You are on page 1of 11

Engineering Structures 182 (2019) 279–289

Contents lists available at ScienceDirect

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

Probabilistic calibration for development length models of deformed T


reinforcing bar
Bo Yua, Ruikai Tanga,b, Bing Lic,

a
Key Laboratory of Engineering Disaster Prevention and Structural Safety of China Ministry of Education, Guangxi Key Laboratory of Disaster Prevention and Engineering
Safety, School of Civil Engineering and Architecture, Guangxi University, Nanning 530004, China
b
Guangxi Road and Bridge Engineering Group Co., Ltd., Nanning 530011, China
c
School of Civil and Environmental Engineering, Nanyang Technological University, Singapore 639798, Singapore

ARTICLE INFO ABSTRACT

Keywords: In order to provide a scientific basis for choosing the appropriate development length models, a probabilistic
Deformed reinforcing bars calibration method was developed to comprehensively evaluate the accuracy and applicability of seven typical
Concrete deterministic development length models for deformed reinforcing bars (rebar) in normal and high-strength
Development length concrete. The influences of important factors (including compressive strength of concrete, concrete cover
Probabilistic calibration
thickness, rebar diameter, tensile stress of reinforcement bar, residual tensile strength of cracked concrete,
Confidence level
Confidence interval
friction on bearing surface and effective bearing angle) on development length were investigated based on the
partly cracked thick-walled cylinder model. Then a probabilistic development length model involving both
aleatory and epistemic uncertainties was proposed based on the Bayesian theory and the Markov Chain Monte
Carlo (MCMC) method. Meanwhile, key probabilistic characteristics of development length were presented and a
probabilistic method was suggested to calibrate available deterministic development length models based on the
confidence interval and the confidence level. Finally, the accuracy and applicability of available deterministic
development length models under different conditions were calibrated comprehensively.

1. Introduction length model of rebar based on the experimental data of 171 beam
specimens, which was adopted as the basis for determining the design
Adequate development length of deformed reinforcing bars (rebar) development length in the ACI 408R-03 [6]; Different from the em-
allows working stresses to be transferred from the rebar to surrounding pirical relationship developed by Orangun et al. [3], it adopts f c 1/4 in
concrete effectively. If the development length insufficient, the rebar place of f c 1/2 to represent the contribution of concrete strength to
would slip with respect to the surrounding concrete which may not only development length, and it also incorporates the influence of the ratio
deteriorate the flexural capacity, shear resistance and deformability of of maximum to minimum concrete cover thickness. Recently, Sajedi
reinforced concrete (RC) structural members [1], but also change the and Huang [2] proposed a development length model by means of
failure mode from ductile failure to brittle failure without warning [2]. explanatory functions considering the influences of effective beam
Therefore, it is essential to accurately evaluate the development length width, relative rib area and distribution of bond stress over the rebar
of rebar in concrete for the analysis and design of RC structural com- surface. It should be noted that the above development length models
ponents. are empirical in nature, since they were developed based on the re-
Up to now, various development length models [3–9] for rebar in gression analysis of different experimental data and they cannot ana-
concrete have been proposed based on different experimental databases lytically explore the mechanisms that different factors influence de-
or model assumptions. Among them, Orangun et al. [3] established an velopment length. In order to overcome this disadvantage, Esfahani and
empirical relationship between the development length and concrete Kianoush [7] proposed an analytical development length model of
compressive strength, concrete cover thickness and rebar diameter rebar based on the partly cracked thick-walled cylinder model [8],
based on the experimental data of 62 beam specimens, which serves as which takes into account the influences of concrete compressive
the basis for determining the design development length in the ACI 318- strength, concrete cover thickness, rebar diameter and distribution of
14 [4]. Subsequently, Zuo and Darwin [5] proposed a development bond stress over the rebar surface. Furthermore, Wang [9] developed an


Corresponding author.
E-mail addresses: gxuyubo@gxu.edu.cn (B. Yu), tangruikai@mail.gxu.cn (R. Tang), CBLi@ntu.edu.sg (B. Li).

https://doi.org/10.1016/j.engstruct.2018.12.047
Received 1 March 2018; Received in revised form 11 November 2018; Accepted 14 December 2018
Available online 29 December 2018
0141-0296/ © 2018 Elsevier Ltd. All rights reserved.
B. Yu et al. Engineering Structures 182 (2019) 279–289

Nomenclature to rebar diameter)


Rlc/ d calculated relative development length of deterministic
Ab cross-sectional area of rebar models
cb clear bottom cover thickness of rebar RlP/ d probabilistic relative development length
cmin minimum cover thickness of rebar RlS/ d specified calculated relative development length
csi 1/2 clear spacing of rebars Rlt/ d tested relative development length
cso clear side cover thickness of rebar Rl(1/ d )
relative development length with specified confidence
Cl / d confidence level of specified calculated relative develop- level
ment length RlL(1
/d
)
lower confidence limit of probabilistic relative develop-
d diameter of rebar ment length with specified confidence level
fc compressive strength of concrete based on 150 × 300 mm RlU(1
/d
)
upper confidence limit of probabilistic relative develop-
cylinders ment length with specified confidence level
fct tensile strength of concrete significant level
fs tensile stress of rebar effective bearing angle (angle between bearing surface
fy yield strength of rebar and bar axis)
f (Rl / d ) probability density function of relative development vector of probabilistic model parameters
length 1, 2 probabilistic model parameters related to aleatory un-
F resultant force of radial pressure and shear stress certainties
F (Rl / d ) cumulative distribution function of relative development 3 probabilistic model parameter related to epistemic un-
length certainties
F 1 (·) inverse function of cumulative distribution function of shape parameter of Gamma distribution
relative development length scale parameter of Gamma distribution
l development length of rebar in concrete µf coefficient of friction between bearing surface and sur-
p bearing pressure from bearing surface to surrounding rounding concrete
concrete µl / d mean value of probabilistic relative development length
pr radial pressure perpendicular to bar surface µ1 mean value of probabilistic model parameter 1
pr,max maximum radial pressure µ 2 mean value of probabilistic model parameter 2
P (Rlt/ d | ) likelihood function of probabilistic model parameters µ3 mean value of probabilistic model parameter 3
P( ) probability density function of prior distribution of prob- standard normal random variable
abilistic model parameters l/ d standard deviation of probabilistic relative development
P ( |Rlt/ d ) probability density function of posterior distribution of length
probabilistic model parameters standard deviation of random normal error
R0 inner radius of partly cracked concrete cylinder hoop tensile stress of surrounding concrete
Rc outer radius of partly cracked concrete cylinder shear stress parallel to bar surface
R ci critical depth of radial splitting cracks u bond strength between rebar and concrete
Ri depth of radial splitting cracks
Rl / d relative development length (ratio of development length

