Engineering Structures 33 (2011) 1491–1502

Contents lists available at ScienceDirect

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

Fatigue behavior of damaged steel beams repaired with CFRP strips

Yail J. Kim a,∗ , Kent A. Harries b
a
Department of Civil Engineering, North Dakota State University, Fargo, ND, United States
b
Department of Civil and Environmental Engineering, University of Pittsburgh, Pittsburgh, PA, United States

article info abstract

Article history: This paper presents the flexural behavior of damaged steel beams repaired with carbon fiber reinforced
Received 1 November 2010 polymer (CFRP) strips. The damage is intentionally created by notching the tension flange of the beams.
Received in revised form Six beams are tested to evaluate the static and fatigue performance of the repaired beams with emphasis
16 December 2010
on local plasticity and the CFRP–steel interface. A three-dimensional finite element analysis (FEA) is
Accepted 20 January 2011
Available online 21 February 2011
conducted to predict the experimental behavior. A modeling approach is proposed to simulate the fatigue
response of the repaired beams, based on the strain-life method and cumulative damage theory. CFRP-
Keywords:
repair results in a recovery of static load-carrying capacity of the damaged beam to that of an undamaged
Beam beam. The stress range at the damage influences the fatigue life, damage propagation, and plastic strain
Fatigue development of the repaired beams. Fatigue-crack propagation across the web of the beams is not
Fiber reinforced polymer significant up to 50% of their fatigue life, whereas brittle web fracture follows beyond the threshold. A
Models bilinear fatigue response is observed at the CFRP-steel interface, whose magnitudes are dependent upon
Repair the number of fatigue cycles and the applied stress range. An empirical model is proposed to predict the
Steel fatigue behavior of the interface.
Strengthening © 2011 Elsevier Ltd. All rights reserved.

1. Introduction strength-to-weight ratios, ease and rapidity of erection, and re-

duced long-term maintenance expenses [7]. A significant advan-
The issue of infrastructure management and rehabilitation is tage of CFRP systems is that they are typically adhesively bonded,
one of the primary interests in civil engineering community. Con- rather than mechanically connected, to the substrate steel, result-
structed bridge members deteriorate because of aging, corrosion, ing in mitigation of the additional stress raisers associated with
increased service loads and traffic volume, use of deicing salts, bolt holes or welds. Although CFRP composites have been primarily
and collision of heavy trucks [1–3]. Over $180 billion is required used for repairing concrete structures [8], there are relatively lim-
to address structurally deficient or functionally obsolete bridges ited data available on strengthening damaged steel members using
in the United States [4]. Of particular interest are steel bridges CFRP [9–15]. Nonetheless, this body of work clearly indicates the
that account for more than 43% of the substandard bridges in the promise of this technique for strengthening damaged steel mem-
nation [5]. Steel bridges are susceptible to corrosion and fatigue
bers subject to monotonic loads. Cadei et al. [16] provides signifi-
cracks. Structural repair of damaged members provides a viable
cant guidance for such applications.
alternative to replacement, reducing downtime and cost. Tradi-
A number of recent studies have investigated the behavior
tional repair techniques for steel bridges include use of steel plates
of steel beams having damaged tension flanges repaired with
bolted or welded to damaged girders. Although the performance
CFRP patches and subject to fatigue loading [5,11,15,17–19].
of such repaired girders is generally acceptable, the repairs them-
selves may introduce new potential problems associated with in- Tavakkolizadeh and Saadatmanesh [5] assessed the fatigue behav-
creased dead load, corrosion (crevice and galvanic corrosion), and ior of intentionally notched steel beams (L = 1.3 m) patched with
the introduction of fatigue-sensitive details at the junction of ex- short CFRP sheets (L = 0.3 m). A design curve was generated to
isting and repair materials [6]. The application of carbon fiber re- estimate the fatigue performance of repaired steel beams. CFRP-
inforced polymer (CFRP) composites is proposed as an alternative patching resulted in improvement in fatigue life of damaged beams
to steel-plating repair methods. The advantages of CFRP materi- of up to 3.4 times over that of unrepaired beams. Nozaka et al. [11]
als include their non-corrosive characteristics, high stiffness- and considered the performance of combinations of two CFRP mate-
rials and five adhesive systems to enhance the fatigue behavior
of steel sections. They reported the greatest increase in fatigue
∗
Corresponding author.
strength resulting from the system combining the CFRP and ad-
E-mail addresses: [email protected] (Y.J. Kim), [email protected] hesive having the lowest moduli of elasticity of those considered.
(K.A. Harries). O’Neill et al. [20] also report improved fatigue performance when
0141-0296/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2011.01.019
1492 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502

