure surface (3). This ex-
sure. Here εp max is determined based on the available experi- pression was modified empirically, by introducing an addi-
mental data and according to the concept of equivalent plastic tional term to account for the increase of residual strength in
work described earlier. From the UT and UC data series, it the material as the lateral confining stress increases. The pro-
was found that this parameter is a function of the maximum posed expression is
volumetric contraction εv max experienced by the material,
which, in turn, is an indirect measure of the effective com- J2 I1
A ⫹ B兹J2 ⫹ C␴1 ⫹ DI1 ⫺ (1 ⫺ r) trans f ⬘c ⫺ rf ⬘c = 0 (9)
paction that the material underwent prior to reversal of the f ⬘c I1

284 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001

 Plastic Potential Function—Direction of Plastic Flow
By definition, the hardening and softening functions of the
model are controlled by the amount of accumulated plastic
deformation in the material. During loading in the plastic
range, the direction of plastic deformation in the material (i.e.,
the unit normal vector in the direction of platic flow) is defined
mathematically through the plastic potential function. Recent
experimental evidence suggests that concrete demonstrates
nonassociativeness (Smith et al. 1989; Imran 1994). For sim-
plicity, a Drucker-Prager type criterion was selected as the
plastic potential function for nonassociated plastic flow in the
model. The merit of that function is that it may easily be ad-
justed to model the plastic dilatancy (volumetric expansion)
observed in experimental studies of concrete. The criterion is
FIG. 3. Evolution of Degrading Surfaces during Softening and expressed in terms of stress invariants
Plastic Potential Function (Drucker-Prager)
g = a␰ ⫹ ␳ ⫺ c = 0 (11)

where I trans = magnitude of volumetric stress at the transition where ␰ and ␳ = hydrostatic and deviatoric components of the
1
point; and r = softening parameter that controls the evolution stress tensor, respectively; and c = constant. Parameter a in
of the degradation surfaces. The value of r ranges from r = 1 (11) is the slope of the Drucker-Prager function defined by the
at the failure surface to r = 0 at the residual strength envelope. ratio ␣ = ⫺d␳/d␰ (lower right of Fig. 3). In terms of strain
To determine the value of I trans , residual strength data from any coordinates it is redefined as (Imran 1994)
1
triaxial test (i.e., below the transition point) are required. Val- d␰ ⬘p 1 dI ⬘1p
ues of I trans derived from the UT series are given in Table 1. a= = (12)
1
d␳ p⬘ 兹6 兹dJ 2p ⬘
The fifth term in (9) represents the mobilization of the fric-
tional strength as volumetric stress is approaching the transi- where dI ⬘1p = first invariant of the incremental strain tensor;
tion value, and the last term models the degradation of cohe- and 兹dJ ⬘2p = second invariant of the deviatoric part of the
sion as damage progresses in concrete. The evolution of the incremental strain tensor. From (12) it is evident that parameter
degradation surfaces given by (9) is illustrated in Fig. 3 for r a basically controls the magnitude of the volumetric compo-
ranging from 1 (i.e., at the failure envelope) to 0 (i.e., at the nent of the plastic-strain increment. Using (12) the functional
residual strength envelope). form of a may be derived from experimental data. This func-
tional form governs the evolution of the Drucker-Prager func-
Softening Function r (␧p) tion throughout the development of plastic deformation in the
material. The following expression was obtained for a from
Contraction of the failure surface as damage progresses is the UT and UC test series (Imran 1994):

