Composites Science and Technology 61 (2001) 491502

www.elsevier.com/locate/compscitech

Biaxial fatigue behaviour of a multidirectional lament-wound

glass-ber/epoxy pipe
F. Ellyin *, M. Martens
Department of Mechanical Engineering, University of Alberta, Edmonton, Alberta, Canada, T6G 2G8
Received 21 January 2000; received in revised form 3 August 2000; accepted 10 October 2000

Abstract
This paper presents the results of an experimental investigation into the biaxial fatigue behaviour of a lament-wound multidirectional glass-ber/epoxy pipe. Tests were conducted to observe the leakage (functional) failure of the composite pipe subjected
to various hoop to axial loadings. The stiness change in the axial and hoop directions, fatigue life curves, biaxial fatigue envelopes
at various lives, and failure mechanisms are discussed. It is noted that optimum ber placement in principal directions may lead to a
weak behaviour for an intermediate loading combination. It is also found that monotonic test data are not a good indicator of the
fatigue behaviour. Furthermore, imperfections in the manufacturing process lead to early damage initiation under cyclic loading,
and result in shorter fatigue lives to leakage. # 2001 Elsevier Science Ltd. All rights reserved.
Keywords: Biaxial cyclic; Fatigue testing; GRP; Filament wound; Leakage failure; Multidirectional lay-up; Pipe

1. Introduction
Fiber-reinforced polymeric composites are seen as an
alternative to products made of conventional materials
(i.e. metals) in pressure retaining components. The
important properties contributing to the above choice
are corrosion resistance and high strength to weight
ratio. However, there is a reluctance in adopting polymeric composite material in high pressure containment
structures due to a limited understanding of the behaviour of these materials under biaxial monotonic and
cyclic loading. In particular, their leakage integrity and
reliability has not been fully investigated. A further
complicating issue concerns ber orientation when the
applied load is biaxial and the biaxial stress ratio varies
during the operation.
The purpose of this study is to investigate the leakage
behaviour and fatigue life of a multidirectional lamentwound glass-ber-reinforced epoxy tube. More specically,
it is the aim of this experimental investigation to determine:
(i) deformation behaviour of the composite pipe under
three dierent applied biaxial stress ratios, (ii) short and
long-term leakage curves for each applied stress ratio,
* Corresponding author. Fax: +1-780-492-2200.
E-mail address: fernand.ellyin@ualberta.ca (F. Ellyin).

(iii) leakage envelopes (functional failure) for all applied

stress ratios, (iv) micro and macro failure modes by
physical observation and measured parameters.
This type of investigation has practical implications, in
the sense that it is a recommended procedure to qualify
pipe products under a cyclic applied load. The results of
biaxial monotonic loading for this particular lay-up
geometry have been reported by Martens and Ellyin [1].
2. Background
A review of the open literature revealed a very limited
number of studies focused on multidirectional pipes, and
even fewer dealing with the fatigue behaviour of these
structures. The majority of the few experimental investigations into lament-wound glass-ber-reinforced-plastic
(GRP) composite pipes under fatigue loading, focused
on angle-ply ber placement (Specically [+55/ 55]n)
of thin-walled asymmetric and symmetric pipes [2,3].
2.1. Angle-ply lay-up of ber-glass pipe
Frost [2] studied the long-term fatigue behaviour of a
[+55 / 55 ]n glass-ber/epoxy lament wound pipe
and concluded that the failure mechanisms for both

0266-3538/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved.
PII: S0266-3538(00)00215-3

492

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

short and long-term fatigue were weepage as a result of

matrix-cracking. The damage found after long-term
cyclic tests was notably less than that of the short-term.
The eect of cyclic frequency on [+55 / 55 ]n glass-ber/
epoxy pipes was investigated by Joseph and Perreux [3].
A review of the experimental investigations of the angle-ply
lay-ups, revealed that this type of structure is strong for
an intended stress ratio, which can be determined by
netting analysis [4]. However, outside the intended stress
ratio the lay-up is prone to transverse cracking. A multidirectional lay-up may be implemented to reinforce a
pipe for other stress ratios.
2.2. Multi-directional lay-up of ber-glass pipe
A multidirectional lay-up is intended to produce a
more damage-tolerant composite pipe under a range of
stress ratios occurring in a structural loading (i.e. internal
pressure, axial bending, etc.).
However, there exists a lack of experimental investigations focusing on multidirectional lay-ups of berglass/epoxy pipe in the literature [5,6].
The long-term fatigue behaviour of a multidirectional
ber-glass/epoxy pipe was investigated by Kujawski et
al. [5]. The lament-wound lay-up [70, 452, 70,0,
+45, 454] was tested at two pressure ratios R=0.05
and R=0.3 R Pmin =Pmax . It was noted that the fatigue
life was about four times smaller for the R=0.3 for the
same applied pressure range. Extensive delamination
through the 0 layer and transverse cracking of the
[+45 / 45 ] layers, was observed in these tests.
Wang et al. [6] investigated the long-term leakage
failure of a [662, 03, 663,0]s lament-wound pipe and
reported that this lay-up showed a better resistance to
leakage under uniaxial loading than in biaxial loading
(i.e. pressure vessel type of loading), which leaves a
concave surface in the positive quadrant of the leakage
envelope. This can be attributed to the high ber content
in the axial and hoop directions.
2.3. Fatigue testing
The procedure for determining the long-term strength
of a composite pipe is based on ASTM Standard D2992
[7]. The ASTM standard stipulates cycling the internal
hydrostatic pressure at a rate of 25 cycles per minute
(frequency of 0.42 Hz) over the full pressure range (i.e.
applied pressure ratio R Pmin =Pmax =0). There are
two loading conditions: a restrained-end condition
(pure hoop stress) and a free-end condition in which the
hoop stress is twice the axial stress (pressure vessel
loading). Some concern has been raised regarding the
above mentioned ASTM Standard for experimentally
determining the fatigue life. Kujawski et al. [8] pointed
out that a composite pipe used in the oil eld is subjected to a constant peak pressure at varying R-ratios,

