Accepted Manuscript

Damage recovery after impact in E-glass reinforced Poly(ε-caprolactone)/epoxy

blends

Amaël Cohades, Véronique Michaud

PII: S0263-8223(17)31361-2
DOI: http://dx.doi.org/10.1016/j.compstruct.2017.08.050
Reference: COST 8808

To appear in: Composite Structures

Received Date: 28 April 2017

Accepted Date: 10 August 2017

Please cite this article as: Cohades, A., Michaud, V., Damage recovery after impact in E-glass reinforced Poly(ε-
caprolactone)/epoxy blends, Composite Structures (2017), doi: http://dx.doi.org/10.1016/j.compstruct.2017.08.050

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Damage recovery after impact in E-glass reinforced Poly(ε-
caprolactone)/epoxy blends

Amaël Cohades, Véronique Michaud*

Laboratory for Processing of Advanced Composites (LPAC), Institute of Materials, Ecole Polytechnique
Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland

* Corresponding author: Tel. +41 21 6934923, e-mail address: veronique.michaud@epfl.ch

Abstract. Damage recovery after low-velocity impact has been assessed in woven E-glass fibre-reinforced
polymer composites with an epoxy matrix and a blend of epoxy and 25vol% of poly(ε-caprolactone) (PCL).
Impact was carried out at three energy levels (8.5, 17, 34 Joules) and composites with epoxy-PCL blends
demonstrated similar energy absorption capacity as compared to pure epoxy composites even though the extent
of damage (quantified by C-scans and optical microscopy) was higher. Ultimate compressive residual strength of
the modified composites was, for the different impact energy levels, 23 to 33% lower as compared to unmodified
composites. Healing efficiency after a thermal mending cycle at 150°C for 30 minutes has been quantified using
three complementary characterization methods; impact damage could be recovered from 20% to 100%
depending on the impact energy level. These modified matrix composites are thus able to fully recover low-
velocity impact damage at energy levels often met in real structures.

Keywords: self-healing materials, fiber reinforced polymers, polycaprolactone, impact testing, compressive
residual strength

1. Introduction

While fibre reinforced polymers (FRPs) display a very good strength to weight ratio and are increasingly used
for many structural applications, their weakness still lies in out-of-plane loading events. As an example, low
velocity impact damage is of main concern in aircrafts where damage events can be attributed to operational or
maintenance activities and tend to decrease strength, durability and stability of the structure [1]. The compressive
residual strength of composite structures is thus often a property of interest which has been standardised with the
development of Compression After Impact (CAI) tests [2]. Several methods to repair damage in composite
structures are currently in practice or under development (see for example [3]) but they are often complicated
and time consuming. In particular, low-velocity impact damage is often barely visible and thus cannot be readily
identified. Because low velocity impact damage generally leads to matrix microcracking and interfacial
debonding, but not to fibre failure, a solution to these problems would be to give the matrix the ability to repair
itself with minimum manual intervention (for example using moderate heat). Over the past 25 years, the field of
self-healing thermoset polymers has been widely investigated with two main approaches: (i) the use of extrinsic
capsules or vascular systems into the thermoset matrix which ruptures during crack propagation and releases a
healing agent (reactive monomers or solvents) that will flow and polymerise into the damage volume [4,5]; (ii)

1
the use of an alternative matrix, with the intrinsic ability to heal [5,6]. While both approaches have demonstrated
their interest into neat resin [4], their integration to composites is more challenging and less investigated so far.

Extrinsic systems, under the form of capsules, have been successfully integrated to FRPs, but demonstrated
limited healing and resulted in a loss of mechanical properties both when assessing interlaminar crack opening
[7,8] and impact induced damage [9,10], mainly because the damage volume is too large to be filled by the
capsule content. On the other hand, vascular systems have demonstrated better healing properties, especially
with impact induced damage. Williams et al. [11] demonstrated, in a carbon fibre reinforced polymer with a
resin filled hollow glass fibre system distributed at specific interfaces, a compressive strength recovery of 84%.
While this value has been measured for impacts at 2.2 Joules energy (i.e. generating small matrix cracks), this
system was not able to efficiently heal larger impact damage and the importance of optimizing the location, the
viscosity and the amount of resin necessary to fill the damage volume was highlighted. Norris et al. [12]
demonstrated that the above mentioned optimizations enabled this system to fully recover compressive strength
after impacts up to 20 Joules. Even though vascularized systems demonstrated the ability to fully heal large
damage extents, using such multifunctional systems in real life structures remains difficult due to their inability
to (i) provide multiple healing to a single damage and (ii) heal low levels of damage that do not rupture a
vascule.

