Composite Structures 272 (2021) 114132

Contents lists available at ScienceDirect

Composite Structures
journal homepage: [Link]/locate/compstruct

Reliability-based assessment of bond strength models for near-surface

mounted FRP bars and strips to concrete
Reza Aghamohammadi a, Kourosh Nasrollahzadeh a, Amir Mofidi b,⇑, Peter Gosling b
a
Department of Civil Engineering, K. N. Toosi University of Technology, P.O. Box 15875-4416, Tehran, Iran
b
Geotechnics and Structures Group, School of Engineering, Newcastle University, Newcastle Upon Tyne, UK

A R T I C L E I N F O A B S T R A C T

Keywords: This article investigates the reliability of three existing near‐surface mounted (NSM) fiber‐reinforced polymer
Reliability‐based analysis (FRP) bond strength models including the ACI 440.2R‐17 model and two fracture mechanics‐based models.
First‐order second‐moment method Various load combinations and strength reduction factors included in the ACI 440.2R‐17 guidelines and the
Near‐surface mounted CSA S806‐12(17) standard are considered. The uncertainty of the bond models is evaluated using the data
Fibre‐reinforced polymers
gathered from 451 experimental specimens available in the literature with careful consideration of the
Bond strength
Design standard codes
reported modes of failure and their combinations. A reliability analysis based on the first‐order second‐
moment (FOSM) method is conducted on the three bond models for both strips and bars against the dead, live,
snow, and wind loads accounting for uncertainties due to geometrical and mechanical properties of the NSM
FRP‐to‐concrete joints. This article provides a platform for choosing the proper strength reduction factors for a
practical range of target reliability indices. The results show that the strength reliability of the considered bond
models with respect to the existing strength reduction factors requires major improvements. The proposed
strength reduction factors in this study are able to improve the reliability indices of the bond models signifi-
cantly for provisions of ACI and CSA when different load combinations are considered.

1. Introduction quality between the FRP and concrete [2,3,4,5]. This is mainly attrib-
uted to the enhanced bonding stress transfer between FRP and con-
The use of fibre‐reinforced polymers (FRP) in rehabilitation of the crete in the NSM FRP with three sides of the FRP and concrete bond
existing infrastructure is a practical and commonly used solution surfaces compared to that of EB FRP with only one FRP to concrete
among other existing traditional methods. Externally‐bonded (EB) bond surface. In the NSM method, a groove or slit is cut in the concrete
FRP was the first FRP rehabilitation method, proposed by researchers cover followed by the installation of FRP bars or strips before the
in the field in the 1990s. In this method, FRP fabrics or laminates are groove/slit is filled with epoxy. Embedment of FRP in the inner layers
adhesively bonded to the body of the existing structure, as an addi- of the concrete cover with a better concrete quality than the external
tional load‐resisting component. Due to strong mechanical properties face of the concrete cover improves the FRP to concrete bond charac-
of advanced FRP materials, debonding due to failure in the concrete teristics [6]. When considering NSM FRP strips, slits with rather high
adjacent to the adhesive‐concrete interface becomes the most probable depth‐to‐thickness ratio, provide additional confinement to increase
failure mode ahead of FRP rupture. Since a thin layer of concrete is the bond quality between FRP and concrete in the NSM method [7].
attached to FRP strip after failure, the failure actually occurs in con- The quality of the bond is generally considered to be the most impor-
crete, and thus, it is not exactly defined as “debonding” although the tant factor in FRP strengthening methods. This is mainly due to the
term has been widely adopted [1]. Also, the failure due to debonding fact that if the bond strength gets compromised the FRP strengthening
at the adhesive‐concrete interface can be avoided when careful prepa- system will be rendered unserviceable.
ration of surface is conducted. Near‐surface mounted (NSM) was pro- When considering force transfer between two materials, the
posed as a more efficient strengthening method, enhancing the bond mechanical properties of, and bond mechanics between, the two mate-

⇑ Corresponding author.
E-mail addresses: [Link]@[Link] (R. Aghamohammadi), nasrollahzadeh@[Link] (K. Nasrollahzadeh), [Link]fidi@[Link] (A. Mofidi), peter.
gosling@[Link] (P. Gosling).

[Link]
Received 7 December 2020; Revised 5 April 2021; Accepted 17 May 2021
Available online 21 May 2021
0263-8223/© 2021 Elsevier Ltd. All rights reserved.
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Nomenclature

A cohesive shear failure within the adhesive PLoad unfactored load caused by loads (kN)
AC debonding at adhesive and concrete interface Pn;A nominal bond capacity calculated by ACI 440.2R‐17 for-
AF debonding at FRP bar/strip and adhesive interface mula (N)
Af cross‐section area of FRP reinforcement (mm2) Pn;S nominal bond capacity calculated by Seracino model (N)
a shear span (mm) Pn;Z nominal bond capacity calculated by Zhang model (N)
b width of the beam (mm) PPre predicted bond strength (kN)
C concrete splitting failure mode Pt full tensile capacity of the CFRP strip (N)
Cf ;S cross‐sectional contour of the failure surface in Seracino Pu factored bond strength (kN)
model (mm) pf perimeter of FRP reinforcement (mm)
Cf ;Z cross‐sectional contour of the failure surface in Zhang mod- Q demand in LSF
el (mm) R capacity in LSF
D dead load (kN) S snow load (kN)
d effective depth (mm) tf FRP strip thickness (mm)
db diameter of FRP bar (mm) W wind load (kN)
Ef modulus of elasticity of FRP reinforcement (GPa) wg groove width (mm)
F FRP rupture failure mode αL;A reduction factor to account for the effect of insufficient
0
fc compressive strength of concrete (MPa) bond lengths in ACI 440.2R‐17 formula
f fu tensile strength of FRP reinforcement (MPa) αL;S reduction factor to account for the effect of insufficient
G limit state function (LSF) bond lengths in Seracino model
Gf interfacial fracture energy (N/mm) αL;Z reduction factor to account for the effect of insufficient
hf FRP strip height (mm) bond lengths in Zhang model
hg groove height (mm) β reliability index
kf height‐to‐width ratio of failure contour βT target reliability index
L live load (kN) δ coefficient of variation
Lb bond length (mm) η the axial stiffness of bonded joints
Le;A effective bond length in ACI 440.2R‐17 formula (mm) λ bias (mean to nominal value)
Le;S effective bond length in Seracino model (mm) μ mean
Le;Z effective bond length in Zhang model (mm) ξm model uncertainty factor
Pb;A bond capacity of failure modes other than FRP rupture in σ standard deviation
ACI 440.2R‐17 formula (kN) τACI shear bond stress of ACI 440.2R‐17 (=6.9 MPa)
Pb;S bond capacity of failure modes other than FRP rupture in τmax maximum shear bond stress (MPa)
Seracino model (kN) ϕ strength reduction factor
Pb;Z bond capacity of failure modes other than FRP rupture in ϕc concrete strength reduction factor
Zhang model (kN) ϕf FRP strength reduction factor
PD unfactored load caused by dead load (kN) ϕm modifying strength reduction factor
Pexp experimental bond strength (kN) χ2 chi square goodness of fit
Pf probability of failure
PL unfactored load caused by live load (kN)

rials are of vital importance. Many valuable analytical or experimental experimental tests conducted on the NSM FRP‐to‐concrete joints, five
studies have been conducted to propose models for the bond beha- different failure modes have been reported [19]. These failure modes
viour of NSM FRP to concrete (e.g., [8,9,10,11]). However, the com- occur in the adhesive layer, in the concrete, or by the FRP rupture.
plexity of the bond behaviour between the FRP composites and the Besides, the failure may happen at the interfaces of adhesive‐
concrete means that several research questions remain unanswered, concrete or adhesive‐FRP. In many cases, the failure was not due to
requiring further investigation [6]. only one of the five failure categories. Rather, it was reported that a
A particular complexity arises from the underpinning uncertainties combination of two or more failure modes contributed to the overall
that are associated with physical and mechanical characteristics of FRP behaviour, providing additional complexity in understanding the
composite [12,13,14,15,16,17]. The behaviour of the adhesive epoxy behaviour of the NSM FRP‐to‐concrete joints. These and other ques-
at the concrete/epoxy and epoxy/FRP interfaces should also be consid- tions surrounding this range of complexities provide the main impetus
ered in the NSM FRP to concrete bond research studies. The role of the for this study, leading to a better understanding and prediction of the
adhesive epoxy layer only adds to the uncertainties linked to the NSM NSM FRP to concrete bond. To answer these questions, an in‐depth
FRP and concrete bond strength since the mechanical and physical analysis has been conducted to explore the link between the failure
properties of the adhesive layer are uncertain in turn [13,14,18]. Such modes and the cross sectional shape of the FRP and the influence of
uncertainties should be considered in the investigations of NSM FRP to these links on the reliability of the existing bond models proposed to
concrete bond behaviour. predict the NSM FRP to concrete bond behaviour.
Additional complexities are associated with the effect of the varia- In general, the bond strength in an NSM FRP strengthening system
tions in FRP material shapes on the bond strength of NSM FRP. So far, is considered to be the maximum tensile force that the FRP‐to‐concrete
NSM FRP materials in the form of bars and strips have been used for joint can resist. More specifically, the bond strength corresponds to the
strengthening of reinforced concrete (RC) elements in flexure and peak load developed in a pullout test of an NSM FRP‐to‐concrete joint.
shear. Intrinsic variations in the cross‐sectional shape of the FRP Investigations by [10,20] indicate that the bond strength increases
increases the level of uncertainties when endeavouring to predict the with an increase in the bond length, up to a limit, beyond which the
bond behaviour of NSM FRP to concrete. In the currently available bond strength remains constant. This important threshold is recog-