using softer adhesive systems, attributing this to the greater redis- 3.1. Materials
tribution of stresses resulting from such systems. Deng and Lee [17]
studied fatigue behavior of small-scale steel beams (L = 1.2 m) W150 × 18 (US designation: W6 × 12) steel beams were used
retrofitted with CFRP patches. The initiation and development of for the test program. The measured yield strength of the beams was
cracks along the CFRP patch were monitored during cyclic loading. 393 MPa with a modulus of elasticity of 200 GPa. The 50 mm wide
An empirical stress–fatigue life (S–N) curve was constructed using by 1.4 mm thick CFRP strips used had a tensile strength of 2800
test results. Bocciarelli et al. [21] tested steel coupons bonded with MPa and modulus of 155 GPa. The epoxy adhesives had a tensile
CFRP plates in fatigue. A variety of stress ranges were simulated strength of 25 MPa and a modulus of 4.5 GPa.
to evaluate bond between the CFRP and steel. Progressive bond
failure at CFRP-steel interface was observed with increased fatigue 3.2. Specimen details
cycles. The importance of interface design to improve fatigue per-
formance of CFRP plates bonded to steel surface was discussed. The beam specimens, shown in Fig. 1, were designed so
Deng [18] proposed a model for crack initiation and crack growth that both steel fatigue and CFRP bond fatigue damage could be
prediction for CFRP-repaired steel beams in flexure, although this investigated. A total of six beams were tested as indicated in
model is calibrated only against cases of moderate cycle fatigue Table 1. A notch [Fig. 1(b)] was machined through the entire
(N < 2 × 105 ). In summary, all studies report an extended fatigue tension flange of Beams B through F at a location 152 mm to
life and a decreased crack growth rate provided the CFRP remained the right of midspan [Fig. 1(a)]. The notch serves as a stress
concentrator to (a) initiate a vertical crack in the steel web and
bonded to the steel substrate.
(b) initiate debonding of the CFRP propagating toward the right
Notwithstanding the foregoing, very limited information is
support [Fig. 1(a)]. The elastic stress concentration factor (typically
available on the fatigue behavior of steel structures repaired with
denoted k) for the notch provided was 3.55 [22]. The corresponding
CFRP composites; indeed the listing given above is believed to
stress concentration factor for fatigue, kf , is typically less than that
be exhaustive at the time of writing. Additionally, there has been for static loading. For the notch geometry provided, the elastic
no attempt to develop a numerical model for steel members stress concentration factor for fatigue, based on the estimation
repaired with CFRP composites in order to predict fatigue behavior, proposed by Neuber [23], is kf = 2.84. Thus, in this case, the
particularly the fatigue behavior of CFRP-steel interface. This paper presence of the notch is essentially equivalent to AISC [24] fatigue
addresses the flexural behavior of damaged steel beams repaired stress category E.
with CFRP strips, including the static and fatigue responses. Prior to bonding the 50.8 mm wide by 1.4 mm thick CFRP strips
The focus of the investigation is the local plasticity adjacent to to Beams C–F, the steel surface was prepared by sanding using
a damage location and the viscoelastic nature of the adhesive a 1500 sfpm (surface feet per minute) belt sander and a 40 grit
bonding agent which is critical to the hysteretic behavior of zirconia alumina belt. This preparation resulted in a sound, slightly
the CFRP-steel interface. A modeling approach is proposed to striated surface for bonding the CFRP strips. The thickness of the
predict the fatigue behavior of CFRP-repaired steel beams and is adhesive layer was approximately 1 mm.
incorporated into a three-dimensional finite element (FE) model.
3.3. Test set-up and instrumentation
2. Research significance
All beams were subject to three-point flexural loading over a
simply-supported span of 1830 mm [Fig. 1(a)]. A neoprene pad
Fatigue damage to steel girders is one of the most significant was placed in between the beam and support to reduce stress
problems affecting the service life of existing bridges. Fatigue- concentrations. Beams A–C were loaded monotonically to failure.
induced cracks are frequently observed in constructed bridge Beams D–F were subjected to a transient load, cycling from a
members and corresponding repair is conducted. The application minimum of 4.4 kN to a maximum of 35.6 kN, 22.2 kN, and 13.3 kN,
of CFRP composites to repair damaged steel girders is an emerging respectively, as listed in Table 1. The load ranges were equivalent
technique offering an attractive alternative to conventional repair to 0.73Py , 0.42Py , and 0.21Py , respectively, where Py was the load
methods such as steel-plating. Limited monotonic test results to cause yield at the root of notch [Fig. 1(b)] of the repaired
show the potential of such a repair method, whereas very limited control Beam C. The loading frequencies varied from 1 Hz to 2.5 Hz
information is available on the behavior of CFRP-repaired steel (Table 1), which are commonly-used ranges for testing steel or
members subjected to transient (fatigue) loads. No research has concrete beams in fatigue [17,25] yet low enough to avoid self
been reported from the perspective of numerically modeling heating of the CFRP. The applied load was recorded by a load cell
fatigue behavior. This paper proposes a systematic modeling in line with the actuator. One strain gage (Gage 1) was located
approach to predict the behavior of damaged steel beams subject horizontally at the root of the notch to examine the local behavior
to fatigue loading conditions, based on the strain-life method and of the critical section, as shown in Fig. 1(b). Additional gages
cumulative damage theory. An empirical model for the behavior (Gages 2–7) were bonded along the CFRP strip at 38 mm centers to
of the CFRP–steel interface in fatigue is proposed. Debonding monitor debonding propagation of the strip in the vicinity of the
propagation of the CFRP and crack growth across the steel section notch [Fig. 1(c)]. Fatigue crack and CFRP debonding propagation
are discussed. were measured manually at an applied load of Pmax at pauses in
the fatigue loading.

3. Experimental program
3.4. Results of monotonic tests

An experimental program, described in full by Harries et al. [15], For Beams A–C, a monotonic load was applied at midspan of
was carried out to validate the proposed modeling approach to the beams until failure occurred. The ultimate behavior of all three
be explained subsequently. Of particular interest was the behavior monotonic tests was flexural torsional buckling (which was not
of the damaged beams at the crack tip and the efficacy of CFRP- restrained due to considerations of companion fatigue tests). The
repair, subjected to monotonic and fatigue loads. The following behavior of interest in this experimental program, however, is
summarizes details of test beams, materials, and instrumentation. that in the elastic range and thus the ultimate behavior is not
 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502 1493

a b

Fig. 1. Experimental test set-up and notch detail: (a) test geometry; (b) machined notch in tension flange and location of Gage 1; (c) strain gages 2–7 located on CFRP
(reverse plan).