Intrinsic healing systems have also been studied both in neat resins [5,6] and when integrated to FRPs [13,14].
An alternative approach to overcome the need to develop a fully new matrix chemistry is to use a miscible
[15,16] or immiscible [17–20] blend between the base epoxy and a low melting point semicrystalline
thermoplastic. Miscible blends, whose healing principle relies on physical and chemical phenomena once crack
faces are in contact, have been integrated to glass fibre reinforced composites by Hayes et al. [15] and
demonstrated an impact damage area recovery of 30% after a thermal cycle of 2h at 130°C and for an impact
energy of 2.6 Joules. For immiscible blends, the healing phenomenon, as compared to miscible blends, moreover
provides melting and volume expansion of the thermoplastic followed by melt flow into the damage volume.
This healing approach, even though it requires external activation such as heat or light to trigger repair, is
repeatable and latent. Depending on the process parameters and the constituent concentrations, immiscible
thermoplastic/thermoset blends may present two limiting morphologies: (i) a particulate thermoset phase
embedded in a thermoplastic matrix (e.g. [17,18]) or (ii) a particulate thermoplastic phase embedded in a
thermoset matrix (e.g. [19]). While polymerisation-induced phase separation (PIPS) of an initially miscible blend
of thermoplastic and uncured resin governs morphology (i) under certain conditions with some crosslinkers and
was demonstrated for many thermoplastics (polycaprolactone (PCL) [21], poly(ether sulfone) [22],
polyetherimide [23], poly(methyl methacrylate) [24] and polystyrene [24]), simple mixing of the thermoplastic
within the thermoset or again the PIPS process lead to morphology (ii). At intermediate blend compositions, the
PIPS process can also give rise to a range of co-continuous morphologies [17].

Immiscible blends consisting of a thermoplastic phase (mainly concerning the ionomeric copolymer
polyethylene-co-methacrylic acid (EMAA) under the form of particles, stitches or layers) embedded in a
thermoset matrix have been integrated to FRPs and demonstrated full recovery in Mode I crack opening after a
thermal cycle of 30 minutes at 150°C while not affecting the structure integrity [20,25,26]. This efficient healing
ability with the EMAA copolymer resides in: (i) its thermal expansion capacity (7 times higher than epoxy [19])

2
which allows the thermoplastic to flow into the crack and bridge its faces, and (ii) its strong adhesion capability
to epoxy [27].

Miscible blends consisting of interconnected thermoset particles embedded in a thermoplastic phase have been
studied for thermal healing applications in neat resins [17,18] and further integrated to FRPs [28]. These studies
used PCL as thermoplastic and epoxy resin cured by 4,4’-diaminodiphenylsulfone (DDS) as thermoset. This
combination, cured at 180°C, allowed the PIPS process. The interest of these systems for thermal healing
application at a thermoplastic concentration above 23vol% in the blend resides in: (i) the interconnection
between epoxy particles which provides load bearing capacity to the structure, and (ii) the expansion capability
of PCL which is around 10 times larger than that of epoxy and provides, as for blends with EMAA, a differential
expansive bleeding mechanism to fill cracks. Neat blends with 25vol% PCL exhibited a healing efficiency after
Mode I crack opening in excess of 70% (after a thermal cycle at 150 °C for 30 min which allows the PCL to
expand by 14% [29]) over multiple cycles, while retaining suitable room-temperature mechanical properties (a
tensile modulus of 1.5 GPa and a tensile strength of about 16 MPa) [17]. The relatively mild healing temperature
has been chosen to stay below epoxy glass transition temperature and thus keep the system integrity [17]. These
blends were further integrated to E-glass fibre reinforced composites [28] through a Vacuum Assisted Resin
Infusion Moulding (VARIM) process and demonstrated similar storage modulus, flexural modulus and strength
as compared to pure epoxy composites. However, similar to neat resins, the composites with epoxy-PCL blends
demonstrated a decrease of 40% in interlaminar fracture toughness as compared to pure epoxy composites,
which has been attributed to the confinement of the ductile PCL phase by the stiff epoxy particles. A comparison
of main mechanical properties between composites made of pure epoxy and blends of epoxy and 25vol% PCL is
given in Table 1. Subjecting the damaged samples to thermal cycles at 150°C for 30 minutes led to 45%
toughness recovery and 100% stiffness recovery over multiple cycles, demonstrating the interest of this system
and the need to assess damage healing created under industrially relevant loading conditions.