analytical model of development length in terms of the function of condition of force equilibrium. It should be noted that both empirical
concrete compressive strength, concrete cover thickness, rebar dia- and analytical models would be classified as median models, since their
meter, effective beam width and relative rib area based on the as- predictive lines usually lie at the center of the experimental points
sumption of fictitious rib-face over the rebar surface. Unlike empirical [10,11]. However, it is necessary to determine the development length
models, analytical models can rationally explore the influential me- with predefined confidence level instead of mean level to prevent the
chanisms of different factors on development length, since they are RC structural members from brittle bond failure. Hence, various design
developed based on various simplified mechanical models including the models of development length have been proposed in different design

Fig. 1. Variation of accuracy of deterministic development length models.

280
B. Yu et al. Engineering Structures 182 (2019) 279–289

codes [4,6,12–16], by replacing the tensile stress with the yield strength Chain Monte Carlo (MCMC) method. Subsequently, a probabilistic
of rebar and introducing the strength-reduction factors for bond and the method was presented to calibrate available deterministic development
correction factors that relate to steel grade, casting position, geometric length models based on the confidence interval and the confidence
size and surface properties of rebar, et al. However, the confidence level. Finally, the accuracy and applicability of available development
levels of various design models are implicit and it is desirable to cali- length models under different conditions were evaluated based on the
brate the confidence levels of various design models under different experimental data of 236 specimens.
conditions comprehensively.
The accuracy and applicability of available development length 2. Deterministic and probabilistic development length models
models (including mean and design models) often change significantly
under different conditions of concrete compressive strength, concrete 2.1. Deterministic development length model
cover thickness, rebar diameter and yield strength of rebar, since they
were developed based on different experimental databases or model When the reinforcement bar exhibits a rigid displacement and the
assumptions. Three deterministic development length models (e.g., Zuo bond-slip law is invariant in space, it can be assumed that bond stress
model [5], Wang model [9] and Sajedi model [2]) were selected to distributes uniformly along the development length [2,9]. Based on the
illustrate this problem, as shown in Fig. 1. It is clear that Zuo model and condition of force equilibrium between the tensile force (which equals
Wang model usually overestimate the relative development length, to fs Ab ) along the axis of the rebar and the bond force (which equals to
while Sajedi model often underestimates the relative development u dl ) parallel to the rebar surface, the relative development length
length. Therefore, it is necessary to calibrate the accuracy and applic- (i.e., ratio of development length to rebar diameter) of rebar in concrete
ability of available development length models comprehensively. Tra- can be obtained as:
ditional calibration methods usually evaluate the accuracy of devel-
l f
opment length models based on the modeling error [5,9,17,18], which Rl / d = = 0.25 s
d (1)
provides an effective approach to calibrate the accuracy of different u

development length models at the average level in comparison with where Rl / d is the relative development length; l is the development
experimental data. However, traditional calibration methods are de- length; d is the rebar diameter; fs is the tensile stress of rebar; u is the
terministic and cannot take into account the aleatory uncertainties that bond strength between rebar and concrete; Ab =0.25 d 2 is the cross-
arise from the material properties, geometric size and reinforcement sectional area of rebar.
arrangement of concrete specimens as well as the epistemic un- According to Eq. (1), bond strength u is a key parameter to calculate
certainties that arise from the introduction of empirical strength-re- Rl / d when fs is predefined. As illustrated in Fig. 2a, bond stress mainly
duction factors or correction factors, application of various model as- consists of the mechanical interlock p radiating perpendicularly out-
sumptions and omission of some influential factors. As a result, it is wards from the bearing surface and the friction µ f p acting parallel
difficult to calibrate the confidence levels of development length along the bearing surface when rebar moves with respect to the sur-
models of rebar in concrete based the traditional deterministic cali- rounding concrete. Concrete paste in front the rib was crushed with a
bration method. Recently, Bilotta et al. [11] proposed a calibration bearing surface inclined of with respect to the axis of the reinforce-
method for the design formula of the maximum axial strain in FRP at ment bar [20–22], which represents the influence of concrete confine-
the onset of intermediate debonding failure based on the design-by- ment and rib height of rebar. The resultant force F induced by p and
testing approach that originally stated in EN-1990 [19]. But it cannot µ f p can respectively be decomposed into the radial pressure pr and the
take into account the aleatory and epistemic uncertainties explicitly. shear stress in the longitudinal direction as:
Hence, it is desirable to develop a probabilistic calibration method to pr = p cos µf p sin ; = p sin + µf p cos (2)
evaluate the accuracy and confidence levels of available development
length models for rebar in concrete by taking into account both the where µ f is the coefficient of friction between bearing surface and
aleatory and epistemic uncertainties. surrounding concrete, which generally varies from 0.4 to 0.6 [22–24];
The main objective of this study is to provide a scientific basis for is the effective bearing angle with respect to the z axis, whose typical
choosing the appropriate development length models based on the range is within 10-40° (0.17–0.70 rad) [20,21,25].
probabilistic calibration of deterministic models. Influential mechan- If the hoop tensile stress induced by the radial pressure pr exceeds
isms of different important factors on development length was in- the tensile strength of concrete fct , radial splitting cracks often initiate
vestigated based on the derived analytical model. Then a probabilistic near the rebar surface and eventually propagate throughout concrete
development length model involving both aleatory and epistemic un- cover. As a result, splitting bond failure occurs when the critical
certainties was developed based on the Bayesian theory and the Markov cracking depth R ci is reached. Concurrently, the radial pressure pr and
the shear stress achieve their maximum values, which are referred to

y y

pr F cso
2csi
Rci
f p Pullout force
o pr p o z x

x
z R0 d 2 pr
Ri
Accumulation zone cb
Rc cmin d 2

(b) Partly cracked thick -walled cylinder


(a) Bonding action
model
Fig. 2. Bonding mechanism of rebar in concrete.

