You are on page 1of 12

Composites: Part B 38 (2007) 265276

www.elsevier.com/locate/compositesb

Analysis of reinforced concrete columns retrotted


with ber reinforced polymer lamina
a,1 b,*
Baris Binici , Khalid M. Mosalam
a _ nu Bulvar, 06531 Ankara, Turkey
Department of Civil Engineering, Middle East Technical University, Ino
b
Department of Civil and Environmental Engineering, University of California at Berkeley, 721 Davis Hall, Berkeley, CA 94720-1710, USA

Received 20 October 2005; accepted 10 January 2006


Available online 3 May 2006

Abstract

Fiber reinforced polymer (FRP) lamina have been used widely in the last decade to enhance strength and deformation capacity of
decient reinforced concrete (RC) columns. Seismic assessment and retrot of existing columns in buildings and bridge piers necessitate
accurate prediction of the available deformation capacity. In this study, a new analytical model is proposed to represent potential plastic
hinge regions of RC columns prior to and after FRP retrot. A recently developed variable conned concrete representation is employed
within the framework of ber-discretized frame elements to model the compression zone of the FRP-conned region. Connement dis-
tribution within this region is included through the use of a bond model, whereas the eect of lap splices are considered using an eective
steel strain concept. Comparisons of analytical estimates with experimentally measured response show that the proposed model is capa-
ble of capturing essential features of the response such as strength degradation due to lap splice slippage, and failure due to FRP rupture.
Furthermore, a detailed sensitivity study is conducted to determine the parameters whose uncertainty signicantly aects the behavior. It
is observed that, in estimating the response of existing decient columns, parameters such as plastic hinge length, concrete strength and
splice length are important sources of uncertainty. While for FRP-retrotted columns, parameters such as jacket stiness, dilatation
strain at splice failure and yield strength of the reinforcing bars are more important sources of uncertainty.
2006 Published by Elsevier Ltd.

Keywords: A. Polymer ber; C. Stress transfer; C. Computational modeling; C. Statistical properties

1. Introduction deformability of the structural members by avoiding or


delaying undesirable phenomena such as concrete crushing,
Use of ber reinforced polymer (FRP) lamina has longitudinal reinforcing bar buckling and lap splice failure.
gained increasing popularity in structural retrot applica- There is a vast amount of experimental work conducted
tions due to advantages such as lightweight, high strength on FRP retrot of columns in the last decade. Experiments
and ease of application. One of the most attractive applica- were performed on concentrically loaded concrete speci-
tions of FRP lamina is wrapping existing decient rein- mens retrotted with FRP lamina to observe the strength
forced concrete (RC) columns or bridge piers to enhance and deformation capacity enhancement in pure compres-
their deformation capacity especially at the potential plas- sion [14]. Test results revealed that behavior of FRP-
tic hinge regions. In this way, it is possible to circumvent conned concrete substantially diers from that of steel
the expected brittle behavior in the absence of properly conned concrete due to dierences in constitutive behav-
designed conning steel reinforcement and to enhance the ior of the two materials [4]. For steel conned concrete,
conning stresses are proportional to the applied axial load
*
up to the yielding of the steel. Beyond yielding of the trans-
Corresponding author. Tel.: +1 510 643 4805; fax: +1 510 643 8928.
E-mail addresses: binici@metu.edu.tr (B. Binici), mosalam@ce.berke-
verse steel reinforcement, conning stresses remain approx-
ley.edu (K.M. Mosalam). imately constant. On the other hand, for FRP-wrapped
1
Tel./fax: +90 312 210 1193. concrete, the level of connement is proportional to the

1359-8368/$ - see front matter 2006 Published by Elsevier Ltd.


doi:10.1016/j.compositesb.2006.01.006
266 B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276