冉 冊
controlled by the softening function. For smooth transition
from hardening prepeak behavior to softening postpeak be- au εp
a= ⫺␩ (13)
havior, this function is assumed to have (1) zero slope at peak (1 ⫺ ␩)(εv max /ε uv max)1/3 εp max
stress (i.e., when r = 1) with respect to the amount of accu-
where a u = value of function a when uniaxially loaded con-
mulated plastic deformation at peak (i.e., εp max); and (2) zero
crete reaches the failure surface; εp = scalar measure of ac-
slope at the residual strength envelope (i.e., when r = 0) to
cumulated plastic strain; and εp max = value of this scalar at the
simulate observed experimental stress-strain curves that tend
failure surface. In (13), ␩ = ratio of εp /εp max at the zero volu-
to reach the zero residual strength level asymptotically (Imran
metric plastic-strain increment [i.e., at the limit point where
1994). Based on these criteria, the following equation is pro-
contractive response (negative volumetric plastic-strain incre-
posed for the softening function:
ment) reverts to dilative response]. This ratio refers to the

r=
1
2
1
⫹ cos
2
冉␲
rep ⫺ 1
r uep ⫺ 1
冊 (10a)
point where concrete reaches maximum contraction under uni-
axial compression. From the UT data series, this was observed
to occur at a value for the hardening parameter k = 0.88 [(4)].
where Substituting this value for k in (7), it was found that ␩ = 0.34.

εp εp ult Constitutive Equations

rep = ; r uep = (10b)
εp max εp max In the elastic range, the incremental stress-strain relationship
of the material is given by
in which the variable εp ult = accumulated plastic strain when
the state of stress in the material reaches the residual strength {d␴} = [D e]{dε} (14)
envelope. It is evaluated from the requirement of the equiva-
lence of the plastic work dissipated by the softening model to where {d␴} and {dε} = stress and strain increment vectors,
that observed in the experimental softening curve. This pro- respectively; and [D e] = matrix of material constants. In the
cedure is necessary to ensure that the amount of energy dis- plastic range, the elastic component [D e] also degrades with
sipated during softening, in the form of plastic work, repre- increasing damage, as described in the following section. Dur-
sents the observed experimental value. In accordance with the ing the occurrence of plastic flow, the total strain increments
proposed softening model, cohesion of concrete materials de- in the material may be decomposed into elastic and plastic
creases after the failure surface has been reached. Cohesion components, i.e.
vanishes when the amount of accumulated plastic strain (i.e., {dε} = {dε e} ⫹ {dε p} (15)
the damage variable) attains its critical value (i.e., when εp =
εp ult). e p
where d␧ and d␧ = vectors of the elastic- and plastic-strain
JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 285
increments, respectively. The elastic component of the strain of orthotropy in the material coincident with the directions of
increment vector in (15) may be related to the total stress principal strains. For axisymmetric problems, and considering
increment using (14), whereas the plastic-strain increment vec- constant values for the Poisson’s ratio, the orthotropic material
tor is evaluated from the stress increment through the use of laws can be written (Elwi and Murray 1979) (Note that Pois-
a flow rule. For nonassociated response, the flow rule is writ- son’s ratio is used here only to represent the cause-effect re-
ten lationship between the elastic components of strain and is not
intended to account for the nonlinear increase of tensile strains
{dε p} = d ␭{b} (16a) after cracking of concrete)
where 1

再冎
D=

冋 册
⭸g (1 ⫹ ␯)(1 ⫺ 2␯)
{b} = (16b)
⭸␴
E1(1 ⫺ ␯) ␯兹E1 E2 ␯兹E1 E3 0
in which parameter d ␭ = positive constant of proportionality.
Its value is determined using the so-called condition of con- ␯兹E1 E2 E2(1 ⫺ ␯) ␯兹E2 E3 0
⭈ (21)
sistency, which states that during plastic loading the state of ␯兹E1 E3 ␯兹E2 E3 E3(1 ⫺ ␯) 0
stress in the material always lies in the current yield surface 0 0 0 G12
(Chen 1982); this condition essentially requires that loading
from a plastic state leads to another plastic state where G12 = 0.25[E1(1 ⫺ ␯) ⫹ E2(1 ⫺ ␯) ⫺ 2␯ 兹E1 E2]. Var-