281
B. Yu et al. Engineering Structures 182 (2019) 279–289

as the maximum radial pressure pr,max and the bond strength u re- for different concrete specimens due to the aleatory uncertainties
spectively. According to Eq. (2), the relationship between u and pr,max arising from the material properties, geometric size and reinforcement
can be expressed as: arrangement as well as the epistemic uncertainties arising from the
µf + tan introduction of model assumptions (such as the assumptions of uniform
u = pr,max distribution of bond stress, smeared cracking of cracked concrete and
1 µf tan (3) approximation of the tensile strength of concrete), selection of em-
According to Eq. (3), pr,max is one of the key parameters to determine pirical model variables (such as the effective value of clear spacing of
rebar) and omission of other influential factors (such as rib spacing and
u. Based on the assumption of identical bonding condition along the
development length [8], the partly cracked thick-walled cylinder model casting position of rebar) during the model derivation. In order to take
considering the residual tensile strength of cracked concrete can be into account the aleatory uncertainties, deterministic model parameters
established by transforming the spatial problem as shown in Fig. 2a into µ f and in Eq. (6) are replaced with probabilistic model parameters 1
the axisymmetric plane strain problem as shown in Fig. 2b. Subse- and 2 respectively. Moreover, a probabilistic model parameter 3 was
quently, based on the assumption of perfectly elastic behavior for the introduced to describe the epistemic uncertainties. Hence, by introdu-
outer ring of concrete cylinder and the assumption of smeared cracking cing the probabilistic model parameters into the deterministic model
for the inner ring of concrete cylinder, pr can be obtained as [26]: described by Eq. (6), the probabilistic model of development length for
rebar in concrete that involves both aleatory and epistemic un-
Rc2 Ri2 1 (R0 / Ri)1.4 Ri certainties can be expressed as:
pr = ( + ) fct
Rc2 + Ri2 1.4 R0 (4)
0.25 fs 1 1 tan 2
RlP/ d = · + +
where Ri is the depth of radial splitting cracks; R c = cmin + d/2 is the [0.75(2
cmin
+ 1) 0.78(2
cmin
+ 1) 0.4] fct 1 + tan 2
3
d d
outer radius of the partly cracked concrete cylinder; R 0 = d/2 is
(7)
the inner radius of the partly cracked concrete cylinder;
cmin =min(cb, cso, csi + 6.4 mm) is the minimum cover thickness of where RlP/ d is the probabilistic relative development length;
rebar [6]; cb and cso are the clear bottom and side cover thickness of = [ 1, 2 , 3]T is vector of probabilistic model parameters; is the
rebar respectively; csi is 1/2 clear spacing of rebars; fct = 0.33 f c is the random normal error of probabilistic model; is a standard normal
random variable; is the standard deviation of random normal error.
tensile strength of concrete [27]; f c is the compressive strength of
concrete based on 150×300 mm cylinders. Statistical characteristics of the prior distribution of can be de-
According to Eq. (4), pr only depends on Ri when the other basic termined based on existing experimental database or engineering ex-
parameters are predefined. Particularly, splitting bond failure occurs perience. Subsequently, the posterior distribution of can be de-
when Ri equals to R ci , and pr concurrently achieves pr,max . Thus, it is termined according to the Bayesian theory [35] as:
essential to determine R ci before calculating pr,max . Setting the first
P ( ) P (Rlt/ d | )
derivative of Eq. (4) with respect to Ri as zero, an equation expressed as P ( |Rlt/ d ) = +
the function of Ri , R c , R 0 and fct can be obtained. Based on the nu- P ( ) P (Rlt/ d | )d (8)
merical solution of this equation and the regression analysis of ex-
where is a vector of tested development lengths; P (
Rlt/ d is the |Rlt/ d )
perimental data of 400 concrete specimens with splitting failure as
probability density function (PDF) of the posterior distribution of ;
listed in Table 1, R ci = 0.81Rc can be determined. Then, replacing Ri in
P (Rlt/ d | ) is the likelihood function of ; P ( ) is the PDF of the prior
Eq. (4) with R ci , pr,max can be obtained as: +
distribution of . P ( ) P (Rlt/ d | )d is the normalization factor. It is
Rc Rc 0.4] difficult to obtain the analytical expression of P ( |Rlt/ d ) since the cal-
pr,max = [0.75 0.78( ) fct
R0 R0 (5) culation of
+
P ( ) P (Rlt/ d | )d involves multidimensional integral.
Substituting Eqs. (3) and (5) into Eq. (1), the deterministic model In order to overcome this difficulty, the Markov Chain Monte Carlo
for relative development length can be expressed as: (MCMC) method [36] was adopted to determine the statistical char-
acteristics of P ( |Rlt/ d ) . The flowchart to determine the statistical
l 0.25fs 1 µf tan characteristics of posterior distribution of based on the MCMC method
Rl / d = = ·
is shown in Fig. 3.
c c
d [0.75(2 min
d
+ 1) 0.78(2 min
d
+ 1) 0.4 ] fct µ f + tan (6)
A sequence of samples for can be generated by the MCMC method
based on the experimental data of 236 concrete specimens (as listed in
2.2. Probabilistic development length model Table 2) collected from different literatures [32–34,37] based on sev-
eral selection criteria (such as beam specimen, without transverse
It should be noted that basic parameters including f c , cmin , d , l and stirrups, dominant by splitting failure, straight anchored deformed re-
fs for each concrete specimen can all be measured during the experi- inforcement bar, without coating on the surface and casted in bottom
ment. However, model parameters µ f and often change significantly position), which cover the typical ranges of basic variables (as shown in

Table 1
Database for determination of critical cracking depth.
References Type of specimen Ranges of basic parameters

f c (MPa) cso (mm) csi (mm) cb (mm) d (mm)

Hadje-Ghaffari et al. [28] (159 specimens) Beam-end specimen 33–89 102–109 N.A.# 17–107 16–36
Darwin and Graham [29] (86 specimens) Beam-end specimen 31–41 102 N.A.# 48–82 25
Mendis and French [30] (15 specimens) Beam-end specimen 39–96 135–143 N.A.# 38–64 19–36
Miller et al. [31] (35 specimens) Beam-end specimen 33–34 105 N.A.# 35–51 19
Esfahani and Rangan [32] (22 specimens) Splice specimen 66–98 24–82 11–73 24–65 19–23
Harajli et al. [33] (8 specimens) Splice specimen 53–57 18–50 18–50 18–50 16–32
fib TG4.5 splice test database [34] (75 specimens) Splice specimen 52–108 19–77 13–103 13–78 10–36

#
Note: only one developed rebar is contained and csi is not applicable for beam-end specimens.

282
B. Yu et al. Engineering Structures 182 (2019) 279–289

coefficients of correlation between probabilistic model parameters are


all very small. Moreover, the standard deviation of random normal
error is 0.01. The prior distribution of probabilistic model parameters
can be updated progressively when the new experimental data were
available, based on the Bayesian theory and the Markov Chain Monte
Carlo method.