and at a higher cyclic frequency than that dened by

the ASTM Standard D2992. The rationale is that the
maximum pressure remains relatively constant, but the
R-ratio and the cyclic frequency vary depending on
the type of pump or compressor used. For example, the
R-ratio and frequency would be lower for a simplex
positive displacement pump than that of a multiplex
pump [9]. The present ASTM Standard D2992 stipulated R-ratio (R=0) species a worst-case scenario for a
piping system [7]. This is not intended to be the operating pump or compressor frequency, but is equivalent to
restarting the piping system 25 times a minute [10].
Nevertheless, in the tests reported here, the ASTM
recommended procedure was used to produce experimental results useful to the present industry practice.
The eect of frequency (0.01, 1.0 and 5 Hz) on GRP
composite pipe with a [+55 / 55 ]3 lay-up has been
reported by Joseph and Perreux [3]. They concluded
that frequency plays a complex role that acts on two
levels: (i) The material life may not be intrinsic to the
specimen as the frequency increases. (ii) The interaction
between fatigue and creep is dominant. At lower frequencies the loading time was greater, thus the specimen
creeped to a greater extent.
The inhomogeneity and anisotropic aspects of this
material play an important part in the ability to resist
the eects of cyclic loading (i.e. creep, internal damage,
etc.). A study conducted by Lockett [11] concluded that
the combined anisotropy and wall thickness resulted in
inner surface strains being 50% greater than that found
on the outer surface, with the wall thickness having
more of an aect than the material anisotropy. It was
also reported that inhomogeneities caused during manufacturing led to variations in local stiness, which
caused bending strains with magnitudes equivalent to
the primary strains, thus reducing the fatigue life of the
pipe [11].
2.4. Scope of the current investigation
This experimental investigation focuses on a multidirectional lay-up to address some of the concerns not
yet reported in the published literature. To meet growing
demands set out by industry for composite pipes such as
higher internal pressures and axial loads with reliable
performance, it is imperative to better understand the
behaviour of multidirectional composite pipes. The multidirectional lay-up investigated here has a dominant
number of plies to resist high internal pressure and
added layers in the axial direction to account for axial
tension or bending. Thus, it is especially strong in both
the hoop and axial directions [1]. However, the large
variation in the ber direction between hoop and axial
plies produces a large change in the stiness of neighboring
plies. This is a cause for concern in terms of damage
leading to leakage.

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

3. Specimen, appartus and experimental procedure

3.1. Specimen
Tubular specimens were cut from 9 m (30 foot) sections of a lament wound GRP pipe to a nished length
of 348.0 mm (13.7 inch). The ends of the specimen were
potted in an epoxy to provide round end for gripping
[13]. The ber volume fraction was determined to be
0.58. The intended manufactured lay-up was:


inside 664 ; 0 ; 663 ; 0 ; 663 ; 0 ; 665 outside
The pipe had an inside diameter of 50 mm (1.950 inch)
with a variation in roundness less than  0.025 mm
(0.001 inch) and a variation from section to section of
0.5 mm (0.020 inch). The consistently round internal
surface with a surface roughness of 0.762 mm (30m inch)
compared to a typical commercial steel pipe with 8.89
mm (350 minch) surface roughness, can be attributed to
the ground mandrel on which the pipe was wound [12].
The outer surface had a nominal diameter of 62 mm
(2.440 inch) with a variation in roundness of the outer
diameter of the pipe of 1.2 mm (0.050 inch), and a
variation from section to section of  5.0 mm ( 0.200
inch). The outer surface was non-circular and often
resembled an elliptical shape. The outer surface shape
and variation in roundness were dependent on the location of resin-rich imperfections between the 66 and
the 0 layers. However, an applied protective resin
coating also caused small variations in roundness.
3.2. Multiaxial testing system
A MTS uniaxial resonant fatigue testing machine was
modied with an internal pressurization system for
biaxial testing. The servo-valve controlled 3.8:1 pressure
intensier, was connected directly to the pipe specimen
and provided the required test pressure, it was also used
to monitor leakage [13].
A data acquisition computer collected 30 data points
on every 100th cycle of the hoop strain, axial strain,
axial load, internal pressure, oil volume and axial
stroke. The internal pressure and axial load were slowly
increased simultaneously to the mean required biaxial
load before cyclic loading was initiated. The pressure
and axial loads were applied until the test was terminated.
Test stoppage occurred after the high-pressure reservoir
in the intensier was depleted, or a signicant amount
of movement in the axial stroke was recorded, both
indicating a functional failure.
Leakage was obtained from the intensier oil volume
corrected for entrapped air, hydraulic oil compression,
and specimen expansion [13]. Specimen expansion was
determined from strains obtained from the extensometers reading. To better visualize the macro-damage