2
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

nized as the effective bond length (Le), which plays a key role in the without any modifications to these original bond models, which are
understanding of the bond behaviour of the NSM FRP to concrete. In described as semi‐empirical in that they use analytical methods in
order to better predict the bond behaviour of EB FRP to concrete, addition to numerical integration and regression of the experimental
researchers have provided fracture mechanics‐based bond models that results. Due to limitations in the experimental results at the time of
can predict the bond strength and Le with reasonable accuracy (e.g., their publication, the bond models include specific idealizations and
[21,22,23]). Also, two fracture mechanics‐based bond models [7,10] simplifications. It is important, therefore, to evaluate their accuracy
exist to predict the behaviour of NSM FRP bonded to concrete. The probabilistically using an extended database of experimental results.
majority of the existing investigations related to the NSM FRP bond A reliability analysis based on the first‐order second‐moment (FOSM)
strength are experimental studies. In addition, most design guidelines method is conducted on the three ACI, Seracino, and Zhang bond mod-
and standards either chose not to include NSM strengthening in their els against the dead, live, snow, and wind loads defined by the ACI
latest versions (e.g., CSA S806‐12 [24]) or use empirically‐based mod- guidelines and the CSA S806‐12(R17), hereafter called the CSA stan-
els which do not clearly reflect the mechanics behind the NSM FRP to dard. The load combinations in the ACI guidelines and CSA standard,
concrete bond (e.g., ACI 440.2R‐17 [25]). ACI and CSA design references hereafter, are provided by the Ameri-
To the best of the authors’ knowledge, Coelho et al. [19] is the only can Society of Civil Engineers (ASCE) 7–16 Standard [29] and the
existing study that applies a probabilistic approach to determining the National Building Code of Canada (NBCC) [30], respectively.
safety of models that seek to predict the NSM FRP bond strength. In this study, the analysis has been conducted for 16 different sce-
Coelho et al. [19] calibrated the safety factors of the bond formulations narios with respect to different failure modes and different FRP cross‐
by the ACI440.2R‐17 [25] guidelines, hereafter called ACI model, and sectional shapes, which is unprecedented among the existing studies. It
the equations by the Australian standard design handbook [26], here- should be noted that amongst the studied FRP design references, one
after called SA model, in order to be consistent with the partial safety uses global reduction factors applied to the design equation (ACI
factors methodology which is commonly used in Eurocode. In doing guidelines), whereas the other defines partial reduction factors applied
so, Coelho et al. employed the procedure of “design assisted by testing” to the strength of materials (CSA standard). Considering this differ-
proposed by Annex D in Eurocode 0 (EC0) [27], hereafter called EC. ence, the strength reduction factors have been tuned to improve the
However, the method of “design assisted by testing” is not able to pro- structural reliability and safety of the studied three bond models, with
vide an assessment of the safety of the bond strength for different load respect to the provisions of the ACI and CSA design references. It is
combinations. Besides, the same method assumes a failure probability clear that among the studied bond models only the ACI model is cur-
of about 0.1% for the design value (corresponding to the product of 0. rently used, and only in the ACI guidelines. The alternative bond mod-
8 × 3.8 = 3.04 in normalized space) according to EC. In the analysis els are studied in this article to enable possible inclusion of the models
method applied by [19], Monte Carlo simulation was utilized to derive in the ACI and CSA design references.
the joint probabilistic distribution of the limit state resistance function. Finally, the safety levels of the three bond models have been inves-
In addition, all the geometric parameters were assumed to be deter- tigated for the ACI and CSA design references by comparing the relia-
ministic, with the mechanical parameters including Ef, ffu , and f’c bility of the existing strength reduction factors and that of the newly
described as uncertain. Meanwhile, in order to calculate the statistical proposed strength reduction factors considering different correspond-
parameters of the model uncertainty variable, Coelho et al. [19] used ing load combinations. It should be noted that all steps in the calibra-
experimental results of 128 tested NSM CFRP strips‐to‐concrete joints. tion of the strength reduction factors are available for target reliability
The results of their study are only applicable to NSM FRP strips, with indices between 3.5 and 4.5, instead of a certain pre‐selected target
existing experimental test data describing NSM FRP bars not included. reliability index.
In their calibration of bond strength equations, the same partial safety
factors (i.e., for ffu and f’c) currently used throughout EC, were main- 2. Failure modes and bond model equations
tained. In the ACI model, only ffu plays a role, i.e., the strength of con-
crete is not considered in the ACI model. Therefore, Coelho et al. [19] Table 1 summarises the either fully empirical or semi‐empirical
reduced the design bond stress (τd) from 6.9 MPa in the ACI model to components of the ACI, Seracino, and Zhang models describing the
1.77 MPa for achieving the required safety. As for calibration of the SA NSM bond strength calculations. Notably, Seracino and Zhang models
model, Coelho et al. adopted global safety factors in addition to the stem from the same fracture mechanics‐based bond equations. Due to
partial safety factors to be incorporated in the bond strength equa- limitations in the experimental data at the publication date of the mod-
tions. Finally, Coelho et al. [19] clearly recommended that there is a els and the unknown nature of the rather new strengthening system,
need to propose new models to predict the bond behaviour of NSM the Seracino and Zhang models are based on assumptions that are sig-
FRP more accurately and by accounting for all possible failure modes. nificantly different as will be discussed later. In addition, different
When using the NSM FRP method to strengthen RC members in experimental test results have been used for the data regression
flexural or shear, one of the most important challenges is to predict analysis.
the contribution of the FRP to the total resistance of the member. This It is clear that several unknowns remain unanswered to achieve a
will only be viable through accurately calculating the NSM bond clear understanding of the bond behaviour of NSM FRP bars and strips
strength due to significant dependency of the FRP contribution to to concrete. For example, it is necessary to provide (initially poten-
the resistance on the bond strength of the NSM FRP to the concrete. tially uncertain) estimated measures of the parameters involved in
The dependency between the bond strength and the resistance of the the mechanics of the NSM strengthening system. Within an uncer-
member/structure leads to complexities in the choice of the target reli- tainty propagation framework, a probabilistic approach may be helpful
ability index. Darwin et al. [28] proposed a bond target reliability in providing insights into the ranges of measured bond behaviour and
index equal to one fifth of the main mechanism’s target reliability the contributions of the model parameters aiming to predict the per-
index when recommending bond strength reduction factors. formances of the NSM FRP strengthening methods and their associated
In the existing article, a reliability analysis on three published NSM failure modes.
bond strength models is conducted based on the ACI model; Seracino An NSM bond system may fail in five different failure modes. The
et al. [10] bond model, hereafter called Seracino model; and Zhang first failure mode takes place due to cohesive failure in the concrete
et al. [7] bond model, hereafter called Zhang model. In the context surrounding the groove, identified as C. The second type of failure
of the limited number of published NSM bond models, those selected mode takes place due to FRP rupture, (F). The third type of failure nor-
in this study are most well‐known and advanced bond models cur- mally occurs due to shear failure in the adhesive epoxy layer, (A). The
rently available. A comprehensive reliability analysis is conducted

3
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Table 1
Bond strength equations of ACI 440.2R-17 [25], Seracino et al. [10] and Zhang et al.
[7].

ACI 440.2R-17 Seracino et al. Zhang et al.

0:138
τACI ¼ 6:9MPa – hg 0 0:613
τmax ¼ 1:15 wg fc
– Cf ;S ¼ 2hg þ wg þ 4 Cf ;Z ¼ 2hg þ wg

0:422
– – hg 0 0:619
Gf ¼ 0:4 wg fc
– kf ¼
hg þ1 τmax 2 Cf ;Z
η2 ¼ 2G
wg þ2 f E f Af
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Le;A ¼
Af f fu
π
0:526
0:976kf E f Af Le;Z ¼ 1:66
η
pf τAci Le;S ¼ 2
ð0:802þ0:078kf Þ fc
0 0:6
Cf ;S


αL;A ¼ LLe;Ab ≤ 1 αL;S ¼ LLe;Sb ≤ 1 αL;Z ¼ LLe;Zb 2:08  1:08 LLe;Zb ≤ 1
Pt ¼ f fu Af Pt ¼ f fu Af Pt ¼ f fu Af
0:33 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pb;A ¼ αL;A Pt Pb;S ¼ 0:85αL;S k0:25 fc
0
Cf ;S E f Af Pb;Z ¼ αL;Z 2Gf Ef Af Cf ;Z
f
Pn;A ¼ Pb;A Pn;S ¼ Pb;S ≤ Pt Pn;Z ¼ Pb;Z ≤ Pt

Note 1: In ACI 440.2R-17, ffu is replaced by 0.7ffu when debonding governs the
behaviour.
Note 2: Units of hg and wg are in mm.

fourth and fifth types of failure mode take place due to debonding in provides the bond resistance strong enough to develop the ultimate
the Adhesive/Concrete interface (type AC) and Adhesive/FRP inter- tensile stress ffu in the FRP. In this capacity, Hassan and Rizkalla
face (type AF). The failure modes play an important role in this anal- [31] (as one of the main references of ACI 440, section 14.3) reported
ysis not only corresponding to the results provided by the bond model the test results on beams strengthened by NSM CFRP strips with
equations but also with respect to the reliability analysis. debonding failure mode, where the strains at failure increase linearly
As an observation common to all bond models, it is important to with the increase of the embedment length, resulting in a considerable
note that only NSM FRP materials that are bonded with a bond length enhancement in the ultimate load carrying capacity of the beams,
equal or longer than Le will reach the peak bond strength provided by whereas for the beam specimens with the rupture of NSM FRP strip,
the models. Therefore, the studied bond models assume that all the increasing the embedment length could not provide extra strength. It
NSM FRP bars or strips are bonded with minimum Le. However, a should be noted that the experimental data employed in the present
decreasing αL coefficient (0 < αL < 1) is used to represent the effect paper contain different failure modes including FRP rupture as well
of NSM FRP with an anchorage length shorter than Le. Generally, αL as debonding. As long as the FRP rupture is concerned, the required
is related to the ratio Lb/Le when Lb < Le. In order to better investigate development length Le,A is determined by considering ffu as the effec-
the studied bond models, details of the ACI, Seracino and Zhang bond tive stress in bar/strip. On the other hand, for the specimens with
models are provided in the following sections. the debonding failure mode, the corresponding development length
Le,A is based on the bar/strip stress equal to 0.7 ffu according to ACI
2.1. ACI 440.2R-17 bond model 440.2R‐17 under flexural design. It is clear that limiting the FRP strain
to 0.7 ffu is not applicable to the cases where FRP rupture (i.e., full uti-
With exception of the FRP rupture failure mode, the ACI model lization of strength of ffu) is observed in the test specimens.
(Table 1) uses a rather simple empirical equation for all the failure According to the ACI model, the bond strength in the FRP‐to‐
modes to calculate the NSM FRP bond strength. The ACI model does concrete joint is equal to pf × Lb × τACI provided that the available
not distinguish between failure mechanics and details in the calcula- bond length Lb is less than or equal to Le,A (i.e., αL;A ≤ 1), where pf is
tion of the bond strength for the C, A, AC and AF failure modes. This the perimeter of FRP reinforcement cross‐section. In the case of
approach reduces the complexity of the model, making it simpler to αL;A > 1 , the bond capacity is limited to either Af × ffu when FRP rup-
adopt for design codes and guidelines. Conversely, neglecting key ture governs or Af × 0.7 ffu when debonding is dominant. It should be
influential parameters across the different failure modes may lead to noted that 6.9 MPa value for τACI was extracted as an average value
increased levels of uncertainty, which potentially results in rather from the experimental results of all failure modes. Therefore, the
lower predictive accuracy (or reliability) of the ACI model as demon- ACI model considers only two categories of failure modes: 1) the
strated by Coelho et al. [19]. The current version of ACI 440.2R‐17 (in FRP rupture failure mode; and 2) all other failure modes including
chapter 10 entitled “flexural strengthening”) recommends the use of A, AC, AF, and C. Among the studied bond models, the ACI model is
0.7εfu as the debonding design strain (designated by εfd) for NSM the only bond model that considers both FRP bars and strips.
FRP, where εfu is the design rupture strain of FRP. Accordingly, to
achieve this strain value, the bonded length should be greater than 2.2. Seracino et al. bond model
the corresponding development length. The development length (Le,
A) of NSM FRP bars/strips is then calculated by Eq. 14.3 in ACI The Seracino model (Table 1) benefits from a fracture mechanics
440.2R‐17 (as appeared in Table 1), which is based on equilibrium equation based on a linear bond‐slip model. Most coefficients and
conditions assuming an average bond stress τACI developed along the exponents corresponded to the fracture mechanics equation are pro-
NSM FRP bars/strips. In order to calculate Le,A, the ACI model employs duced based on regression of the available experimental results. Due
an empirically‐based bond stress equal to τACI = 6.9 MPa. On the other to limited number of the experimental results at the time of publica-
hand, when the failure mode coincides with the FRP rupture, the FRP tion, the equation has been calibrated based on results of fourteen
cross‐section reaches its maximum tensile force equal to Af × ffu, experimental test specimens of a separate study by the researchers
where Af is the area of cross‐section and ffu is the ultimate tensile [32]. Seracino bond model has been calibrated with FRP strips only.
strength of the FRP. According to the ACI model, this will only take However, studies related to NSM strengthening of beams using Sera-
place when a sufficient bond length (Le,A) exists. Within this scenario cino model (e.g., [5]) suggested that the model can also be used for
in which the FRP rupture occurs, Le,A is the minimum bond length that FRP bars with reasonable outputs for RC beams strengthened with