3. Probabilistic calibration of deterministic development length


models

3.1. Development length with specified confidence level

Based on the assumption that probabilistic model parameters are


independent of random normal error , according to the probabilistic
development length model defined in Eq. (7) and statistical character-
istics of the posterior distribution of as listed in Table 4, mean value
µl / d and standard deviation l / d of probabilistic relative development
length can be determined by:
1 µ 1 tan µ 2 2R P
1 3 3 l/d
µl / d = K µ 1 + tan µ 2
+µ3+ 2 i=1 j=1
cov( i, j)
i j
S
= K S 13 [S22 + sec2 2 (C11 + 2C12) + S3 C22] + µ
2
3 (9)

3 3 RlP/ d RlP/ d 2
l/d = i=1 j =1
cov( i , j) +
i j

sec4 µ 2 C11 + 2 sec2 µ 2 S3 C12 + S32 C22 sec2 µ 2 C13 + S2 C23 2


= K2 S24
2K S23
+ C33 +

(10)
where µ 1, µ 2 , and µ 3 are mean values of 1, 2 and 3 respect-
Fig. 3. Flowchart to determine the statistical characteristics of posterior dis- ively; Cij = cov( i , j ) is covariance coefficient of i and j ;
tribution of . S1 = 1 µ 1 tan µ 2 , S2 = µ 1 + tan µ 2 , S3 = (1 + µ 1)2sec2 µ 2 and
K = [5.97(
cmin
+ 0.5) 2.35(
cmin
+ 0.5) 0.4 ] 1 fs .
d d fct
Fig. 4). Subsequently, the empirical distribution of can be determined One of the advantages of the proposed probabilistic development
by the Kolmogorov-Smirnov (K-S) test with the generated samples. The length model is that it can provide an effective way to calibrate the
K-S test is a nonparametric test of the null hypothesis that the popu- accuracy of available deterministic development length models. The
lation of cumulative distribution function (CDF) of the generated accuracy of five deterministic mean models for development length
samples is equal to the hypothesized CDF. The statistic of the K-S test is reported by Refs. [2,3,5,7,9] are calibrated based on the proposed
defined as the maximum absolute difference between the empirical CDF probabilistic development length model. Descriptions of the above five
calculated from the generated samples and the hypothesized CDF. If the deterministic mean models are summarized in Table 5. Based on the
statistic of the K-S test is greater than the critical value, the null hy- statistical characteristics of the posterior distribution of and 236
pothesis is rejected. Otherwise, the null hypothesis is accepted. The concrete specimens as listed in Table 2, a large number of samples of
statistic of the K-S test are summarized in Table 3, wherein × indicates Rl / d can be generated by means of the Monte Carlo simulation (MCS)
that the generated samples do not coincide with the physical meaning with Eq. (7). According to the results of K-S test, it was observed that
of the assumed empirical distribution. When the sample size is 1000 Rl / d does not reject to obey the Gamma distribution. Hence, the prob-
and the significant level is 0.05, the critical value of the K-S test is ability density function (PDF) f (Rl / d ) and the cumulative distribution
0.0430. As shown in Table 3, both 1 and 2 do not reject to obey the function (CDF) F (Rl / d ) of relative development length can be described
Lognormal distribution and the Gamma distribution and 3 does not as:
reject to obey the Normal distribution. According to the statistics of the
( Rl / d) 1exp( Rl / d ) ( Rl / d ) 1exp( Rl / d)
K-S test, it is assumed that both 1 and 2 are the Lognormal random f (Rl / d ) = = +
variables and 3 is a Normal random variable. ( )
0
exp( u) u 1du
(11)
Statistical characteristics of the posterior distribution of probabil-
istic model parameters are listed in Table 4. It is observed that both ( , Rl / d)
Rl / d
exp( u) u 1du
0
mean values of 1 and 2 are within their typical ranges. Meanwhile, F (Rl / d ) = = +
( ) exp( u) u 1d u
(12)
0

Table 2
Database for determination of statistical characteristics for posterior distribution of .
References Type of specimen Ranges of basic parameters

f c (MPa) cso (mm) csi (mm) cb (mm) d (mm) l (mm) fs (mm)

Esfahani and Rangan [32] (22 specimens) Splice specimen 66–98 24–82 11–73 24–65 19–23 200–550 199–503
Harajli et al. [33] (10 specimens) Splice specimen 33–57 18–50 0–34 18–50 16–32 80–160 83–253
fib TG4.5 splice test database [34] (199 specimens) Splice specimen 30–109 13–167 0–119 13–140 10–36 140–1676 165–687
Hassan et al. [37] (5 specimens) Splice specimen 41–78 37–38 30–45 76–82 29–64 813–5969 407–758

283
B. Yu et al. Engineering Structures 182 (2019) 279–289

Fig. 4. Frequency distributions of basic variables for 236 sets of experimental data.

Table 3 Rl(1/ d )
=F 1 (1 ) (14)
Statistics of K-S test of probabilistic model parameters.
where is the characteristic value of development length with
Rl(1/ d )
Model parameters Distribution type specified confidence level 1 ; F 1 (·) is the inverse cumulative dis-
Lognormal Gamma Normal Weibull
tribution function of relative development length. According to Eqs.
(12) and (14), the characteristic value of development length with
1 0.0151 0.0209 0.0571 0.0638 specified confidence level 1 can be obtained. However, Eq. (14) can
2 0.0141 0.0259 0.0645 0.0634 only be solved numerically, since it involves the gamma distribution.
3 × × 0.0193 ×

3.2. Calibration based on confidence interval


where and are shape and scale parameters of the Gamma dis-
According to F (Rl / d ) defined in Eq. (12), the lower confidence limit
tribution respectively, which can be obtained from:
RlL(1
/d
)
and the upper confidence limit RlU(1 /d
)
of the confidence in-
2
µl / d l/d terval of relative development length with specified confidence level
=( )2 ; =
l/d µl / d (13) 1 can be expressed as:

Based on the CDF of relative development length as defined in Eq. RlL(1 )


=F 1( ) ; RlU(1 )
=F 1 (1 )
/d
2 /d
2 (15)
(12), the characteristic value of development length with specified
confidence level 1 can be obtained by: where F 1 (·)
is the inverse function of F (Rl / d ) ; is the significant level.