493

through leakage, 5 ml of uorescent penetrant (Magnaux

Zyglo ZL-56) was added to the uid within the specimen,
which made the damaged areas highly visible under
ultraviolet light.
3.3. Testing procedure
The biaxial fatigue leakage life envelope was developed
under stress control, cycled over the full positive range
of stress. The following are the hoop to axial stress
ratios applied to the specimen to dene the leakage life
envelopes.
3.3.1. 0H:1A
The zero hoop to one axial stress ratio simulates a
pipe receiving a small internal pressure (applied to
enable detection of leakage failure) and a cyclic axial
load. An internal pressure of 2.07 MPa (300 psi) was
applied to monitor leakage development. This small
internal pressure translates into a 1H:9A applied stress
ratio or greater.
3.3.2. 2.5H:1A
The 2.5 hoop to one axial stress ratio simulates a
pressure vessel type of loading. The pipe ends are capped
and a cyclic internal pressure is applied.
The slight dierence in the pressure vessel loading
stress ratio of 2H:1A for monotonic loading, reported
by Martens and Ellyin [1], and 2.5H:1A for fatigue
loading is due to the wall thickness of the pipe and the
stress equation being dened by the mean radius rather
than the inside radius. The 2.5H:1A applied stress ratio
used in the fatigue loading is obtained by applying an
internal pressure without making an axial load correction.
3.3.3. 1H:0A
The one hoop to zero axial stress ratio simulates a line
pipe with a cyclic internal pressure. The cyclic internal
pressure determines the maximum functional capability
of the pipe.
The hoop and axial stresses were cycled at various
stress levels with a R-ratio of about zero ( 0.05) at 25
cycles per minute (0.42 Hz).
3.3.4. Stressstrain response
The fatigue life of a composite is limited by the ability
of the matrix to resist damage (i.e. matrix cracking, ber
debonding, and delamination) during cyclic loading
[14]. Upon the initiation of damage, the stiness properties degrade [15]. This degeneration of the stiness
properties can be monitored by the reduction in the
secant modulus (some times referred to as the dynamic
stiness), which is the slope of the extremities of a
stress/strain hysteresis loop. The method of monitoring
the reduction of the secant modulus has long been used
to obtain an indication of the damage accumulation in

494

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

composite materials [1619]. For the multidirectional

ber-wound pipe, the reduction in the secant modulus
can be used to determine the level of micro-structural
cracking.
The secant modulus was determined from the linear
regression calculated from the stress/strain hysteresis
loop, along with the stress intercept, as shown in Fig. 1.
This method was used in order to avoid the small
amount of shifting of the maximum and minimum
stress/strain data points collected by the data acquisition
system. The purpose of the linear regression is to obtain
a secant modulus based on the entire hysteresis loop as
well as to obtain an indication of the global trends. This
is a good indicator of monitoring the change in the
shape of the hysteresis loops as cracking progresses, and
is less sensitive to variations of the signal readings due
to noise and digital resolution.
3.3.5. Leakage curves
An important aspect in the fatigue testing of composite
pipes is damage detection. To determine the state of
damage in which micro-cracks coalesce to form larger
cracks through the wall thickness, leakage was monitored
from the oil volume of the internal pressure intensier
and correlated with changes seen in the secant modulus.
For biaxial fatigue tests of composite pipes, the axial
load and internal pressure are ramped up to a mean
loading condition, consequently compressing the
hydraulic uid. When a cyclic loading is initiated, the
uid volume decreases and increases as a result of the
compression. The minimum and maximum uid volume
remains relatively constant until creep occurs and/or
until leakage initiates. However, to ensure a stable
volume reading, independent of the variances in the
system (i.e. line noise, resolution, etc.), the average of
the total oil volume was determined.
Two stages of leakage were observed which describe a
state of damage, independent of the biaxial loading

Fig. 1. Illustration of the method used to determine the secant modulus by a linear regression t through the stress/strain data.

conditions. They are: (i) The initiation point on the

corrected leakage curve which is indicative of the
amount of micro cracking and crack coalescence that is
sucient to allow uid loss. This is a good indication of
functional failure, but it is sensitive to factors such as
changes in the specimen shape and intensier leakage.
(ii) The point of intense leakage, also known as the knee
on the corrected leakage curve, which is found from the
slope of the curve before and after the change in the
leakage rate. This change in the leakage rate is a good
indicator of the state of macro cracking through the
wall thickness. As well, the point of intense leakage
describes a state of damage independent of failure mode
(i.e. delamination causing pinholes, sudden bursting,
and uniform matrix cracking).
4. Test results and interpretation
The fatigue tests conducted along with the corresponding loading information and leakage life are summarized in Tables 13, for the 0H:1A, 2.5H:1A and
1H:0A applied stress ratios, respectively.
4.1. Stress/strain response
A typical cyclic axial stress/strain diagram for the
0H:1A applied stress ratio with a corresponding maximum axial stress level of 117.40 MPa (17,000 psi) is
illustrated in Fig. 2. The change in slope of the cycle
curve is an indication of the stiness loss of the specimen with increasing number of cycles. This is illustrated
in Fig. 3, where the secant modulus was determined
from the linear regression of the stress/strain hysteresis
loop, (Fig. 1), and is plotted along with the stress intercept versus number of cycles. An initial reduction in the
axial secant modulus is noted, which stabilizes after the
rst 2000 cycles. A reduction in the axial secant modulus for the maximum stress level of 68.84 MPa (9984
psi) also occurred within the rst 20,000 cycles before
stabilizing, gure not shown.
For the 2.5H:1A applied stress ratio, an initial reduction in both the hoop and the axial secant moduli was
noted before stabilizing. A greater reduction was
observed initially for the hoop secant modulus followed
by a gradual decrease for the applied maximum hoop
stress of 103.27 MPa (14,980 psi). The reduction in the
axial secant modulus remained gradual and both moduli stabilized around 65,000 cycles for the applied maximum hoop stress of 103.27 MPa (14,980 psi).
Only a slight reduction in the hoop secant modulus
was observed for the 1H:0A applied stress ratio.
The secant modulus determined from linear regression
of the stress/strain hysteresis loop provides a good indication of the initial stiness and the amount of stiness
reduction. The secant modulus is based on the applied