Table 1
Beam details.
Specimen Notch CFRP-repair Moment of Method Monotonic to Fatigue
inertia (106 mm4 ) failurea
Py Pu Stress ranged and loading Load (kN) Stress range Fatigue life
(kN) (kN) frequency (MPa)e

Exp –b 102.3 – – – –
Beam A None None 9.20
FEA 104.0 116.4 – – – –
Exp 31.1c 81.8 – – – –
Beam B Yes None 3.34
FEA 78.0 97.6 – – – –
Exp 42.6c 84.5 – – – –
Beam C Yes Yes 4.54
FEA 105.0 123.0 – – – –
Exp – – 0.73Py @ 1 Hz 4.4 – 35.6 274 20,000
Beam D Yes Yes 4.54
FEA – – N/A 4.4 – 35.6 274 15,000
Exp – – 0.42Py @ 2 Hz 4.4 – 22.2 158 152,380
Beam E Yes Yes 4.54
FEA – – N/A 4.4 – 22.2 158 150,000
Exp – – 0.21Py @ 2.5 Hz 4.4 – 13.3 81 1,703,020
Beam F Yes Yes 4.54
FEA – – N/A 4.4 – 13.3 81 2,100,000
a
Py = yield load; Pu = ultimate load.
b
lateral torsional buckling failure.
c
yield load at root of notch.
d
stress range based on yield of Beam C (i.e.: Py = 42.6 kN).
e
stress range at notched location.

investigated. Table 1 provides a summary of monotonic test results to cause 1% strain at the notch, was only 123% of that of Beam
as well as the moment of inertia, I, of the beams calculated at the B, while the ultimate load, following considerable debonding, was
location of the notch. For Beam C, this value includes the presence only 103%.
of the 50 mm CFRP strip. The presence of the notch in Beam B Debonding of the CFRP was identified from the gages arrayed
resulted in a beam stiffness of only 83% of that of the unnotched on the CFRP [Fig. 1(c)]. The stress in the CFRP developed rapidly,
Beam A. The addition of the FRP in Beam C restored the stiffness within 38 mm of the notch. Beyond this, stress was redistributed
only marginally to a value of 86% of the unnotched beam stiffness. back into the steel tension flange and the CFRP stress was reduced.
This result is consistent with those reported previously indicating Debonding resulted in the stress transfer from the CFRP moving
that such a repair has little impact on beam stiffness [26–28]. away from the notch and was evidenced by the relatively constant
The CFRP repair of Beam C increased the load required to distribution of stress across a longer length of debonded CFRP
cause yield at the root of the notch, Py , to 137% of that of the ‘bridging’ the notch. The CFRP debonding strain was approximately
unrepaired Beam B. This value is consistent with the increase in 3300 µε which is consistent with observations from other studies
elastic section modulus, Sb , affected by the CFRP. This improvement using the same CFRP/adhesive system on a steel substrate [29].
was consistent until the CFRP began to debond at which point, The maximum adhesive stress developed occurred between Gage
predictably, its efficacy began to deteriorate. For example, the load 3 (38 mm) and Gage 4 (76 mm) at an applied load of 66.7 kN. This
1494 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502

a b

c d
Fig. 2. Predictive model: (a) constitutive behavior of steel and CFRP; (b) interface model; (c) developed FEA model: (d) sensitivity analysis (Beam A).

value, calculated from the difference in CFRP stress between these has three translational degrees of freedom. An elastic-perfectly-
locations, was 11.6 MPa and is also consistent with findings from plastic material stress–strain relationship, having the properties
related studies [20]. given above, was used to model the steel. A linear stress–strain
relationship was established for the CFRP. Both material models
3.5. Results of fatigue tests are shown in Fig. 2(a). Poisson’s ratios used were 0.3 and
0.27 for the steel and CFRP, respectively. A nonlinear interface
Table 1 provides a summary of fatigue test parameters and element (COMBIN39) was used to model the behavior of the
results. The fatigue stress range is calculated based on strains CFRP–steel interface. The element includes two nodes whose initial
obtained in the initial cycle (N = 2; N = 1 is used as a shakedown relative distance is zero, having a bilinear bond-slip relationship
cycle to ensure the specimen is well seated). The final failure of established between them, as shown in Fig. 2(b).
the Beam occurred at N = Nu . All three specimens behaved in
an essentially similar manner. Fatigue induced cracking of the 4.2. Mesh formulation, boundary conditions, and validation of
steel was observed relatively early in the fatigue history (earlier interface model
than 0.25Nu ) followed by CFRP debonding initiating later in the
fatigue history (later than 0.70Nu ). This behavior will be discussed A constructed FE model is shown in Fig. 2(c). A relatively fine
at greater length in subsequent sections. mesh, having a maximum element length of 38 mm (1.8% of
the beam length) was used based on the results of a mesh size
4. Model development sensitivity study [Fig. 2(d)]. The simply-supported beam condition
was represented by restraining the nodes at supports. Fig. 3 shows
A three-dimensional nonlinear FE model was developed to pre- the validation of the proposed interface modeling approach, based
dict experimental findings. The general-purpose FE package ANSYS on an experimental program. Fawzia et al. [30] conducted simple
was used to model the test beams. The following summarizes de- tension tests using steel coupons bonded with CFRP. The predicted
tails of the proposed modeling approaches, including constitutive responses agreed well with those of the experiments, including
characteristics of the materials and CFRP-steel interface. the load–displacement [Fig. 3(a)] and CFRP-strain development
[Fig. 3(b)].
4.1. Material modeling and elements
4.3. Monotonic loading model
The steel beam was modeled with three-dimensional structural
solid elements (SOLID45). These eight node elements have three An incremental load was applied for the monotonic model until
translational degrees of freedom per node and are capable of failure occurred. The failure of the beams was modeled when
simulating a large plasticity problem making them suitable for either the steel section showed significant plasticity or rupture
examining local behavior near the notched location of the test of the CFRP was predicted for the repaired beams. The failure of
beams. Two-node three-dimensional spar elements (LINK8) were the unidirectional CFRP was determined by the maximum stress
used to represent the unidirectional CFRP strip. Consistent with failure criterion. Further details on the monotonic model may be
the steel substrate elements, each node of the CFRP element available in Kim and Harries [31].
 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502 1495

a b

Fig. 3. Validation of the proposed approach for the CFRP–steel interface using simple tension tests: (a) load–displacement using Specimen SN80 [30]; (b) CFRP strains using
Specimen SN80 [30].

