system ductility and redundancy of frp beam structures with ductile adhesive joints
TRANSCRIPT
System ductility and redundancy of FRP beam structures
with ductile adhesive joints
Thomas Keller*, Julia de Castro
Composite Construction Laboratory CCLab, Swiss Federal Institute of Technology EPFL, BP Ecublens, Station 16, CH-1015 Lausanne, Switzerland
Received 3 December 2004; revised 28 April 2005; accepted 1 May 2005
Available online 24 June 2005
Abstract
This paper reports on a structural concept for engineering structures composed of FRP components to provide system ductility that
compensates for the lack of material ductility inherent to FRP materials. The concept includes the use of redundant structural systems and
ductile or flexible adhesive joints. To demonstrate the feasibility of the proposed concept, quasi-static experiments on pultruded GFRP beams
were performed. The two-span beams were connected with flexible adhesive joints at the middle support. The flexible joints from highly non-
linear adhesives provided a favorable redistribution of the internal and external forces in the statically indeterminate system compared to
single-span and continuous beams, which were also examined. In the case of adhesive joint failure, structural collapse was prevented because
of system redundancy. Due to the stiffness-governed design of the GFRP beams, the stresses in the flexible adhesive joints were small and
creep deformations in the joints could be controlled.
q 2005 Elsevier Ltd. All rights reserved.
Keywords: B. Plastic deformation; E. Pultrusion; E. Joints/joining; Adhesive
1. Introduction
Fiber-reinforced polymer (FRP) composites are being
used increasingly in engineering structures thanks to their
advantageous material properties. Applications for strength-
ening purposes are accepted worldwide. For new
construction, structural FRP components such as profiles
and sandwiches are also starting to be used. The main
applications for these components are pedestrian bridges
and bridge decks where the properties of these materials,
such as high specific strength, insensitivity to frost and
de-icing salts, and rapid installation of components are
advantageous [1].
There are, however, also some material properties that
still hinder the widespread acceptance of new FRP
construction by structural engineers familiar with traditional
construction materials such as steel or reinforced concrete.
One of these disadvantages is the lack of ductility inherent
to FRP materials. Ductile materials allow the redistribution
1359-8368/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compositesb.2005.05.001
* Corresponding author.
E-mail address: [email protected] (T. Keller).
of internal forces leading to an increase in structural safety,
the dissipation of energy from impact or seismic actions and
give warnings of a possible structural problem due to large
plastic or inelastic deformations before failure.
A second disadvantage of structural FRP components
concerns the difficulty of joining the elements due to the
brittle fibrous and anisotropic character of the materials. The
current practice of bolting is not material-adapted and leads,
in most cases, to an over-sizing of the components [1].
Adhesive bonding is more appropriate for FRP materials.
The relatively high stiffness of the brittle epoxy adhesives
currently used, however, leads to high shear and peeling
stress peaks at the edges of the bonded joints [2].
Furthermore, the surface preparation and process of
applying the adhesives are very demanding, and simple
non-destructive quality control procedures do not yet exist.
Therefore, in an engineering structure composed of FRP
components and adhesive joints, the unexpected failure of a
bonded joint cannot be ruled out and must be considered in
the structural concept.
To overcome the drawbacks of bolted or epoxy-bonded
FRP engineering structures, the authors propose a new
concept for structures composed of brittle FRP components
that includes system ductility through the use of ductile
Composites: Part B 36 (2005) 586–596
www.elsevier.com/locate/compositesb
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596 587
adhesive joints and statically indeterminate structural
systems. The proposed concept envisages adhesives with
initial elastic behavior that is sufficiently stiff to meet short-
and long-term serviceability requirements. However, when
the serviceability (SLS) and ultimate (ULS) loads are
exceeded, the behavior of the adhesives should change and
become plastic or at least highly non-linear/inelastic with a
much smaller stiffness. In the highly non-linear/inelastic
case, the behavior is designated in the following as flexible
and not ductile. The ductile or flexible joints can
compensate for the lack of material ductility of FRP
components by providing ductility in the structural system,
called system ductility, which offers the same advantages as
the aforementioned advantages for material ductility. In the
case of unexpected joint failure, the redundant (statically
indeterminate) system provides alternative load paths and
redistribution of the cross-section forces due to the presence
of other ductile or flexible joints. In this way, a structural
collapse can be prevented.
Furthermore, the elasto-plastic or highly non-linear/
inelastic behavior of the adhesives prevents the occurrence
of high stress peaks. Shear and peeling stresses are much
more evenly distributed in the bonded surface, which leads to
less sensitive and more robust joints with regard to premature
and unexpected failure. This concept has already been
introduced in Keller et al. [2] in order to control the stresses in
the flanges of adhesively bonded sandwich girders.
The behavior of bolted joints is often described as ‘semi-
rigid’ [3–5]. This term was intentionally avoided in this
study, because it is too general and does not differentiate
between elastic, flexible or ductile joint behavior. To
demonstrate the feasibility of the proposed concept of
FRP structures with system ductility, experiments were
performed on continuous FRP beams with flexible adhesive
joints. The results of these experiments are presented in this
paper.
2. System ductility in FRP structures
System ductility, as described in the preceding section,
can comprise several distinct ‘sub-concepts’ that were
developed in earlier works by different researchers. Table 1
summarizes and classifies these sub-concepts as follows.