In the present work, we investigate healing of impact damage in FRPs with phase separated epoxy-PCL blends
as a matrix and woven glass fibres as reinforcement. FRPs were manufactured with a 25vol% PCL blend (as well
as with pure epoxy for comparison) through VARIM at high temperature. Different damage severities have been
imparted to the samples by low-velocity impact tests up to 34 Joules, and the healing capacity, after a thermal
cycle at 150°C for 30 minutes, has been measured with three complementary methods: (i) recovery in impact
damage area by means of C-scans, (ii) theoretical crack thickness closure by optical microscopy observations
and (iii) recovery in compressive residual strength after an impact event.

2. Materials and methods

2.1. Materials

Epon™ 828EL (Momentive), a widely available diglycidyl ether bisphenol A (DGEBA) resin with a molar mass
of 340.41 g/mol and a molar mass per epoxide group of 185-192 g/eq was blended with PCL (CapaTM 6500, Mw
≈ 50,000 g/mol, Perstorp) and then cured with DDS (DDS 98 %, ABCR, molar mass = 248.3 g/mol, 2:1 molar
ratio with respect to the epoxy). Two types of resin have been prepared: (i) pure epoxy resin, (ii) epoxy-PCL
blends containing 25vol% of PCL. The PCL volumetric ratio was determined for the liquid blends, assuming a
density of 1.145 g/cm3 for the PCL and a density of 1.34 g/cm3 for the epoxy-DDS. The glass fibre

3
reinforcement was a woven twill 2x2 E-glass fabric, with a nominal areal weight of 390 g/m2, 6 end/cm for warp
fibres and 6.7 picks/cm for weft fibres, fibre diameter of 9 µm, yarn thickness of 0.45 mm, warp tex of 68x5 and
weft tex of 272, from Suter-Kunststoffe AG.

2.2. Samples preparation and processing

The composite plates were processed by Vacuum Assisted Resin Infusion Moulding (VARIM). Sixteen layers of
fibre reinforcement were cut in 350x250 mm rectangles and stacked with a sequence of [(+45/-45)/(0/90)2/(+45/-
45)]4, similar to our previous study [28]. A target fibre volume fraction (
 ) of 50 vol% and a final plate
thickness of 5 mm were sought. The preforms to be infused were prepared following the lay-out as depicted in
Figure 1. In order to ease resin infusion, the inlet has been placed at the middle of the infusion design and the
flow medium both at the top and bottom of the reinforcement.

Blends of epoxy with 25vol% PCL were prepared following a similar procedure to that of Luo et al. [18] and of
our previous study [17,28]. The liquid epoxy and the PCL pellets were first melt-mixed in glass jars at 120 °C by
mechanical stirring at 100 rpm for 1 h with a three-blade propeller (Caframo, Canada, 1”, 5/16” bore). After 1
hour, the temperature was increased to 140 °C, the stirring rate was increased to 400 rpm, and DDS (in powder
form) was slowly added to the mixture. When the mixture became transparent, it was degassed at 120 °C for 20
min and infused into the composite in an oven set to 140°C, which ensured a blend viscosity below 1 Pa·s [28].
The plates then underwent a curing treatment at 180°C for 3h, which previously showed full resin
polymerization [17,28]. Samples were cut from the plates (dimensions are given in Figure 1) using a Maïco saw
equipped with a diamond blade. For each sample, the weight was recorded, the thickness measured in 3 points,
averaged and the fibre volume fraction calculated. In addition, for every epoxy-PCL composite process, a small
droplet of the blend was placed in between two glass slides in order to control by optical microscopy (Olympus
BX-60) that the same morphology as in our previous study [17,28] (e.g. epoxy particles with a surrounding PCL
matrix) was obtained.

Composites made of pure epoxy are named "Plain" in the following and contained a fibre volume fraction of
45.4 ± 0.7% whereas composites made of epoxy-PCL blends are named "PCL(25") in the following and
contained a fibre volume fraction of 48.0 ± 0.7%.