axial load up to the point where FRP ruptures and failure basic response quantities such as strength and drift ratio.
occurs in a sudden and brittle manner. Other studies con- It should be emphasized that this novel method of analysis
centrated on FRP retrot of decient RC columns for seis- incorporating a realistic distribution of connement in the
mic strengthening [59]. It has been shown that FRP compression zone and tracing the jacket strains during
retrot can signicantly improve the lateral deformation nonlinear analysis using ber-discretized frame elements
capacity of columns keeping the plastic hinge regions intact is employed for the rst time in the literature. It is believed
even at large deformation cycles. Furthermore, it is pointed that the methods described in this paper will facilitate the
out that the poor seismic performance due to short splice accurate estimation of the deformation demands of de-
lengths of column longitudinal reinforcing bars at plastic cient and FRP-retrotted structural systems.
hinge regions can be improved by limiting concrete dilata-
tion through the use of FRP jackets [79]. 2. Modeling aspects
In light of the experimental studies summarized above, a
better understanding of FRP connement mechanism was This section describes the constitutive relationships and
achieved and many FRP-conned concrete models were the methodology of modeling FRP-wrapped RC columns
developed. An extensive review of the literature on FRP- having conning steel and lap splice deciencies. The
conned concrete can be found in [10,11]. Most of these important class of columns with deciency in shear capac-
models are empirical in nature and employ best-t expres- ity is out of the scope of the present research. The modeling
sions (e.g. functions of the jacket properties) of the exper- approach described below is valid for RC columns having
imentally obtained stressstrain curves. Other analytical circular sections. However, this approach can be easily
models [1214] dene the axial and lateral stressstrain extended to members with other sections by introducing
relationships of concrete for dierent levels of connement. the well-known section eciency factors [16] to reect the
By matching the expansion of concrete to the straining of reduced connement eciency of non-circular sections.
the jacket, FRP-conned concrete response is obtained
from a family of active conned concrete curves. The 2.1. Connement model
advantages of these latter models include: (1) the exibility
of introducing dierent material models for the jacket and Consider a circular RC section wrapped with an FRP
(2) the ability to directly dene failure of the jacket from jacket subjected to the combined eect of axial force and
the transverse jacket strains. Although these models have uniaxial bending moment (Fig. 1). Following the classical
been shown to estimate the axial response of FRP-conned beam theory, a linear axial strain distribution is assumed
concrete accurately, they have not been used widely for the for the section under the given loading conditions. In the
analysis of columns subjected to combined axial loads and compression zone, it is reasonable to consider that conn-
bending moments. ing stress distribution will not be uniform. The conning
The main objective of this paper is to propose a novel stress is maximum at the extreme compression ber,
analytical model that can be used in the framework of non- decreases with the decreasing distance to the neutral axis
linear analysis making use of the ber-discretized frame ele- and vanishes at the neutral axis location. In order to esti-
ments [15]. Modeling of concrete in compression and of the mate the non-uniform conning stress distribution, an elas-
reinforcing steel lap splice regions prior to and after tic bond model is utilized (Fig. 1). This model assumes that
upgrades with FRP jacket is discussed in detail. Subse- the FRP and the adhesive are thin compared to the con-
quently, experimental validation of the proposed computa- crete section and carry only axial and shear stresses, respec-
tional models to estimate the load-deformation response of tively. Furthermore, it is assumed that perfect bond exists
structural elements is presented. Finally, a deterministic between the jacket and the concrete section. The equilib-
sensitivity analysis is conducted to observe the major rium condition for an innitesimal element along the arc
parameters whose uncertainty may signicantly alter the of the jacket surrounding the circular section is

Fig. 1. Conning stress distribution on a reinforced concrete column cross-section.


B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276 267

drf tf sdx 1 1

where rf is the FRP stress, tf is the FRP thickness and s is

Relative Distance from Neutral Axis


0.8
the interface shear stress. The assumed elastic material 5
behavior of the adhesive and the FRP yields 4 3
0.6
rf Ef du=dx and s uGa =ta 2
0.4 2 1 parabolic
where Ef is the modulus of elasticity of the FRP, u is the
distribution
axial elongation of the FRP (or the shear deformation of
0.2
the adhesive), Ga is the shear modulus of the adhesive,
and ta is the thickness of the adhesive. Combining Eqs.
(1) and (2) and using the transformation x = Rh, the fol- 0
0.00 0.20 0.40 0.60 0.80 1.00
lowing governing dierential equation of the bond model Relative Confining Stress
is obtained
Fig. 2. Conning stress distribution for dierent values of bond parameter
d2 rf =dh2  A2 rf 0 3 (values shown on curves are bond parameter, A values).

where A is the bond stress parameter with


A2 R2 Ga =Ef tf ta . Solution of Eq. (3) is in the form
rf C sin hAh D cos hAh with C and D obtained sive thickness and the shear modulus to compute A. For
from the boundary conditions: typical column retrots, values of A range from about 1
to 5. It can be observed that for this range, conning stress
rf 0 at h 0 rf rfmax at h hc 4 distribution can be reasonably approximated as propor-
Above conditions imply that maximum jacket stress tional to the square of the distance from the neutral axis.
occurs at the extreme compression ber and it vanishes at The eect of this distribution on the global response of
the neutral axis location. Therefore, the conning stress FRP-retrotted RC columns is studied later in the deter-
distribution becomes: ministic sensitivity analysis section.
In order to determine rfmax for a prescribed compressive
rf rfmax sin hAh= sin hAhc 5 strain at the outermost compressive ber, a variable con-
Once the maximum jacket stress rfmax is known, conn- ned concrete constitutive model is adopted for concrete
ing stress at any location in the compression zone can be in compression [12]. The conned concrete model describes
found using Eq. (5). Fig. 2 shows conning stress distribu- the axial stressstrain behavior of conned concrete
tion for dierent values of A. In fact, it is very dicult to (Fig. 3) using three continuous functions for the ascending
precisely estimate some of the parameters such as the adhe- and descending branches given by

Fig. 3. Variable conned concrete model.