再 冎
iables E1, E2, and E3 represent the damaged elastic stiffness
⭸f ⭸f along the major, intermediate, and minor principal strain di-
d f = {a}{d␴} ⫹ {dε p} ⫹ d␬ = 0 (17)
⭸ε p ⭸␬ rections, respectively. They are defined as Ei = ␤i E0 , where E0
where the hardening parameter ␬ = function of plastic-strain is the initial elastic stiffness of the material (representing the
vectors; and {a} = {⭸ f/⭸␴}. By substitution of (15) and (16) stiffness of undamaged concrete) and ␤i is a function that
into (17) and solving for {d␴}, it follows: quantifies the extent of stiffness degradation in the direction
of interest.
{d␴} = 冋[D e] ⫺
[D e]{b}{a} T[D e]
h ⫹ {a} T[D e}{b} 册 {d ε} (18)
Degradation of elastic stiffness in the direction of compres-
sion is related to the development of lateral expansion of the
strut cross section bearing the load (area strain). This strain
Parameter h = scalar function related to the hardening rule quantifies development and propagation of microcracks (dam-
(=0 for perfectly plastic material models) age) suffered by the area supporting the load. Based on the

冋再 冎 再 冎册
T T experimental evidence of Imran and Pantazopoulou (1996), the
⭸f ⭸f ⭸␬
h=⫺ ⫹ {b} (19) following function has been used to relate the evolution of
⭸ε p ⭸␬ ⭸ε p elastic stiffness of concrete along the direction of principal
Hence, the elastoplastic [D ep] matrix of material coefficients compression to the development of lateral expansion:
may be expressed Ei 1
= f (εai) = (22)
[D ep] = [D e] ⫺ [D p] (20) E0 1 ⫹ 具εai典/εs