4
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

NSM FRP. Meanwhile, a confinement ratio was used to propose a gen- 3.1. Statistical and probabilistic characteristics of variables
eric debonding bond‐slip model for both EB and NSM strengthening
methods. It should be noted that Seracino et al. [10] considers the con- Given that the parameters used to predict the NSM FRP bond
crete cohesive failure mode to be the desirable failure mode, given that strength are not deterministic, levels of uncertainty should be propa-
in this failure mode (C) the failure takes place within the original gated in predicting outputs of the bond models. The uncertain param-
structure, with the strengthening system resisting having capacity eters (basic variables) are divided into two separate categories,
beyond the ultimate failure of the repaired structure. In the present namely, resistance parameters and load effect (Table 2). Meanwhile,
article, the potential of the Seracino model in predicting the strength the resistance variables are divided into three subcategories including
of repairs using FRP bars will be assessed numerically using uncer- dimensions, materials, and the bond model uncertainty‐related vari-
tainty propagation. Given that the number of the experimental results ables. In this study, the dimension‐related variables are db, tf, hf, hg,
for all failure modes is significantly greater than the data used to cal- wg, and Lb , describing the diameter of the FRP bar, thickness and
ibrate the model by Seracino et al. [10], the accuracy of the bond will height of the FRP strip, depth of the concrete groove, width of the con-
be evaluated in this article with a larger test database for each failure crete groove, and the available bond length, respectively. In this con-
model. text, the material variables are f’c, ffu, and Ef , introduced as the
compressive strength of concrete, ultimate tensile strength of FRP,
2.3. Zhang et al. bond model and modulus of elasticity of FRP, respectively. In order to get a mea-
sure of the reliability index, β, of each model, a third category of the
This model proposes a rational method to calculate the bond resistance‐related uncertain variables, namely the model uncertainty
strength of NSM FRP laminates only under cohesive concrete failure, factor (ξm) is considered. It should be noted that ξm is the ratio
indicating that the strengthening system has sufficient capacity and between the experimental bond strength (PExp) over the predicted
effectiveness to induce failure in the original structure [10]. Accord- bond strength (PPre) by each model. In order to calculate ξm, it is nec-
ing to Zhang et al. [7], proposing an analytical model for the failure essary to gather a database of existing NSM bond strength test
mode C is relatively straightforward since only one material is results. In this study, a comprehensive experimental test database of
involved in this mode. Conversely, in failure modes AC and AF, 451 specimens is extracted from 22 existing studies on NSM
interaction of at least two materials should be considered to develop FRP direct pullout tests in the literature. The list of the experimen-
a logical and accurate analytical bond model. Meanwhile, Zhang tal studies included in the database is as follows: [2,3,9,32,41, 42,43
et al. [7] believe that adequate preparation of the concrete and ,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59]. Details of the exper-
FRP surfaces at the interface in combination with an appropriately imental results are provided in Table 3 that also includes the number
selected epoxy, failure mode C will be predominate. Therefore, a of specimens failed in a particular mode among the five reported fail-
bond slip model based on the failure mode C is proposed in Zhang ure modes in the literature. From the 451 investigated specimens, 137
et al. [7,33], assuming an effective bond length is provided for the failed with failure mode C. Failure mode F was the least identified fail-
FRP‐to‐concrete joint. A decreasing coefficient (αL,Z), accounting for ure mode with 44 reported specimens. The most common failure mode
the effect of insufficient bond length for the cases where the bond was A with 183 reported instances. The numbers of reported failure
length is shorter than the effective bond length, is considered by modes at the interfaces for failure mode AC and AF were 93 and 53
Zhang et al. [33]. Similar to the Seracino model, the majority of specimens, respectively. In the reported studies, for 111 specimens,
model coefficients and exponents are estimated using regression more than one failure mode was reported. In total, by adding the
analyses of experimental data or the finite element model outputs. tested specimens with no reported failure modes, 451 specimens are
Zhang et al. [7] evaluated the accuracy of the proposed bond considered in this study’s database. This database is the largest ever
strength equation [33] using 51 experimental specimen datasets col- used on direct pullout testing of NSM FRP bonded to concrete. As com-
lated from 7 published studies. It should be noted that Zhang model pared to other published papers, Coelho et al. [19] employed 128 test
(summarised in Table 1), similar to Seracino model, is only proposed data while Zhang and Xue [60] utilized 145 specimens. It is noticed
to predict the bond strength of FRP strips. The accuracy of the Zhang that some of the experimental data reported in the very recent litera-
model to predict the bond strength of NSM FRP bars is evaluated in ture related to the bond of NSM FRP are not in line with the objective
this paper using available experimental data. The purpose of this of the current work because their conducted tests have dealt with
evaluation is mainly to validate analytically the existence of an either beam test setup or some particular parameters such as various
NSM bond model that can provide accurate predictions for both types of adhesives (e.g., cement‐based adhesives), application of
NSM FRP bars and strips. Such a model can be used as the universal anchorage systems, effects due to adjacent NSM bars/strips, or fatigue
bond model of choice in the future strengthening design equations loading, all of which are beyond the scope of this paper, and also,
using NSM FRP bars and strips. Therefore, the behaviour of all three beyond the current guidelines or the existing bond models; and
NSM bond models will be evaluated for both NSM FRP thereby, they are excluded from the present experimental database.
cross‐sectional shapes with regard to different reported failure It should be noted that the NSM FRP bar and strip specimens are ana-
modes. lyzed separately in this study. Fig. 1 illustrates the experimental bond
strength values versus the predicted bond strength by each of the
3. Reliability analysis details studied models.
The second category of the variables is related to the uncertainties
Reliability analyses provide a practical means to evaluate the safety corresponded to the loads. In each design guideline or standard code,
of structural engineering designs, through relations between uncertain the load variables are given a specific level of uncertainty, which is
parameters that influence structural safety, performance measures, defined by the increased load coefficients, corresponded to each load
and design factors. The reliability index, β, resulting from a reliability case. Table 4 shows the load factors used in different load combina-
analysis enables decisions to be made about the implied safety level of tions of principal load cases that are used in the ACI guidelines and
the design. In this article, structural reliability analyses are used to CSA standard. According to Turkstra load combination rule [61], only
estimate strength reduction factors for the identified NSM bond mod- one principal load should be considered with its maximum value in
els for load factors given in the ACI and CSA design references consid- combination with the permanent dead load. The other loads in such
ering different levels of safety corresponded to β between 3.5 and 4.5. combinations should be considered with their mean value as a load
The details of the reliability analysis of the three NSM bond models in combination with the principal load and the permanent dead load.
have been provided in the following sections. This is due to the assumption that the probability that two principal

5
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Table 2
Probabilistic characteristics of different variables.

Source of uncertainty Name (Unit) Nominal Value Mean (μ) Bias (λ) SD (σ) COV (δ) References

Fabrication db (mm) db – 1 – 0.015 Zadeh et al. [34]

t f (mm) tf – 1 – 0.02 Okeil et al. [35] a
hf (mm) hf – 1 – 0.02
hg (mm) hg hg  4:826 – 12.7 – Mirza and MacGregor [36] b

wg (mm) wg wg þ 2:286 – 4.83 –

Lb (mm) Lb – 1 17.53 –
Material 0
f c (MPa)
0
fc – * – 0.101 Nowak and Szerszen [37]
f fu (MPa) ðEf Þ (GPa) μ −3σ(μ) 597c(40) – – 0.08(0.08) El-Sayed et al. [38]
692c(41.3) – – Michaluk et al. [39]
769c(1 3 5) – – El-Sayed et al. [40]
2068 d(1 3 1) – – Peng et al. [41]
2040 d(1 2 3) – – Kalupahana et al. [42]
2100 d(1 3 0) – – Shield et al. [43]
Load e
Dead (kN) D 1.05 – – 0.10 Nowak and Collins [44]
Live (kN) L 1.00 – – 0.18
Snow (kN) S 0.82 – – 0.26
Wind (kN) W 0.78 – – 0.37

0 3
0 2
0
*2:413  105 f c þ 3:1743  103 f c  1:3543  101 f c þ 3:0649
Note: All variables have a lognormal distribution.
a
The reference is used for both tf and hf .
b
For hg , wg and Lb the statistical parameters of effective depth (d), width (b) and span (a) of the beam are used, respectively.
c
Bar
d
Strip
e
Maximum 50 years load

Table 3
Statistical properties of experimental dataset.

Parameter C*,† (137 Specimens) F*,† (44 Specimens) A*,† (183 Specimens) AC*,† (93 Specimens) AF*,† (53 Specimens) Total*,† (451 Specimens)

Min. Max. Med. Min. Max. Med. Min. Max. Med. Min. Max. Med. Min. Max. Med. Min. Max. Med.

db (mm) 7.5 13.0 9.0 6.0 12.7 10.0 6.0 13.0 9.1 6.0 12.0 8.0 6.0 12.0 8.0 6.0 13.0 9.0
t f (mm) 1.2 10.0 1.3 1.2 4.5 1.40 1.3 10.0 10.0 1.4 10.0 2.5 2.0 10.0 4.5 1.2 10.0 2.0
hf (mm) 10.0 40.0 16.0 10.0 20.5 16.0 10.0 20.4 10.0 10.0 16.0 15.0 10.0 16.0 16.0 10.0 40.0 16.0
hg (mm) 11.3 41.0 18.0 10.0 30.0 20.0 11.3 40.0 16.0 10.0 40.0 16.0 12.0 30.0 16.0 10.0 41.0 18.0
wg (mm) 3.0 30.0 15.0 3.2 25.4 10.0 3.3 30.0 15.9 5.0 25.0 14.0 10.0 25.0 14.0 3.0 30.0 14.0
Lb (mm) 38.0 508.0 250.0 60.4 478.5 250.0 30.0 508.0 228.0 30.0 300.0 300.0 40.0 400.0 192.0 30.0 508.0 228.0
0
f c (MPa) 19.0 64.8 28.5 19.0 60.0 40.0 22.0 64.8 28.5 19.0 60.1 22.0 15.0 42.2 23.5 15.0 64.8 30.0
f fu (MPa) 749.0 2800.0 2010.0 749.0 2800.0 2144.5 749.0 3100.0 1676.0 873.0 3100.0 2068.0 1250.0 2800.0 2068.0 749.0 3100.0 2068.0
E f (GPa) 371.7 174.0 125.0 41.0 177.0 131.0 37.2 174.7 124.0 37.2 182.0 124.0 51.0 170.0 131.0 37.2 182.0 130.0
Pexp (kN) 114.1 197.4 55.3 27.7 134.0 64.7 8.7 116.9 44.9 88.0 7.2 40.3 103.7 84.0 389.2 7.2 197.4 46.7

Note: For 111 specimens, more than one failure mode is reported.
*
C: Concrete splitting, F: FRP rupture, A: Cohesive shear failure within the adhesive, AC: Debonding at adhesive and concrete interface, AF: Debonding at FRP
bar/strip and adhesive interface, Total: The entire database including all mentioned failure modes in addition to specimens with unknown failure modes
†
The number of FRP bars in the failure modes of C, F, A, AC, AF and Total is 98, 16, 156, 70, 39, and 278, respectively; and the number of FRP strips is 39, 28, 27,
23, 14, and 173, respectively.

loads reach their maximum at the same time is equal to zero. In this other research works, is that the probabilistic behaviours of all the
study, load combinations corresponded to the principal load have been involved variables in a bond strength model are taken into account.
considered and the companion loads are neglected for the sake of sim- Considering the fact that NSM method is a rather new method, the
plicity. Therefore, a load combination is consisted of summation of a statistics and probabilistic characteristics of some of the parameters
factored dead load, as the permanent load, and a factored principal used to calculate the NSM bond strength are not available in the liter-
load, either live, snow or wind loads. To this end, the assumed load ature. For such variables including hg, wg, and Lb, probabilistic beha-
combinations corresponded to the live load as the principal load for viour of similar variables (i.e., the effective depth, the cross section
the ACI and CSA design references are respectively, 1.2D + 1.6L width, and the span length, respectively) used in RC members have
and 1.25D + 1.5L. Meanwhile, according to Szerszen and Nowak been selected, as reported in Table 2.
[62], the probable ratio of the live load to the total load, i.e., L/
(D + L), can be considered to be within the range 0.3 to 0.7. Consid- 3.2. Design space
ering the mentioned ratios, the nominal values of the actual loads are
calculated and then their probabilistic characteristics are determined The main objective of considering a design space is to guarantee
using the given values in Table 2. The probabilistic characteristics of the safety of the equations in all design case scenarios [35]. Besides,
the loads are based on the maximum loads applied to the structure the design space, which provides sets of nominal values for the design
over a period of fifty years. Afterward, the computed probabilistic variables, is used for calculating the statistical parameters for each of
characteristics are used to calculate β for each case. One of the main the uncertain variables since the values of these parameters such as
goals of the current study, which separates the existing study from mean or bias are dependent on the nominal value as seen in Table 2.