System ductility can include two different sub-concepts:
ductility and pseudo-ductility. Ductile concepts
Table 1
System ductility and redundancy—overview
System ductility System redu
Concepts Description Concepts
Ductility Combinations of brittle/ductile
components or connections
Structural r
indetermina
Pseudo-ductility Combinations of brittle/brittle
components or connections
Cross-sectio
include combinations of ‘brittle’ and ‘ductile’ structural
components or connections at the system level. Examples
are GFRP bridge decks adhesively bonded to steel girders.
Keller and Gurtler [6] reported on ductile failures of hybrid
FRP/steel girders. Brittle GFRP bridge decks failed during
the yielding of the bottom steel flanges. Other examples are
brittle FRP bars in concrete elements where the (small)
plasticity of the compressed concrete is used to provide a
ductile failure [7,8]. Ductile adhesive joints, as described in
this paper, can develop plastic hinges and are also classified
under ductile concepts.
Pseudo-ductility is the term used for the behavior of
combinations of ‘brittle’ and ‘brittle’ structural components
or connections at the system level. Non-linear, inelastic
load–deformation behavior, showing decreasing stiffness
with increasing load similar to the behavior of ductile
structures, can be achieved by progressive failure of brittle
components or connections. To prevent structural collapse,
however, structural redundancy is required. Combinations
of rigid carbon fiber laminates and flexible glass fiber
sections have been proposed by Deskovic et al. [9]. The
carbon fiber laminates are used to increase the stiffness
during the serviceability limit state, while the glass fiber
sections increase the deformations before failure (and ensure
the strength). Based on a similar concept, Razaqpur and Ali
[10] developed concrete beams reinforced externally by
bonded carbon fiber sheets and internally by a low modulus
polypropylene fiber grid.
Comparable non-linear, inelastic load–deformation
behavior can also occur on the component level by means
of a non-linear, inelastic increase in deformation caused by
progressive internal failures. Keller and Gurtler [11]
reported on experiments on bridge deck specimens
subjected to in-plane shear loading. Transverse bending in
the deck webs led to progressive local delamination failures
and non-linear, inelastic behavior, similar to the behavior of
a ductile material. In contrast to ductile structures, however,
pseudo-ductile structures can only dissipate energy within a
limited number of loading cycles, depending on the degree
of static indeterminacy. After a certain number of failures,
pseudo-ductile structures become statically determinate,
whereby a further failure would lead to a brittle collapse.
Analogous to system ductility at the system level, the
term system redundancy is proposed and discussed in the
following as a requirement for pseudo-ductile behavior.
Table 1 also gives an overview of possible sub-concepts of
ndancy
Description
edundancy (statically
te systems)
Systems with alternative load paths,
back-up systems
nal redundancy Multi-component cross-sections
Fig. 2. Bonded joint over middle support of beam PH1 with strain gages and
clinometers.
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596588
system redundancy. Concepts of structural redundancy
(statically indeterminate systems) and cross-sectional
redundancy are proposed. Statically indeterminate systems
provide alternative load paths in the case of failure of a
component or a connection. An example is the truss
structure of the Pontresina Pedestrian Bridge with adhe-
sively bonded joints in one span [2]. The crossed diagonals
offer alternative load paths in the case of failure of one or
several of the components or bonded joints. Furthermore,
the addition of bolts in the adhesive joints provides a
mechanical back-up system. The second concept of
redundancy, called cross-sectional redundancy, is provided
in multi-component cross-sections. The sandwich girders
with adhesively bonded flanges or the built-up sections of
the Eyecatcher building, reported in [2], are examples of
cross-sectional redundancy. In both the Pontresina and
Eyecatcher examples, single cross-sectional components or
the adhesive bond between single components can fail
without failure of the whole section.
3. Experiments on bonded pultruded beams
3.1. Beam description
As already introduced, experiments on FRP beams with
flexible adhesive joints were carried out and can be seen in
Fig. 1. Four two-span beams, PH1–PH4, were examined.
The beams were built up from two single span beams. The
spans of the built-up beams were two times 3.60 m.
Pultruded beams with square box cross-sections of 240!240 mm and a 12 mm wall thickness were used. The depth/
span-ratio was 1/15.
The beam flanges were connected at the middle support,
i.e. at the location of the maximum negative bending
moment, with adhesive strap joints, as shown in Fig. 2. The
strap joints had different overlap lengths: 2!100 mm
(PH3), 2!200 mm (PH1 and PH2) and 2!300 mm
(PH4), as listed in Table 2. Due to the size of the support
Fig. 1. Loading test set-up for bonded beam PH2.
surface required at the middle support, the beam with upper
overlaps of 2!100 mm (PH3) had lower overlaps of 2!200 mm. The flanges were connected with cover plates of
the same thickness (12 mm) and width (240 mm), as shown
in Fig. 2. The cover plates were cut from identical beam
sections. The average gap between the connected beams
was between 12 and 16 mm on the centerline. The beam
webs were not connected. The bonded joints, therefore,
were able to transfer bending moments over the middle
support, but no vertical shear forces. The chosen adhesive
thickness was 2 mm due to the tolerances of the sections and
the wires of the strain gages bonded on the flanges in the
joints.