where [D p ] = plastic stiffness matrix representing the degra- in which Ei = current stiffness along principal direction i; E0
dation of the stiffness of the material because of plastic flow. = stiffness of undamaged concrete, εai = area strain in the plane
normal to the direction i; and εs = calibrated constant that
Degradation of Elastic Modulus normalizes the area strain εai. Symbol 具x典 in (22) is the
McAuley bracket, which has zero value if x < 0. Thus, (22)
Isotropic linear elasticity is often assumed in calculating the implies that the degradation of elastic stiffness along any prin-
elastic component of total deformation in the material during cipal strain direction occurs only when the area strain of the
plastic flow. However, the experimental results of Imran and specimen cross section normal to the direction considered is
Pantazopoulou (1996) clearly demonstrate that concrete under positive. Based on the experimental data of Imran and Pan-
load exhibits a gradual decrease of elastic stiffness near and tazopoulou (1996), the value of εs was found to be approxi-
beyond peak stress because of development and propagation mately 0.05 (with an R-square value of 0.65). The proposed
of microcracks (damage) as the level of induced deformation function is plotted along with the corresponding experimental
increases. Evidently the initially isotropic material law needs data in Fig. 4.
to be modified to account for the effects of progressive damage For concrete under uniaxial tension, tensile strain develops
buildup on stiffness. In this paper, damage in concrete is eval- parallel to the load. When this strain exceeds its limiting value
uated with reference to the direction of principal strains. Upon (cracking strain), cracks are assumed to occur in planes normal
initiation of damage, it is assumed that the initially isotropic to the direction of the load. Degradation of concrete stiffness
material response is transformed to orthotropic with the axes in the direction of tensile load has been related to the increase
of coaxial tensile deformation (Reinhardt 1984). Based on this
study, the following relationship is proposed for modeling the
stiffness degradation of concrete under tension:
Ei 1
= g(εi) = (23)
E0 1 ⫹ 具εi ⫺ εcr典/εcr
in which εi and εcr = tensile strains in the direction of the load
and at peak uniaxial tensile stress, respectively. Under triaxial
tension, concrete in any principal strain direction experiences
both lateral expansion and tensile straining along the direction
FIG. 4. Degradation of Stiffness with Increasing Area Strain of considered. Because all those measures of deformation con-
Compressive Strut tribute to stiffness degradation [as expressed by (22) and (23)],
286 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001
it is concluded that, for concrete subjected to multiaxial ten- where f ⬘cc = axial compressive strength of confined concrete
sion, this process is basically a function of (1) the strain along (positive number = ⫺␴3); f ⬘c = uniaxial compressive strength
the direction of the load; and (2) the area strain of the cross of concrete (here taken as a positive number); and fl = lateral
section normal to the direction considered. The relationship confining stress (=⫺␴1). Note that (25) represents a confine-
used to determine degradation of concrete under combined tri- ment model, relating concrete compressive strength to confin-
axial tension is written ing pressure. A similar model is that derived by Mander et al.
(1988) starting from a different failure criterion: K = 2.254 ⭈
Ei 1 1 (1 ⫹ 7.94( fl /f ⬘))1/2
⫺ 2( fl /f c⬘) ⫺ 1.254. Although the empirical
= ⭈ (24) c
E0 1 ⫹ 具εi ⫺ εcr典/εcr 1 ⫹ 具εai典/εs constants differ because of fitting different criteria to the avail-
able test data, it is worth noting the similarities between the
CALIBRATION OF MODEL WITH PUBLISHED two independent expressions for K, namely, in both cases K =
EXPERIMENTAL SUMMARIES a( fl /f c⬘) ⫹ b(c ⫹ d( fl /f c⬘))1/2 ⫹ e. In the range of low confine-
ment pressures ( fl /fc < 0.5), the two models yield very similar
The total number of tests used to calibrate the model pa- results. Results from the two models deviate at higher confin-
rameters (Table 1) is 130 (all specimens used are from the UT ing pressures. In the first case (Imran 1994), fitting of the
test series). Test data of Xie et al. (1995) (that have not been model was carried out using results from actively confined
used for derivation of model parameters) are used to evaluate concrete tests, whereas in the second case (Mander et al.
the model’s performance. Fig. 5 presents the plain concrete 1988), passively confined specimens had been used for cor-
model estimations for the triaxial test data of Xie et al. (1995). relation. Results of (26) along with the corrected test data of
Results are obtained using the parameter values shown in Ta- Ansari and Li (1998) and Xie et al. (1995) are plotted in Fig.
ble 1. Computed stress-strain curves are generally in good 6, confirming that the strength increase described by (25) is in
agreement with the test results with regards to both reported good agreement with the test results.
strength and deformation values (Fig. 5). A similar approach was used to derive the residual strength
For axisymmetric triaxial compression (i.e., ␴1 = ␴2 > ␴3) envelope. As described in the previous section, confined con-
with lateral confining stress ␴1 = fl , the proposed model is crete experiences postpeak softening until it reaches the resid-
compared with the high-strength concrete test results of Ansari ual strength envelope (when r = 1). Hence, the residual
and Li (1998) and Xie et al. (1995) (Table 2). Concrete strength envelope (9) simplifies to (26) below, where I trans is
1
strength of Xie’s specimens was in the range 60.2–119 MPa the volumetric stress at the transition point from brittle to duc-
(Xie et al. 1995). Substituting ␴1 = ␴2 = fl , the maximum tile response
strength envelope [(3)] may be expressed

冉 冊 冑
AJ2 I1
f ⬘cc fl fl ⫹ B兹J2 ⫹ C␴1 ⫹ DI1 ⫺ trans f ⬘c = 0 (26)
K= = ⫺ 0.021 ⫹ 1.043 ⫹ 10.571 (25) f ⬘c I1
f ⬘c f ⬘c f ⬘c
For axisymmetric compression, it can be shown that the
residual strength envelope may be expressed
fr
f ⬘c
= 冉 fl
f ⬘c
冊 冑
⫺ 0.021 ⫹ (0.021)2 ⫹ s
fl
f ⬘c
ⱕ
f ⬘cc
f ⬘c
(27)

where fr = residual strength of confined concrete; and s =

⫺0.0348f ⬘c ⫹ 15.147. The proposed residual strength envelope is
also plotted in Fig. 6 along with the high-strength concrete test
results of Xie et al. (1995), with f ⬘c = 92.2 MPa. Note the marked
presence of a transition point from brittle to ductile response in
Fig. 6. The good correlation of (26) and (27) with the available
FIG. 5. Comparison of Triaxial Responses between Analytical test data for high-strength concrete specimens encourages their
and Experimental Results of Xie et al. (1995) use in modeling applications of confined stress states for this type
of concrete in particular.