268 B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276

8
>
> E c e1 for e1 6 e1e
>
>  
>
< r r  r e1  e1e r
1e 10 1e for e1e 6 e1 6 e10
r1 e10  e1e r  1 e1  e1e =e10  e1e r 6
>
>   
>
>
> e1  e10 2
: r1r r10  r1r exp  for e10 6 e1
a

where r1 and e1 are the axial stress and strain, respectively, e10  e1e tl  0:5
D p where b 12
Ec is the modulus of elasticity of concrete. The elastic stress  ln b t l  t0
limit, r1e, ultimate strength, r10, and residual strength, r1r,
The limiting secant strain ratio, tl, can be obtained from:
for a given lateral conning stress r3 are computed using
the LeonPramono [17] failure criterion given by 1
 p  tl 0:5 13
/ 0:854
r1 fc0 k c m/  1  k/2 / 7
Further details of the model summarized above and its
where m fc02  ft02 =fc0 ft0 , ft0 is the concrete uniaxial ten- extensive experimental validation can be found in [12].
sile strength, k is the hardening parameter and is equal to For FRP-conned concrete, strain compatibility in the
0.1 at elastic stress limit and 1.0 at ultimate strength and lateral direction is enforced for the imposed axial strain
beyond, c is the softening parameter and is equal to 1.0 such that the jacket and the lateral concrete strains are sim-
prior to softening and 0.0 at residual strength, and ilar. Strain compatibility is schematically shown in Fig. 3.
/ r3 =fc0 is the connement ratio, where fc0 is the 28-day The connement provided by the jacket on the concrete
concrete uniaxial compressive strength of the standard cyl- section or the jacket strain is computed for a prescribed
inder. The elastic strain limit, e1e, and strain at ultimate axial strain by solving the following nonlinear equation
strength, e10, are computed by in the inelastic range
 
r1e r10
e1e and e10 5e0  0:8 8 e1 ts r3  r3 =D 0 14
Ec fc0
where ts(r3) is the secant strain ratio given by Eq. (11) in
where e0 is the strain at peak stress under uniaxial compres-
the inelastic range, and it is a function of r3 since tl is in
sion. The other constant r in Eq. (6) is computed by:
turn a function of /. The parameter D is the eective jacket
Ec r10  r1e
r where Es 9 rigidity given by
Ec  E s e10  e1e
Ej t Ej qj
Parameter a in Eq. (6) is calibrated such that the area D 15
under the softening region is equal to the compressive frac- R 2
ture energy obtained from a uniaxial compression test where Ej is the modulus of elasticity of the jacket in the
divided by the characteristic length of the specimen in the hoop direction, t is the thickness of the jacket, R is the ra-
loading direction, Gfc/lc, and is given by: dius of conned concrete section, and qj is the volumetric
! ratio of the jacket (ratio of the transverse reinforcement
2
1 2Gfc r10  r1r
a p  10 volume to the ratio of the concrete volume).
pr10  r1r lc Ec For the cross-section subjected to combined axial force
It can be observed that the conned concrete model and bending moment (Fig. 1), the lateral strain compatibil-
described above includes the gauge length of the test speci- ity is enforced for the extreme concrete ber in compression
men, lc. Therefore, stresses and strains refer to average val- using Eq. (14). Once the connement at this extreme ber is
ues over the specied length in the presented constitutive known, the connement at any level in the compression
model. zone can be computed using Eq. (5). The modeling
Deformations in the transverse direction are described approach described above incorporates the variation in
using the secant strain ratio, ts = e3/e1, given by: the connement distribution for the compression zone,
8 and it is capable of monitoring the expansion of concrete
< t0    for e1 6 e1e and accordingly the jacket strains. Therefore, the jacket
ts e1  e1e 2
: ts tl  tl  t0 exp  for e1e 6 e1 rupture, which generally dictates the failure of a retrotted
D
column, is directly computed from the known FRP rupture
11 strain, erup.
In the elastic region, the secant strain ratio is equal to
the Poissons ratio of concrete, t0, that is usually between 2.2. Splice model
0.15 and 0.20. At ultimate strength, secant strain ratio is
0.5, whereas it approaches a limiting value, t1, for strains An eective steel stress approach is used to model the
well beyond the strain at the peak stress. The parameter longitudinal reinforcing bar lap splices in the plastic hinge
D is computed by: region with and without conning FRP jacket. Consider
B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276 269

Fig. 4. Longitudinal reinforcing bars lap splice model.