6
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Table 4
Various load combinations.

ACI 440.2R-17 CSA S806-12(R17)

1.2D + 1.6L 1.25D + 1.5L

1.2D + 1.6S 1.25D + 1.5S
1.2D + 1.6W 1.25D + 1.4W

develop the design space, three nominal values (i.e., minimum, maxi-
mum, and median values in the practical range of data) are considered
for each variable. These variables include db, tf, hf, hg, wg, fʹc, ffu, Ef, and
L/(L + D). It should be noted that the three values for ffu (and the cor-
responding Ef) were chosen with respect to the reported values for bars
and strips in Table 2. Meanwhile, seven nominal values are taken for Lb
such that the ratios of Lb/Le include three values smaller than 1, one
value equal to one, and three values greater. Using the produced val-
ues for the involved variables, 5103 virtual NSM FRP bar‐to‐concrete
and 15,309 NSM FRP strip‐to‐concrete joints have been developed.
The number of virtual strip‐to‐concrete joints is three times greater
than that of bar‐to‐concrete joints since there is one more variable in
the case of strips when compared to the number of variables in the
case of bars. The developed design space can safely guarantee that
most design case scenarios are covered. Such a design space developed
in this study is unprecedented among similar existing articles.

3.3. Reliability study method

In order to calculate the β associated with each prediction model,

different reliability‐based methods can be utilized. Each method has
its own advantages and disadvantages. Meanwhile, the First‐Order
Second‐Moment (FOSM) method of structural reliability analysis is
one of the most recognized and widely used reliability analysis meth-
ods among engineers in the field [44]. Most of the sections in the dif-
ferent revisions of the ACI 318 code that deal with the reliability‐based
analyses have used FOSM methods [62]. The FOSM method has been
used in order to calculate the reliability index of the bond models
described in this paper.

3.4. Target reliability index

When conducting a reliability analysis, it is necessary to use a tar-

get reliability index (βT) as a reference level to determine the allowable
failure probability. However, choosing a suitable βT for a rather new
material (FRP material) used in a relatively new bond system (NSM)
can be challenging. In the presence of a proper statistical database,
the target reliability level is determined considering consequences of
failure, and cost of increasing/decreasing the safety margin by a unit
[62]. For a structure with 50‐years of reference period, the Annex G
of ISO 2394 [63] and the Annex B of EC0 for the reliability class of
RC2 consider βT equal to 3.8. However, the probability of failure in
the NSM FRP bond should be less than that of the principal load‐
carrying mechanism such as bending and shear mechanisms in beams
[28]. To this end, Darwin et al. [28] proposed that the probability of
the failure in bond should be assumed to be 20% of that of the princi-
pal mechanism. It is noted that the definition of βT is not included in
the scope of this study. Therefore, for each principal mechanism, its
corresponding βT commonly used in standards has been adopted, such
that the failure probability of the bond strength is one fifth of that of
Fig. 1. Comparison between experimental data and predictions by the studied the principal mechanism, resulting in alternative values of βT of
models. NSM FRP bond, according to the principal mechanisms and their cor-
responding probability of failure as given in Table 5. Values of βT for
the principal mechanisms are assumed to be between 3.0 and 4.0. Fol-
Since for the reliability analysis, the statistical parameters of the uncer- lowing Darwin et al. [28] a range of 3.5 to 4.4 for the βT of NSM FRP
tain variables are required, the nominal values of the variables need to bond performance is proposed. Therefore, in this study, all the results
be assumed in advance through building a design space. In order to are presented for βT of the NSM FRP bond in the range 3.5 to 4.4.

7
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Table 5 presented in Table 6. In addition, considering the non‐zero nature of

Target reliability index βT . ξm, a lognormal distribution is assumed. The validity of such assump-
βT for the main mechanism Pf 0:2Pf βT for the bond tion has been verified using a chi square test for the selected number of
strength evaluation states as illustrated in Fig. 2.
As mentioned earlier, the experimental database used in this paper
3.0 1.349810-3 2.699810-4 3.5
3.5 2.326310-4 4.652610-5 3.9 is the largest one available in the literature. Furthermore, the number
3.8 7.234810-5 1.447010-5 4.2 of test data employed in each state of this study to evaluate the model
4.0 3.167110-5 6.334210-6 4.4 uncertainty factor, ξm , is sufficient in terms of reliability aspects. In
order to indicate the sufficiency of the number of data, the following
approach is taken [12]. The mean of ξm is considered as a normally dis-
tributed random variable, of which the mean is the same as mean of ξm
3.5. Limit state function while the standard deviation, s, is equal to s = σ/n , where σ is the
standard deviation of ξm , and n is the number of data. Basically, when
The principle of limit states design is based on the principal that the n increases, the value of s decreases, leading to less dispersion. Given a
resistance of the member (R) should always be greater than or equal to specific value of n, the probability for mean of ξm to fall below 90% of
the applied load effects (Q). The difference between the two variables the mean can be determined, and it is intended to be small enough to
(R ─ Q) is defined as the limit state function (G), indicating the margin assure the sufficiency of the number of data in evaluating the mean
between safe and unsafe zones. The design equations should keep the value of the modeling uncertainty factor. The 90% threshold (i.e.,
member design within a desirable margin of safety with respect to the 10% decrease allowance) is assumed here as an acceptable tolerance.
limit state function. In equation (1), the general form of the limit state For the states corresponding to the combined failure modes considered
function is written as, in this paper, the results of calculated 90%‐probability indicate that
the highest probability is equal to 2.54% related to 15‐S (with 67 data
G¼RQ ð1Þ points) whereas the lowest one is 0.0002% related to 16‐B (with 278
where R is calculated as, test data), both of which are negligible. Thereby, the number of test
data used herein is sufficient as the probability of deviation from the
R ¼ ξm Ppre ð2Þ
mean value of the model uncertainty factor due to the finite number
where Ppre is the NSM FRP bond capacity predicted by an existing of data is negligible. In line with the above‐mentioned explanation, a
model as listed in Table 1. Meanwhile, Q is the summation of all the reference can be made to Annex D in Eurocode [27], where the statis-
unfactored loads applied to bond joint as mentioned in Eq. (3): tical parameters, particularly the coefficient of variation, obtained
based on 30 samples are taken almost equal to those by infinite num-
Q ¼ PLoad ð3Þ
ber of samples. To this end, the validity of the results presented for
combined failure modes in this study is not affected by increased num-
3.6. Different evaluation states and the corresponding model uncertainties ber of test data, if any of such data become available in the future.

In this paper, the reliability analysis is conducted for various eval- 3.7. Reliability index for different evaluation states
uation states which are different in terms of: 1) the FRP cross sectional
shape; 2) failure modes; and 3) the used bond model. In Table 6, for When evaluating the reliability of the bond strength for any given
each bond model, 16 reliability analysis evaluation states have been evaluation state in Table 6, the corresponding resistance of the given
considered, predicting all possible failure modes for both NSM FRP state (Pn) will be used as Ppre in Eq. (2). For example, in the state 1‐
bars and strips. Under the “State Name” column in Table 6, the desig- B, Pn = Pb,A, that is the ACI bond model as reported earlier in Table 1,
nation of each state is provided. The state designations corresponding will be used as Ppre in Eq. (2). In general, Pn will be calculated using the
to NSM FRP bars include the letter B, whereas states related to NSM nominal values of the variable in the design space that will be applied
FRP strips are denoted using S. Five separate failure modes and three to the corresponding bond model of each evaluation state. It should be
combined failure modes are considered, with each assigned a number noted that each state has been evaluated separately based on ACI and
corresponded to the failure modes in their state designations. The CSA design references load factors. When the loads are calculated for
same sequence in numbering of the failure modes or their combina- the combination of dead and live loads, PD and PL are the tensile dead
tions are used for each of the three bond models (1–8, 9–16, 17–24) and live loads applied to the NSM FRP joints. The nominal values for
as a function of the FRP cross‐section. For example, the state 1‐B cor- PD and PL are determined based on two conditions as follows. Firstly,
responds to NSM bars that failed due to concrete cohesive failure while the factored load combination (Pu) is taken equal to the design resis-
the bond strength is calculated using the ACI model. tance, that is Pu = ϕ Pn , where ϕ is the resistance reduction factor.
In order to evaluate the model uncertainty factor ξm , that is the For instance, for ACI bond model under ACI reference load and resis-
ratio between the experimental bond strength over the bond strength tance factors, we have Pu = 1.2 PD + 1.6 PL , Pn = Pb,A , and ϕ = 1. As
predicted by a bond model, a number of test data are needed. The explained earlier, Pb,A is calculated accounting for the chosen nominal
experimental database collected in this paper is divided into certain values for the variables in the design space. Secondly, a certain ratio
categories on the basis of reported failure mode and the FRP cross‐ between the live load and the total load is chosen, that is PL /(PD + PL)
sectional shape. Thereby, for each evaluation state in Table 6, the cor- is a given value. Then, the unknowns of PD and PL are readily deter-
responding number of the experimental datasets is listed under “Num- mined by use of the two above‐mentioned conditions and for each vir-
ber of Test Data” column. As an example, the number of test data tual NSM FRP bonded joint in the design space. The process is also
related to state 1‐B in Table 6 indicates that 98 specimens out of the illustrated in a flowchart in Fig. 3. Following the calculation of the
existing database on pullout test of NSM FRP bars failed in C mode. nominal values of loads, probabilistic parameters of the loads are gen-
For each of the test data and for a particular bond model (Table 1), erated using Table 2. The probabilistic characteristics of ξm, are calcu-
the predicted bond strength is calculated and compared with the cor- lated using the dataset of the experimental results corresponding to
responding experimental value, yielding a value for ξm. Finally, the cal- each failure mode and cross‐section shape for each state, as explained
culated mean value (μ) and the coefficient of variation (δ) of ξm using earlier. The β for each virtual bond system in the design space related
the corresponding experimental results of the failure mode and FRP to an evaluation state will be calculated using the FOSM method.
cross‐section for each state and the corresponding bond model are Finally, the mean β value for all virtual bond systems of an evaluation

8
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Table 6
Statistical parameters of model uncertainty factor (ξm ) and calculated reliability index (β).