Table 2
Fatigue properties of steel beam.
K′ n′ σ ′f ε ′f b c

Value 796 0.14 800 1.23 −0.10 −0.62

′
agreed well with those predicted using Eq. (1) (e.g., Ktest = 796
′
and Kpredict = 777) and thus the properties given in Table 2 were
accepted for the present study. An expression may then be
established to predict the cyclic stress–strain behavior of structural
steel, given that cyclic stress–strain relationships of a material may
be different from monotonic behavior [32]:
σ  σ 1/n′
Fig. 4. S–N curve of steel beam.
εtot = εel + εpl = + (2)
E K′
4.4. Fatigue loading model where εtot is the total strain; εel and εpl are the elastic and plastic
strains, respectively; σ is the stress of the member; and E is the
The fatigue model of the test beams was much more elastic modulus. To determine the response of a notched location
complicated than the monotonic due to the need to change subjected to a specific stress state [Fig. 1(a)], including local stress
constituent properties with increased fatigue cycles. A typical concentrating effects, Neuber’s rule is accepted and used in this
stress range–fatigue life (S–N) relationship of structural steel study [23]:
used for this study is shown in Fig. 4: Category E in AISC [24].
(Kf 1S )
As described above, the stress concentrating effect of the notch 1σ 1ε = with 1S = Smax − Smin (3)
provided is essentially equivalent to a Category E detail. The initial E
endurance limit of the steel beams was 32 MPa. A systematic where 1σ and 1ε are the local stress and corresponding strain,
approach to model the fatigue behavior of steel beams repaired respectively; Kf is the notch factor; 1S is the stress range; and
with CFRP strips is proposed as follows. Smax and Smin are the maximum and minimum stresses applied at
the level of the notch, respectively. As noted above Kf = 2.84
4.4.1. Fatigue life was used for this study. The stress range at the fatigue-critical
location of each test beam is given in Table 1. An intersection
The fatigue life of the test beams was predicted using the strain-
at the cyclic stress–strain curve (Eq. (2)) and the Neuber’s rule
life approach. Such a method is particularly relevant to a member
(Eq. (3)) may provide specific stress and strain values at the
demonstrating significant plasticity induced by hysteretic loads.
root of notch subjected to an arbitrary stress range, as shown
A brief description of the strain-life method is given here, while
in Fig. 5(a). Massing’s hypothesis may be used to determine a
the theoretical background is available elsewhere [32]. Table 2
stabilized hysteresis loop for cyclic load [35]:
summarizes material constants necessary to predict the fatigue life
of the beams (detailed definitions are given below). The adequacy 1/n′
1σ 1σ

of fatigue properties shown in Table 2 [33] was confirmed using 1ε = +2 . (4)
Eq. (1) [34]: E 2K ′
Massing’s hypothesis can predict hysteretic stresses in com-
′
σf′ ′ b plete reversals. Mean stress effects therefore need to be accounted
K = with n = (1)
(εf′ )n′ c for in the predicted fatigue life of the test beams since the fatigue
properties shown in Table 2 were obtained from tests having com-
where K ′ and n′ are the coefficient of cyclic strength and the plete load reversals. An expression proposed by Smith et al. [36] is
cyclic strain hardening exponent, respectively; σf′ and εf′ are the used to estimate the fatigue life of the test beams:
coefficients of fatigue strength and fatigue ductility, respectively;
and b and c are the exponents of fatigue strength and fatigue 1ε (σf′ )2
σmax = + σf′ εf′ (2Nf )b+c
ductility, respectively. These properties may be obtained from 2 E
multiple coupon tests subject to complete stress reversal fatigue 1σ
loading and corresponding curve-fitting. The referenced test values with σmax = + σ0 . (5)
2
1496 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502

a b

Fig. 5. Strain-life method to predict fatigue life of the beams: (a) cyclic stress–strain relationship (Beam D at 1S = 274 MPa); (b) stress range versus fatigue life.

The number of fatigue cycles to cause failure (Nf ) is finally obtained

by iterating Eq. (5), as shown in Fig. 5(b).