the spliced bars consisting of starter and spliced bars r3


r r0  13 P 1:1 23
(Fig. 4). For any location within the splice, the equilibrium fc0
equations of the bars are
where r0 is 2 for Grade 60 steel and 1.5 for Grade 40 steel.
 2
1 d Once the conning stress, r3, is known, a unique bond
si pd b li rsi p b for i 1;2 16 stressslip curve can be constructed using the above expres-
2 4
sions. Experiments conducted in [7,9] reveal that after a
where si is shear resistance between the bar and the con- concrete dilatation strain, edl, of about 0.00100.0015, con-
crete, li is the length shown in Fig. 4, db is the spliced bar nement has no eect on clamping the splices. Therefore,
diameter, and rsi is the steel stress at the considered loca- an average strain limit of 0.00125 is used to compute r3
tion. It is assumed that the distributions of si for the two due to the presence of the FRP jacket and the transverse
bars are similar and follow a linear variation given by steel reinforcement. Hence, r3 can be calculated from
li  
si sm for i 1;2 17 Ef tf Es Ash
Ls r3 edl 24
R sRc
where sm is the maximum shear resistance along the splice
where Rc is the radius of the column concrete core, Ash and
length Ls, Fig. 4. Adding the two equations for i = 1,2 gi-
s are the transverse steel area and spacing, respectively, and
ven in Eq. (16) for the two bars and making use of Eq.
edl = 0.00125.
(17) with l2 = Lsl1, the following equilibrium equation is
At the considered location within the splice, which can
obtained for the splice
be the integration point for a ber element discretization,
rs d b total steel strain, est, is dictated by the strain prole of
sm ! 18
l21 Ls  l1 2 the section at the corresponding analysis increment, and
2 it is assumed to be decomposed into slip strain ess = u/Ls
Ls
and elongation strain ese, i.e. est = ess + ese. The strain com-
where rs is the summation of the two reinforcing bar stres- ponents ess and ese are computed iteratively such that equi-
ses, rs1 and rs2. The approach presented above is dierent librium (Eq. (18)) and the bond stressslip relationship
from the splice model presented in [9] in the sense that it (Eqs. (19)(24)) are simultaneously satised. Once ese is
assumes a bond stress distribution, which satises the zero computed, the employed steel constitutive model (e.g. elas-
stress condition at the free ends of the bars. tic perfectly plastic with yield (plastic) strength fy) can be
The bond stressslip relationship proposed in [9] is utilized to compute the eective steel stress.
adopted in this study to compute the eective steel stress
taking into account bar slip. This relationship is given by 3. Fiber-discretized frame element implementation
smax ru=umax
sm r 19 Constitutive models outlined above are implemented
r  1 u=umax within the framework of ber-discretized frame elements.
where u is the splice slip and the rest of the parameters are A retrot plastic hinge element using distributed plasticity
dened using Eqs. (20)(23). with displacement-based frame elements is developed using
the ber discretization. The modeling approach adopted
smax s0 1:4r3 20 for a cantilever circular column is shown in Fig. 5. The
p
s0 20 fc0 =d b in MPa and mm units 21 plastic hinge region is modeled using a single displace-
  ment-based element with linear curvature distribution
r3
umax 0:25 1 75 in mm 22 and ve integration points along its length. In this way, it
fc is possible to approximate the non-uniform curvature
270 B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276

Fig. 5. Fiber frame element modeling of a cantilever column.

distribution and the stiness degradation in an accurate were conducted in a single curvature deformed congura-
manner. The above models, implemented in a nonlinear tion by applying reversed lateral deformation cycles under
frame element analysis program [18], are employed within constant axial load [5,9]. The simulations of the experi-
the constitutive driver for steel and concrete bers. Con- ments were conducted under constant axial load and
crete in tension is modeled using a linear elastic behavior imposing monotonically increasing lateral displacement
up to cracking followed by a linear descending branch history for a cantilever column model using ve elements
(Fig. 5). The strain corresponding to complete loss of ten- as illustrated in Fig. 5. The length of the plastic hinge ele-
sile resistance is determined from the tensile fracture ment used to model the FRP connement inuence on the
energy, Gft, which is assumed to be a known material prop- critical region was selected using the following equation
erty. Furthermore, the characteristic length, lc, is taken as
Lp 0:077L 8:16d b 25
the length of the plastic hinge, which equals the element
length in the present study, times the weight of the integra- where L is the column length and db is the longitudinal steel
tion point under consideration as suggested in [19]. bar diameter. Eq. (25) was found to yield perfect correla-
For nonlinear analysis, the forcedisplacement relation- tion with the mean of the experimentally measured plastic
ship at the element level is commonly expressed in an incre- hinge lengths [22]. The eects of FRP wrapping on conne-
mental form using the tangent stiness matrix and ment and splice clamping, if there is any, were only consid-
incremental deformations. From the section stiness and ered for the plastic hinge element. The remaining elements
resisting forces, the element stiness matrix and resisting were modeled using frame elements discretized with con-
force vector are obtained. Nonlinear geometric eects are crete bers having elastic behavior in compression and
incorporated through the addition of the geometric sti- elastic-brittle behavior in tension (Fig. 5). The elastic-brit-
ness matrix to the element stiness matrix. The individual
element stiness matrices and individual resisting force vec-
tors are assembled by the conventional nite element Table 1
assembly procedure to determine the global stiness matrix Data used in numerical simulations
and global resisting force vector. For nonlinear response, Properties Experimental program Aa Experimental program Bb
equilibrium is usually not satised in a single iteration. fc (MPa) 45 40
Therefore, an incremental-iterative numerical technique is fy (MPa) 462 450
needed. In this study, the minimum unbalanced displace- c (mm)c 38 20
ment norm method [20] is adopted because of its simplicity Ls (mm) 381 None
and capability of capturing the softening behavior in the s (mm) 127 300
ql (%)d 2 1
force-deformation relationship. Fiber-discretized element Ef (MPa) 48,000 54,000e
formulations are well-established in the literature and rele- 20,700f
vant references can be consulted for further details [18,21]. erup 0.011 0.0147e
0.0200f
a
4. Experimental validation Experiments from [9].
b
Experiments from [5].
c
Concrete cover.
The results of two experimental programs, named here- d
Reinforcement ratio of column longitudinal steel.
after as A and B are used for validation of the developed e
Values for CFRP lamina.
f
analytical model. All the specimens in these experiments Values for GFRP lamina.
B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276 271