Table 4
Statistical characteristics of posterior distribution of probabilistic model parameters.
Model parameters Mean value Standard deviation Distribution type Coefficients of correlation

1 2 3

1 0.4528 0.1415 Lognormal 1.0000 −0.0053 0.0073


2 (rad) 0.3066 0.0983 Lognormal −0.0053 1.0000 −0.0148
3 0.8986 0.9097 Normal 0.0073 −0.0148 1.0000

Note: 1 rad = 57°.

284
B. Yu et al. Engineering Structures 182 (2019) 279–289

Table 5
Descriptions of five deterministic mean models for development length.
References Models Descriptions

Orangun et al. [3] Orangun model fs


50

Rl / d =
l
=
4 fc
(BS units) , where cmin /d 2.5 , cs = min(cso, csi, cb)
c
d 1.2 + 3 min
d
Zuo and Darwin [5] Zuo model fs c
2350(0.1 max + 0.9)
l fc 1/4 cmin
cmax
Rl / d = = c (BS units) , where (cmin/d + 0.5)(0.1 + 0.9) 4.0 , cmax = max(cs , cb) ,
d 76.1(c min / d + 0.5)(0.1 max + 0.9) cmin
cmin
cmin = max(cs , cb) , cs = min(cso, csi + 0.25 in.)
Esfahani and Kianoush Esfahani model l fs
Rl / d = = , where cmin = min(cb, cso, csi + d/2) , cmed = median(cb, cso, csi + d/2) ,
[7] d c
10.8 min
/ d + 0.5 c
(0.88 + 0.12 med )
1+1/M
fc
c min / d + 3.6 cmin 1.85 + 0.024 M
3fc
M = cosh(0.0022l )
d
Wang [9] Wang model l 5.55fs
Rl / d = = 3c + be
, where cmin = min(cb, cso, csi) is adopted for bars being developed; cmin = min(cb, cso, csi + d) is adopted
d
( min ) fc
d
for bars being spliced; be is the effective beam width
Sajedi and Huang [2] Sajedi model l 0.25fs
Rl / d = = , where cmin = min(cb, cso, csi) is adopted for bars being developed;
d c b µ + Rr 8d
exp[ 0.728 + (0.476 min + 0.119 e ) ] fc
d d 1 µRr l
cmin = min(cb, cso, csi + d) is adopted for bars being spliced; be is the effective beam width

Based on the lower and upper confidence limits defined by Eq. (15), randomness of development length. However, more than half of tested
the confidence interval of relative development length with specified data fall within the 50% confidence interval and almost all of tested
confidence level can be described as [RlL(1 /d
)
, RlU(1
/d
)
]. Taking the data fall within the 95% confidence interval, which implies that the
confidence intervals of 50% and 95% (i.e., 1 are taken as 0.50 and proposed probabilistic model not only possesses satisfied calculating
0.95 respectively) as examples, distributions of tested data and calcu- accuracy, but also provides a rational way to describe probabilistic
lated results of relative development lengths obtained by the above five characteristics of development length.
deterministic mean models (as listed in Table 5) within the confidence In order to quantitatively evaluate the accuracy of the above five
intervals are shown in Fig. 5. It can be observed that the distribution of deterministic mean models as listed in Table 5, percentages of calcu-
tested data shows considerable scattered, which indicates significant lated relative development lengths in different ranges of confidence

80 80 80
Confidence interval of 95% Confidence interval of 95% Confidence interval of 95%
Confidence interval of 50% Confidence interval of 50% Confidence interval of 50%
Tested data Tested data Tested data
60 Orangun model 60 60
Zuo model Esfahani model
l/d

l/d

40 40
l/d

40
R

20 20 20

0 0 0
0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250
Sample number Sample number Sample number
(a) Orangun model (b) Zuo model (c) Esfahani model

80 80
Confidence interval of 95% Confidence interval of 95%
Confidence interval of 50% Confidence interval of 50%
Tested data Tested data
60 Wang model 60 Sajedi model
l/d

l/d

40 40
R

20 20

0 0
0 50 100 150 200 250 0 50 100 150 200 250
Sample number Sample number
(d) Wang model (e) Sajedi model
Fig. 5. Distribution of tested data and calculated results of relative development lengths within 50% and 95% confidence intervals.