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

495

Table 1
Summary of test results for the applied 0H:1A stress ration
Test

R-ratio

Cyclic
frequency
(Hz)

Maximum
hoop stress
(MPa)

Minimum
hoop stress
(MPa)

Maximum
axial stress
(MPa)

Minimum
axial stress
(MPa)

Leakage
(initiation)
cycle

Leakage
(intense)
cycle

Last
recorded
cycle

1
2
3
4
5

0.05
0.05
0.05
0.05
0.05

0.417
0.417
0.417
0.417
0.417

8.493
8.763
8.449
8.442
11.523

8.493
8.763
8.449
8.442
11.523

117.396
100.905
91.376
79.041
68.844

5.870
8.570
4.569
3.952
3.901

6000
3000
35,000
10,000
220,000

19,000
10,000
55,000
72,000
270,000

39,000
33,000
182,300
301,200
275,800

Table 2
Summary of test results for the applied 2.5H:1A stress ratio
Test

R-ratio

Cyclic
frequency
(Hz)

Maximum
hoop stress
(MPa)

Minimum
hoop stress
(MPa)

Maximum
axial stress
(MPa)

Minimum
axial stress
(MPa)

Leakage
(initiation)
cycle

Leakage
(intense)
cycle

Last
recorded
cycle

1
2
3
4
5
6
7

0.05
0.05
0.05
0.05
0.05
0.05
0.05

0.417
0.417
0.417
0.417
0.417
0.417
0.417

219.653
218.816
164.426
137.730
127.424
110.537
103.273

13.723
13.305
10.459
8.786
6.371
7.113
5.164

83.441
83.432
62.423
52.293
49.832
42.029
41.309

4.536
4.808
3.294
2.631
2.492
2.611
2.065

100
100
1300
4400
6000
28,000
47,000

160
170
1500
5800
7800
34,000
77,000

450
350
2600
9500
13,500
50,000
124,560

Table 3
Summary of test results for the applied 1H:0A stress ratio
Test

R-ratio

Cyclic
frequency
(Hz)

Maximum
hoop stress
(MPa)

Minimum
hoop stress
(MPa)

Maximum
axial stress
(MPa)

Minimum
axial stress
(MPa)

Leakage
(initiation)
cycle

Leakage
(intense)
cycle

Last
recorded
cycle

1
2
3
4
5

0.05
0.05
0.05
0.05
0.05

0.417
0.417
0.417
0.417
0.417

219.110
166.772
142.289
118.924
71.092

10.955
8.339
7.114
5.946
3.555

0
0
0
0
0

0
0
0
0
0

400
6500
9500
15,000
n/a

800
14,600
52,000
80,000
n/a

2300
28,500
71,300
226,000
1,200,000

cyclic load and the measured strain. Since the tests were
conducted under load control, the reduction in the
stiness is mainly due to the increase in the cyclic strain.
This increase in the cyclic strain has been linked to the
initiation of cracking, which induces a non-uniform
strain in the plies constraining the crack [20]. For the
multidirectional ber-wound pipe, the stabilization of
the secant modulus can be interpreted as the saturation
limit of micro cracking within the plies.
For the 0H:1A applied stress ratio there was an initial
drop in the axial secant modulus and a stable value was
reached after a few thousand cycles, implying that a
substantial amount of damage occurs in the initial few
thousand cycles. A comparison of the rate of reduction
of the secant modulus indicates that for this loading the
drop in the secant modulus is dependent on the magnitude of the applied cyclic load.
For the 2.5H:1A applied stress ratio there is a slight
drop in both the axial and the hoop secant moduli

before stabilization. For one case a greater initial drop

was seen in the hoop secant modulus. Again, this
implies that most of the damage occurs within a few
thousand cycles.
To summarize, the largest initial reduction with quick
stabilization was seen in the axial secant modulus under
pure axial loading and the hoop secant modulus under
pressure vessel type loading, in contrast to pure internal
pressure loading. In all cases, a decrease in the secant
modulus was followed by a stabilized response, implying
a saturated crack state.
4.2. Leakage curves
Two points were recorded from the corrected leakage
curves as listed in Tables 13, the initiation point of
leakage and the initiation point of intense leakage.
Typical leakage curves are presented in Fig. 4(a), (b)
and (c) for 0H:1A, 2.5H:1A and 1H:0A applied stress

