Professional Documents
Culture Documents
of Concrete Structures
Frédéric Légeron, M.ASCE1; Patrick Paultre, M.ASCE2; and Jacky Mazars, M.ASCE3
Abstract: Performance-based design of structures is becoming the preferred seismic design method. Its use requires special numerical
Downloaded from ascelibrary.org by Li. Co.Sa 8181901/mi/155985 on 03/24/20. Copyright ASCE. For personal use only; all rights reserved.
programs capable of predicting the performance of structures during a seismic event well into the nonlinear range. Seismic analysis results
obtained from these programs depend on the types of elements and constitutive material laws used. For one type of element used, the
results can be sensitive to the size of the elements. This paper presents a simplified finite-element analysis program based on multilayer
elements and damage mechanics modeling of concrete behavior. A method to identify the various parameters required to define the
behavior of the different materials is presented and some guidance on structural modeling using this type of program is provided. This
program is used to predict the behavior of three different structures: overreinforced normal-strength concrete and high-strength concrete
共HSC兲 beams tested monotonically, HSC columns tested under constant axial load and cyclic flexure, and bridge piers subjected to
earthquake-type loading by the pseudodynamic test method. It is shown that predictions are in excellent agreement with experimental
results.
DOI: 10.1061/共ASCE兲0733-9445共2005兲131:6共946兲
CE Database subject headings: Damage; Seismic response; Concrete structures; High strength concretes; Confinement; Finite
elements; Localization.
冉 冊
strain and corresponding stress as well as the strain at peak stress.
Assuming a pure compression test 共D1 = 0兲, when = 0, the strain
− f ⬍ ⬍ 0 → F共兲 = 1 − , F⬘共兲 = 1 − 共6兲
2 f f determined by Eq. 共1兲 is
Fig. 2. Cusson and Paultre stress–strain model for confined concrete 1 = 0.5 + 0.35f t⬘ 共16兲
where f ⬘t = concrete tensile strength.
 2D 2 In monotonic tension, it is assumed that there is no damage up
i = 共11兲 to the tensile strength. The threshold of damage Y 01 is determined
E共1 − D2兲
directly from Eq. 共8兲
where D2 = damage value, which is constant along the unloading
path since no damage is added during unloading. The strain, u, f 2t 1 f t
just before unloading is Y 01 = + 共17兲
2E E
u It is difficult to experimentally obtain the complete tensile
u = + i 共12兲
E共1 − D2兲 stress–strain curve of concrete with a stable postpeak behavior. It
is accepted that postpeak behavior of concrete in tension is related
The Dodd and Cooke 共1994兲 model provides a relation between i
to the amount and distribution of reinforcement. The parameters
and u as
defining the postpeak softening are fitted to the tension stiffening
共u − a兲u behavior of reinforced concrete presented by Vecchio and Collins
i = u − 共13兲 共1986兲. As for the case of compression, the constants A1 and B1
u − Ea
are determined by fitting the response of a “numerical cylinder”
where using the damage model to the stress–strain curve predicted with
a = a冑u⬘cc
Vecchio and Collins’s tension stiffening model. When the post-
共14兲
peak behavior in tension is not critical, which is generally the
⬘cc is the strain at peak of the confined stress–strain curve 共see case with seismic analysis, A1 = B1 = 1 can be used. These values
Fig. 2兲 and give a sharp postpeak softening. It must be noted that, under
冉 冊
cyclic loading, the crack interface deteriorates faster than under
⬘cc u monotonic loading and therefore A1 = B1 = 1 is generally a good fit
a = max ;0.09 共15兲
+ ⬘cc
u
⬘cc for cyclic and seismic loading.
u is determined for a certain u from the confined stress–strain The mechanism of crack closure is controlled by the so-called
relationship of concrete in compression proposed by Légeron and “crack closure stress” f . The crack closure stress is very much
Paultre 共2003兲 described below. i is determined by Eq. 共13兲. affected by concrete strength and placement method 共crack rough-
These three values are replaced in Eq. 共12兲 to determine D2. 2 is ness兲. For monolithic structures, f is in the range of the tensile
obtained from Eq. 共11兲. The procedure can easily be used to find strength and can be taken as
2 for a wide range of unloading strain u. Based on the Dodd and f ⬘c
Cook model, 2 varies very slightly, which reinforces the concept f = − 共18兲
10
of a material constant. In this research program, an average value
over an appropriate strain range is used. Based on a wide range of For concrete with dry joints, f can be significantly lower.