4.4.2. Cumulative damage

Cumulative damage theory was used to model material
degradation of the steel beams. Simple linear damage rules
(e.g., the Miner–Palmgren rule) are widely used but often present
significant errors. Therefore, an approach proposed by Henry [37]
was adopted for this study. The concept of Henry’s damage theory
is based on the change of an endurance limit depending upon
the number of fatigue cycles; in other words, a decrease in the
endurance strength of a material represents the damage level of
the material at a certain number of fatigue cycles. A damage index
(DI) that is a measure of the progressive damage in the steel beam Fig. 6. Damage accumulation of steel beams.
may be expressed as Eq. (6).
σeo − σe (a rapid increase in the index) was noticed beyond this limit until
DI = (6) fatigue failure of the beams occurred. The level of damage index for
σeo the beams was influenced by the stress range. For example, a dam-
where σeo and σe are the initial endurance limit and the limit age index of 0.6 was achieved at fatigue cycles of 1 × 104 , 1 × 105 ,
after an arbitrary number of fatigue cycles, respectively. The and 1.5 × 106 for Beams D, E, and F, respectively (Fig. 6).
number of cycles to failure (Nf ) and the remaining fatigue life (Nr )
following a particular number of cycles are given in Eqs. (7) and (8), 4.4.3. CFRP–steel interface
respectively.
A bilinear relationship for bond stress versus slip of the
k1 CFRP–steel interface was modeled for this study [Fig. 2(b)], based
Nf = (7) on experimental observations (to be discussed). A regression
S − σeo
analysis provided Eqs. (11) and (12) to predict the bond-slip of the
k2 CFRP–steel interface subjected to fatigue loads:
Nr = Nf − n = (8)
S − σe
s1 = 0.0243(1S )0.2040 and τ1 = 1.3187(1S )0.3555 (11)
where k1 and k2 are material constants; and n is the present
0.3009
fatigue cycle at a stress amplitude of S. The fatigue life (Nf ) s2 = 0.0158(1S ) and τ2 = 0 (12)
may be determined from the strain-life method described above
(Eq. (5)). By assuming k2 /k1 = σe /σeo [37], an endurance limit of where s1 is the slip at the maximum bond stress (τ1 ) and s2 is the
the steel beam subjected to n fatigue cycles may be obtained with slip at the bond stress of τ2 [Fig. 2(b)]. By definition, s2 is greater
a combination of Eqs. (7) and (8): than or equal to s1 . The fatigue damage to the CFRP is negligible,
given that the CFRP–steel interface buffers damage transfer from
S (1 − n/Nf ) the beam to the CFRP and the level of fatigue damage accumulation
σe = . (9)
(S − σeo )/σeo + (1 − n/Nf ) for the CFRP is insignificant [38]. In any event, CFRP is very tolerant
to fatigue damage [39]. Fatigue effects on the behavior of the
The damage index of the steel beam at a specific number of fatigue
adhesive are included in Eqs. (11) and (12) which account for the
cycles, n, is obtained by substituting Eq. (9) to Eq. (6). The damage
interface deterioration of the repair system. Further details on the
index is then incorporated into the classical concept of a material
interfacial behavior between the CFRP and steel are discussed in
degradation relationship:
the following section.
En = (1 − DI )E0 (10)
where En and E0 are the elastic modulus of the steel beam sub- 5. Results and discussion
jected to n cycles of fatigue and the initial modulus of the beam,
respectively. The accumulated damage indices for Beams D, E, This section provides a comparison of the experimental and
and F obtained from the theories described above are shown in predicted results of the beams with emphasis on fatigue responses,
Fig. 6. A gradual increase in damage index was observed up to ap- such as fatigue life and bond-slip characteristics at the CFRP–steel
proximately DI = 0.2, whereas significant damage accumulation interface. A summary of the monotonic test is given here to provide
 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502 1497

a b

c d

Fig. 7. Static behavior of test beams: (a) predicted load–displacement response at midspan; (b) strain development at the root of notch (line = experiment;
mark = prediction); (c) CFRP strain profile along the beam (Beam C); (d) strain development of CFRP near the damage (Beam C).

reference behavior to the fatigue beams; while, further details are damaged beam (Beam B) showed a rapid strain-development
found in Kim and Harries [31], including an extensive parametric because of the stress concentration affected by the notch. The
study. repaired beam (Beam C) exhibited an improved behavior, although
the effect of the notch was clearly evident and not mitigated by
5.1. Monotonic behavior the presence of CFRP. This observation indicates that, although the
CFRP-repair provides a noticeable strength gain, local plasticity at a
Table 1 reports the flexural capacity of the monotonic beams. damaged location may still exist and need additional or continued
The undamaged control beam (Beam A) failed at a load of 102.3 kN, examination. Fig. 7(c) shows strain profiles of the CFRP strip along
whereas the damaged (Beam B) and repaired beams (Beam C) failed the beam at selected service load levels. The strains increased
at loads of 81.8 kN and 84.5 kN, respectively. All of the test beams gradually from the termination point of the CFRP towards midspan
failed by torsional buckling because of insufficient lateral bracing. of the beam except at the notched location where significant
Such incomplete flexural responses were not a critical issue strain localizations were observed. The CFRP-strain development
because the primary purpose of the monotonic tests was to provide in the vicinity of the notch (Gages 2–4) is shown in Fig. 7(d). The
reference responses for the fatigue beams and the predictive predictions agreed well with the measured strains prior to the
models. Fig. 7(a) shows the predicted load–displacement response stability failure. CFRP rupture was predicted to follow debonding
of the monotonic beams. The predicted ultimate load of Beam A of the strip, as shown in the strain of Gage 2.
was 116.4 kN, while that of Beam B was 16.2% lower (97.6 kN).
After CFRP-repair, the capacity was fully recovered up to 123.0 kN 5.2. Fatigue behavior
(Beam C). The local strain behavior at the root of notch [Gage 1 in
Fig. 1(b)] is shown in Fig. 7(b). A linear response was observed for 5.2.1. Fatigue response of repaired beams
the undamaged beam (Beam A) until failure occurred at midspan Table 1 summarizes the fatigue test results. The stress range was
where significant yielding of the steel section took place. The a governing factor for fatigue life of the repaired beams. Beam D
1498 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502