The layout of the bonded beams and the experimental
set-up were primarily chosen from a scientific point of view
to demonstrate the feasibility of the global structural
concept. The practical application of this layout with the
bonded joints over the supports was not the main concern.
For comparison purposes, three single-span beams,
PS1–PS3, with 3.60 m span as well as three continuous
two-span beams, PC1–PC3, without bonded joints over the
middle support were also examined. Table 2 gives details of
the beams and the experimental results at failure.
3.2. Materials
The beams were composed of approximately 45%
E-glass fibers by volume and an isophthalic polyester
resin. The architecture consisted of approximately 67%
UD-rovings and two outer and two inner combined mats
(CSM and 08/908 fabrics stitched together). The overlaps of
the combined mats at the corners of the cross-section were
visible. The wall thickness was slightly increased in the
overlap region on a width of approximately 30 mm on each
side of the corners. On the outside, a polyester surface veil
was added. The stiffness of the beams was determined
experimentally with the single-span beams (PS1–PS3) by
three-point and four-point bending tests and compared to
Table 2
Overview experimental beams and results at failure
Beam Specification Ultimate fail-
ure load (kN)
Maximum
deflection at
failure (mm)
Rotation at
middle sup-
port (8)
Reaction at
middle sup-
port (kN)
Moment at
middle sup-
port (kNm)
Moment at
jack location
(kNm)
Failure
location
PS1–PS3 Single span 155G5 66G3 2.9G0.1 103G3 0 125G4 Loading point
PC1–PC3 Continuous 146G11 32G6 0 242G17 K86G7 60G5 Middle sup-
port
PH1 Adhesive joint
overlap 2!200 mm
180 45 1.6 278 K67 99 Loading point
PH2 Adhesive joint
overlap 2!200 mm
178 50 1.8 267 K53 107 Loading point
PH3 Adhesive joint
2!100 mm
(top) 2!200 mm
(bottom)
(1) 135 (1) 38 (1) 1.5 200 (1) K36 (1) 84 (1) Adhesive
joint
(2) 154 (2) 55 (2) 2.2 206 (2) 0 (2) 122 (2) Loading
point
PH4 Adhesive joint
overlap 2!300 mm
178 47 1.4 274 K66 98 Loading point
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596 589
the values given in the pultruder’s design manual [12].
Table 3 gives an overview of the results. The full-section
elastic and shear moduli were calculated in different ways
from the measured deflections, rotations and strains. The
shear modulus, determined from the strains, was calculated
from readings of gages in the 458 direction on the centerline
of the webs. The shear coefficient, K (synonymous with the
shear correction factor), was estimated according to Bank
[13] and Omidvar [14]. The properties used to predict the
behavior of the beams were 30.0 GPa for the elastic
modulus and 3.0 GPa for the shear modulus.
The acrylic-based two-component SikaFast 5221 struc-
tural adhesive, based on the ADP (acrylic double
performance) technology developed by Sika [15], was
used in bonding of the specimens. The adhesive choice was
based on previous experiments on bonded GFRP double-lap
joints with epoxy, polyurethane and acrylic adhesives. The
shear stress–strain behavior, t–g, of the ADP adhesive was
determined from napkin-ring tests [16], which are based on
EN 11003-1 specifications [17]. The adhesive behavior was
highly non-linear, as shown in Fig. 3. The stress–strain
curve can be modeled as bi-linear with two different shear
Table 3
Properties of GFRP beams
Data origin Maximum stress (MPa) E
Design Manual [12] 240 (tension and compression
strength)
2
Eurocomp [19] 100 (onset of buckling)
Obtained from beams PS1–PS3 99/153 (onset/post-buckling)
From deflections 3
From rotations (E)
and deflections (G)
2
From strain gages
08 (E) and 458 (G)
3
Used for calculations 3
moduli Ge (elastic) and Gp (plastic), where Gp is
approximately 11 times smaller than Ge. The properties of
the ADP adhesive are listed in Table 4.
3.3. Experimental set-up and instrumentation
The experimental set-ups can be seen in Figs. 1 and 4.
The two-span beams were loaded by a hydraulic jack of
500 kN capacity in each span. The jacks were each located
at a distance of one-third of the span from the middle
support. For the single-span beams, in addition to the three-
point load configuration, a four-point load configuration was
used to determine the beam properties given in Table 3.
Timber plates and 5 mm thick neoprene pads were placed
between the beams and the jacks and supports. The timber
plates at the most critical middle supports were 300 mm
long, 260 mm wide and 27 mm thick. In addition, the edges
of the plates were chamfered to a 10 mm radius to relieve
stress concentrations, as shown in Fig. 2.
Each beam was instrumented with load and displacement
sensors in the jacks, load sensors at each support, vertical
displacement transducers (up to 12), electronic clinometers
lastic modulus, E (GPa) Shear modulus, G (GPa)
3.0 3.0
2.1G1.8 2.1G0.5
9.8G0.2 3.3G0.2
0.3G0.4 4.3G0.1
0.0 3.0
0
2
4
6
8
10
0.0 0.5 1.0 1.5 2.0 2.5
Shear strain [-]
She
ar s
tres
s [M
Pa]
Fig. 3. Shear stress–strain behavior of ADP adhesive.