identifications, a starting value of 0.6f ⬘cc is recommended, where The identification procedure may seem tedious, although, with
f ⬘cc = strength of confined concrete as described in Légeron and some experience, it can be done quite fast, since only A2, B2, and
Paultre 共2003兲. Y 02 require a trial-and-error procedure. All the other parameters
The model is set up for the general 3D case, but is used in a are calculated directly. To simplify the task for users, sets of val-
uniaxial formulation incorporating confinement due to lateral ues corresponding to typical concretes and typical levels of con-
pressure that may come, in a passive way, from transverse rein- finement are available in a design process. The identification step
forcement. When confinement is considered, the constants A2, B2, is reduced to selecting a set of parameters that correspond to the
and Y 02 must be appropriately selected. Parameter Y 02 controls the concrete used. For cases for which data are not provided, using a
damage threshold in compression. Increasing Y 02 results in set of values close to the case at hand as a starting point acceler-
greater stress range over which the behavior of concrete is linear. ates the parameter identification process.
It also increases the maximum strength without modifying peak
strain. Parameters A2 and B2 control the maximum stress and
Reinforcing Steel
ductility, respectively. Increasing A2 decreases the strength and
the strain at peak. Increasing B2 decreases ductility while increas- A damage mechanics behavioral law can be use to model the
ing slightly the strength. These constants are identified by fitting cyclic behavior of reinforcing steel 共Ragueneau 1999兲. However,
冤 冥
where
k1 0
k4 − k1 0 − k4
e = Es 共20兲 k3 k5 0 − k3 k5
k6 − k4 − k5 k6/2
If the strain is the maximum experienced strain in that direction, Ke = 共24兲
the stress is found from the bilinear curve. Otherwise SYM. k1 0 k4
k3 − k5
p = 冉
− a
b − a
冊 p
关Es共b − a兲 − b兴 共21兲
where
k6
where N
1
兺
冉 冊
k1 = E j共D j兲b jh j
c L j=1
Es 1 − 共b − a兲
Es
p= 共22兲 N
Es共b − a兲 − b 12
k3 = 兺
L3 j=1
E j共D j兲b jh j y 2j
All the variables in Eqs. 共19兲–共22兲 are described in Fig. 3. The
algorithm requires storing the last point of crossing with the x
axis 共strain- or zero-stress axis兲 a and the last maximum strain −1
N
e = 关f 1
FiT 0 f3 − f1 0 − f 3兴 共28兲
where
N
兺
Fig. 7. Concentration of plastic deformation in bottom element
f1 = L 关共1 − D j兲Ecjij + pj j兴b jh j 共29兲
i=1
more smaller elements. However, for nonlinear analysis, this is would have the highest accuracy. Yet this does not reflect what
not necessarily true, since localization phenomena may result in actually happens.
large errors in predicting structure responses. Consider, for ex- Inadequate element size for the material stress–strain law used
Downloaded from ascelibrary.org by Li. Co.Sa 8181901/mi/155985 on 03/24/20. Copyright ASCE. For personal use only; all rights reserved.
ample, a 2 m high column under high axial load and subjected to overcomes any effort to precisely model material behavior. Two
monotonically increasing flexure. The column is discretized with methods are commonly used to adequately predict the behavior of
equal length elements. Three models are obtained with element softening structures: 共1兲 adapt the stress–strain law of material to
length equal to 100, 250, and 500 mm, resulting in three models element size 共Coleman and Spacone 2001兲 or 共2兲 use a predeter-
with 20, 8, and 4 elements, respectively. EFiCoS analysis consists mined element size corresponding to conventional stress–strain
of imposing an incremental displacement at the tip of the columns laws defined in standardized tests. The second approach is more
up to rupture. The result of the analysis is presented in Fig. 6 practical and is adopted here. A stability approach 共Légeron 1998兲
showing tip displacement versus horizontal force at the base for or an energy approach 共Bažant 1979兲 can be used to determine
the three models. It can be seen that, although moment capacity is element length. An easier approach is to use an equivalent plastic
marginally affected by the element size, displacement capacity hinge length giving comparable results in practical situations
varies considerably for the three models. To understand the prob- 共Légeron 1998兲. An appropriate expression for the equivalent
lem, assume that the column shown in Fig. 6 has a perfectly plastic hinge length is given by Priestley et al. 共1996兲 as
bilinear moment–curvature response. The curvature distribution
when the maximum curvature u is reached in the most loaded l p = 0.08L + 0.022db f y 共m,MPa兲 共32兲
element is presented in Fig. 7 for all three models. From this where L = length between the end of the member and the point of
simplified curvature distribution, the ultimate displacement ⌬u contraflexure; and db and f y = diameter and yield strength of the
can be calculated as longitudinal bars, respectively.