State Name Bond Model Failure Mode FRP Cross-section Number of Test Data ξm β

μ δ (%) ACI CSA

1-B ACI 440.2R-17 C bar 98 1.246 34.3 1.3 1.6

1-S C strip 39 1.226 40.8 1.0 1.2
2-B F bar 16 0.750 52.2 0.1 0.9
2-S F strip 28 0.936 14.3 1.9 4.0
3-B A bar 156 1.222 41.7 1.1 1.3
3-S A strip 27 1.228 36.3 1.2 1.4
4-B AC bar 70 1.106 44.6 0.8 1.0
4-S AC strip 23 0.924 33.6 0.5 0.8
5-B AF bar 39 1.154 26.7 1.5 1.8
5-S AF strip 14 1.195 25.6 1.5 1.8
6-B C, A, AC, AF bar 261 1.201 42.5 1.0 1.2
6-S C, A, AC, AF strip 98 1.165 38.2 1.0 1.2
7-B C, F bar 109 1.241 34.0 1.4 1.6
7-S C, F strip 67 1.171 34.6 1.1 1.3
8-B all bar 278 1.194 42.3 1.0 1.2
8-S all strip 173 1.149 36.6 1.0 1.2
9-B Seracino et al. [10] C bar 98 1.293 27.4 1.2 2.4
9-S C strip 39 1.380 47.5 0.9 1.8
10-B F bar 16 0.750 52.2 0.1 0.9
10-S F strip 28 0.936 14.3 1.9 4.0
11-B A bar 156 1.272 36.3 0.9 1.9
11-S A strip 27 1.101 27.6 0.8 2.1
12-B AC bar 70 1.503 35.5 1.3 2.3
12-S AC strip 23 0.910 16.7 0.6 2.2
13-B AF bar 39 1.204 25.3 1.0 2.3
13-S AF strip 14 0.967 18.1 0.7 2.3
14-B C, A, AC, AF bar 261 1.341 36.1 1.0 2.1
14-S C, A, AC, AF strip 98 1.155 42.4 0.6 1.6
15-B C, F bar 109 1.309 28.7 1.4 2.6
15-S C, F strip 67 1.283 41.9 1.1 2.1
16-B all bar 278 1.351 35.7 1.3 2.3
16-S all strip 173 1.097 38.4 0.9 2.0
17-B Zhang et al. [7] C bar 98 1.213 28.5 1.0 1.5
17-S C strip 39 1.147 28.0 0.9 1.6
18-B F bar 16 0.750 52.2 0.1 0.9
18-S F strip 28 0.936 14.3 1.9 4.0
19-B A bar 156 1.036 36.6 0.5 1.0
19-S A strip 27 0.850 23.9 0.2 0.9
20-B AC bar 70 1.173 45.2 0.6 1.0
20-S AC strip 23 0.888 14.9 0.5 1.4
21-B AF bar 39 0.972 33.0 0.4 0.9
21-S AF strip 14 0.816 26.2 0.1 0.7
22-B C, A, AC, AF bar 261 1.085 40.1 0.5 1.0
22-S C, A, AC, AF strip 98 0.969 29.6 0.5 1.1
23-B C, F bar 109 1.229 31.6 1.2 2.1
23-S C, F strip 67 1.160 25.6 1.5 2.6
24-B all bar 278 1.111 40.6 0.7 1.5
24-S all strip 173 0.988 29.6 0.9 1.9

Note: For ACI, the above results are presented under load combination of 1.2D + 1.6L; and ϕ = 1; and for CSA, the calculations are for 1.25D + 1.5L and
ϕc = ϕf = 0.65.

state is calculated and reported for each state. Fig. 3 shows the details In brief, the main concept developed herein is to employ the relia-
of the systematic procedure leading to calculation of β for each state in bility analysis technique (here, FOSM) to enhance the safety of three
the form of a flowchart. bond strength equations existing in the literature on NSM FRP. This
In Table 6, for each of the evaluation states, the mean (μ) and the goal is achieved by proposing new strength reduction factors for a fea-
coefficient of variation in percentage (δ) of ξm are reported, along with sible range of target reliability indices varying from 3.5 to 4.4. In order
the calculated β using the load and resistance factors of the ACI guide- to calibrate the resistance models, the concepts of both global as well
lines and CSA standard, respectively. Thereby, for each state, two val- as partial reduction factors are utilized according to ACI and CSA ref-
ues of β are reported under ACI and CSA columns in Table 6, erences, respectively. The probabilistic characteristics of the variables
accounting for the differences existing between ACI and CSA in load involved in the problem are extracted from the available literature.
combination factors as well as resistance reduction factors. The calcu- Besides, the statistical parameters of the model uncertainty are evalu-
lated values of β in Table 6 correspond to the basic load combination ated based on a large experimental database collated from the open lit-
including dead and live loads related to the guidelines and standard erature. The test database is carefully selected to include only direct
code as provided in Table 4. The resistance factor is assumed to be pullout tests while the data related to beam tests, mechanical anchor-
ϕ = 1 for the current ACI guidelines, with the material resistance fac- age, effects due to adjacent NSM bars/strips, and fatigue loading are
tors for concrete and FRP taken as ϕc = ϕf = 0.65 as suggested by the excluded. The load combination factors for dead, live, snow, and wind
CSA standard. loadings are taken in accordance with ACI and CSA reference codes.

9
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Fig. 2. χ2 Goodness-of-fit test of model uncertainty factor ξm .

The earthquake load effect, which is mainly resisted by a parallel sys- provided. In this regard, it is assumed that the failure probability in
tem consisting of several members rather than a single component, is bond should be one fifth of that in the principal mechanism, that gives
beyond the scope of this paper. The present article is not intended to βT in the range of 3.5–4.4 for bond, corresponding to βT = 3.0–4.0 for
suggest a specific target reliability index although some guidance is the main mechanism. The obtained results are presented for bars and

10
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Fig. 3. Diagram of calculation of β.

11
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

strips, separately, and under various individual or combined failure to the experimental data. It is expected that the reliability of the bond
modes. models improve after recalibration of the models take place in future
using the existing larger dataset in the literature. Alternatively, new
4. Safety evaluation of the existing bond models bond formulations can be established based on a better understanding
of the underlying mechanics. Such analysis is not in the scope of this
Table 6 summarises the probabilistic parameters of the model article; however, it could be a suggestion for a possible future study.
uncertainty factor and the reliability index. The three combined failure In this study, the original equations of the existing bond models are
modes in Table 6 recognise that the Seracino and Zhang models kept unchanged, and to reach a desirable level of safety, the β values
neglect the adhesive related failure modes, namely A, AF, and AC, con- are adjusted through calibrating the strength reduction factors. The
sidering only C and/or F modes separately or in a combined manner. proposed strength reduction factors are eventually multiplied by the
The combined mode of C and F is reported under states 7, 15, and original equations of the existing models to give the design value of
23 in Table 6. Conversely, the ACI model considers failure modes in the bond strength. In order to provide better comparisons among sim-
two categories including F and/or all other modes (i.e., C, A, AC, ilar states with different cross section shapes, the numerical values of
and AF), as reported under states 6, 14, and 22 in Table 6. In addition, Table 6 are illustrated in Fig. 4, which presents the β values of different
all five combined failure modes are reported under states 8, 16, and evaluation states for bars versus those for strips.
24. In brief, the combined failure modes related to Seracino and Zhang When considering the states with respect to the failure modes, for
models are C and F, whereas those related to ACI model are all of five failure mode F, given that the mechanics behind this failure mode is
modes or all modes without F. Therefore, the majority of the results rather straightforward, the three bond models give the same equation
presented in the upcoming sections are developed for states 6, 7, 8,
14, 15, 16, 22, 23, and 24.
The broad range of calculated μ and δ for ξm across the evaluation
states (Table 6) can be analyzed based on certain categories of varia-
tion when comparing similar states. The significant differences
observed in the ξm value of similar states with different bond models
can be due to major dissimilarities in the idealizations, assumptions,
and the approaches with regards to the mechanics of NSM FRP bond
among the studied bond models.
The mean value of ξm varies from 0.75 for the FRP rupture failure
mode for FRP bars for all three models to a value equal to 1.50 for state
12‐B which corresponds to the AC failure mode for the Seracino
model. A value of ξm less than 1 implies that the experimental value
is less than the model prediction, meaning an unsafe situation of the
bond model. It is worth noting that the coefficient of variation of ξm
for all states differs from 14.3% to 52.2%, with the lower and upper
bounds corresponding to states associated with strips and bars in the
FRP rupture failure mode, respectively. The β value for the load and
resistance factors calculated based on ACI guidelines is between 0.1
and 1.9, whereas β values for those of the CSA code are between 0.8
and 4.0. The higher values for β of CSA code compared to those of
ACI guidelines are expected since the resistance reduction factors in
ACI guidelines are taken equal to unity whereas those in CSA standard
are equal to 0.65. The difference in the β value implies different levels
of safety in the relations for each mode of failure and FRP cross‐
sectional shape. These significant differences in the β value for differ-
ent states and codes indicate the importance of a comprehensive reli-
ability analysis of the NSM bond failure. In the following sections a
more detailed analysis considers the parameters of the study included
in the evaluation states, separately.
Regardless of the differences between the β values, for almost all
states, the produced β values are clearly low compared to those of
the conventional failure mechanisms (examples in Table 5) with a sub-
stantial distance away from 3.8 or 4.2. This is attributed to two factors:
the inaccuracy associated with the existing bond models for NSM FRP,
which requires development of new bond models with a higher accu-
racy; and also, the lack of proper values for the strength reduction fac-
tors in the current guidelines on NSM FRP bond equations. It is
however known that even for the models with an acceptable level of
accuracy, a reliability analysis is inevitable in order to tune the
strength reduction factors to achieve the target reliability index. The
deviation of the existing bond models from the experimental results
can be seen from a rather high value of δ of ξm and low value of μ of
ξm. In general, the cases with μ of ξm as close as to unity and δ of ξm
as low as possible are more desirable. In meeting these conditions, Fig. 4. Reliability index (β) calculated for strips vs. bars for different models
one possible way is to refine the bond model by altering its formula- and various states (a) under ACI load and resistance factors (b) under CSA load
tion or the associated coefficients such that its predictions get closer and resistance factors.