Fig. 4. Loading positions: (a) single beams PS, (b) continuous beams PC,
(c) bonded beams PH (dimensions in mm).
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596590
at the supports (4) and strain gages (up to 20) at different
cross-sections. The strain gages were used to measure the
axial strain distributions through the depth and across the
width of the beams. In addition, strain gages were placed
in the bonded joints of beams PH1–PH4 on the top and
bottom flanges (40 gages in each joint). The strain
gage arrangement in the upper joint of beam PH2 with
2!200 mm overlap length is shown in Fig. 5 (24 gages).
12span 1 (North)span 2 (South)
3.4. Experimental program
In the first phase of the experimental program, each beam
was subjected to several load cycles at different load levels
to observe the behavior of the beams and the instrumenta-
tion. These cycles were carried out under load-control at a
rate of 5 kN/min. In the second phase, the beams were
loaded up to failure under displacement-control at rates of
1.2–2.3 mm/min. The single-span beams were loaded in the
three-point configuration up to failure with the load at the
one-third span position. With regard to the flexible adhesive
used, a creep experiment was performed on the bonded
beam PH2 between the first and second phases of the
experimental program. The beam was subjected to a load of
40 kN per jack for a period of 7 days. The resulting
deflection-to-span ratio of 1/400 was approximately equal to
the admissible deflection ratio at the serviceability limit
state in building construction.
240
108
108
18overlap edge
4. Experimental results
The results of the first phase of the experimental program
showed that the axial strains through the depth of all beams
Table 4
Properties of ADP adhesive (Napkin-ring test)
Shear strength,
tu (MPa)
Shear strain at
failure, gu (%)
Shear modulus
Ge (GPa) Gp (GPa)
8.3 2.3 0.033 0.003
remained linear. The neutral axis was approximately 2 mm
below the centerline in the spans. This small shift was likely
due to the slightly different elastic tensile and compressive
moduli. (The values from 08 strain gages given in Table 3
are average values from the tension and compression
flange.) It was also seen that the compressive strains were
slightly higher than the tensile strains. Furthermore, the
strains across the width of the flanges remained constant. No
reductions in strain due to shear lag were observed. Only the
results of the failure experiments and the creep experiment
on beam PH2 are presented in the following sections.
4.1. Behavior of single-span beams
The load–deflection curves of the three single-span
beams, PS1–PS3, are shown in Fig. 6. The behavior of the
beams was identical and linear-elastic up to a load of
approximately 100 kN per jack (approximately 65% of the
ultimate failure load). At this load level, a slight decrease in
the stiffness was observed. The stiffness decrease coincided
with the onset of buckling of the compression (top) flange of
200
20 20 20 40 40 40 40 20
12
40 40 40 40
200
Fig. 5. Strain gage arrangement in upper adhesive joint with 200 mm
overlap (dimensions in mm).
0
50
100
150
200
0 10 20 30 40 50 60 70Vertical deflection [mm]
Load
per
load
ing
poin
t [kN
]
PS1-3
PC1-3
PH1 PH2
PH3
PH4
PH1
0
25
50
0 5 10
Fig. 6. Load–deflection curves of all beams, deflections at loading section.
Fig. 8. Failure of beam PS1 near patch load.
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596 591
the beams, which could be traced back with the strain gages
distributed across the width of the top flanges. Fig. 7 shows
the strain measurements across the top and bottom flanges
of beam PS2, 50 mm from the edge of the patch load. The
strains in the tension (bottom) flange increased linearly up to
failure, whereas the strains in the compression (top) flange
began to decrease non-linearly as the load exceeded
approximately 65% of the ultimate failure load.
Subsequent to the onset of buckling, the stiffness of the
beams continued to decrease slightly up to failure
(cf. Fig. 6). The top flange failed in beams PS1 and PS3,
as shown in Fig. 8. At the same time, horizontal cracks
formed in the webs at the locations of the overlap edges of
the combined fiber mats. In beam PS2, only the webs failed
and two horizontal cracks formed on each side, one of them
again at the locations with the overlap edges of the
combined mats. The failure modes of the top flanges and
webs were interconnected and very complex.
Fig. 9 shows the measured load–rotation behavior of the
beams at the support closer to the jack in the three-point
bending tests. The rotation curves showed the same elastic/
non-linear behavior as the deflection curves. The measured
ultimate failure loads, maximum deflections at the load
application points and maximum rotations at the support
Fig. 7. Strain distribution across tension and compression flanges of beam
PS2 (dimensions in mm).
closer to the jack (location of middle support for the
continuous beams) are listed in Table 2 (average values and
standard deviations).
4.2. Behavior of continuous beams
The load–deflection curves of the three continuous
beams, PC1–PC3, are shown in Fig. 6. The average values
of the two jack loads and of the deflections at the load
application points are indicated. In general, the measured
values of the two spans correspond well; the maximum
variations were approximately 2%. The behavior of the
beams was identical and linear-elastic up to a load of
approximately 100 kN per jack (approximately 70% of the
ultimate failure load). As for the single-span beams, the
stiffness began to decrease at this point due to the onset of
buckling of the compression flange and, particularly, of the
webs. In the continuous beam configuration, the most highly
compressed flange and webs were the bottom flange and
webs at the middle support. Ultimate failure occurred in the
webs at the middle support. Horizontal cracks formed
30 mm above the lower corners in the region where the
overlaps of the combined fiber mats ended. The support
plate punched into the beam, as can be seen in Fig. 10 for
0
50
100
150
200
0.0 0.5 1.0 1.5 2.0 2.5 3.0Rotation [˚]
Load
per
load
ing
poin
t [kN
] PH2
PH3
PH4 PH1
PC1-3
Middle support
PS1-3
Fig. 9. Load–rotation curves at middle support.