tle tension behavior was used to accurately model stiness 1960s and 1970s construction practice. One reference spec-
degradation due to progressive cracking with the increase imen and two FRP-retrotted specimens were analyzed
of the applied lateral deformation. Hence, the need for using the developed computational models. The axial load
using eective rigidity was eliminated. Concrete
p compres- was about 8% of the axial load carrying capacity of the col-
sive and tensile fracture energies, Gfc 8:8 fc0 (mm and umn, which represents well the axial load level in bridge
N units) and Gft = Gfc/250 were adopted from [23] based piers. For the two FRP-retrotted specimens, two dierent
on uniaxial compression test results. pConcrete tensile FRP amounts were used, namely four and ve layers of
strength was assumed to be ft0 0:35 fc0 (MPa units) in FRP with a layer thickness of 3.2 mm, to investigate the
all analyses. Properties of the test specimens for the two eect of FRP amount on strength and deformation capac-
experimental programs are given in Table 1. ity enhancement. Comparisons of monotonic analyses
Experimental program A [9] was conducted on 1/2-scale results with the experimentally measured cyclic load-defor-
bridge columns with lap splice deciencies common to mation relationships are presented in Figs. 6 and 7. It can

Fig. 6. Comparison of experimental cyclic load-deformation behavior [9] with estimated analytical response for no retrot specimen.

Fig. 7. Comparisons of experimental cyclic load-deformation behavior [9] with estimated analytical response for FRP retrot specimens.
272 B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276

300 400
ST2NT
300 P/Po =0.54
S 3NT 2 layers GFRP
200 P/Po = 0.54
200

Moment (kNm)
No Retrofit
Moment (kNm)

100
100
0

0 -100 Analysis
Experiment
-200 Detected FRP
rupture initiation
-100 (analysis)
-300

Analysis -400
-200 Experiment -250 -200 -150 -100 -50 0 50 100 150 200 250
Curvature (1/km)
-300
-80 -60 -40 -20 0 20 40 60 80 400
Curvature (1/km) ST3NT
300 P/Po =0.54
1 layer CFRP

Moment (kNm)
300 200

S 4NT 100
200 P/Po = 0.27
No Retrofit 0
Moment (kNm)

-100
100
Analysis
Experiment
-200 Detected FRP
rupture initiation
0 -300 (analysis)

-400
-100 -250 -200 -150 -100 -50 0 50 100 150 200 250

Curvature (1/km)
Analysis
-200 Experiment Fig. 9. Comparisons of experimental momentcurvature response [5] with
analytical estimations for FRP-retrotted specimens under high axial
-300 load.
-80 -60 -40 -20 0 20 40 60 80
Curvature (1/km)
dicted. Since specimens were not tested up to complete col-
Fig. 8. Comparisons of experimental momentcurvature response [5] with
analytical estimations for decient columns.
lapse, a clear comparison on the descending branch of the
momentcurvature responses can not be made. The exper-
be observed that for the as-built decient column, signi-
400
cant strength degradation occurs after reaching the peak ST4NT
P/Po = 0.27
300
strength. For the FRP-retrotted columns, strength degra- 1 layer CFRP

dation is delayed resulting in stable behavior at large 200


Moment (kNm)

deformations. The model predictions of strength and 100

deformation at ultimate load are in good agreement with 0

those obtained from the test whereas strength degradation -100


Analysis
in the numerical simulation occurs more rapidly than that -200
Experiment
Detected FRP

observed in the experiment. For the FRP-retrotted col- -300


rupture initiation
(analysis)

umns, the backbone curves of the cyclic load-deformation


-400
relationships are in excellent agreement with the predicted -250 -200 -150 -100 -50 0 50 100 150 200 250

response. Curvature (1/km)


Experimental program B [5] was conducted to observe
400
the eect of two types of FRP lamina, namely carbon ST5NT
300
and glass, on the behavior of retrotted building columns P/Po =0.27
1 layer GFRP

with conning transverse steel deciency. Two dierent 200


Moment (kNm)

axial load levels (27% and 54% of the axial load carrying 100

capacity, Po) and two dierent FRP lamina were taken as 0

the test parameters. Momentcurvature response obtained -100


Analysis
from LVDTs located within the plastic hinge region were -200
Experiment
Detected FRP

reported for all the specimens. The momentcurvature at -300


rupture initiation
(analysis)

the bottom integration point of the plastic hinge element


-400
is used for validation of the analytical results. Comparisons -250 -200 -150 -100 -50 0 50 100 150 200 250

of the behavior of the two reference specimens (as built Curvature (1/km)
with axial load = 27% and 54% Po) with analytical estima- Fig. 10. Comparisons of experimental momentcurvature response [5]
tions are presented in Fig. 8. It can be observed that the with analytical estimations for FRP-retrotted specimens under low axial
stiness and strength of the test specimens are closely pre- load.
B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276 273