12
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

for failure mode F that is applied to both bars and strips (see Table 1). In Figs. 5 and 6, variations of β are provided versus changes in the
Thus, μ and δ of ξm of all three bond models are the same for a partic- strength reduction factors for the basic load combination of dead and
ular cross‐sectional shape, with correspondingly similar values of reli- live loads by use of load factors of ACI and CSA reference codes,
ability index, β . These results are valid for load and resistance factors respectively. In these figures, both FRP cross sectional shapes are con-
of both ACI and CSA guidelines. Moreover, for failure mode F, the sidered for states of C&F and All failure modes. The presented figures
mean of ξm , μ, is significantly lower than 1 for the states corresponded will help to identify strength reduction factors in relation to the
to FRP bars as compared to strips (Table 6). In fact, in this failure desired βT. In addition, in Table 7, the strength reduction factors cor-
mode, the coefficient of variation ,δ, of ξm is significantly greater for responded to different target reliability indices presented in Table 5
FRP bars. These statistical measures lead to significantly dissimilar are revealed. Although here a βT equal to 4.2, that is a failure probabil-
reliability levels, as indicated by the reliability index β , for the three ity equivalent to one‐fifth of βT = 3.8 (see Table 5), has been proposed
models in failure mode F for FRP bars and strips (Table 6 and and utilized in different analyses of this study, the results of the study
Fig. 4). This result may be affected partly by the significantly limited have been presented in a way that can be used for a feasible range of
numbers of test data available for FRP bars in this particular failure βT.
mode. In Table 7, and Figs. 5 and 6, it can be seen that the strength reduc-
In case of the ACI model, the β values for FRP bars and strips are the tion factors are proposed in a very strict manner. This is expected con-
same for the state numbers 6, and 8, which are the failure modes that sidering the rather low reliability of the existing bond models, raising a
the ACI model aims to consider. This holds true for both ACI and CSA serious need to develop more accurate bond models in the future. In
design references. This could be due to the fact that the ACI model is the following paragraphs, detailed discussions will be provided regard-
the only model among the three studied bond models that was cali- ing the load and resistance factors of ACI and CSA references in this
brated to be used both for FRP bars and strips. As mentioned earlier, matter.
Seracino and Zhang models were developed for FRP strips only. As for the ACI bond model, two combined failure modes are of
In case of the Seracino model, in all failure modes except for the interest, including states 6 and 8. The difference between the two
failure mode F, the β value for bars is greater than that for strips for states 6 and 8 is that the latter considers all failure modes but the for-
the ACI load factors (Fig. 4). In particular, for the state numbers 15 mer excludes the mode of FRP rupture while considering all other
and 16 (the main failure modes considered by Seracino and Zhang modes. In order to address the limitation put forth by the ACI
models), the β values for bars is greater than those for strips, for both 440.2R‐17 on the tensile strain of NSM FRP bar/strip in order to
ACI and CSA references, despite the fact that the Seracino model is account for debonding, the state 6 is evaluated by the ACI bond
mainly calibrated for strips. strength equation using the development length based on the FRP
Failure modes 23 and 24, which are the basis of developing the stress of 0.7ffu in accordance with ACI 440.2R‐17. As mentioned ear-
Zhang model, the β values for strips are greater than those for bars lier, the use of FRP stress limit of 0.7ffu is applicable only when FRP
for both ACI and CSA load and resistance factors, as illustrated in rupture is not the case since when FRP rupture is included such as
Fig. 4, and consistent with the Zhang model developed for the strip in state 8, the full tensile strength of FRP (i.e., ffu) is utilized. Table 8
case originally. However, for the other evaluation states (except for compares reliability indices and ϕ values of the ACI bond model under
F failure mode) and for the ACI load factors, the situation alters as the ACI reference load for state 6 in two conditions: a) the FRP stress
the β values for strips are no longer greater than those for bars. As limit taken as 0.7ffu ; b) the FRP stress limit equal to ffu , in which the
for the CSA load and resistance factors, the β values of strips in all latter is presented for the sake of comparison. It is concluded from
evaluation states are either greater than or very close to those of Table 8 that when the required development length decreases due to
bars. lesser value of the effective stress in NSM FRP bar/strip (i.e., 0.7ffu ver-
In this study, when considering all the states related to all failure sus ffu), the reliability index increases from 1.0 to 1.1. However, it is
modes, it can be seen that the ACI model is resulting in similar β values still far from the acceptable target reliability indices. Furthermore, to
for FRP bars and strips. However, the results of this study show that achieve the target reliability indices listed in Table 8, the suggested
this is not exactly the case for Seracino and Zhang models. It is inter- reduction factor ϕ for “case a” (either 6‐B or 6‐S) is slightly greater
esting to note that the β values related to the Seracino model is greater than that for “case b”, of which the difference is in order of 0.01, that
for FRP bars rather than that of the states corresponded to FRP strips, is, however, negligible for practical use of ϕ factors. It is noted that by
as explained earlier. This shows that Seracino model is more reliable to inclusion of 0.7ffu versus ffu, the associated uncertainty is not affected
predict the results corresponded to NSM FRP bars rather than those of although the nominal value is altered. In addition, with reference to
NSM FRP strips. However, the reliability of Zhang model is higher to Fig. 5, which compares states 6 (case a) and 8, it is seen that to arrive
predict the NSM FRP strips bond strength than that of NSM FRP bars, at a specific target reliability index, the ϕ factor required in state 8 is
when considering all failure modes. When comparing the δ of ξm in the less than that in state 6 (case a) for FRP bars, whereas ϕ for state 8 is
case of all failure modes, it can be seen that Seracino model leads to slightly less than that for state 6 (case a) in the case of FRP strips. With
the minimum δ of ξm and the ACI model leads to maximum δ of ξm the aim of proposing an individual ϕ to cover both states, the lesser
for the case of bars. As for the strips, δ of ξm for the Zhang model is value of ϕ (i.e., corresponding to state 8) is suggested since it lies on
the lowest while that for the Seracino model is the greatest one. the conservative side. Thus, in the rest of the paper, the results related
In the following sections of analysis, the results will be mainly to only state 8 are presented.
focused on the C&F (state designations start with 7, 15, and 23) and In this article and under the ACI reference, for ϕ used in the ACI
All failure modes (i.e., the states 8, 16, and 24). This is due to the fact model and for βT = 4.2, values between 0.24 and 0.34 have been cal-
that the probabilistic outputs produced by these states correlate better culated for different evaluation states. When considering All failure
with the approaches of the studied bond models while covering a great modes in specimens tested with NSM FRP strips (state 8‐S), ϕ is calcu-
number of tested specimens. lated equal to 0.26. It should be noted that ϕ values for Seracino and
Zhang models and for βT = 4.2, are calculated between 0.23 and 0.33
5. Calibrations of strength reduction factors and 0.18 to 0.41, respectively, for the studied states. These ϕ values
are eventually multiplied by the existing bond equation. It is can be
In this section, with regards to the inconsistencies in β for the cho- seen that different assumptions and limitations of each model leads
sen states corresponded to C&F and All failure modes, a reliability‐ to various ranges of ϕ for the three models.
based analysis will be provided to calibrate the strength reduction For the CSA code, certain limitations exist for calibration of the
factors. resistance or strength reduction factors. In this code, partial strength

13
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Fig. 5. Reliability index (β) vs. strength reduction factor (ϕ) in ACI 440.2R-17 provisions (a) bars (b) strips.

Fig. 6. Reliability index (β) vs. strength reduction factor (ϕm) in CSA S806-12(R17) provisions (a) bars (b) strips.

Table 7
Proposed strength reduction factor (ϕ and ϕm ) for ACI 440.2R-17 and CSA S806-12(R17) provisions.

State Name βT ¼ 3:5 βT ¼ 3:9 βT ¼ 4:2 βT ¼ 4:4

ϕ(ACI) ϕm (CSA) ϕ(ACI) ϕm (CSA) ϕ(ACI) ϕm (CSA) ϕ(ACI) ϕm (CSA)

7-B 0.44 0.47 0.38 0.40 0.34 0.36 0.30 0.33

7-S 0.38 0.40 0.32 0.34 0.28 0.29 0.25 0.27
8-B 0.32 0.34 0.27 0.29 0.24 0.25 0.21 0.23
8-S 0.35 0.36 0.29 0.31 0.26 0.27 0.23 0.25
15-B 0.43 0.71 0.37 0.61 0.33 0.54 0.29 0.50
15-S 0.33 0.53 0.27 0.44 0.24 0.39 0.33 0.35
16-B 0.36 0.59 0.30 0.49 0.26 0.43 0.23 0.39
16-S 0.31 0.51 0.26 0.43 0.23 0.38 0.20 0.34
23-B 0.37 0.56 0.31 0.47 0.27 0.42 0.24 0.38
23-S 0.52 0.76 0.46 0.67 0.41 0.61 0.38 0.57
24-B 0.25 0.38 0.21 0.31 0.18 0.27 0.15 0.24
24-S 0.39 0.56 0.34 0.49 0.30 0.44 0.27 0.41

Note: For ACI, the above results are presented under load combination of 1.2D + 1.6L; and for CSA, the calculations are based on 1.25D + 1.5L and
ϕc = ϕf = 0.65.

Table 8
The effect of FRP stress limit on the reliability of ACI bond model.

State Name FRP stress limit ξm β ϕ

μ δ (%) βT ¼ 3:5 βT ¼ 3:9 βT ¼ 4:2 βT ¼ 4:4

6-B 0.7 ffu* 1.229 40.5 1.1 0.34 0.28 0.25 0.23
ffu† 1.201 42.5 1.0 0.33 0.27 0.24 0.22
6-S 0.7 ffu * 1.250 37.0 1.1 0.35 0.29 0.26 0.24
ffu† 1.165 38.2 1.0 0.33 0.28 0.25 0.23

Note: The above results are presented under load combination of 1.2D + 1.6L according to ACI reference. The combined failure mode of C, A, AC, AF is
considered.
* The FRP stress limit is taken as 0:7f fu as per ACI 440.2R-17 to account for debonding.
†
The FRP stress limit is equal to f f u , that is full utilization of FRP strength.

14
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

reduction factors (i.e., ϕc for concrete and ϕf for FRP materials) have the reliability of the two design references reach a consistent level;
been used that are fixed in all sections of the code. Since changing 3) when comparing different load conditions for each standard, sepa-
of the partial strength reduction factors is not a viable option, a mod- rately; the reliability indices show a steady behaviour; and 4) the reli-
ifying strength reduction factor (ϕm) has been used in this article to ability indices for each state to reach the target reliability level for
address the expected reliability of the NSM FRP bond strength. The each standard.
modifying coefficient is multiplied to the bond strength equation while Figs. 9 and 10 verify the effect of the proposed strength reduction
ϕc and ϕf are used in the equation with unchanged values. In this way, factors to reach the prescribed targeted objectives with respect to the
ϕm proposed for the ACI, Seracino, and Zhang models is calculated reliability of the design references, with the curves shown in Figs. 9
between 0.25 and 0.36, 0.38 to 0.54, and 0.27 to 0.61, respectively, and 10 correspond to those illustrated in Figs. 7 and 8, respectively.
for the selected states and for βT = 4.2 when the Canadian standard The major difference between Figs. 9 and 10 from Figs. 7 and 8 is that
is considered (Table 7). It should be noted that similar approaches to the results shown in Figs. 9 and 10 are produced after the newly pro-
solve the issue with fixed partial strength reduction factors have been posed calibrated strength reduction factors for the two design refer-
selected by other researchers [12]. For the CSA standard, ϕm can easily ences are applied. It should be noted that these figures are prepared
be applied to the Pn equation and can be used in combination of ϕc and for βT = 4.2, with alternative values of βT displaying similar features.
ϕf to provide a reasonable level of safety to the equations. Meanwhile, Figs. 9 and 10 clearly show that all the targeted objectives are
if the Canadian standard is planning to use either the Seracino or the achieved by proposing the new strength reduction factors. The reliabil-
Zhang bond model in the future revisions of CSA S806, the modifying ity levels for different evaluation states and for various load ratios are
strength reduction factors in Table 7, and Figs. 5 and 6, are suggested. now consistent and comparable. In addition, the reliability levels of
As a guiding principle [28], the design should force the probability the two design references, despite the differences in the order of
of the failure in bond to be less than the probability of failure in the involvement of the strength reduction factors, attain a consistent level.
main load‐carrying mechanism. In this context, it is important to know Moreover, the reliability of the studied states with respect to different
if the NSM FRP is being used for flexural or shear strengthening of a load combinations are now reached a consistent level of reliability.
reinforced concrete member. By selecting the target reliability index, This shows that by calibrating the strength reduction factors for the
the corresponding strength reduction factors may be extracted from load combination related to the live load, the reliability levels of the
Figs. 5 and 6. other load combinations corresponded to the snow and wind load also
are improved. Finally, the proposed strength reduction factors yield
the target reliability indices.
6. Reliability evaluations with respect to load combinations
7. Design example
In this section, the reliability of the studied bond models is investi-
gated before and after the calibrations of the strength reduction factors The results presented in this paper enable the designer to select a
are applied. This takes place by considering different load combina- strength reduction factor to meet a user‐defined target reliability index
tions of live, snow, and wind loads as recommended by the ACI and when utilizing each of the three bond models considered herein. Also,
CSA design references with respect to the selected failure modes corre- this can be done based on either ACI or CSA reference accounting for
sponded to C&F and All failure modes. Figs. 7 and 8 show the variation differences in the load factors and in the global or partial resistance
of the reliability index in the aforementioned states versus changes in factors. The obtained results presented in the form of Figs. 5 and 6
the loading based on the ACI and CSA provisions, correspondingly. or Table 7 provide the means to choose ϕ for a practical range of βT
Figs. 7 and 8 are presented before applying the calibrated strength varying between 3.5 and 4.4. This work is not, however, intended to
reduction factors for the two design references. The changes in the propose a specific target reliability index although some recommenda-
loading are defined as follows for each figure. In Fig. 7‐a and 7‐b, vari- tions were conducted in the previous sections (i.e., Section 3.4).
ations of β are illustrated versus the changes in the ratio of the live In this section, the NSM FRP bond equation currently incorporated
load to the total load applied to NSM FRP, calculated using in the ACI design code (see Table 1) is used for a numerical example.
1.2D + 1.6L load combination. In Fig. 7‐c and 7‐d, the horizontal axis Among the three bond strength equations and two design codes con-
corresponds to the ratio of the snow load to the total load applied to sidered in this paper, the ACI bond model described in the ACI 440
NSM FRP calculated using 1.2D + 1.6S. Similarly, in Fig. 7‐e and 7‐ is the only equation available in guidelines, as stated earlier. Given
f, the horizontal axis shows the variations to the ratio of the wind load PD and PL as the dead and live loads, respectively, which are tensile
to the total load applied to NSM FRP calculated using 1.2D + 1.6 W. forces applied to an NSM FRP bonded joint, the required diameter of
Fig. 8‐a to ‐f follow the same pattern used for the load ratios in Fig. 7‐a FRP bar is to be determined. Subsequently, the required bond length
to ‐f, respectively, with the exception that CSA standard load combina- of NSM FRP bar is also calculated. The example includes three bond
tions, as shown in Table 4, are used instead of the used load combina- models (i.e., ACI, Seracino, and Zhang) under both ACI and CSA refer-
tions of the ACI guidelines in Fig. 7. ences and for two levels of target reliability index, namely 3.9 and 4.2.
When comparing the curves in Figs. 7 and 8 separately for different The factored load combination according to ACI design reference is
states, it can be seen that the values of the corresponding curves for equal to 1:2PD þ 1:6PL . In order to reach an optimum design, the fac-
each figure are very different from one another. The differences are tored load is set equal to the design strength with the expressions pre-
less significant between the curves corresponded to the ACI guidelines sented in Table 1, as follows:
compared to those related to the CSA code. However, the reliability
level in the ACI guidelines is less than that in the CSA code with αL;A ¼1
1:2PD þ 1:6PL ¼ ϕPn;A ¼ ϕαL;A Pt ! 1:2PD þ 1:6PL ¼ ϕf fu Af
respect to the load combination in all studied states, with many of  
the calculated values of β for the figures corresponded to the ACI db 2
¼ ϕf fu π ð4Þ
guidelines close to one. Considering the fact that the scope of the 4
two standards is similar, it was expected that the safety levels of them
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
are more or less similar. By calibrating the strength reduction factors 4ð1:2PD þ 1:6PL Þ
proposed in this study, four major improvements in the reliability of db ¼ ð5Þ
ϕπf fu
the studied standards were targeted as follows: 1) the reliability
indices for varying load ratios and different evaluation states follow where ϕ is the strength reduction factor calculated in this study for a
a consistent pattern when each reference is considered separately; 2) specific target reliability index as listed in Table 7. It is also assumed