Fig. 10. Failure of beam PC3 at middle support. Fig. 11. Deformation of bonded beam PH1 at onset of failure.
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596592
beam PC3. Furthermore, two longitudinal cracks formed in
the lower flange of all three beams 30 mm from the lower
corners (edges mat overlaps).
No rotations occurred over the middle support. The main
experimental results are listed in Table 2. The ultimate
failure loads were slightly below the values of the single-
span beams. The maximum deflections varied between 27
and 38 mm (approximately span/110) and reached approxi-
mately 50% of the single-span beam values.
0.00
0.02
0.04
0.06
0.08
0.0 0.2 0.4 0.6 0.8 1.0Normalized overlap length [-]
Axi
al s
trai
n [%
]
PH1PH2PH3PH4
Fig. 12. Measured axial strain distributions on upper flanges in adhesive
joints of beams PH1–PH4 at load of 50 kN per jack.
4.3. Behavior of bonded beams with 200 mm overlap (PH1
and PH2)
The load–deflection responses of the bonded beams PH1
and PH2 with 2!200 mm joint overlap are shown in Fig. 6.
The beams exhibited tri-linear behavior. In the first part, up
to approximately 125 kN (70% of the ultimate failure load),
the behavior can be described as bi-linear, as it was for the
ADP adhesive. In the load range up to approximately 20 kN
per jack, the stiffness dropped to a value which,
subsequently, remained constant. In the second part,
above 125 kN, a further stiffness decrease was seen that
coincided with the onset of buckling of the top flanges and
the webs below the patch loads, and the bottom flanges and
the webs at the middle support. Beam PH1 was slightly
stiffer than beam PH2, which may have been due to the
different temperatures in the laboratory during the exper-
iments (PH1 228, PH2 288). The failures in both beams
occurred in the webs below one of the jacks, similar to what
was observed for the single-span beams. Fig. 11 shows the
deformed beam PH1 at the onset of failure. The negative
curvature over the middle support due to the continuity
effect provided by the bonded joint is visible. Below the left
jack, buckling of the web is also visible. The subsequent
failure occurred at this location in the area of the edge of the
mat overlap. No damage or failure was observed in the
bonded joints from visual inspection and strain
measurements.
Fig. 9 shows the measured load–rotation behavior of the
beams at the middle support. The rotation curves showed the
same bi-linear behavior as the deflection curves in the first
phase of the experiments. However, the stiffness decrease in
the second phase due to buckling could not be detected. The
main experimental results are listed in Table 2 for both
beams.
Fig. 12 shows the measured axial strains along the upper
strap joints at a load of 50 kN for both beams PH1 and PH2.
The symbols represent average values from four strain gage
readings from gages at equal positions in the joint. The
strain distribution was nearly linear along the major part of
the overlap. The strains increased non-linearly only in the
first and last 20 mm. A linear least-squares fit is also shown
in Fig. 12.
As mentioned above, beam PH2 was subjected to 7 days
of creep loading. The first hour of the measured deflection–
time curve is shown in Fig. 13. The loading curve up to
9 mm deflection is not entirely linear due to the manually
driven loading system. When subjected to constant loads,
the maximum deflection increased from 9.1 to 11.6 mm
(approximately 27%) during 7 days. Seventy percent of the
increase occurred within the first 5 min.
0
5
10
15
20
0 15 30 45 60
Time [min]
Ver
tical
def
lect
ion
[mm
]
North span
South span
~ 5 min
Fig. 13. Creep deformations of bonded beam PH2 during first hour of
loading.
Fig. 14. First failure of bonded beam PH3 at middle support (strain gages
staggered in 100 mm overlap).
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596 593
4.4. Behavior of bonded beam with 300 mm overlap (PH4)
The bonded beam with 2!300 mm overlap lengths,
PH4, showed almost the same load–deflection and load–
rotation behavior as the beams PH1 and PH2 with 200 mm
overlaps, as shown in Figs. 6 and 9. In the buckling phase
(above 155 kN per jack), the stiffness loss was less
pronounced than in beams PH1 and PH2. The failure
occurred through buckling of the webs below one of the
jacks and was similar to the failures of beams PH1 and PH2.
The bonded joint remained undamaged.
The main experimental results are listed in Table 2. The
ultimate failure load was 178 kN per jack and the maximum
deflection reached 47 mm. Both the load and deflection
values corresponded well with those of beam PH2. The
maximum rotation at the middle support was 1.48, which is
12% less than that of beam PH1 and 23% less than that of
beam PH2.
The measured axial strain distribution in the joint was
approximated by a linear least-square fit, as was done for the
beams PH1 and PH2. The slope of the straight line fit is
smaller than that of beams PH1 and PH2, as shown in
Fig. 12. The symbols represent average values from four
strain gage readings.