285
B. Yu et al. Engineering Structures 182 (2019) 279–289

intervals are summarized in Table 6. According to Fig. 5 and Table 6, the ACI 408R model, and are taken as 1.00 and 0.82 respectively;
the percentages of calculated relative development lengths of the Or- For the GB 50010 model, s and c are taken as 1.10 and 0.86 respec-
angun model and the Esfahani model within 50% confidence interval tively. Furthermore, the lower bounds of the ratio of development
(from RlL(50%)
/d to RlU(50%)
/d ) are 67.37% and 55.51% respectively, which length to rebar diameter (l/ d ) were not restricted in the ACI 318 model,
indicates that the above two models can provide an approximation of the CSA A23.3 model and the GB 50010 model. The lower bound of l/ d
development length at the average level. For the Zuo model and the for the ACI 318 model and the CSA A23.3 model are taken as 16 (i.e.,
Wang model, the percentages within the confidence interval from l/ d 16) and the lower bound of l/ d for the GB 50010 model is taken as
RlU(50%)
/d to RlU(95%)
/d are 51.27% and 69.07% respectively, which indicates 14 (i.e., l/ d 14 ). Confidence levels of the above seven design models
that the above two models usually overestimate the development length under different conditions of yield strength of rebar ( fy are 300, 400
at the average level. For the Sajedi model, the percentage within the and 500 MPa), ratio of cover thickness to rebar diameter
confidence interval from RlL(50%)/d to RlL(95%)
/d is 54.66%, which implies (0.5 cmin /d 3.0 ) and compressive strength of concrete
that the Sajedi model usually underestimate the development length at (30 MPa ≤ f c ≤ 80 MPa) are shown in Figs. 7–13. As illustrated in
the average level. According to above analysis, it is clear that the Fig. 7, confidence level of the ACI 318 model is not sensitive to f c and
proposed probabilistic model provides an efficient way to calibrate the fy , but it increases with the increase of cmin / d . Confidence level of the
accuracy of deterministic mean models for development length based ACI 318 model is about 50–75% when cmin / d < 1.5, while it increases to
on the confidence interval. 75–95% when cmin / d > 1.5. Furthermore, confidence level of the ACI
318 model is greater than 95% when fy = 300 MPa , cmin / d > 2 and
3.3. Calibration based on confidence level f c > 45 MPa. As depicted in Fig. 8, confidence level of the ACI 408R
model is not sensitive to f c , but it increases with the increase of cmin / d
Another advantage of the proposed probabilistic development and fy . Particularly, it is greater than 95% when fy = 500 MPa and
length model is that it makes it convenient to determine the confidence cmin / d > 1.0 , while it decreases to 53–75% when fy = 300 MPa and
levels of various design models for development length in different cmin / d < 1.0 . Furthermore, confidence level of the ACI 408R model is
design codes. Based on F (Rl / d ) defined in Eq. (12), the confidence level greater than 95% when cmin / d > 2 and f c > 45 MPa. As shown in
Cl / d of specified calculated development length RlS/d can be expressed as: Fig. 9, confidence level of the CEB 1990 model is not sensitive to fy , but
RlS/ d 1d u
it increases with the increase of f c and the decrease of cmin / d respec-
( , RlS/ d ) 0
exp( u) u tively. Confidence level of the CEB 1990 model is smaller than 50%
Cl / d = F (RlS/ d) = = +
( )
0
exp( u) u 1du
(16) when cmin / d < 1.5, while it increases to 75–95% when cmin / d > 2.0 and
f c < 60 MPa. According to Fig. 10, confidence level of the CEB 2010
Confidence levels of seven typical design models for development model increases with the decrease of f c and the increase of fy and
length adopted in different design codes [4,6,12–16] were evaluated cmin / d . Confidence level of the CEB 2010 model is greater than 95%
using the proposed probabilistic development length model. Descrip- when fy = 500 MPa and cmin / d > 1.25, while it falls within 75–95%
tions of seven typical design models for development length under when fy = 300 MPa and cmin / d > 1.25. As shown in Fig. 11, confidence
specific conditions (e.g., normal-weight concrete, uncoated and bottom- level of the CSA A23.3 model is generally greater than 90%. Further-
cast rebar and rebar diameter is smaller than 36 mm) are listed in more, it is not sensitive to f c and fy but slightly increases with the
Table 7. Two specimens with the same basic parameters (i.e., decrease of cmin / d . Confidence level of the CSA A23.3 model falls within
cmin =12.70 mm, f c = 30.68 MPa, d = 12.7 mm, fy = 344.74 MPa) 90–95% when cmin / d > 1.0 , while it is generally greater than 95% when
were collected from the literature [34] to calibrate seven different de- cmin / d < 1.0 . Moreover, it is greater than 95% when fy = 300 MPa ,
sign models. Mean value and standard deviation of relative develop-
cmin / d > 2.25 and f c > 50 MPa. As illustrated in Fig. 12, confidence
ment length determined by Eqs. (9) and (10) are 33.38 and 12.60 re-
level of the AS 3600 model increases with the decrease of fy and the
spectively. The relative development length calculated by seven design
models are 33.61, 44.77, 27.63, 35.87, 44.81, 29.00 and 23.87 re- increase of f c and cmin / d . Confidence level of the AS 3600 model falls
within 75–95% when cmin / d > 1.5 and fy = 500 MPa , while it is gen-
spectively. The distribution of calculated relative development length of
design models in the probability density function determined by Eq. erally greater than 95% when cmin / d > 1.5 and fy = 300 MPa . As shown
(11) is shown in Fig. 6. It is obviously that notable discrepancies of in Fig. 13, confidence level of the GB 50010 model is not sensitive to f c
confidence level of design models are observed even when the basic and fy , but it increases with the increase of cmin / d . Confidence level of
parameters are the same, which indicates that influences of aleatory the GB 50010 model is lower than 50% when cmin / d < 1.25, while it is
and epistemic uncertainties are significant. Hence, it is necessary to generally beyond 95% when cmin / d > 2.5.
calibrate the accuracy and applicability of development length models According to the above analysis, it is clear that the proposed
comprehensively based on the confidence level. probabilistic development length model provides an efficient approach
In order to evaluate the confidence levels of different design models, to evaluate the confidence levels of available design models for devel-
the strength-reduction factors and the correction factors are all taken as opment length under different conditions of compressive strength of
the most conservative values. For the ACI 318 model, the CEB 1990 concrete, ratio of cover thickness to rebar diameter and yield strength
model, the CEB 2010 model, the CSA A23.3 model and the AS 3600 of rebar.
model, s are taken as 1.00, 1.04, 0.90, 1.00 and 1.04 respectively; For

Table 6
Percentages of calculated relative development lengths in different ranges of confidence intervals.
Models Ranges of confidence intervals

Below RlL(95%
/d
)
RlL(50%
/d
)
RlL(95%)
/d RlL(50%
/d
)
RlU(50%)
/d RlU(50%)
/d RlU(95%)
/d Above RlU(95%)
/d

Orangun model 6.35% 13.56% 67.37% 12.29% 0.42%


Zuo model 5.39% 6.78% 30.93% 51.27% 5.08%
Esfahani model 2.97% 22.46% 55.51% 13.14% 5.93%
Wang model 0.00% 0.42% 17.80% 69.07% 12.71%
Sajedi model 5.51% 54.66% 38.98% 0.85% 0.00%

286
B. Yu et al. Engineering Structures 182 (2019) 279–289

Table 7
Descriptions of design models for development length in different design codes.
References Models Descriptions