APPLICATION: STUDY OF REINFORCED-CONCRETE

COLUMNS
The model derived in this paper is implemented into a host
finite-element (FE) package, FEAP, originally developed at the
University of California, Berkeley (Zienkiewicz and Taylor
1989). Integration of the elastoplastic constitutive equations
was carried out using a scheme developed by Mitchell and
Owen (1988). Here the proposed constitutive model is used to
analyze the mechanical response of reinforced-concrete col-
umns with a circular cross section. This problem was selected
because it represents a real-life example of the stress states
considered by the background triaxial tests. The last part of
this section presents results of a parametric study done with
the model, investigating the effect of various design variables
on the behavior of circular columns.

Corroboration with Experimental Results

FIG. 6. Maximum and Residual Strength Envelope of Axisym-
metric Triaxial Compression: Comparison of Eq. (25) with Test The circular column tests done by Sheikh and Toklucu
Results of Xie et al. (1995) and Ansari and Li (1998) (1993) were used to validate the performance of the present
JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 287
 FIG. 7. (a) FE Mesh of Test Columns; (b) Analytical versus Experimental Results of Columns 1 and 14

TABLE 3. Material Properties of Analyzed Columns (Sheikh and Toklucu 1993)

Experiment Analysis
Specimen f ⬘co
a
Ecob
␳s fys ␳l fyl fl fl
ID (MPa) (GPa) (%) (MPa) (%) (MPa) (MPa) K ke (MPa) K ke
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
1 30.5 27.7 2.3 452 2.5 509 6.9 1.7 3.10 5.75 1.65 3.4
14 30.2 27.7 1.15 607 3.0 478 3.5 1.4 3.86 4.1 1.40 2.9
19 30.2 27.7 1.7 607 3.0 478 5.5 1.6 3.25 6.1 1.60 3.0
28 29.7 27.7 1.15 629 3.1 484 3.6 1.4 3.40 4.2 1.38 2.7
In analysis, 85% of reported value was used for f c⬘ for consistency with procedures adopted in reference study.
a