4.5. Behavior of bonded beam with 100 mm overlap (PH3)
The bonded beam with 2!100 mm top overlap lengths,
PH3, showed load–deflection and load–rotation responses
similar to beams PH1, PH2 and PH4 up to approximately
120 kN. The stiffness was slightly lower than that of the
other beams, as shown in Fig. 6. At higher than 120 kN, the
stiffness also began to decrease due to the onset of buckling
in the compression flange and webs below the jacks and at
the middle support. At a load of 135 kN, however, the
behavior of beam PH3 changed distinctly from that of the
others. One part of the adhesive joint in the upper flange
failed, as shown in Fig. 14. Failure occurred partially in the
outer mat of the pultruded beam (interlaminar failure) and
partially in the interface between the beam and the adhesive
(adhesion failure). The load dropped slightly after this first
failure due to the displacement-control being used in the
loading (cf. Fig. 6). Subsequently, however, the load
increased again with increasing deflection. The load–
deflection path progressed then parallel to the single-span
beams up to an ultimate failure load of 154 kN. The ultimate
failure occurred below one of the jacks, as was observed for
the single-span beams and the other bonded beams. The
webs buckled and the same horizontal cracks formed, as
already described.
The load–rotation measurements, illustrated in Fig. 9,
showed an analogous response to Fig. 6. After the first
failure, the slope changed and the curve developed parallel
to the single-span beams. The main experimental results are
listed in Table 2. The measured axial strain distribution in
the joint was approximated by a linear least-squares fit, as
for the other beams. Fig. 12 shows that the slope of the fitted
straight line is higher than that of the other beams.
5. Discussion
5.1. Load–deflection and load–rotation behavior
The full-section elastic and shear moduli calculations
indicated that approximately 83% of the vertical deflections
were due to bending and 17% due to shear. As expected, the
deflections of the bonded beams were between the values of
the single-span beams (28% smaller) and the continuous
beams (49% higher) (refer to Table 5). The bonded beams
showed a tri-linear behavior with a first loss of stiffness due
to the bi-linear behavior of the adhesive and a second loss
due to the onset of buckling in the compression flange.
Beam PH3 showed an additional loss of stiffness due to the
failure of the bonded joint. Therefore, the bonded beams
exhibited system ductility at first (concept of ductility,
according to Table 1), however, the effect was quite
small because the adhesive was initially too flexible.
Table 5
Comparison of beams with adhesive joints and single/continuous beams
Beams Ultimate failure load Maximum deflection at failure
Beams with
adhesive
joints
Single span
(reference
155 kN)
Continuous
beam (refer-
ence 146 kN)
Single span
(reference
66 mm)
Continuous
beam (refer-
ence 32 mm)
PH1
(200 mm)
C16% C23% K32% C42%
PH2
(200 mm)
C15% C22% K24% C58%
PH3
(100 mm)
K1% C5% K17% C74%
PH4
(300 mm)
C15% C22% K29% C48%
Average
(PH3
excluded)
C15.3G0.6% C22.3G0.6% K28.3G4.0% C49.3G8.1%
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596594
Ideally, the adhesive should be much stiffer at the beginning
(elastic modulus comparable to an epoxy adhesive) and then
should behave plastically. The loss of stiffness due to local
buckling was small and it was not possible to detect if the
deformations were elastic and reversible or not. According
to [18], losses of stiffness due to local buckling may not be
elastic and reversible in pultruded sections in the post-
buckling range. The additional loss of stiffness due to the
failure of the bonded joint of beam PH3 can be considered
as additional system ductility (concept of pseudo-ductility,
according to Table 1).
The creep experiment showed that, even though a very
flexible adhesive was used, the creep deformation was kept
within admissible values due to the design-governing
deflection limit for the GFRP beams and the associated
low shear stresses of approximately 1.1 MPa in the adhesive
joint. The shear stress was calculated according to Eq. (1)
(see Section 5.2).
5.2. Estimation of the rotation angles and moment
distributions
The measured axial strains of the flanges in the joints of
beams PH1–PH3 showed a linear distribution with a
noticeable slope, with the exception of small regions at
the beginning and at the end (approximately 5% of the joint
length each, cf. Fig. 12). This indicates that the shear
stresses in the adhesive layer were almost constant over the
joint length and the magnitude of constant shear stress was
proportional to the slope of the straight-line fit to the axial
strains. Only beam PH4, which had the longest overlap
Table 6
Estimation of rotation angles and moments at middle support
Beam t (MPa) g (%) qcalc. (8) qmeas. (8)
PH1 5.7 1.0 1.3 1.5
PH2 5.7 1.0 1.3 1.8
PH3 6.2 1.2 1.2 1.5
PH4 4.3 0.5 1.1 1.4
lengths of 2!300 mm, deviated slightly from this con-
clusion. The slope of the straight-line fit for beam PH4 was
quite small and, therefore, a considerable part of the shear
stresses were transmitted towards the edges of the overlaps.