15
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Fig. 7. Reliability index (β) vs. load variables by ACI 440.2R-17 provisions in the original conditions.

that αL;A ¼ 1 for an economical design since the bond length beyond the state No. 7‐B in Table 7) are to take place. For other cases such as
effective length (i.e., Le,A in Table 1) no longer increases the bond FRP strips or consideration of all failure modes, the relevant ϕ values
capacity while the existing length shorter than that results in less bond can be taken from Table 7. For the above‐mentioned ϕ values, the cor-
strength. responding diameters of FRP bars are determined from Eq. (5) as:
0
As for numeric values, PD = 10 kN, PL = 20 kN, f c ¼ 25 MPa, db = 9.9 and 10.5 mm, respectively. The effective lengths of bonded
ffu = 1500 MPa, and Ef = 100 GPa are considered in this example. FRP bars are calculated as: Le,A = 539 and 570 mm, whereas the con-
The current version of ACI 440 does not offer any strength reduction sumed amounts of FRP materials become as follows: Vf = 41602 and
factor, that is ϕ = 1. This yields to db = 6.11 mm from Eq. (5). In addi- 49155 mm3, respectively. As seen, the FRP consumption increases
tion, the required length of bonded NSM FRP is calculated as Lb = Le, with the third power of FRP bar diameter in the ACI bond model. Also,
A = 332.2 mm with respect to αL;A ¼ 1 . Hence, the total volume of the
if the FRP stress is limited to 0.7 ffu, then the required diameter is

2 0 pffiffiffiffiffiffiffi
consumed FRP bars becomes V f ¼ π d4b :Le;A = 9745 mm3. If the increased as db ¼ db = 0:7 . Thus, for βT = 3.9 and 4.2, we have:
0 0 0
db ¼ 11.8 and 12.5 mm; Lb = 451 and 477 mm; and V f ¼ 49,711
debonding governs the failure, the effective stress of FRP bars is lim-
ited to 0.7ffu as per ACI 440.2R‐17, as explained earlier. In the case and 58737 mm3, respectively.
that the FRP stress is reduced from ffu to 0.7ffu, the cross section area Similar calculations can be conducted for Seracino and Zhang bond
of FRP bar should be increased such that the same pullout force can models when incorporated in the ACI design guideline. The required
be carried by NSM FRP system. Thereby, the required diameter of diameter of FRP bars for Seracino bond model (see Table 1) is deter-
0 pffiffiffiffiffiffiffi mined by the following expression, which can be solved based on a
FRP bar becomes equal to db ¼ db = 0:7 ¼ 7:3mm . Accordingly, the
trial and error procedure:
required bond length is reduced to
pffiffiffiffiffiffiffi  0:25
0 0
Lb ¼ Le;A ¼ Af  0:7f fu =ð6:9pf Þ ¼ 0:7Lb ¼ 278mm (see Table 1), and 2db þ 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1:3275ð1:2PD þ 1:6PL Þ
db ð6db þ 4Þ ¼ 0 0:33 pffiffiffiffiffi
ð6Þ
0
the total FRP volume equals to V f ¼ 11645 mm3 . 2db þ 2 ϕf c Ef
In the next step, various values of strength reduction factors includ-
As for Zhang bond model, the FRP bar diameter is calculated as
ing ϕ = 0.38 and 0.34 are considered that in turn correspond to
follows:
βT = 3.9 and 4.2, according to Table 7. These values are related to
the case of FRP bars and when combined failure modes C and F (i.e.,

16
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Fig. 8. Reliability index (β) vs. load variables by CSA S806-12(R17) provisions in the original conditions.

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u guidelines for βT = 3.9 and for bond models of ACI, Seracino and
u
3 ð1:2PD þ 1:6PL Þ
2
db ¼ t ð7Þ Zhang are as follows: db = 11.8, 15.8, and 16.6 mm; Lb = 416, 369,
2 0 0:619
1:2πϕ E f f c and 281 mm; and Vf = 45346, 72120, and 60552 mm3, respectively.
It is noticed that in both above‐mentioned equations, the dimen- As for βT = 4.2, the CSA results for bond models of ACI, Seracino and
sions of the groove are taken as hg ¼ wg ¼ 2db . Moreover, as stated Zhang are as follows: db = 12.4, 17.1, and 17.9 mm; Lb = 439, 385,
earlier, the existing bond length is assumed to be equal to the effective and 292 mm; and Vf = 53110, 88727, and 73037 mm3, respectively.
bond length, i.e., αL;S ¼ αL;Z ¼ 1 . For the numeric values mentioned
above, and for the example values of βT = 3.9 and 4.2, the results 8. Conclusions
are as follows. For Seracino bond model (state No. 15‐B in Table 7),
ϕ = 0.37 and 0.33, db = 16.9 and 18.3 mm, Le,S = 309 and In this paper, three bond strength models available in the literature
321 mm, and Vf = 69341 and 84159 mm3 , respectively, whereas for evaluation of the capacity of NSM FRP to concrete are investigated
for Zhang bond model (state No. 23‐B in Table 7), ϕ = 0.31 and within a structural reliability context. The studied models include ACI,
0.27, db = 19.4 and 21.3 mm, Le,Z = 245 and 256 mm, and Seracino, and Zhang models. The reliability analysis is conducted
Vf = 72279 and 90993 mm3, respectively. It is seen that in three con- under two design reference codes namely, ACI and CSA, which intro-
sidered bond models, by reducing ϕ value to achieve higher reliability, duce different load factors used in combinations of dead, live, snow,
the diameter of FRP bar increases, as expected. Consequently, the and wind loads along with different strength reduction factors. The
effective bond length, which is proportional to the FRP bar diameter analysis accounts for differences between the approach of ACI and that
(see Table 1), increases too. This procedure can also be employed of CSA as the former considers global reduction factors (ϕ) applied to
for the bond models when incorporated to CSA design guidelines to the model equation while the latter employs the partial reduction fac-
a similar manner. It is noted that in the CSA design reference, the fac- tors (ϕc and ϕf , which are both equal to 0.65) applied to f’c and ffu. The
tored load is equal to 1:25PD þ 1:5PL and the material strength reduc- ACI equation is the only bond model among the three considered mod-
tion factors are set as ϕc = ϕf = 0.65 while the modifying strength els that is currently incorporated in a design code (i.e., ACI guidelines)
reduction factor (i.e., ϕm) should be taken from Table 7. In the case with ϕ = 1. In reliability‐based assessment of the models, the FOSM
of the numeric example mentioned above, the results of CSA design method is utilized taking into account the uncertainties inherent in

17
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Fig. 9. Reliability index (β) vs. load variables by ACI 440.2R-17 provisions in the calibrated conditions.

the geometrical and mechanical parameters included in each model in model the β values of strips are greater. These results hold true in
addition to the probabilistic nature of the loads. The model uncertainty both cases when the ACI and CSA design references are considered.
is evaluated by use of a large database of the experiments collated However, the results may change depending on the other individ-
from the open literature with careful considerations of reported failure ual failure modes under study.
modes. The reliability index is determined for each model in various 3. In the proposed approach, the strength reduction factors associated
evaluation states corresponding to individual or combinatory failure with the studied bond models are calibrated for a feasible range of
modes. The results are presented for NSM FRP bars and strips, sepa- βT under load and resistance reduction factors of both ACI and CSA
rately. The conclusions are as follows: design references. In doing so, a strength reduction factor ϕ is
applied to the bond strength equation for the ACI guidelines
1. The calculated β values for the studied bond models and for differ- whereas a so‐called modifying strength reduction factor (ϕm) is
ent failure modes and cross‐section shapes vary in the range of multiplied by the model equation for CSA design code while the
0.1–1.9 under the ACI load and strength reduction factors (i.e., partial strength reduction factors (i.e., ϕc for concrete and ϕf for
ϕ = 1), and in the range of 0.8–4.0 under the CSA design reference FRP materials) are kept unchanged. The proposed factors ϕ and
load and resistance reduction factors (i.e., ϕc = ϕf = 0.65), imply- ϕm for ACI and CSA design references, respectively, are determined
ing significant differences in the safety levels offered by the exist- for each model in the case of strips and bars, separately, for various
ing bond models in different evaluation states. In most cases, target reliability indices. Depending on the desirable value of
these β values are clearly inadequate compared to the common tar- βT, the proper value for the strength reduction factors can be
get reliability indices, raising the need for development of more selected.
accurate bond models, and meantime, the need for calibration of 4. The reliability‐based investigation of the bond strength models
the proper strength reduction factors to be incorporated in the cur- before and after applying the proposed strength reduction factors
rent design codes. for βT = 4.2 (as an example target reliability index) against differ-
2. When considering all failure modes together, the β values for ACI ent load combinations including dead, live, snow and wind indicate
model are the same for bars and strips. For Seracino model, the β that the proposed strength reduction factors in the current study
values of bars are greater than those of strips despite the fact that developed a consistent level of reliability for varying load ratios
Seracino model has been calibrated for strips, whereas for Zhang under different evaluation states and different load combinations.

18
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

Fig. 10. Reliability index (β) vs. load variables by CSA S806-12(R17) provisions in the calibrated conditions.