496

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

ratios, respectively. Because of the varying total oil

volume, which is the summed volume of the specimen
and the pressurization system, the leakage curves have
been made dimensionless, by dividing the average oil
volume per cycle by the average initial oil volume and
then subtracting one.
The fatigue tests were allowed to run until the oil
reservoir of the intensier was exhausted (approximately 75 ml of oil volume), and leakage curves were
plotted to observe the change in the leakage rate.
For the 0H:1A applied stress ratio, a constant internal
pressure was added for determining the leakage. The
observed initiation point of leakage, which is not
marked on the leakage plots, and the starting point of
intense leakage (NL), are recorded in Table 1. The
characteristic leakage curves for the axial cyclic loading

shows a slow initiation into leakage with a slow increase

in the leakage rate reaching a constant rate [Fig. 4(a)].
The observed initiation point of leakage and the
beginning of intense leakage (NL) for the 2.5H:1A
applied stress ratio are recorded in Table 2. The characteristic leakage curves for the pressure vessel type of
cyclic load shows a slow initiation into leakage followed
by a rapid increase in the leakage rate [Fig. 4(b)]. The
dierence in the characteristic leakage curve for the
purely axial type of loading and the pressure vessel type
of loading is noted in the transition zone [compare
Fig. 4(a) and (b)].
For the 1H:0A applied stress ratio the observed
initiation point of leakage and the starting point of
intense leakage (NL) are listed in Table 3. The characteristic leakage curve for the pure pressure type of cyclic
loading is rather dicult to predict due to the change in
the mode of failure. For a hoop stress of 219.11 MPa
(31,700 psi), a bursting failure with no initiation into
leakage was observed. With a hoop stress of 166.77 MPa
(24,190 psi), a slow initiation into leakage followed by a
rapid increase in leakage rate, is noted as shown in
Fig. 4(c). With a hoop stress of 118.92 MPa (17,250 psi)
and 71.09 MPa (10,300 psi), one observes a slow leakage
initiation followed by a rapid increase in the leakage
rate, similar to Fig. 4(a).

Fig. 2. Recorded applied axial stress versus axial surface strain

response of the composite pipe with a 0H:1A applied stress ratio. Note
the change in slope with loading cycle.

Fig. 3. Variation of secant moduli and stress intercept with the number of cycles for the 0H:1A applied stress ratio, corresponding to the
stress/strain response presented in Fig. 2. This test had a maximum
axial stress level of 117.4 MPa (17,000 psi).

Fig. 4. Leakage versus the number of cycles: (a) specimen no. 4 for the
0H:1A applied stress ratio; (b) specimen no. 7 for the 2.5H:1A applied
stress ratio; (c) specimen no. 2 for the 1H:0A applied stress ratio.

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

497

Fig. 5. Applied maximum axial stress versus number of cycles to

leakage (in loglog scale) for the 0H:1A stress ratio.

applied stress ratios, respectively. The monotonic

initiation point of intense leakage, reported in Martens
and Ellyin [1], has also been added to each gure for the
comparison. In these gures the maximum applied cyclic stress is plotted versus the number of cycles to leakage failure (NL). Also depicted in the gures is the least
square regression line (best t) of the data.
The maximum applied axial cyclic stress is plotted
versus number of cycles to leakage, in a loglog scale in
Fig. 5, for the 0H:1A applied stress ratio. Note that the
monotonic leakage failure point is only slightly higher
in comparison to the short-term fatigue life (NL=103
cycles).
For the 2.5H:1A applied stress ratio, two fatigue life
curves are plotted in Fig. 6. The maximum applied cyclic stresses in the ordinate versus number of cycles to
leakage failure in abscissa. The applied hoop to the
axial stresses (see Table 2) have a constant ratio, however, the slopes of the best ts dier slightly.
Fig. 7 shows the maximum applied hoop cyclic stress
plotted versus number of cycles to leakage failure, in a
loglog scale for the 1H:0A applied stress ratio. The
monotonic leakage failure point is marked with a continuation arrow because leakage failure was not
obtained due to the system's pressurization limitation.
A single long-term fatigue point is also depicted for
which no leakage was observed.
In comparing the leakage life curves depicted in Figs. 5
and 6, it is seen that the fatigue lives are governed by the
axial stress component. The 2.5H:1A and the 1H:0A
applied stress ratio have comparable slopes, but for the
same maximum hoop stress, the 1H:0A applied stress
ratio has a longer life. For the latter applied stress ratio
failure is governed by the hoop component, and due to
the predominant ber content in the hoop direction
longer life is achieved.

Fig. 6. Applied maximum axial and hoop stress versus number of

cycles to leakage (in loglog scale) for the 2.5H:1A stress ratio.

Fig. 7. Applied maximum hoop stress versus number of cycles to

leakage (in loglog scale) for the 1H:0A stress ratio.

The slow leakage initiation followed by a rapid

increase in leakage rate is a characteristic leakage curve
for the applied stress ratios with a substantial hoop
loading. This indicates that once micro-cracks have
been formed as a result of cyclic loading, aided by cyclic
internal pressure, they coalesce which leads to an
increase in the leakage rate. Increasing the internal
pressure allows the uid to penetrate through the
microcracks, thus causing the cracks to open and coalesce at a faster rate. Hence, a rapid increase in leakage
rate follows a slow leakage initiation stage.
4.3. Fatigue life curves
The fatigue life curves, based on the initiation point of
intense leakage, listed in Tables 13, are presented in
Figs. 5, 6 and 7 for the 0H:1A, 2.5H:1A and 1H:0A