Based on the assumption of a constant shear stress
distribution in the adhesive bond, an estimate of the
resulting angles of rotation at the middle support at failure
can be made. The constant shear stress, t, is calculated from
the axial force, F, in the beam flange while the axial force,
F, can be obtained from the negative bending moment over
the middle support, M, according to Eq. (1)
t ZF
lbZ
M
ðh K tÞlb(1)
Due to the eccentricity of the cover plate, small secondary
moments were produced in the cover plate. These secondary
moments, however, were neglected. In Eq. (1) the maximum
negative bending moment is used and any decrease along
the length of the overlap is neglected. The denominator in
Eq. (1) comprises: the section height, hZ240 mm, the
flange thickness of the beam, tZ12 mm, the section width,
bZ240 mm, and the overlap length, lZ100/200/300 mm.
The elongation of the upper overlap, Dl, is composed of
the shear deformations of the adhesive layer, Dl1, and the
axial elongation of the cover plate and the top flange, Dl2, as
follows
Dl Z Dl1 CDl2 (2)
Dl1 Z agðtÞ (3)
Dl2 ZMl
ðh K tÞtbE(4)
where a is the adhesive layer thickness (2 mm), g(t) is the
shear strain of the adhesive according to Fig. 3, and E is the
full-section elastic modulus (30.0 GPa). From these
elongations, the total rotation angle over the middle support,
qcalc., is calculated as follows:
qcalc: Z arctan2 Dl
h Ca(5)
Now, the bending moment distributions are calculated for a
static system consisting of two single-span beams connected
by a non-linear rotational spring at the middle support (i.e.
the adhesive joints). The stiffness of the spring, kq, is
calculated as follows:
Dq (%) Mcalc. (kNm) Mmeas. (kNm) DM (%)
K13 K62 K67 K5
K28 K62 K53 C15
K20 K34 K36 K6
K21 K71 K66 C8
0.4
0.8
1.2
1.6
2.0
0 50 100 150 200Load per loading point [kN]
Ben
ding
mom
ent r
atio
M- /M
+ [-]
PH3
PH1PH2
PH4
Fig. 16. Redistribution of bending moments from middle support (MK) to
loading point cross-sections (MC).
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596 595
kq ZM
qcalc:
(6)
For the pultruded beams, the full-section elastic and shear
moduli given in Table 3 were considered. Due to the non-
linear behavior of the adhesive, an iterative calculation
procedure was necessary.
The results calculated for the angles of rotation (qcalc.)
and the negative bending moments at the middle support
(Mcalc.) are shown in Table 6 and are compared to the
measured values (qmeas., Mmeas.). The percentage difference
between measured and calculated values, Dq and DM, are
also indicated. The calculated rotations underestimate the
measured rotations by 13–28%. The variations from the
‘measured’ moments at the middle support (that is,
the moments calculated from the measured support reactions)
are between K5 and C15%. The largest deviations between
measured and calculated values were observed in beam PH2,
which was loaded at a different temperature.
5.3. Moment redistribution and failure behavior
Fig. 15 shows the bending moment diagrams along the
beam length for all beams at 135 kN per jack, which
corresponds to the lowest failure load (beam PC2). The
results for the single-span beams are mirrored. The
moments were calculated from the measured support
reactions. Compared to the single-span and continuous
beams, the flexible adhesive joints led to a more uniform
distribution of moments and support reactions. The
maximum moments and the support reactions at the middle
support were considerably lower as compared with the
continuous beam (PC2), which failed at this location.
Compared to the single-span beams, the cross-sections
below the jacks were more lightly loaded due to the partial
fixity at the middle support. The loads could thus be
increased in the bonded beams beyond the failure loads of
the single-span and continuous beams due to the remaining
flexural capacities in the load and support cross-sections. As
given in Table 5, the ultimate failure loads of the bonded
beams were increased by 15% as compared with the single-
span beams and 22% compared with the continuous beams
-100
-50
0
50
100
150
Ben
ding
mom
ent [
kNm
]
7.2
PC1-3
PS1-3
PH1,4
PH3
1.2 2.4 4.8 6.0PH2
Axis
Fig. 15. Bending moment distributions along beam length (at 135 kN load
per jack).
(average values excluding beam PH3). In this respect,
however, a direct comparison of the failure loads is not
possible, since the bottom cover plate additionally
reinforced the support region of the bonded beams.
Fig. 16 shows that a redistribution of the moments from
the support (MK) to the loading point regions (MC) in the
PH beams occurred. The reason for this redistribution is the
non-linear behavior of the adhesive joint. As expected,
the redistribution effect increased with decreasing overlap
length. However, the effect was again smaller than
expected, because the adhesive was too flexible during the
early load stages. Ideally, the bending moment ratio should
be constant up to the ULS load (approximately 1.5!40 kNZ60 kN, 40 kNZSLS load), and then start to
decrease.
The maximum compressive flange stresses at failure
were approximately 153 MPa (PS beams, post-buckling)
and, therefore, far below the value of 240 MPa given in the
pultruder’s design manual [12]. The short overlap of the
combined fiber mats proved to be a weak area in the beams.
The post-buckling value of 153 MPa was approximately
50% above the compressive flange stress at the onset of
flange buckling of 100 MPa, calculated according to the
Eurocomp Design Code and Handbook ([19], Eq. 4.9, with
ExZ30.0 GPa, EyZ8.5 GPa, GxyZ3.0 GPa, nxyZ0.27,
nyxZ0.09, cf. Table 3). The Eurocomp value, however,
exactly matched the calculated compressive flange stress at
the onset of buckling of the PS beams (99 MPa at a load of
100 kN, cf. Table 3).