In the current study, the existing equations for the studied bond draft, Writing ‐ review & editing, Visualization, Supervision. Amir
models were kept unchanged and the desired level of reliability was Mofidi: Writing ‐ review & editing, Resources, Visualization, Project
achieved by proposing new strength reduction factors. Since the reduc- administration. Peter Gosling: Writing ‐ review & editing, Visualiza-
tion factors derived in this study have relatively low values to achieve tion, Supervision.
the target reliability indices, it is concluded that the model errors asso-
ciated with the current bond equations are relatively high, implying Declaration of Competing Interest
the need to develop new bond models with an improved accuracy.
The enhanced accuracy of new bond models can be achieved by devel- The authors declare that they have no known competing financial
oping equations based on better understanding of the mechanics interests or personal relationships that could have appeared to influ-
underlying the NSM FRP bond behaviour while accounting for the ence the work reported in this paper.
existing larger dataset in the literature. Such analyses could be a sug-
gestion for a possible future study. References
Funding
[1] Yao J, Teng JG, Chen JF. Experimental study on FRP-to-concrete bonded joints.
This research received no specific grant from any funding agency in Compos B Eng 2005;36(2):99–113.
the public, commercial, or not‐for‐profit sectors. [2] De Lorenzis L, Nanni A. Bond between near-surface mounted fiber-reinforced
Data availability polymer rods and concrete in structural strengthening. ACI Struct J 2002;99
(2):123–32.
The raw/processed data required to reproduce these findings can- [3] Rashid R, Oehlers DJ, Seracino R. IC debonding of FRP NSM and EB retrofitted
not be shared at this time due to technical or time limitations. concrete: plate and cover interaction tests. J Compos Constr 2008;12(2):160–7.
[4] Chaallal O, Mofidi A, Benmokrane B, Neale K. Embedded through-section FRP rod
method for shear strengthening of RC beams: performance and comparison with
CRediT authorship contribution statement existing techniques. J Compos Constr 2011;15(3):374–83.
[5] Mofidi A, Chaallal O, Cheng L, Shao Y. Investigation of near surface-mounted
Reza Aghamohammadi: Methodology, Validation, Formal analy- method for shear rehabilitation of reinforced concrete beams using fiber
reinforced–polymer composites. J Compos Constr 2016;20(2):04015048.
sis, Investigation, Writing ‐ original draft, Writing ‐ review & editing, [6] Mofidi A, Cheng L, Chaallal O, Shao Y. Shear strengthening of RC beams with NSM
Visualization. Kourosh Nasrollahzadeh: Conceptualization, Method- FRP—Influencing parameters and a theoretical model. Am Concr Institute Specl
ology, Validation, Formal analysis, Investigation, Writing ‐ original Publicat 2018;327:31–41.

19
R. Aghamohammadi et al. Composite Structures 272 (2021) 114132

[7] Zhang SS, Teng JG, Yu T. Bond strength model for CFRP strips near-surface [37] Nowak AS, Szerszen MM. Calibration of design code for buildings (ACI 318): Part
mounted to concrete. J Compos Constr 2014;18(3):A4014003. 1-Statistical models for resistance. ACI Struct J 2003;100(3):377–82.
[8] de Sena Cruz JM, Oliveira de Barros JA. Bond between near-surface mounted [38] El-Sayed A, El-Salakawy E, Benmokrane B. Shear strength of one-way concrete
carbon-fiber-reinforced polymer laminate strips and concrete. J Compos Constr slabs reinforced with fiber-reinforced polymer composite bars. J Compos Constr
2004;8(6):519–27. 2005;9(2):147–57.
[9] De Lorenzis L, Rizzo A, La Tegola A. A modified pull-out test for bond of near- [39] Michaluk CR, Rizkalla SH, Tadros G, Benmokrane B. Flexural behavior of one-way
surface mounted FRP rods in concrete. Compos B Eng 2002;33(8):589–603. concrete slabs reinforced by fiber reinforced plastic reinforcements. ACI Struct J
[10] Seracino R, Raizal Saifulnaz MR, Oehlers DJ. Generic debonding resistance of EB 1998;95(3):353–65.
and NSM plate-to-concrete joints. J Compos Constr 2007;11(1):62–70. [40] El-Sayed AK, El-Salakawy EF, Benmokrane B. Shear strength of FRP-reinforced
[11] Wang X, Cheng L. Bond characteristics and modeling of near-surface mounted concrete beams without transverse reinforcement. ACI Struct J 2006;103
CFRP in concrete. Compos Struct 2021;255:113011. [Link] (2):235–43.
[Link].2020.113011. [41] Peng H, Hao H, Zhang J, Liu Y, Cai CS. Experimental investigation of the bond
[12] Nasrollahzadeh K, Aghamohammadi R. Reliability analysis of shear strength behavior of the interface between near-surface-mounted CFRP strips and concrete.
provisions for FRP-reinforced concrete beams. Eng Struct 2018;176:785–800. Constr Build Mater 2015;96:11–9.
[13] Zhang D, Gu X-L, Yu Q-Q, Huang H, Wan B, Jiang C. Fully probabilistic analysis of [42] Kalupahana WKKG, Ibell TJ, Darby AP. Bond characteristics of near surface
FRP-to-concrete bonded joints considering model uncertainty. Compos Struct mounted CFRP bars. Constr Build Mater 2013;43:58–68.
2018;185:786–806. [43] Shield C, French C, Milde E. The effect of adhesive type on the bond of NSM tape to
[14] Shi J, Wu Z, Wang X, Noori M. Reliability analysis of intermediate crack-induced concrete. Special Publication. 2005;230:355–72.
debonding failure in FRP-strengthened concrete members. Struct Infrastruct E. [44] Nowak AS, Collins KR. Reliability of structures. CRC Press; 2012.
2015;11(12):1651–71. [45] De Lorenzis L, Lundgren K, Rizzo A. Anchorage length of near-surface mounted
[15] Triantafillou TC, Plevris N. Strengthening of RC beams with epoxy-bonded fibre- fiber-reinforced polymer bars for concrete strengthening-experimental
composite materials. Mater Struct 1992;25(4):201–11. investigation and numerical modeling. ACI Struct J 2004;101(2):269–78.
[16] Huang X, Sui L, Xing F, Zhou Y, Wu Y. Reliability assessment for flexural FRP- [46] Novidis D, Pantazopoulou SJ, Tentolouris E. Experimental study of bond of NSM-
Strengthened reinforced concrete beams based on Importance Sampling. Compos FRP reinforcement. Constr Build Mater 2007;21(8):1760–70.
Part B-Eng. 2019;156:378–98. [47] Novidis DG, Pantazopoulou SJ. Bond tests of short NSM-FRP and steel bar
[17] Liang H, Li S, Lu Y, Yang T. Reliability analysis of bond behaviour of anchorages. J Compos Constr 2008;12(3):323–33.
CFRP–concrete interface under wet–dry cycles. Materials. 2018;11(5):741. [48] Soliman SM, El-Salakawy E, Benmokrane B. Flexural behaviour of concrete beams
[18] He Z, Tian G-W. Reliability-based bond design for GFRP-reinforced concrete. strengthened with near surface mounted fibre reinforced polymer bars. Can J Civil
Mater Struct 2011;44(8):1477–89. Eng. 2010;37(10):1371–82.
[19] Coelho M, Neves L, Sena-Cruz J. Designing NSM FRP systems in concrete using [49] Bilotta A, Ceroni F, Di Ludovico M, Nigro E, Pecce M, Manfredi G. Bond efficiency
partial safety factors. Compos B Eng 2018;139:12–23. of EBR and NSM FRP systems for strengthening concrete members. J Compos
[20] Teng JG, De Lorenzis L, Wang Bo, Li R, Wong TN, Lam L. Debonding failures of RC Constr 2011;15(5):757–72.
beams strengthened with near surface mounted CFRP strips. J Compos Constr [50] Palmieri A, Matthys S, Taerwe L. Experimental investigation on fire endurance of
2006;10(2):92–105. insulated concrete beams strengthened with near surface mounted FRP bar
[21] YUAN H, WU Z, YOSHIZAWA H. Theoretical solutions on interfacial stress transfer reinforcement. Compos B Eng 2012;43(3):885–95.
of externally bonded steel/composite laminates. Doboku Gakkai Ronbunshu. 2001 [51] Ceroni F, Pecce M, Bilotta A, Nigro E. Bond behavior of FRP NSM systems in
(675):27–39. [Link] concrete elements. Compos B Eng 2012;43(2):99–109.
[22] Chen JF, Teng JG. Anchorage strength models for FRP and steel plates bonded to [52] Lee D, Hui J, Cheng L. Bond characteristics of NSM reinforcement in concrete due
concrete. J Struct Eng 2001;127(7):784–91. to adhesive type and surface configuration. In 6th International conference on FRP
[23] Yuan H, Teng JG, Seracino R, Wu ZS, Yao J. Full-range behavior of FRP-to-concrete composites in civil engineering (CICE2012). Rome, Italy; 2012.
bonded joints. Eng Struct 2004;26(5):553–65. [53] Lee D, Cheng L, Yan-Gee Hui J. Bond characteristics of various NSM FRP
[24] CAN/CSA S806-12 Design and Construction of Building Components with Fibre- reinforcements in concrete. J Compos Constr 2013;17(1):117–29.
Reinforced Polymers. Canadian Standards Association, Ontario;2017. [54] Palmieri A, Matthys S, Barros J, Costa I, Bilotta A, Nigro E, et al. Bond of NSM FRP
[25] ACI 440.2R-17. Guide test methods for fiber-reinforced polymer (FRPs) strengthened concrete: round robin test initiative. In 6th International conference
composites for reinforcing or strengthening concrete structures. American on FRP composites in civil engineering (CICE2012). Rome, Italy; 2012.
Concrete Institute 2017. [55] El-Gamal S, Al-Salloum Y, Alsayed S, Aqel M. Performance of near surface
[26] Oehlers DJ, Haskett M, Wu C, Seracino R. Embedding NSM FRP plates for mounted glass fiber reinforced polymer bars in concrete. J Reinf Plast Comp
improved IC debonding resistance. J Compos Constr 2008;12(6):635–42. 2012;31(22):1501–15.
[27] EN-1990, EC0 (Eurocode 0) – Basis of structural design; 1990 (2002). [56] Seo S-Y, Feo L, Hui D. Bond strength of near surface-mounted FRP plate for retrofit
[28] Darwin D, Idun EK, Zuo J, Tholen ML. Reliability-based strength reduction factor of concrete structures. Compos Struct 2013;95:719–27.
for bond. University of Kansas Center for Research Inc; 1995. [57] Sharaky IA, Torres L, Baena M, Miàs C. An experimental study of different factors
[29] ASCE/SEI (Structural Engineering Institute). “Minimum design loads for buildings affecting the bond of NSM FRP bars in concrete. Compos Struct 2013;99:350–65.
and other structures.” ASCE/SEI 7-16, Reston, VA, 2016. [58] Sharaky IA, Torres L, Baena M, Vilanova I. Effect of different material and
[30] National Building Code of Canada. Institute for Research in Construction. Ottawa, construction details on the bond behaviour of NSM FRP bars in concrete. Constr
Ont: National Research Council of Canada; 2015. Build Mater 2013;38:890–902.
[31] Hassan T, Rizkalla S. Investigation of bond in concrete structures strengthened [59] Bilotta A, Ceroni F, Nigro E, Pecce M. Strain assessment for the design of NSM FRP
with near surface mounted carbon fiber reinforced polymer strips. J Compos systems for the strengthening of RC members. Constr Build Mater
Constr 2003;7(3):248–57. 2014;69:143–58.
[32] Seracino R, Jones NM, Ali MS, Page MW, Oehlers DJ. Bond strength of near- [60] Zhang R, Xue X. A predictive model for the bond strength of near-surface-mounted
surface mounted FRP strip-to-concrete joints. J Compos Constr 2007;11(4):401–9. FRP bonded to concrete. Compos Struct 2021;262:113618. [Link]
[33] Zhang SS, Teng JG, Yu T. Bond–slip model for CFRP strips near-surface mounted to 10.1016/[Link].2021.113618.
concrete. Eng Struct 2013;56:945–53. [61] Turkstra CJ 1970 theory of structural design decisions [Waterloo, Ont.], [Solid
[34] Jawaheri Zadeh H, Mejia F, Nanni A. Strength Reduction Factor for Flexural RC Mechanics Division, University of Waterloo], [1970] University of Waterloo.; Solid
Members Strengthened with Near-Surface-Mounted Bars. J Compos Constr Mechanics Division.; SM study
2013;17(5):614–25. [62] Szerszen MM, Nowak AS. Calibration of design code for buildings (ACI 318): Part
[35] Okeil AM, Belarbi A, Kuchma DA. Reliability assessment of FRP-strengthened 2-Reliability analysis and resistance factors. ACI Struct J 2003;100(3):383–91.
concrete bridge girders in shear. J Compos Constr 2013;17(1):91–100. [63] International Organization for Standardization. General principles on reliability of
[36] Mirza SA, MacGregor JG. Probabilistic study of strength of reinforced concrete structures. ISO2394:2015, Geneva, Switzerland; 2015.
members. Can J Civ Eng 1982;9(3):431–48.

20