(1S = 274 MPa) testing was stopped at Nf = 20,000; this was a

shortly before the final fatigue failure which was extrapolated to
occur only shortly thereafter [15]. Beams E (1S = 158 MPa) and
F (1S = 81 MPa) failed at Nf = 152,380 and Nf = 1,703,020,
respectively. Fundamental mechanics was used to determine the
stress range at the notched location of the beam (i.e., S = My/I
where M is the applied bending moment at the location of the
notch and y is the distance from the neutral axis to the notched
location). The predicted fatigue life showed reasonable agreement
with the test data, including a maximum error of 25%. Such a
difference in fatigue life is attributed to a number of uncertainties
in fatigue tests, for example, nominal versus actual material
properties (Table 2), theoretical assumptions in the strain-life
approach, use of an ideal loading environment in the model, and
so on. The change in flexural compliance of the beams is shown b
in Fig. 8(a). The compliance (C ) is a measure of damage level in the
beam and was calculated as: C = ∆/P where ∆ is the displacement
at midspan of the beam when subjected to a load of P that is the
upper limit of a fatigue load range. The compliance of the repaired
beams was almost constant prior to fatigue failure, irrespective
of the stress ranges. The magnitudes of compliance at failure
were, however, affected by the ranges, as shown in Fig. 8(a). The
threshold limits for Beams D, E, and F, where a sudden increase
in the compliance was observed, were 50.0%, 65.6% and 58.7% of
their fatigue lives, respectively. This observation is consistent with
the development of damage indices shown in Fig. 6, indicating
that insignificant damage has occurred prior to the threshold.
Fig. 8(b) shows the predicted notch opening displacement of each
test beam until failure occurred. The initial opening of Beam D was c
0.09 mm, while those of Beams E and F were 0.06 mm and 0.03 mm,
respectively. Such differences in notch opening are attributed to
the development of curvature of the beams, which depended upon
the level of applied load ranges. The measured strains at the root
of the notch of Beam E are shown in Fig. 8(c). The predicted
strains agreed well with those obtained experimentally. An abrupt
increase in strain was observed near cycle 50,000 (i.e., 32.8% of the
fatigue life of Beam E) for both the experimentally obtained and
predicted curves. Other beams showed similar behavior.
Fig. 9 shows the measured vertical crack propagation up the
web with the increased number of fatigue cycles. The propagation
was measured from the bottom of the web (i.e., tip of the notch).
The prediction of crack propagation was not available because
the developed model was based on a continuum approach so Fig. 8. Fatigue response of steel beams repaired with CFRP strip: (a) predicted
compliance; (b) predicted notch opening displacement; (c) strain development at
that discrete cracks did not form. The crack propagation was the root of notch (Beam E).
insignificant up to 50.0%, 49.9%, and 38.2% of the fatigue life for
Beams D, E, and F, respectively, and corresponding peak crack
responses, as previously presented in Fig. 2(b), while the maximum
depths were 49.4 mm, 87.0 mm and 122.1 mm, respectively, as
stress was dependent upon the level of stress ranges and fatigue
shown in Fig. 9(a). This observation illustrates that the repaired
cycles [Fig. 10(a)–(c)]. For example, the maximum shear stress of
beam subjected to a high load range, such as in the case of Beam D,
tends to show a large amount of brittle web fracture (i.e., relatively Beam D (1S = 274 MPa) was maintained at around 1 MPa until a
insignificant fatigue crack propagation) when compared to those fatigue cycle of 5000, whereas an abrupt increase was noticed up to
cases subjected to low load ranges that are associated with 9.8 MPa when the cycles further increased, as shown in Fig. 10(d).
substantial fatigue crack growth prior to the brittle fracture This observation is attributed to local bond failure of the CFRP strip
[Fig. 9(b)]. due to fatigue damage. A trend of decreasing the maximum shear
stress was observed with decreased stress ranges [Fig. 10(d)].
5.2.2. Bond-slip behavior
Fig. 10 reports the measured bond-slip response of the 5.2.3. Fatigue response of the CFRP–steel interface
CFRP–steel interface subjected to various fatigue stress ranges. The A regression analysis was conducted to predict the fatigue
shear stress was obtained using Eq. (13). properties of the CFRP–steel interface that had exhibited bilinear
responses [Fig. 10(a)–(c)]. The proposed model simplified the
1εfrp Efrp tfrp
τ= (13) bilinear curve with three coordinates, as shown in Fig. 2(b),
1x depending upon the stress range considered. The experimentally
where τ is the average stress between two arbitrary points obtained data (Fig. 10) were presented with respect to the
separated by a distance of 1x; 1εfrp is the strain increment within stress range and were fitted to generate a regression line that
1x; and Efrp and tfrp are the tensile modulus and thickness of the could represent the fatigue properties of a CFRP-steel interface.
CFRP, respectively. All of the beams exhibited essentially bilinear All of the fitted curves demonstrated good accuracy with the
 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502 1499

a b

Fig. 9. Crack propagation across the web: (a) measured values; (b) view of Beam F following failure.

a b

c d

Fig. 10. Bond-slip response of the CFRP–steel interface subject to fatigue: (a) Beam D; (b) Beam E; (c) Beam F; (d) variation of the maximum shear stress.

minimum coefficient of determination (R2 ) of 0.9765, as shown cycles increased further, plastic strains were observed [Fig. 13(b)].
in Fig. 11(a)–(c). The established equations have been already Beam D exhibited the greatest plastic strain of 0.009, while Beams
presented in Eqs. (11) and (12). E and F exhibited maximum plastic strains of 0.001 and 0.0008,
respectively. This observation indicates that applied stress ranges
5.2.4. Strain development are a critical factor affecting the development of plastic strains at
Fig. 12 shows the measured strain development along the fatigue-critical locations of CFRP-repaired steel beams; however,
CFRP strip, depending upon the fatigue cycles. Beams D and E the magnitude of the plastic strains may not be significant for the
exhibited a typical strain progression of the CFRP, whereas Beam F beams subjected to a fatigue stress range below 40% of the yield
showed a complete debonding history (i.e., initiation, propagation, strength (Fy = 393 MPa).
and debonding). The peak strain of Beam F gradually increased Fig. 14(a) compares the predicted strain profiles along the
near the notch until approximately N = 1.2 × 106 (70.6% of the CFRP strip of Beam E with counterpart experimental strains. Three
experimental fatigue life) and debonding of the strip initiated at locations were arbitrarily selected to show the strain variation
N = 1.4 × 106 (82.4% of the life). With the increased fatigue cycles, from locations close to and farther from the notch (i.e., Gages 2, 4
the debonding propagated towards the near termination point of and 7 in Fig. 1(c): 38 mm, 76 mm, and 190 mm from the notch,
the CFRP strip and, finally, complete debonding of the strip was respectively). The measured CFRP strains were almost constant
observed near the fatigue life of the beam, as shown in Fig. 12(c). from 50% to 95% of the fatigue life of the beams, depending
Fig. 13 shows the measured strains of the test beams at the fatigue- on the location of the gages. For example, the strains at Gage
critical location (i.e., the root of notch in Fig. 1). The elastic strains 2 showed insignificant changes until a cycle of 76,000 (50% of
of the beams were constantly maintained up to approximately the fatigue life); however, those at Gages 4 and 7 showed little
10% of their fatigue life, as shown in Fig. 13(a). When the load change until 100,000 and 144,000 cycles (65% and 95% of the
1500 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502

a a

b b

c c

Fig. 12. CFRP-debonding due to fatigue cycles (Beam F).