b
⬘ /εco.
Computed as 2f co

model. A total of 27 circular columns with diameters of 356, strains, the spiral reinforcement fractured in the experiments
254, and 203 mm had been tested under either load control or because of longitudinal reinforcement buckling [Fig. 7(b)].
displacement control to study the effect of various design pa- Thus, the actual effectiveness of the spiral in confining the
rameters on the behavior of columns. In all cases the height- core concrete was reduced; therefore, the strength of the core
to-diameter ratio was 4. Columns were reinforced with lon- concrete also decreased. This aspect of spiral fracture was not
gitudinal and spiral or hoop reinforcement. modeled in the analysis because of the smeared representation
The typical column FE mesh consists of 16 eight-noded of the reinforcement, and this is why no degradation of core
axisymmetric quadrilateral elements [Fig. 7(a)]. The height of concrete stress is observed in the analytical results at a high
the FE model represents the test region of the columns. A level of axial deformation (i.e., at axial strains >1%). In the
representative sample of the reference test series were ana- model, the response of core concrete at this level of defor-
lyzed (Specimens 1, 14, 19, and 28). Reported material prop- mation is characterized by the residual strength envelope [(9)].
erties for these members are given in Table 3. Values for the No further strength degradation will occur in the model after
additional parameters of the proposed model were taken from concrete reaches its residual strength envelope under increas-
the preceding calibrations (Table 1) for normal strength con- ing deformation, unless the magnitude of the applied lateral
crete. Reinforcement was modeled as a layer of equivalent stress generated from the confining reinforcement decreases.
thickness, smeared through a concrete cross section. The Another factor that may have contributed to the discrepan-
stress-strain curves of steel in compression and in tension were
cies between computed and experimental results is smearing
taken to be identical and were idealized by a trilinear model
of the spiral reinforcement in the FE idealization [shaded area
(elastic, plastic, and strain hardening). Perfect bond was as-
in Fig. 7(a)]. When smeared, transverse reinforcement is ide-
sumed between steel and concrete, whereas instability effects
ally effective in confining the core concrete, whereas in the
associated with buckling of reinforcement were neglected. All
analyses were carried out simulating displacement control. actual specimens confinement effectiveness was <100% (Man-
Computed results are compared with their experimental der et al. 1988). With discrete hoops or stirrups, there is an
counterparts in Fig. 7(b) (the axes represent normalized con- unconfined part of core concrete in the columns that usually
crete stress f ⬘cc /f ⬘co and axial strain—here, f cc
⬘ is the stress of spalls off once it reaches its ultimate strain. Thus the bearing
core concrete). Overall, the proposed model simulates suc- area of core concrete supporting the load is reduced, causing
cessfully the trends of the experimental results. In addition, a consequent loss of column strength, an effect not adequately
the model produces a reasonable estimation of the strength and modeled by the smeared reinforcement idealization.
deformability increase caused by confinement. However, some The estimated lateral stresses fl (in megapascals) developed
discrepancies are observed in the postpeak regime. Here, after in the columns at peak stress for all the analyzed columns are
the spiral has yielded, the model estimates a constant stress in compared with the experimental values in Table 3. Also in-
the core concrete. On the other hand, the experimental results cluded in the table are the normalized strength of the core
show that the stress of core concrete decreases at the postpeak ⬘ /f co
concrete, K = f cc ⬘ and the confinement effectiveness coeffi-
regime. This discrepancy is attributed to the fact that, at large cient ke [defined by Sheikh and Toklucu (1993) as ke = (K ⫺
288 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001
 ⬘ /fl]. It is evident from this comparison that the estimated
1)f co CONCLUSIONS
values are generally in good agreement with the test results.
A constitutive model of the mechanical behavior of concrete
under load was developed using the theoretical framework of
Parametric Study of Circular Columns strain hardening/softening plasticity. In the model, the re-
sponse of concrete was regulated by appropriately selected
The increase in strength and ductility of core concrete in hardening/softening functions, defined in terms of a plastic de-
passively confined reinforced-concrete columns depends upon formation measure characterizing the state of internal damage
the geometry, amount, and material properties of the confining in the material. Failure was modeled as a continuous process,
steel. In this section, the effect of various design variables on the failure surface either continuously expanding (prepeak
the behavior of circular columns is investigated by means of range) or continuously contracting (postpeak range). Thus, for
a parametric study. Ranges considered for each variable are any level of applied load, a certain amount of irrecoverable
(numbers in parentheses represent the respective parameter (plastic) deformation was produced, building up with increas-
value of the reference case study) ␳s = (0.023)–0.0058; fys = ing load intensity. Plastic dilatancy was obtained from a plastic
(452)–600 MPa; ␳l = (0.079)–0.02; fyl = (509)–350 MPa, potential function (of Drucker-Prager type) properly calibrated