498

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

4.4. Biaxial fatigue failure envelopes

A biaxial fatigue leakage failure envelope can be constructed to present the compiled data from the fatigue
leakage life curves of Figs. 57. The biaxial fatigue
leakage failure envelope illustrates the characteristic
cyclic behaviour of this multidirectional lay-up composite pipe at a constant life. The very short-term leakage
life 103 cycles through to the long-term leakage life, 106
cycles are extracted from the fatigue life curves in
Figs. 57 and are illustrated in Fig. 8. The latter gure
can then be used for qualifying a multidirectional pipe.
These envelopes represent the initiation of intense
leakage under a cyclic biaxial state of stress for a cyclic
load ratio, R Pmin =Pmax =0.05.
4.5. Macro failure observations
The complexities of the multidirectional pipe failure
modes are shown in Figs. 911. The addition of the
uorescent penetrant in the pressurizing uid makes the
cracks highly visible under ultraviolet light. The failure
observations of the lament wound pipe specimen are
based on visual inspection during and after testing.
Note the highly visible area at the border of the potted
epoxy end and the specimen, where a small amount of
penetrant entered underneath a small area of the
debonded epoxy (Fig. 9): however, no damage was
observed at this location.
The uniform matrix cracking of the multidirectional
pipe at a 0H:1A applied stress ratio with a maximum
cyclic axial stress of 79.04 MPa (11,460 psi) is illustrated
in Fig. 9. Uniform matrix cracking developed parallel to
the 66 angle-plies, eventually leading to leakage, as
noted on the outer surface of Fig. 9. First, small oil droplets
formed on the outer surface (weepage), subsequently wetting the entire surface. The penetrant provides a clear

Fig. 8. The biaxial fatigue leakage envelopes illustrating the characteristic cyclic behaviour of this multidirectional lay-up composite
pipe for very short-term (103) to long-term (106) cycles to leakage.

visualization of the damage. A small amount of acoustic

emission was detected during testing. It is to be noted
that the specimens maintained structural integrity after
leakage.
Leakage failure through combined matrix cracking
and delamination of the pipe for the 2.5H:1A applied
stress ratio, with a maximum cyclic hoop stress of
103.27 MPa (14,980 psi), is illustrated in Fig. 10. This
gure shows a representative sample of the combined
failure mode, which was typical for this stress ratio.
Fig. 10 illustrates the matrix cracking parallel to the
66 angle-plies, which eventually led to delamination
and leakage failure from pinhole bursting.
A dierent mode of macro damage was observed for
the 1H:0A applied stress ratio. A uniform type of matrix
cracking, parallel to the 66 angle-plies throughout the
length of the specimen, for the maximum applied cyclic
hoop stress of 219.11 MPa (31,780 psi) was observed
which is similar to the damage associated with the
0H:1A applied stress ratio (Fig. 9). Also matrix cracking, parallel to both the 66 angle-plies and 0 ber
layer was noted, which led to delamination and pinhole

Fig. 9. Photograph of the observed macro damage for the 0H:1A

stress ratio with a maximum cyclic axial stress of 79.04 MPa (11,460
psi). Note the uniform matrix cracking.

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

499

bursting. The latter is illustrated in Fig. 11, for the

maximum applied cyclic hoop stress of 118.92 MPa
(17,250 psi). This mode of damage was also seen for the
maximum cyclic hoop stress of 166.77 MPa (24,180 psi)
and 142.29 MPa (20,640 psi).
4.6. Micro failure observations
The complex ply interactions at dierent stress ratios
causes a change in the failure mode, and the results of
the accumulation of damage seen in the macro level,
originate at the micro-structural scale.
A micrograph taken of the cross-section at the center
of the gauge length, seen in Fig. 9, is illustrated in
Fig. 12. The majority of the cracking is seen as hairline
cracking in the 66 angle-plies, with a few larger
cracks running through the 66 plies. A small amount
of interface cracking between the 0 and the66 plies
was also observed. However, a greater concentration of
cracking was observed in the thinner section of the wall
thickness.

Fig. 11. Photograph of the observed macro damage for the 1H:0A
stress ratio with a maximum cyclic hoop stress of 142.28 MPa (20,640
psi). Note delamination and matrix cracking leading to pinhole
bursting.

Fig. 10. Photograph of the observed macro damage for the 2.5H:1A
stress ratio with a maximum cyclic hoop stress of 103.27 MPa (14,980
psi). Note the matrix cracking parallel to the 66 bers, and delamination.

Fig. 12. A micrograph taken of the observed uniform matrix cracking

of the 0H:1A stress ratio, corresponding to Fig. 9, magnied 40.
Observe the cracks in the 66 plies.

500

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

Fig. 13 shows a micrograph of the combined matrix

cracking and delamination of the pipe for the 2.5H:1A
applied stress ratio. The majority of the cracking did
take place in the 0 plies branching into the 66 plies.
In a few locations cracks were observed in the 66
plies parallel to the cracks in the 0 bers, as seen in the
micrograph. Even though hairline cracks were noted
throughout the 0 and 66 plies, the majority of the
cracking took place in the 0 plies. For the 2.5H:1A
applied stress ratio, cracks were more noticeable in the
thicker section of the wall thickness, in the resin rich
regions of the 0 plies.
For the 1H:0A applied stress ratio a micrograph of
the observed matrix cracking, parallel to both the 66
and 0 bers is illustrated in Fig. 14. This led to delamination and pinhole bursting. The micrograph was
taken of the cross-section through the center of the
delaminated section as seen in Fig. 11. Extensive cracking
could be seen in all of the 0 plies extending into the
66 plies. Once sectioned, the delamination cracks
could be seen visually without the aid of magnication,
Fig. 15. There was a greater concentration of cracking
in the thicker section of the wall thickness. However, the
extent of the damage observed in 66 plies is considerably less than that observed for the 2.5H:1A
applied stress ratio.