6. Conclusions
Quasi-static and creep experiments on brittle GFRP
beams connected with flexible adhesive joints were carried
out and were compared to quasi-static experiments on
T. Keller, J. de Castro / Composites: Part B 36 (2005) 586–596596
single-span and continuous beams. The adhesively con-
nected beams showed the following characteristics:
(1) The flexible joints composed of a highly non-linear
adhesive provided a favorable redistribution of the
internal and external forces in the redundant system.
(2) In the case of adhesive joint failure, structural collapse
was prevented due to structural redundancy of the
statically indeterminate system. After joint failure, the
load was increased by 14% up to the ultimate load.
(3) Due to the almost constant shear stress distribution in
the adhesive joints arising from the low stiffness and
highly non-linear behavior of the adhesive, the rotations
in the joints could be estimated using a simple
analytical model. From these rotations, the internal
forces in the statically indeterminate system were
calculated using beam theory. Results from measure-
ments and calculations showed a good agreement.
(4) Due to the stiffness-governed design of GFRP beams,
the stresses in flexible adhesive joints were small and
creep deformations in the joints could be controlled.
The results of this research confirm the feasibility of the
proposed concept for beam structures composed of brittle
FRP components. The concept combines redundant beam
systems, and ductile or flexible adhesive joints to provide
system ductility that compensates for the lack of material
ductility inherent to FRP beams. Further research will be
carried out to define the adhesive properties required to
enhance the ductile behavior of the joints.
Acknowledgements
The authors wish to acknowledge the support of the
Swiss Innovation Promotion Agency CTI (Contract
No. 4676.1 KTS), Fiberline Composites A/S Denmark
(supplier of the pultruded profiles) and Sika AG Zurich
(supplier of the adhesives).
References
[1] Keller T. Use of fiber reinforced polymers in bridge construction.
Structural engineering documents.: International Association for
Bridge and Structural Engineering, IABSE; 2003 p. 131.
[2] Keller T, De Castro J, Schollmayer M. Load-carrying behavior of
adhesively bonded GFRP-sandwich girders. J Compos Construct
2004;8/5:461–70.
[3] Bank LC, Mosallam A, Gonsior HE. Beam-to-column connections
for pultruded FRP structures. In: Suprenant BA, editor. Service-
ability and durability of construction materials, Proc., ASCE
first material engineering congress. Proceedings of the ASCE
first material engineering congress, Denver, CO, vol. 35/4, 1990. p.
804–13.
[4] Mottram JT, Zheng Y. State of the art review on the design of beam-
to-column connections for pultruded frames. Compos Struct 1996;
35/4:387–401.
[5] Turvey GJ. Analysis of pultruded glass reinforced plastic beams with
semi-rigid end connections. Compos Struct 1997;38(1–4):3–16.
[6] Keller T, Gurtler H. Composite action and adhesive bond between
FRP bridge decks and main girders. J Compos Construct; in press.
[7] Naaman AE, Jeong SM. Structural ductility of concrete beams
prestressed with FRP. In: Taerwe L, editor. Proceedings of the
second international RILEM symposium. London: E. and F.N. Spon;
1995. p. 379–86.
[8] Morais M, Burgoyne C. Energy dissipation in sections prestressed
with FRP tendons. In: Figueiras J et al, editor. Proceedings of the
composite in construction International conference CCC 2001. The
Netherlands: Balkema/Lisse; 2001. p. 421–6.
[9] Deskovic N, Triantafillou TC, Meier U. Innovative design of FRP
combined with concrete: short-term behavior. J Struct Eng 1995;
121(7):1069–78.
[10] Razaqpur AG, Ali MM. Ductility and strength of concrete beams
externally reinforced with CFRP sheets. In: El-Badry MM, editor.
Advanced composite materials in bridges and structures. Montreal,
Canada: Canadian Society for Civil Engineering; 1996. p. 505–12.
[11] Keller T, Gurtler H. In-plane compression and shear performance of
FRP bridge decks acting as top chord of bridge girders. Compos
Struct; in press (available online).
[12] Fiberline design manual. http://www.fiberline.com/gb/home/index.
asp; July 6, 2004.
[13] Bank LC. Flexural and shear moduli of full-section fiber reinforced
plastic (FRP) pultruded beams. J Test Eval 1989;17(1):40–5.
[14] Omidvar B. Shear coefficient in orthotropic thin-walled composite
beams. J Compos Construct 1998;2(1):46–55.
[15] SikaFast 5221 fast curing 2-C structural adhesive. Technical data
sheet; 2004.
[16] Schmid M, Kieselbach R. Adhesive bonding-a challenge for design.
J Eng Integrity Soc 2001;10:24–33.
[17] European Committee for Standardization (CEN). Adhesives-
determination of shear behaviour of structural bonds: torsion
test method using butt-bonded hollow cylinders. EN ISO 11003-1,
Brussels; 2002.
[18] Bank LC, Nadipelli M, Gentry TR. Local buckling and failure of
pultruded fiber-reinforced plastic beams. J Eng Mater Technol 1994;
116:233–7.
[19] Clarke JL. Structural design of polymer composites. EUROCOMP
design code and handbook. London: E and FN Spon; 1996.