ACI 318-2014 [4] ACI 318 model l fy


Rl / d = = 0.9 c min
s
, where f c 69 MPa , cmin /d 2.5 , l 300 mm ; cmin = min(cb, cso, csi) ; when d 19 mm, s = 0.9 ;
d + 0.5
fc
d
when d 22 mm, s = 1.0
ACI 408R-2003 [6] ACI 408R fy
57.4
model l f c 1/4 cmax
Rl / d = = , where f c , (cmin + 0.5d) 4d , l max(16d, 300 mm) ; = (0.1 + 0.9) , cmax = max(cs , cb) ,
d 1.83(c min / d + 0.5) c min
cmin = max(cs , cb) , cs = min(cso, csi + 6.4 mm)
CEB-FIP Model Code 1990 CEB 1990 l c min s f yd
Rl / d = = 0.175(1.15 0.15 ) , where 1.0 cmin/ d 3.0 , l max(10d, 100 mm) ; cmin = min(cb, cso, csi) ; when d 32
[12] model d d fc 8
( )2/3
10
mm, s = 1.0 ; when d > 32 mm, s = 100/(132 d)
CEB-FIP Model Code 2010 CEB 2010 l f yd c
Rl / d = = 0.715 , where 1.0 cmax /cmin 5.0 , 0.5 cmin /d 3.5 , 15 MPa fc 110 MPa ,
[13] model d c min
(f c
c
8) ( max )0.15
s y d cmin
l max(10d, 100 mm) , cmin = min(cb, cso, csi) , cmax = min(cso, csi) ; when d 25 mm, s = 1.0 ; when d > 25 mm, s = (25/d )0.3
CSA A23.3-2014 [14] CSA A23.3 l c min 1 fy
Rl / d = = 0.9 s( d ) , where f c 64 MPa , cmin /d 2.5 , l 300 mm ; cmin = min(cb , cso , csi ) ; when d 20 mm, = 0.8 ;
model d
fc
s

when d 25 mm, s = 1.0


AS 3600-2009 [15] AS 3600 model l c min s f yk
Rl / d = = 0.5(1.15 0.15 ) , where f c 65 MPa , 1.0 cmin/ d 3.0 , l 29d ; cmin = min(cb, cso, csi) , s = 100/(132 d)
d d
fc
GB 50010– GB 50010 l f yd
2010 [16] model
Rl / d =
d
= 0.14 s c f , where ftd 2.04 MPa , s c 0.6 , l 200 mm ; when d > 25 mm s = 1.10 , s = 1.0 for other diameters of
td
rebar; when the minimum cover thickness equals to 3d c = 0.80 , and when it equals to 5d c = 0.70 ; for other cases of minimum
cover thickness, c is determined by linear interpolation

Note: cb and cso are the clear bottom and side cover thickness of rebar respectively, csi is 1/2 the clear spacing of rebars, cb and cso are the distance measured from the
bottom and lateral surface of the concrete specimen to the center of rebar respectively, csi and csi are 1/2 and 2/3 the center to center spacing of rebars respectively,
s is a coefficient related to the diameter of reinforcing bar, y is a coefficient related to the yield strength of rebar, is a strength-reduction factor related to the load
factor and capacity-reduction factor for section in tension, c is a coefficient related to the minimum cover thickness, fy is the yield strength of rebar, fyk is the
characteristic yield strength of rebar, fyd is the design yield strength of rebar, ftd is the design tensile strength of concrete, c is a safety factor for bond strength.

0.04 4. Conclusions
= 33.38 ACI 318
= 12.60 ACI 408R
fib 1990
In order to provide a scientific basis for choosing the appropriate
0.03 fib 2010 development length models, a probabilistic calibration method was
Probability density

CSA A23.3 developed to evaluate the accuracy and applicability of seven typical
AS 3600 deterministic models based on the confidence interval and the con-
GB 50010 fidence level. The following conclusions can be drawn:
0.02
(1) Mean models of development length can only provide an approx-
imation at the average level in comparison with experimental data,
0.01 and the accuracy of different mean models change significantly
under different conditions of concrete compressive strength, cover
thickness, rebar diameter and yield strength of rebar.
0 (2) Probabilistic calibration of different design models indicates that
0 20 40 60 80 100
R the confidence level of the CSA A23.3 model is generally greater
l/d
than 90%, while the confidence levels of the other six deign models
Fig. 6. Distribution of calculated relative development length of design models are affected by compressive strength of concrete, yield strength of
in the probability density function. rebar and ratio of cover thickness to rebar diameter. Particularly,

(a) f y = 300 MPa (b) f y = 400 MPa (c) f y = 500 MPa


Fig. 7. Confidence level of the ACI 318 model.

287
B. Yu et al. Engineering Structures 182 (2019) 279–289

(a) f y = 300 MPa (b) f y = 400 MPa (c) f y = 500 MPa


Fig. 8. Confidence level of the ACI 408R model.

(a) f y = 300 MPa (b) f y = 400 MPa (c) f y = 500 MPa


Fig. 9. Confidence level of the CEB 1990 model.

(a) f y = 300 MPa (b) f y = 400 MPa (c) f y = 500 MPa


Fig. 10. Confidence level of the CEB 2010 model.

(a) f y = 300 MPa (b) f y = 400 MPa (c) f y = 500 MPa


Fig. 11. Confidence level of the CSA A23.3 model.

(a) f y = 300 MPa (b) f y = 400 MPa (c) f y = 500 MPa


Fig. 12. Confidence level of the AS 3600 model.

288
B. Yu et al. Engineering Structures 182 (2019) 279–289

f y = 300 MPa (b) f y = 400 MPa (c) f y = 500 MPa


Fig. 13. Confidence level of the GB 50010 model.