For the 2.5H:1A applied stress ratio the observed

characteristic type of damage is a combination of matrix
cracking and delamination. Unlike the 0H:1A applied
stress ratio, where the damage mode is not dependent
on the applied maximum stress, there is a dependency in
the case of 2.5H:1A applied stress ratio. Once the maximum applied hoop stress is below 127.42 MPa (18,480
psi) the damage type changed from a combined matrix
cracking and delamination to a delamination with pinhole
bursting.
The 1H:0A applied stress ratio did not seem to have a
distinct type of damage characteristic, the observed
damage appears to depend on the maximum applied

4.7. Correlation between macro and micro behaviour

In summary, the characteristic type of damage
observed in the multidirectional pipe for the 0H:1A
applied stress ratio is uniform matrix cracking, which
develops parallel to the 66 bers with a greater concentration of cracking in the thinner section of the wall
thickness.

Fig. 13. A micrograph taken of the observed matrix cracking and

delamination section of the 2.5H:1A stress ratio with the corresponding maximum cyclic hoop stress of 103.27 MPa (14,980 psi) magnied
40. Observe the cracks in the 0 ply extending into the 66 plies.

Fig. 14. A micrograph taken of the observed matrix cracking and

delamination section seen in Fig. 11 of the 1H:0A stress ratio magnied 40. Observe the extensive cracking in both the 0 ply and the
66 plies.

Fig. 15. A photograph of the observed matrix cracking for the 1H:0A
stress ratio, corresponding to Fig. 11. The photograph was taken
under ultra-violate light to observe cracking.

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

stress. Damage changed from uniform matrix cracking

to delamination with pinhole bursting to a combination
of both. Similar to the 2.5H:1A applied stress ratio, the
cracks were more notable in size in the thicker section of
the pipe wall, in the resin rich regions of the 0 plies.
Imperfections are obviously a contributor to the
initiation of internal cracking and other forms of
damage. However, some of the eects of imperfections
are dicult to determine from the micrographs and
photographs of the cross-section, like the misalignment
of bers and partial layers. The micrographs and photograph of the cross-section do give an indication of the
proximity of the ber waviness, resin rich regions in the
0 plies and voids (porosity) to the areas of cracking.
4.8. Comparison of fatigue and monotonic results
The axial strain versus hoop strain linear limit and
leakage failure points found from monotonic testing,
reported in Martens and Ellyin [1], for the multidirectional lay-up under study have been superimposed
onto the biaxial fatigue leakage life envelope, as shown
in Fig. 16. The monotonic axial strain versus hoop
strain linear limit points seem to correlate with the
short-term leakage life, 104 cycles, with the exception of
a single data point. It is dicult at this stage to determine if this relationship is coincidental or has some signicance.
However, the monotonic test data fail to predict the
long-term leakage life, 106 cycles or greater which is
needed to determine a maximum operating pressure for
the composite pipe. This combined biaxial fatigue leakage
life and strength envelope, however, does give an
insight into the complex behaviour of this multidirectional
lay-up.

Fig. 16. The linear limit of strain/strain plots and the leakage failure
points from monotonic testing are superimposed on to biaxial fatigue
life envelope. Note the relative position of the linear limit of strain/
strain and the short-term fatigue leakage life (104 cycles).

501

5. Summary and conclusions

Biaxial fatigue testing of lament wound, multidirectional ber-glass reinforced epoxy pipes yielded a
number of interesting results. The results from fatigue
testing showed that the lay-up had a greater resistance
to fatigue loading in both the axial and hoop directions
than a pressure vessel type of loading (2.5H:1A applied
stress ratio).
The small reduction in the stiness of this multidirectional lay-up made it dicult to visualize the global
trends found in the stress/strain hysteresis loops.
Instead, the secant modulus was determined from linear
regression calculated from the stress/strain hysteresis
loop, along with the stress intercept.
The following conclusions are drawn from the
experimental results and failure observations of biaxial
fatigue tests:
. The type of accumulated damage, the decrease in
secant modulus with cyclic loading and the leakage
are shown to be dependent on both the applied
maximum biaxial stress and the applied biaxial
stress ratio.
. In all cases there was a noticeable decrease in the
secant modulus which stabilized after a certain
period. The level of reduction in the secant modulus
and the rate at which it stabilizes, indicate that
there is a connection between the stress ratio and
the maximum applied stress. The largest initial
reduction was seen in the axial secant modulus
under pure axial loading and in the hoop secant
modulus under pressure vessel type loading
(2.5H:1A). The stabilized response appears to be
indicative of cracking reaching a saturation state.
. Each stress ratio produced a characteristic leakage
curve. The 0H:1A applied stress ratio had a slow
leakage initiation with a slow increase in the leakage
rate until rapid leakage. The 2.5H:1A applied
stress ratio had a slow leakage initiation followed
by a rapid increase in the leakage rate. However,
the 1H:0A applied stress ratio had a change in the
leakage curve, ranging from a burst type of leakage
to slow leakage initiation with a slow increase in
the leakage rate until rapid leakage, the latter
being similar to the 2.5H:1A applied stress ratio.
Increasing the internal pressure allows the uid to
penetrate through the micro-cracks, causing the
cracks to open and coalesce, which leads to a rapid
increase in the leakage rate after a slow leakage
initiation.
. The fatigue leakage life curves for the 0H:1A and
2.5H:1A applied stress ratios indicate that leakage
in both these ratios is governed by the axial stress
component. The 2.5H:1A and the 1H:0A applied
stress ratio have comparable slopes, but the 1H:0A