5. Deterministic sensitivity analyses

Evaluation of the structural response, namely the engi-


neering demand parameter (EDP), e.g. strength and ulti-
Probability distribution of mate drift ratio, of components and systems due to
random variable X severe loading such as seismic action includes various
XLB XUB uncertain input parameters. Parameter uncertainty causes
Lower bound of X Upper bound of X uncertainty of the response upon which damage and loss
analyses are based. A deterministic sensitivity analysis is
Structural analysis performed to determine the relative signicance of each
(e.g. pushover analyses) uncertain variable in the developed computational models
on the response measures of decient columns with and
without FRP retrot. In that regard, the so-called tornado
diagram, commonly used in decision analysis [24], is
employed. Recently, the tornado diagram has been used
in sensitivity analysis in earthquake engineering [2527].
Probability distribution of This tornado diagram consists of a set of horizontal bars,
EDP (unknown) referred to as swings, one for each random variable (RV).
EDP(XLB.) EDP(XUB.) The length of each swing represents the variation in the
output due to the variation in the respective RV. Therefore,
Swing = [EDP(XUB.) - EDP(XLB.)]
a variable with larger eect on the output has larger swing
Fig. 11. Procedure of developing a swing in the tornado diagram. than those with smaller eects. In a tornado diagram,
swings are displayed in the descending order of the swing
size from top to bottom. This wide-to-narrow arrangement
of swings resembles a tornado.
imental and analytical momentcurvature responses for In the tornado analysis, the output variable (strength or
FRP-retrotted specimens having high axial loads (54% ultimate drift ratio in this study) is assumed to be a known
Po) are presented in Fig. 9. It can be observed that the deterministic function (developed using the computational
strength, deformation capacity and FRP rupture deforma- models within the ber element method) of a set of input
tion levels are in excellent agreement. The experimental and RVs whose probability distributions are assumed by the
analytical momentcurvature responses for FRP-retrot- analyst. For each input variable, two extreme values corre-
ted specimens having low axial loads (27% Po) are pre- sponding to predened upper and lower bounds of its
sented in Fig. 10. It can be observed that strength of the probability distribution (e.g. mean 2 standard deviation
test specimens are estimated with a reasonable accuracy as used in this study) are selected. For each input RV,
for both retrot schemes. However, the deformation capac- the deterministic function is evaluated twice, using the
ity is underestimated by about 30%. This discrepancy is two extreme values of the selected input RVs while the
attributed to the uncertainty associated with model param- other input RVs are set to their best estimates such as
eters whose sensitivity might aect the analyses results. the means. Fig. 11 schematically shows the procedure for
Detailed deterministic sensitivity analyses are presented in developing a swing. This procedure yields two bounding
Section 5 to shed light on the sources of uncertainty for col- values of the output for each input RV. The absolute
umns prior to and after FRP retrot. dierence of these two values is the swing of the output

Random Structural
Swing Tornado diagram
Variable Analysis
X2 ..Xn at means
X1

X1 , X3 ..Xn at means

X1 , X2 , X4 ..Xn at means
X3 Sorting

X1 .. Xn-2 , Xn at means EDP corresponding to


X1 mean of all RVs
X1 .. Xn-1 at means
Xn

Fig. 12. Procedure of developing the tornado diagram.


274 B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276

corresponding to the selected input RV. This process is Pa


repeated for all other input RVs to compute the swings Ls
of the output. Finally, one builds the tornado diagram by Lp
arranging the obtained swings in a descending order,

Parameters
fc '
Fig. 12. ft
dl
5.1. Selected uncertain variables s
c
The selection of the input variables depends on the out- Gf c
Gf t
put variable of interest. In this study, we selected the No Retrofit
fy
strength (peak lateral load) and ultimate drift ratios (corre-
sponding to 85% of the peak load on the descending 0.6 0.7 0.8 0.9 1 1.1 1.2
branch of the force-deformation relationship and to the Swing of Normalized Strength
rupture of the FRP jacket for the as-built and retrotted
Ls
columns, respectively). Note that the ultimate drift ratio Ef
is dened as the ultimate lateral displacement of the col- dl

umn tip normalized by the column height. The columns Pa


selected for the tornado analyses have the mean geometri- fc'