Fig. 11. CFRP-steel interface subject to fatigue: (a) slip at peak shear stress; (b) peak
shear stress; (c) maximum slip.
loads. A total of six beams were tested to examine the local
plasticity near the damaged location and the fatigue responses of
life), respectively. Exponential damage accumulation was noticed
the CFRP–steel interface. A modeling approach was proposed to
beyond the constant-strain range. The predicted strains were
predict the monotonic and fatigue behavior of the test beams. For
reasonable when compared to those of the experiment [Fig. 14(a)];
the fatigue model, a combination of the strain-life method and
however, the peak strains of the model were higher near fatigue
cumulative damage theory was used. The following is concluded.
failure of the beams. This observation is due to the fact that
predicted strains are, in general, more sensitive than measured • The monotonic capacity of the damaged beam was predicted
strains at a location where a numerical singularity takes place to be improved to that of the undamaged beam due to CFRP-
(i.e., stress concentration). The predicted strain development of the repair; however, local plasticity near the damage was not
CFRP at the notched location is shown in Fig. 14(b). It should be proportionally improved. The strain of the CFRP strip along
noted that strain gages were not bonded at the notched location of the beam span gradually increased from the termination point
the test beams. All of the beams exhibited exponential responses, to midspan except at the damaged location where significant
whereas the magnitude of the CFRP strains was affected by the concentrations were observed.
stress ranges in both of the constant strain and the peak strain • The proposed modeling approach showed good agreement
zones. For instance, Beam F (1S = 81 MPa) showed decreases of with the test data. The fatigue life of the repaired beams
67.0% and 67.5% in the initial and the peak strains, respectively, in was significantly influenced by the stress range. The damage
comparison to Beam D (1S = 274 MPa), as shown in Fig. 14(b). propagation of the beams was abrupt and the level of damage
was not influenced by the stress range up to approximately 60%
6. Summary and conclusions of the fatigue life, whereas the stress range affected damage
levels with further increases in fatigue cycles. The vertical crack
This paper has presented the behavior of damaged steel beams propagation in the web was not substantial up to 40%–50%
repaired with CFRP strips subjected to monotonic and fatigue of the fatigue life but increased substantially following this,
 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502 1501

a b

Fig. 13. Strain development at fatigue-critical location: (a) elastic strain; (b) plastic strain.

a b

Fig. 14. Fatigue behavior of CFRP strip: (a) strains along the strip (Beam E); (b) strain development at notched location.

eventually resulting in brittle web fracture. The stress range [4] American Society of Civil Engineers (ASCE). Report card for America’s
affected the crack propagation rate of the repaired beams. infrastructure. Reston (VA, USA): American Society of Civil Engineers; 2005.
[5] Tavakkolizadeh M, Saadatmanesh H. Fatigue strength of steel girders
• The fatigue response of the CFRP–steel interface was bilinear strengthened with carbon fiber reinforced polymer patch. J Struct Eng ASCE
and was dependent upon the stress range and the number of 2003;129(1):186–96.
fatigue cycles. The local bond failure of the CFRP strip due to [6] Lenwari A, Thepchatri T, Albrecht P. Debonding strength of steel beams
fatigue damage caused a sudden increase in the bond stress and strengthened with CFRP plates. J Compos Constr ASCE 2006;10(1):69–78.
[7] Bakis CE, Bank LC, Brown VL, Cosenza E, Davalos JF, Lesko JJ, Machida A,
corresponding slip of the strip. The proposed empirical model Rizkalla SH, Triantafillou TC. Fiber–Reinforced polymer composites for
may be useful for estimating the fatigue behavior of CFRP–steel construction-state-of-the-art review. J Compos Constr ASCE 2002;6(1):73–87.
interfaces. [8] Teng JG, Chen JF, Smith T, Lam L. Behavior and strength of FRP-strengthened
RC structures: a state-of-the-art review. P I Civil Eng-Str B 2003;156(1):51–62.
• The fatigue cycles resulted in a gradual increase in CFRP strains [9] Mertz DR, Gillespie JW, Chajes MJ, Sabol SA. The rehabilitation of steel
near the damage and debonding strains were noticed when bridge girders using advanced composite materials. NCHRP-IDEA Project 51.
the cycles further increased. The elastic strains at the root of Transportation Research Board. Washington (DC, USA); 2002.
notch were essentially constant up to 10% of the fatigue life [10] Al-Saidy AH, Klaiber FW, Wipf TJ. Repair of steel composite beams with carbon
fiber-reinforced polymer plates. J Compos Constr ASCE 2004;8(1):163–72.
of the repaired beams, irrespective of the stress range. The [11] Nozaka K, Shield CK, Hajjar JF. Effective bond length of carbon–fiber-reinforced
development of the plastic strains was influenced by the stress polymer strips bonded to fatigued steel bridge I-girders. J Bridge Eng ASCE
range; however, their magnitude was not significantly affected 2005;10(1):195–205.
[12] Schnerch D, Rizkalla S. Flexural strengthening of steel bridges with high
when the stress range was below 40% of the yielding strength
modulus CFRP strips. J Bridge Eng ASCE 2008;13(1):192–201.
of the steel (Fy = 393 MPa). [13] Shaat A, Fam A. Repair of cracked steel girders connected to concrete slabs
using carbon–fiber-reinforced polymer sheets. J Compos Constr ASCE 2008;
12(3):650–9.
Acknowledgements [14] Harries KA, El-Tawil S. Review of steel-FRP composite structural systems. In:
Proc. 5th int. conf. on composite construction. Tabernash, CO. USA; 2008.
The authors would like to acknowledge Michael J. Richard and [15] Harries KA, Richard M, Kim YJ. Fatigue behaviour of CFRP retrofitted damaged
steel beams. In: Proc. 13th int. conf. on structural faults and repair. Edinburgh
the Watkins–Haggart Structural Engineering Laboratory at the (UK); 2010.
University of Pittsburgh for the experimental contribution. Test [16] Cadei JMC, Stratford TJ, Hollaway LC, Duckett WG. Strengthening metallic
materials were donated by Triad Metals, Fyfe Company, and SIKA structures using externally bonded fibre-reinforced polymers. CIRIA Pub. No.
USA. C595. London (UK); 2004.
[17] Deng J, Lee MMK. Fatigue performance of metallic beam strengthened with a
bonded CFRP plate. Compos Struct 2007;78: 777-231.
References [18] Deng J. Fatigue life prediction of steel beams strengthened with a carbon fibre
composite plate. In: 5th int. conf. on frp composites in civil engineering. Beijing
[1] Kim YJ, Green MF, Fallis GJ. Repair of bridge girder damaged by impact loads (China); 2010.
with prestressed CFRP sheets. J. Bridge Eng ASCE 2008;13(1):15–23. [19] Wu G, Liu H, Wu Z, Wang H. Experimental study on fatigue behavior of I-
[2] Kim YJ, Yoon DK. Identifying critical sources of bridge deterioration in cold steel beam strengthened with different FRP plates. In: 5th int. conf. on FRP
regions through the constructed bridges in North Dakota. J Bridge Eng ASCE composites in civil engineering. Beijing (China); 2010.
2010;15(5):542–52. [20] O’Neill A, Harries KA, Minnaugh P. Fatigue behavior of adhesive systems used
[3] Harries KA. Structural testing of prestressed concrete girders from the Lake for externally-bonded FRP applications. In: Proc. 3rd int’l conf. durability &
View Drive bridge. J Bridge Eng ASCE 2009;14(1):78–92. field applications of FRP for const. Quebec City (QC, Canada); 2007.
1502 Y.J. Kim, K.A. Harries / Engineering Structures 33 (2011) 1491–1502