For structures with hardening behavior, localization has only a
⌬u = ⌬y + le共u − y兲共H − le/2兲 共31兲
limited influence on global behavior. However, a certain strain
where H = height of the column; le = element length; ⌬y = yield dis- localization always occurs and the curvature ductility demand
placement; and y = yield curvature. As shown in Fig. 7, ⌬y and will also depend on element length, although in a less pronounced
y are not affected by element length since the layered beam way. In this case, it is also necessary to use fixed element length
element is exact within the Euler–Bernouilli beam theory for lin- as for softening structures.
ear elastic behavior. However, ⌬u is directly related to element
length le as evidenced by Eq. 共31兲. The maximum displacement of
the model with 500 mm long elements is roughly three times the Comparison to Test Data
maximum displacement obtained with the model with 100 mm
long elements. Hence, using a fine mesh to get a more accurate To validate the methodology and the model, predictions using
response may result in a higher ductility demand in the critical EFiCoS are compared to three different sets of experimental data
section for a constant displacement ductility. This mesh depen- covering a wide range of applications: 共1兲 tests on over-reinforced
dency is certainly not acceptable in performance-based design. beams with normal-strength concrete 共NSC兲 and high-strength
Reasoning along the same lines as for linear elastic analysis concrete 共HSC兲 共Van Mier and Ulfkjaer 2000兲; 共2兲 cyclic quasi-
would lead to assuming that the model with 100 mm elements static tests on HSC columns 共Légeron and Paultre 2000 and
Fig. 9. Prediction of small overreinforced high-strength concrete and Fig. 10. Prediction of small overreinforced fiber high-strength con-
normal-strength concrete beams crete beams
Fig. 12. Tests of high-strength concrete columns under axial load and reversed flexure
Fig. 14. Bridge tested at Ispra ferent damping values, but only the results with the values of the
Rayleigh parameters used during the tests are presented, disre-
garding the additional damping due to the pseudodynamic test.
pier, i.e., shear at the top of a pier is transferred into horizontal The predictions are excellent, even at high nonlinear demand,
shear in the girder; flexure at the top of the pier is transferred into considering that only the data available to an engineer at the
horizontal shear and torsion in the girder 共as the actual bearing project level were used.
did not transfer any moment兲.
The displacements at the top of the central pier and at one of
the lateral piers recorded during the test are compared with the Conclusion
predicted displacements using EFiCoS. Damage parameters used
are presented in Table 3. The results shown in Figs. 15 and 16 In this paper, a complete approach to the prediction of concrete
were obtained for a generated accelerogram having a peak ground structures subjected to cyclic and earthquake-type loadings is pre-
acceleration twice the design peak ground acceleration. This high sented. This approach is based on a 2D damage mechanics lay-
acceleration level implies that the piers are subject to high ductil- ered finite-element program. Some guidance is provided on how
ity demand. This translates into a highly nonlinear pier response. to use such modeling techniques in terms of material constitutive
A set of stationary artificial accelerograms fitting the Eurocode 8 laws and element length to regularize the response predictions of
response spectrum has been generated and one of them was cho- nonlinear softening structures such as highly compressed rein-
sen for the pseudodynamic tests 共Pinto et al. 1996兲. The acelero- forced concrete columns. The program was used to predict the
gram has a duration of 10 s and a nominal peak ground accelera- behavior of a large range of structural components and structures
tion of 0.35g. This accelerogram was scaled for the test model tested under monotonic, cyclic, and seismic loading. It was shown
according to the similitude rules chosen 共Cauchy similitude, ve- that the global response of the test specimens was predicted quite
locity preserved兲. The resulting accelerogram used in the PSD well. An important point in performance-based design is that local
tests had a peak ground acceleration of 0.35⫻ 2.5= 0.875g and a behavior such as spalling of concrete cover, cracking, and yield-
duration of 10/ 2.5= 4 s. The complete acceleration time history ing of longitudinal bars were taken into account as well as global
was used in the prediction with EFiCoS. As can be seen from behavior. This approach is a powerful tool for seismic analysis
Figs. 15 and 16, the predicted responses are in good agreement and other engineering problems in which nonlinear structure re-
with the test data. Small differences between the predicted and sponse is dominated by flexure. Moreover, damage mechanics
measured responses can be explained in part by the integration provide a good picture of the level of damage that a structure has
method used during the tests. Indeed, a special algorithm was accumulated during an earthquake, which is of prime importance
developed at ISPRA for the nonlinear analysis and the substruc- for performance-based earthquake engineering. The method de-
ture technique which was not used in the analysis with EFiCoS. scribed in this paper is a basis from which researchers and prac-
This integration method introduced a certain numerical damping titioners can develop guidance to help engineers in modeling
into the system, which could not be reproduced by the traditional structures for PBEE. Such approaches are critical if the engineer-
Rayleigh damping. A parametric study was carried out with dif- ing community wants to successfully implement PBEE. Further