502

F. Ellyin, M. Martens / Composites Science and Technology 61 (2001) 491502

applied stress ratio has a longer life. This is

because there is a predominant ber content in the
hoop direction, hence a longer life is achieved.
. The characteristic type of damage observed in the
multidirectional pipe for the 0H:1A applied stress
ratio was a uniform matrix cracking, which developed parallel to the 66 bers. The characteristic
type of damage observed for the 2.5H:1A applied
stress ratio was a combined matrix cracking and
delamination, which was dependent on the applied
load. The 1H:0A applied stress ratio did not seem
to have a distinct characteristic type of damage.
The observed damage, which ranged from a uniform
matrix cracking to a delamination with pinhole
bursting to a combination of both, appeared to be
dependent on the maximum applied stress. The
observed damage indicated that the amount of
axial tension in the specimen governs the uniformity
of the matrix cracking, where internal pressure
governed the amount of delamination.
. Imperfections contributed to the initiation of internal
cracking and other forms of damage. However, the
eect of some defects was dicult to determine from
the micrographs and photograph of the cross-section.
The imperfections included the misalignment of
bers from the intended angle of the lay-up and
the existence of partial layers. The micrographs
and photograph of the cross-section did give a
good indication of the proximity of damage region
to the ber waviness caused by resin rich regions in
the 0 plies and voids (porosity) of a certain size.
Acknowledgements
The results reported here are part of a larger research
program of the NOVA (TCPL)/NSERC Senior Industrial Research Chair. The Chairholder (F. Ellyin)
Wishes to thank the industrial partners and researchers
in his group who contributed to this investigation. In
particular, we acknowledge the contribution of the
technical sta, B. Faulkner and I. Butter to this project.
References
[1] Martens M, Ellyin F. Biaxial monotonic behaviour of a multidirectional glass ber epoxy pipe. Composites Part A
2000;31:100114.
[2] Frost SR. Predicting the long-term fatigue behaviour of lament
wound glass-ber/epoxy matrix pipes. Proceeding of ICCM-10,
vol. I: fatigue and fracture for publication, p. 64956.

[3] Joseph E, Perreux D. The eect of frequency on the fatigue performance of lament-wound pipes under biaxial loading: experimental results and damage model. Composites Science and
Technology 1997;57:35364.
[4] Tew BW. Preliminary design of tubular composite strucres using
netting theory and composite degradation factors. Journal of
Pressure Vessel Technology, Transations of the ASME
1995;117:3904.
[5] Kujawski D, Ellyin F, Culen M. The fatigue behaviour of lament-wound berglass/epoxy tubes under cyclic pressure. Journal
of Reinforced Plastic and Composites 1998;17:26881.
[6] Wang SS, Srinivasan S, Cole B. Long-term leakage failure and
life prediction of ber composite laminate tubing under combined
internal pressure and axial loading: theory and experiments. In:
Second International Conference on Composite Material for
Oshore Operations. CMOO-2, University of Houston, Houston,
TX, 1997.
[7] Standard Practice for obtaining hydrostatic or pressure design
basis for ``berglass'' (glass-ber-reinforced thermosetting resin)
pipe and ttings. American Society for Testing Materials
(ASTM) Designation: D2992-91.
[8] Kujawski D, Chiu AS, Ellyin F. An eective stress parameter for
long-term fatigue strength of ber-reinforced plastic pipes. Journal of Composite Technology & Research 1996;18(3):21520.
[9] Karassik IJ, Kratzsch WC, Fraser WH, Messine JP. Pump
handbook. 2nd ed. New York: McGraw-Hill, 1986.
[10] Nova Pipeline. Functional loads. Unpublished manuscript from
Nova Gas Transmission Ltd. (NGTL), Calgary, Alberta,
Canada, 1990.
[11] Lockett FL. Eects of anisotropy and inhomogeneity on GRP
pressure pipe. Composites 1980;20916.
[12] Wess M. Personal communination, 2 November 1998. NOVA
Research and Technology Corporation, Fluid Dynamic Group,
Calgary, Alberta, Canada.
[13] Martens M. Biaxial monotonic and fatigue behaviour of a multidirectional glass ber reinforced epoxy pipe. Master's thesis.
Department of Mechanical Engineering, University of Alberta,
Edmonton, Alberta, Canada, 1999.
[14] Chamis CC, Smith GT. Resin selection criteria for ``tough''
composite structures. AIAA Journal 1985;23(6):90211.
[15] O'Brien TK, Reifsnider KL. Fatigue damage evaluation through
stiness measurements in boronepoxy laminates. Journal Composite Material 1981;15:5570.
[16] Hwang W, Han KS. Fatigue of composite fatigue modulus
concept and life prediction. Journal of Composite Materials
1986;20:15465.
[17] Hwang W, Han KS. Fatigue of composite materials-damage
model and life prediction. In: Lagace PA, editor. Composite
material: fatigue and fracture, vol. 2. ASTM STP 1012 Philadelphia: American Society for Testing and Material, p. 87102.
[18] Lee LJ, Yang JN, Sheu DY. Prediction of fatigue life for matrixdominated composite laminates. Composites Science and Technology 1993;46:218.
[19] Reifsnider KL, Schulte K, Duke JC. Long-term fatigue behaviour of composite materials. In: O'Brien TK, editor. Long-term
behaviour of composites. ASTM STP 813. Philadelphia (PA):
American Society for Testing and Material, 1983. p. 13659.
[20] Talreja R. Transverse cracking and stiness reduction in composite laminates. Journal of Composite Material 1984;19:355
75.