confidence levels of the ACI 408R model and the CEB 2010 [10] Monti G, Alessandri S, Santini S. Design by testing: A procedure for the statistical
model are generally greater than 95% when fy = 500 MPa and determination of capacity models. Const Build Mater 2009;23(4):1487–94.
[11] Bilotta A, Faella C, Martinelli E, Nigro E. Design by testing procedure for inter-
cmin / d > 1.25. Meanwhile, confidence level of the AS 3600 model is mediate debonding in EBR FRP strengthened RC beams. Eng Struct
generally greater than 95% when fy = 300 MPa and cmin / d > 1.5. 2013;46(1):147–54.
Moreover, confidence level of the GB 50010 model is beyond 95% [12] CEB-FIP. Model code 1990. Lausanne, Switzerland: Comite Euro International Du
Beton; 1993.
when cmin / d > 2.5. [13] CEB-FIP. fib model for concrete structures 2010. Lausanne, Switzerland:
(3) The proposed probabilistic calibration method provides a scientific International Federation for Structural Concrete; 2013.
basis for choosing the appropriate development length model. [14] Canadian Standards Association. Design of concrete structures (CSA A23.3-14).
Mississauga, Ontario; 2014.
According to the probabilistic calibration of deterministic models
[15] Committe BD-002. Concrete structures (AS 3600–2009). North Sydney, Australia:
under different conditions of compressive strength of concrete, ratio Standards Australia; 2010.
of cover thickness to rebar diameter and yield strength of rebar, the [16] GB 50010. Code for design of concrete structures (GB 50010-2010) (in Chinese).
CSA A23.3 model would be preferred since its confidence levels are Beijing, China; 2010.
[17] Hwang SJ, Leu YR, Hwang HL. Tensile bond strengths of deformed bars of high-
larger than 90% in most cases. strength concrete. ACI Struct J 1996;93(1):11–20.
(4) Based on the proposed probabilistic model of development length, [18] Darwin D. Tension development length and lap splice design for reinforced concrete
the uncertainties of basic parameters (such as concrete compressive members. Prog Struct Eng Mater 2005;7(4):210–25.
[19] European Commitee for Standardization, EN 1990 – Eurocode – Basic of structural
strength, minimum concrete cover thickness, rebar diameter and design; 2002.
tensile stress of reinforcing bar) can further be considered by de- [20] Lutz LA, Gergely P. Mechanics of bond and slip of deformed bars in concrete. J Am
scribing them as random variables. Concrete Instit 1967;64(11):711–21.
[21] Darwin D, Graham EK. Effect of deformation height and spacing on bond strength of
reinforcing bars. ACI Struct J 1993;90(6):646–57.
Acknowledgments [22] Choi OC, Lee WS. Interfacial bond analysis of deformed bars to concrete. ACI Struct
J 2002;99(6):750–6.
[23] Idun EK, Darwin D. Bond of epoxy-coated reinforcement: coefficient of friction and
The financial supports received from the National Natural Science rib face angle. ACI Struct J 1999;96(4):609–15.
Foundation of China (Grant No. 51668008 and Grant No. 51738004) [24] Lundgren K. Pull-out tests of steel-encased specimens subjected to reversed cyclic
are gratefully acknowledged. loading. Mater Struct 2000;33(7):450–6.
[25] Xu YL. Experimental study of anchorage properties for deformed bars in concrete
PhD thesis Beijing: Department of Civil Engineering, Tsinghua University; 1990.
Appendix A. Supplementary material [26] Wang XH, Liu XL. Bond strength modeling for corroded reinforcement in reinforced
concrete. Struct Eng and Mech 2004;17(6):863–78.
Supplementary data to this article can be found online at https:// [27] Pantazopoulou SJ, Papoulia KD. Modeling cover cracking due to reinforcement
corrosion in RC structures. J EngMech-ASCE 2001;127(4):342–51.
doi.org/10.1016/j.engstruct.2018.12.047. [28] Hadje-Ghaffari H, Darwin D, McCabe SL. Effects of epoxy coating on bond of re-
inforcing steel to concrete. SM Report No. 28, University of Kansas Center for
References Research, Lawrence, Kansas; July, 1991, 288 pp.
[29] Darwin D, Graham EK. Effect of deformation height and spacing on bond strength of
reinforcing bars. SL Report 93-1, University of Kansas Center for Research,
[1] He Z, Tian GW. Probabilistic evaluation of the design development length of a GFRP Lawrence, Kansas; January, 1993, 68 pp.
rod pull-out from concrete. Eng Struct 2011;33(10):2943–52. [30] Mendis PA, French CW. Development lengths of reinforcement in high-strength
[2] Sajedi S, Huang QD. Probabilistic prediction model for average bond strength at concrete. Aust Civil Eng Trans 1997;39(1):27–34.
steel–concrete interface considering corrosion effect. Eng Struct 2015;99:120–31. [31] Miller GG, Kepler JL, Darwin D. Effect of epoxy coating thickness on bond strength
[3] Orangun CO, Jirsa JO, Breen JE. A reevaluation of test data on development length of reinforcing bars. ACI Struct J 2003;100(3):314–20.
and splices. J Am Concrete Instit 1977;74(3):114–22. [32] Esfahani MR, Rangan BV. Bond between normal strength and high-strength con-
[4] ACI Committee 318. Building code requirements for structural concrete (ACI crete (HSC) and reinforcing bars in splices in beams. ACI Struct J
318–14) and commentary (ACI 318R–14). Farmington Hills, Michigan: American 1998;95(3):272–80.
Concrete Institute; 2014. [33] Harajli MH. Comparison of bond strength of steel bars in normal and high-strength
[5] Zuo J, Darwin D. Splice strength of conventional and high relative rib area bars in concrete. J Mater in Civil Eng-ASCE 2004;16(4):365–74.
normal and high-strength concrete. ACI Struct J 2000;97(4):630–41. [34] fib TG4.5 splice test database. < http://fibtg45.dii.unile.it/files%20scaricabili/
[6] ACI Committee 408. Bond and development of straight reinforcing bars in tension Database_splice test%20Stuttgart%20sept% 202005.xls > .
(ACI 408R–03). Farmington Hills, Michigan: American Concrete Institute; 2003. [35] Bernardo JM, Smith AFM. Bayesian theory. John Wiley & Sons; 2008.
[7] Esfahani MR, Kianoush MR. Development/Splice length of reinforcing bars. ACI [36] Laine M. Adaptive MCMC methods with applications in environmental and geo-
Struct J 2005;102(1):22–30. physical model. Soc Sci Electron Publ 2008;30(1):89–100.
[8] Tepfers R. Cracking of concrete cover along anchored deformed reinforcing bars. [37] Hassan TK, Lucier GW, Rizkalla SH. Splice strength of large diameter, high strength
Mag Concrete Res 1979;31(106):3–12. steel reinforcing bars. Const Build Mater 2012;26(1):216–25.
[9] Wang HZ. An analytical study of bond strength associated with splitting of concrete
cover. Eng Struct 2009;31(4):968–75.

289

You might also like