[21] Bocciarelli M, Colombi P, Fava G, Poggi C. Fatigue performance of tensile [31] Kim YJ, Harries KA. Modeling of steel beams strengthened with CFRP
members strengthened with CFRP plates. Compos Struct 2009;87:334–43. strips including bond-slip properties. In: 5th international conference on
[22] Young WC. Roark’s formulas for stress and strain. 6th ed. McGraw Hill; 1989. FRP composites in civil engineering (CICE 2010). Beijing (China); 2010.
[23] Neuber H. Theory of notch stresses. Ann Arbor (MI, USA): J.W. Edwards Inc.; p. 873–6.
1946. [32] Bannantine JA, Comer JJ, Handrock JL. Fundamentals of metal fatigue.
[24] American Institute of Steel Construction (AISC). ANSI/AISC 360–05 specifica- Englewood (NJ, USA): Prentice Hall; 1990.
tions for structural steel buildings. Chicago (IL, USA): American Institute of [33] American Society of Metals (ASM). Metals handbook Vol. 1. 10th Edition.
Steel Construction; 2005. Materials Park (OH, USA): American Society of Metals; 1990.
[25] Enberson NK, Mays GC. Significance of properties mismatch in the patch repair [34] Heffernan PJ, Erki MA, DuQuesnay DL. Stress redistribution in cyclically loaded
of structural concrete. Mag Concrete Res 1996;48(174):45–57.
reinforced concrete beams. ACI Struct J 2004;101(1):261–8.
[26] Miller TC, Chajes MJ, Mertz DR, Hastings JN. Strengthening of a steel bridge
[35] Massing G. Eigenspannungen und verfestigung beim messing. In: Proc. 2nd int.
girder using CFRP plates. J Bridge Eng ASCE 2002;6(3):514–22.
cong. for applied mechanics. Zurich (Switzerland); 1926. p. 332–5.
[27] Sen R, Liby L, Mullins G. Strengthening steel bridge sections using CFRP
laminates. Compos Part B 2001;32:309–22. [36] Smith KN, Watson P, Topper TH. A stress–strain function for the fatigue of
[28] Patnaik AK, Bauer CL. Strengthening of steel beams with carbon FRP laminates. metals. J Mater 1970;5(2):767–78.
In: Proc. 4th int. conf. on advanced composites for bridges and structures. [37] Henry DL. A theory of fatigue damage accumulation in steel. ASME Trans 1955;
Calgary (Canada); 2004. 77(3):913–8.
[29] Harries KA, Webb P. Experimental assessment of bonded FRP-to-steel [38] Aidoo J, Harries KA, Petrou MF. Fatigue behavior of carbon fiber reinforced
interfaces. P I Civil Eng-Str B 2009;162(2):233–40. polymer strengthened reinforced concrete bridge girders. J Compos Constr
[30] Fawzia S, Al-Mahaidi R, Zhao X-L. Experimental and finite element analysis of ASCE 2004;8(3):501–9.
a double strap joint between steel plates and normal modulus CFRP. Compos [39] Curtis PT. Fatigue Behavior of Fibrous Composite Materials. J Strain Anal Eng
Struct 2006;75(1–4):156–62. 1989;24(2):235–44.