Parameters
c
cal and material properties of the specimens C1-A and Lp
C2-RT4 for the as-built and FRP-retrotted cases, respec- fy
tively, in [9]. s
Gft
Uncertainty in the structural strength and stiness has A
been of major interest by a number of probabilistic studies ft
of RC members and systems. A short review can be found rup
With FRP Retrofit
Gfc
in [18]. Sources of uncertainty in dierent parameters of the
developed computational models are selected to be as com- 1.1 1.2 1.3 1.4 1.5 1.6 1.7
prehensive as practically possible considering available Swing of Normalized Strength
information from literature on the expected variability of Fig. 13. Deterministic sensitivity analyses results for strength variation.
these parameters. Table 2 summarizes the statistical prop-
erties of the selected RVs in the present study. The data
is based on published literature in most part [28,29] and retrotted cases. For the strength, all results are normal-
engineering judgment where data are not available in the ized with respect to the mean strength value of the as-built
literature. It is to be noted that all variables in Table 2 case. Accordingly, one infers a 50% increase of the mean
are dened previously in the text. strength due to the FRP jacket. For the ultimate drift ratio,
no normalization is introduced where the mean is low
5.2. Tornado diagram results (0.6%) for the as-built case compared to the much higher
mean (3.4%) for the FRP-retrotted case.
Figs. 13 and 14 show the results of the sensitivity studies The following remarks summarize observations and
for the strength and ultimate drift ratios, respectively. Each implications of the tornado diagrams:
gure illustrates results for both the as-built and the FRP-
(1) For the as-built column, uncertainty in the strength is
more sensitive to uncertainty in axial load, lap splice
Table 2 length, plastic hinge length, and concrete compressive
Parameters used in deterministic sensitivity analysis
strength. Changes in the other variables have much
Parameters Mean Coecient of variation smaller eect on the column strength.
Gfc (N/mm) 59 0.20 (2) For the FRP-retrotted column, uncertainty in the
Gft (N/mm) 0.24 0.20 strength is more sensitive to uncertainty in lap splice
fc (MPa) 45 0.10
length, stiness of the FRP jacket, concrete dilatation
ft (MPa) 2.35 0.20
fy (MPa) 462 0.05 strain in the lap splice region, axial load, and concrete
c (mm) 38 0.20 compressive strength. Changes in the other variables
Ls (mm) 381 0.15 have much smaller eect on the column strength.
Lp (mm) 345 0.20 (3) For the as-built column, uncertainty in the ultimate
s (mm) 127 0.10
drift ratio is more sensitive to uncertainty in plastic
edl (mm) 0.00125 0.10
Pa (kN) 712 0.25 hinge length, concrete compressive strength, axial
Ef (MPa) 48,000 0.15 load, concrete cover, and lap splice length. Changes
erup 0.011 0.15 in the other variables have much smaller eect on
A 3 0.33 the column drift capacity.
B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276 275

It should be noted that the tornado analyses results pre-


Lp
sented above do not reect possible correlations among
fc '
Pa
uncertain variables. However, considering correlations
c may aect sensitivity results. A more accurate sensitivity
analysis method using, e.g. mean value rst order second
Parameters

Ls
s moment method (MVFOSM), can overcome this short-
ft coming if combined with the methodology presented in this
dl study [27].
Gf t
fy
Gf c No Retrofit 6. Concluding remarks
0.5 0.6 0.7 0.8
Swing of Drift Ratio (%)
This paper presents a sound computational model for
decient and FRP-retrotted RC columns having conn-
Ef ing steel and lap splice deciencies. A bond stress model
dl is employed to account for variable conning stress distri-
fy bution in the compression along with an active conned
Ls
fc'
concrete stressstrain model. Splice presence within the
Parameters

Lp plastic hinge region is considered by using an eective steel


c stressstrain model. Furthermore, a deterministic sensitiv-
s
ft ity analysis is conducted using a wide range of variables
A that are of practical importance for the developed compu-
rup
tational model. From the discussion presented above, the
Pa
Gfc
following can be inferred:
Gft With FRP Retrofit

1 2 3 4 5 6
(1) The novel computational model, which employs a
Swing of Drift Ratio (%) variable connement model by taking into account
the non-uniform connement distribution in the com-
Fig. 14. Deterministic sensitivity analyses results for drift ratio variation. pression zone was employed successfully for the rst
time in the literature. In this way, it was possible to
(4) For the FRP-retrotted column, uncertainty in the trace lateral strains that can characterize the onset
ultimate drift ratio is more sensitive to uncertainty of rupture of the FRP lamina during nonlinear anal-
in stiness of the FRP jacket, concrete dilatation ysis. Hence, the developed model can be regarded as a
strain in the lap splice region at which bond failure pseudo three-dimensional ber-discretized frame ele-
initiates, yield strength of the longitudinal reinforce- ment approach.
ment, lap splice length, concrete compressive strength (2) The developed computational model is capable of
and the plastic hinge length. Changes in the other tracing the load-deformation response of columns
variables have much smaller eect on the column having deciencies in splices and conning steel with
drift capacity. sucient accuracy. Furthermore, it is observed that
(5) Although the bounds of RVs are selected to be sym- the presence of FRP retrot within the plastic hinge
metric around the mean (2 standard deviation), region can be modeled accurately using the proposed
tornado-diagram results for EDPs are skewed around models.
the mean. The use of upper bound values for param- (3) The deterministic sensitivity analysis results reveal
eters such as plastic hinge length, concrete uniaxial that for columns with lap splice deciencies, plastic
compressive and tensile strengths are found to have hinge length, splice length, concrete strength and
negligible inuence on the strength of the as-built axial load level are the most important random vari-
decient columns compared to mean value analy- ables that need to be considered in reliability studies.
sis. On the other hand, the use of lower bound values On the other hand, for columns with FRP retrot,
for parameters such as splice length and yield stiness of the jacket and dilatation strain of concrete
strength of longitudinal reinforcement has negligible at onset of bond failure are found to be important
eect on the ultimate drift ratios for FRP-retrotted sources of uncertainty.
columns. This latter observation states that for a
successful FRP-retrot design, uncertainty in these The extensions of variable conned concrete model and
two parameters no longer has a detrimental eect its validation with experimental results for cyclic loading
on the deformation capacity, since the FRP con- situations is needed in future studies. In this way, it will
nement suppresses the splice deciency to a great be possible to simulate the demand and response more real-
extent. istically by conducting nonlinear dynamic analysis together
276 B. Binici, K.M. Mosalam / Composites: Part B 38 (2007) 265276