Espiral = (200)–350 GPa; f ⬘c = (35.9)–60 MPa, f ⬘/f t ⬘ = (0.1)–
c to model the experimental trends. As in the case of soils and
0.05; ε 0 = (0.022%)–0.04%; ␯ = (0.2)–0.35. other granular frictional materials, moisture content was shown
Results of the study are summarized in Fig. 8. Response to significantly affect the characteristics of the failure surface
indices considered were the normalized confined concrete for concrete. Experimental data used from the literature were
strength (=K ) and the axial strain at peak core concrete stress correlated with and without prior correction of the stress state
that occurs prior to spalling of the cover ( y- and x-axes in the to account for the fraction of hydrostatic pressure that is car-
plot, respectively). The common point of all lines in Fig. 8 ried by the water in the pores of saturated concrete. By com-
represents the properties of the reference case study. Deviation parison of mechanical responses of dry and saturated concrete,
from the reference case was obtained by varying any single the effect of this variable on calculated response was evaluated
parameter. The arrows and the (⫹) or (⫺) signs illustrate the and modeled. In simulating the experimental response of re-
consequences of increasing or decreasing the magnitudes of inforced-concrete columns with circular cross section, the
the respective parameters from their nominal values. Note that model produced strength and deformation estimates consistent
reducing the volumetric ratio of spiral reinforcement ␳s causes with observed values, whereas the rate of postpeak softening
a decrease in both the strength and the peak strain of the core was somewhat affected by such limitations in the idealization
concrete (Fig. 8). From this analysis it is also found that the of the columns as smearing of transverse reinforcement and
normalized strength enhancement K of the core concrete is neglecting buckling of longitudinal steel and hoop fracture.
mainly characterized by the yield stress of the spiral and by Through evaluation of the parametric sensitivity of the model,
the residual strength of concrete. Increasing the yield stress of it was shown that the properties of both lateral reinforcement
the spiral fys causes an increase in both the values of K and ⬘ , f t⬘/f c⬘, ␯, and ε 0) influenced the
(␳s, fys, and Es) and concrete f co
the strain at peak (Fig. 8), whereas softening response is more strength, ductility, and shape of the stress-strain curves of these
gradual (with milder descending slope and higher residual elements in a manner consistent with the familiar experimental
strength). insights.
Increasing the modulus of elasticity Es of the spiral steel
ACKNOWLEDGMENTS
affected slightly the normalized strength of core concrete K
and the strain at peak (Fig. 8). Variation of other properties of The research presented in this paper was partially funded by Natural
the reinforcing steel in the columns, such as the amount ␳l and Sciences and Engineering Research Council of Canada, Ottawa. The fi-
nancial support provided to the first writer by the government of Indo-
the yield stress of longitudinal reinforcement fyl had no effect nesia (through the University Research for Graduate Education Project)
on the peak stress and strain of core concrete, whereas re- is gratefully acknowledged.
sponse was affected profoundly by changes in the concrete
material properties (the uniaxial concrete strength f ⬘co and ratio APPENDIX. REFERENCES
of f ⬘/f
t ⬘co). Increasing the value of f co
⬘ caused a reduction in both Ansari, F., and Li, Q. (1998). ‘‘High-strength concrete subjected to triaxial
the normalized strength K and the peak strain, whereas reduc- compression.’’ ACI Mat. J., 95(6), 747–755.
ing the ratio of f ⬘/f t ⬘co increased the value of K but decreased Chen, W. F. (1982). Plasticity in reinforced concrete, McGraw-Hill, New
the strain at peak stress (Fig. 8). Variation of the Poisson’s York.
ratio ␯ produced only a marginal increase in the value of K Elwi, A. A., and Murray, D. W. (1979). ‘‘Constitutive model for concrete
but a substantial increase in the value of the strain at peak structures.’’ J. Engrg. Mech. Div., ASCE, 105(4), 623–641.
Finite element analysis of reinforced concrete: State-of-the-art report.
(Fig. 8). Finally, increasing the uniaxial compressive strain at (1982). ASCE, New York.
peak εco caused a corresponding increase in both strength and Hsieh, S. S., Ting, E. C., and Chen, W. F. (1988). ‘‘Applications of a
strain at peak. plastic-fracture model to concrete structures.’’ Comp. and Struct., 28(3),
373–393.
Imran, I. (1994). ‘‘Applications of non-associated plasticity in modeling
the mechanical response of concrete.’’ PhD thesis, Dept. of Civ. Engrg.,
University of Toronto, Toronto.
Imran, I., and Pantazopoulou, S. (1996). ‘‘Experimental study of plain
concrete under triaxial stress.’’ ACI Mat. J., 93(6), 589–601.
Jansen, D. C., and Shah, S. P. (1997). ‘‘Effect of length on compressive
strain softening of concrete.’’ J. Engrg. Mech., ASCE, 123(1), 25–35.
Lade, P. V. (1988). ‘‘Effects of voids and volume changes on the behavior
of frictional materials.’’ Int. J. Numer. and Analytical Methods in Geo-
mech., 12, 351–370.
Launay, P., and Gachon, H. (1972). ‘‘Strain and ultimate strength of con-
crete under triaxial stress.’’ ACI SP 34-13, American Concrete Institute,
Detroit, 269–282.
Li, Q., and Ansari, F. (1999). ‘‘Mechanics of damage and constitutive
relationships for high strength concrete in triaxial compression.’’ J.
FIG. 8. Parametric Sensitivity Analysis of Circular Columns Engrg. Mech., ASCE, 125(1), 1–10.