with MVFOSM, which will help to establish performance [14] Spoelstra MR, Monti G. FRP-conned concrete model. ASCE J
based design procedures for FRP retrots within a simpli- Compos Constr 1999;3(3):14350.
[15] Spacone E, Ciampi V, Filippou FC. Mixed formulation of nonlinear
ed probabilistic framework. beam nite element. Comput Struct 1996;58(1):7188.
[16] Mander JB, Priestley MJN, Park R. Theoretical stress-strain model
References for conned concrete. ASCE J Struct Eng 1988;114(8):180426.
[17] Pramono E, Willam K. Fracture-energy based plasticity formulation
[1] Xiao Y, Wu H. Compressive behavior of concrete conned by carbon of plain concrete. ASCE J Eng Mech 1985;115(8):1183204.
ber composite jackets. ASCE J Mater Civ Eng 2000;12(2):13946. [18] Lee T-H, Mosalam KM. Probabilistic ber element modeling of
[2] Harries KA, Kharel G. Experimental investigation of the behavior of reinforced concrete structures. Comput Struct 2004;82(27):228599.
variably conned concrete. Cem Concr Res 2003;33:87380. [19] Coleman J, Spacone E. Localization issues in nonlinear frame
[3] Lam L, Teng JG. Design-oriented stress-strain model for FRP elements. In: Shing B, Tanabe T, editors. Modeling of inelastic
conned concrete. Constr Build Mater 2003;17:47189. behavior of RC structures under seismic loads; 2001. p. 40320.
[4] Samaan M, Mirmiran A, Shahawy M. Model of conned concrete by [20] Clarke MJ, Hancock GJ. A study of incremental iterative strategies
ber composites. ASCE J Struct Eng 1998;124(9):102531. for nonlinear analysis. Int J Numer Methods Eng 1990;29:136591.
[5] Sheikh SA, Yao G. Seismic behavior of concrete columns conned [21] Maekawa K, Pimanmas A, Okamura H. Nonlinear mechanics of
with steel and ber-reinforced polymers. ACI Struct J reinforced concrete. Spon Press; 2003, p. 721.
2002;99(1):7280. [22] Lu Y, Gu X, Guan J. Probabilistic drift limits and performance
[6] Iabucci RD, Sheikh SA, Bayrak O. Retrot of square concrete evaluation of reinforced concrete columns. ASCE J Struct Eng
columns with carbon ber-reinforced polymer for seismic resistance. 2005;131(6):96678.
ACI Struct J 2003;100(6):78594. [23] Nakamura H, Higai T, Localization issues in nonlinear frame
[7] Seible F, Priestley MJN, Hegemier GA, Innamorato D. Seismic elements. In: Shing B, Tanabe T, editors. Modeling of inelastic
retrot of RC columns with continuous carbon ber jackets. ASCE J behavior of RC structures under seismic loads; 2001. p. 47187.
Compos Constr 1997;1(2):5262. [24] Clemen RT. Making hard decisions: an introduction to decision
[8] Saadatmanesh H, Ehsani MR, Jin L. Seismic strengthening of analysis. 2nd ed. Belmont, CA: Duxbury; 1996, p. 688.
circular bridg pier models with ber composites. ACI Struct J [25] Porter KA, Beck JL, Shaikhutdinov RV. Sensitivity of building loss
1996;93(6):63947. estimates to major uncertain variables. Earthquake Spectra
[9] Xiao Y, Ma R. Seismic retrot of RC circular columns using 2002;18(4):71943.
prefabricated composite jacketing. ASCE J Struct Eng [26] Lee T-H, Mosalam KM. Sensitivity of seismic demand of a reinforced
1997;123(10):135764. concrete shear-wall building. Proceedings of the Ninth International
[10] Lam L, Teng JG. Strength models for ber-reinforced plastic- Conference on Applications of Statistics and Probability in Civil
conned concrete. ASCE J Struct Eng 2002;128(5):61223. Engineering, ICASP9, San Francisco; July 69, 2003.
[11] Teng JG, Chen JF, Smith ST, Lam L. FRP strenghened RC [27] Lee T-H, Mosalam KM. Seismic demand sensitivity of reinforced
structures. New York: Wiley; 2002, p. 202. concrete shear-wall building using FOSM method. Earthquake Eng
[12] Binici B. An analytical model for stress-strain behavior of conned Struct Dyn; 2005;34(14):17191736.
concrete. Eng Struct 2005;27(7):104051. [28] Mirza SA, MacGregor JG. Variability of mechanical properties of
[13] Kazunori F, Mindness S, Xu H. Analytical model for concrete reinforcing bars. ASCE J Struct Div 1979;105(ST5):92137.
conned with ber reinforced polymer concrete. ASCE J Compos [29] Mirza SA, Hatzinikolas M, MacGregor JG. Statistical description of
Constr 2004;8(4):34151. strength of concrete. ASCE J Struct Div 1979;105(ST6):102137.

You might also like