JOURNAL OF ENGINEERING MECHANICS / MARCH 2001 / 289

Mander, J. B., Priestley, M. J. N., and Park, R. (1988). ‘‘Theoretical material like concrete.’’ Heron, TNO Building and Construction Re-
stress-strain model for confined concrete.’’ J. Struct. Engrg., ASCE, search, Delft, The Netherlands, 29(2), 1–42.
114(8), 1804–1826. Richart, F. E., Brandtzaeg, A., and Brown, R. L. (1928). ‘‘A study of the
Mirmiran, A., and Shahawy, M. (1997). ‘‘Behavior of concrete columns failure of concrete under combined compressive stresses.’’ Engrg. Ex-
confined by fiber composites.’’ J. Struct. Engrg., ASCE, 123(5), 583– periment Station Bull. No. 185, University of Illinois, Urbana, Ill.
590. Sheikh, S. A., and Toklucu, M. T. (1993). ‘‘Reinforced concrete columns
Mitchell, G. P., and Owen, D. R. J. (1988). ‘‘Numerical solutions for confined by circular spirals and hoops.’’ ACI Struct. J., 90(5), 542–553.
elastic-plastic problems.’’ Eng. Comput., 5(4), 274–284. Smith, S. S., William, K. J., Gerstle, K. H., and Sture, S. (1989). ‘‘Con-
Mizuno, E., and Hatanaka, S. (1992). ‘‘Compressive softening model for crete over the top, or: Is there life after peak?’’ ACI Mat. J., 86(5),
concrete.’’ J. Engrg. Mech., ASCE, 118(8), 1546–1563. 491–497.
Toutanji, H. A. (1999). ‘‘Stress-strain characteristics of concrete columns
Ohtani, Y., and Chen, W. F. (1989). ‘‘A plastic-softening model for con-
externally confined with advanced fiber composite sheets.’’ ACI Mat.
crete materials.’’ Comp. and Struct., 33(4), 1047–1055.
J., 96(3), 397–404.
Pantazopoulou, S. J., and Mills, R. (1995). ‘‘Microstructural aspects of Van Mier, J. G. M. (1986). ‘‘Fracture of concrete under complex stress.’’
the mechanical response of plain concrete.’’ ACI Mat. J., 92(6), 605– Heron, 31(3), 1–90.
616. Xie, J., Elwi, A. E., and MacGregor, J. G. (1995). ‘‘Mechanical properties
Peterson, P. E. (1981). ‘‘Crack growth and development of fracture zones of three high-strength concretes containing silica fume.’’ ACI Mat. J.,
in plain concrete and similar materials.’’ Rep. No. TVBM-1106, Div. of 92(2), 135–145.
Build. Mat., University of Lund, Lund, Sweden. Zienkiewicz, O. C., and Taylor, R. L. (1989). The finite element method,
Reinhardt, H. W. (1984). ‘‘Fracture mechanics of an elastic softening Vol. 1, McGraw-Hill, New York.

290 / JOURNAL OF ENGINEERING MECHANICS / MARCH 2001