2.3. Impact testing

The low-velocity impact behaviour of the produced samples was assessed following ASTM 7136 [30] with a
Rosand impact testing machine. The load imparted by the 5.5 kg impactor was recorded by a 60 kN Kistler load
cell. Thus, the velocity, impactor displacement as well as energy absorbed as a function of time could be
calculated. Also, the velocities before and after impact were measured through light gates placed just above the
samples and the energy dissipated within these samples deduced. The potential energy, E, imparted to the
specimen prior to drop was measured following the ASTM standard [30]:

 =  ℎ (1)

with CE the specified ratio of impact energy to specimen thickness, 6.7 J/mm, and h the nominal thickness of the
specimen. In order to observe different damage areas, three energies have been tested, corresponding to E, E/2

4
and E/4 which represents respectively around 34, 17 and 8.5 Joules energies and around 3.0, 2.2 and 1.5 m/s
velocities. Notice that because healing is the main concern of this study, the energy was each time adapted to
each specimen thickness in order to create the same damage amount. At least 5 samples were tested for each
energy and system.

2.4. C-scans

The damage areas were quantified by means of C-scan tests. Each sample (protected with a Teflon tape on the
sides to prevent water penetration) was placed in a water bath and scanned with a 2.5 MHz piezoelectric
transducer at steps of 0.2 mm. 2D-maps of the sample (squares of 50x50 mm2) were generated, analysed with
TomoView software to show the variation in the sound wave attenuation when damage was present. Damage was
defined as locations where 80% of the sound wave was attenuated. To quantify healing, a first C-scan was taken
after impact, then a thermal cycle was performed at 150°C for 30 minutes before a second C-scan. Healing
efficiency was calculated as follows:

  
 =1− (2)


with Ahealed and Avirgin the damage area in the virgin and healed state respectively. At least 3 samples were
scanned for each energy, condition and system.

In order to further use these samples for CAI tests, they were placed after C-scan at 40°C for 1h in order to dry
potential infiltrated water. DMA measurements in single-cantilever mode (TA Q800, TA Instruments) of the
samples as produced, and compared to samples placed in water for 30 minutes, then dried for 1h at 40°C showed
similar storage modulus over a temperature range of 0°C to 200°C, thus demonstrating that C-scanning followed
by a drying procedure did not affect the samples mechanical properties.

2.5. Morphological characterization

Transverse cuts at the impact location of PCL(25) specimens impacted at the three different energies (i.e. 34 J,
17 J and 8.5 J) were embedded in Epofix Resin, then polished until 1 µm size diamond disc polishing, and
observed with a reflexion optical microscopy (Olympus BX61) having an automated plate displacement. A map
of the impact damage was built and the crack size distribution measured with AnalySIS pro software. Knowing
(i) the crack size distribution, (ii) the PCL expansion capacity over free surfaces at the healing temperature [29]
and (iii) the damage size for each impact measured from C-scans (thus the amount of free surfaces over the 15
composite interfaces), a theoretical ratio of cracks that could be healed was determined.

2.6. Compression After Impact (CAI)

The ultimate compressive residual strength, FCAI of the impacted samples was assessed following the ASTM
standard D7137 [2] with a 600 kN Schenk universal testing machine. At least 3 samples were tested for each
energy, condition and system. Compressive properties of undamaged samples were also tested as reference.
Healing efficiency was defined as:

5
  ,    −  ,  
= (3)
,  − ,  

with respectively FCAI,virgin, FCAI,impacted and FCAI,healed, the compressive residual strength of the samples before
(virgin) and after impact (impacted), and after healing (healed) at 150°C for 30 minutes.

3. Results and Discussion

3.1. Impact testing

The impact force as well as impact energy levels during tests are presented on Figure 2 for the Plain and
PCL(25) samples. As expected, as the impact energy increased, the peak load as well as the energy during test
increased. The load increased for both cases with some oscillations until the peak load before a smooth decrease
until reaching zero. This behaviour, typical of conventional FRPs [1,30–33] is attributed to growing ply
delamination in the sample when the impactor penetrates while during rebound no further delamination and thus
no peak in the load curve is present. The first peak in the load response is representative of the first delamination
within the sample (i.e. Hertzian failure [31,34]) and appeared earlier (at lower load) for the PCL(25) sample.
Indeed, resistance to crack propagation (i.e. toughness) is lower for the PCL(25) composite as compared to the
Plain case [28]. This lower resistance to crack propagation and thus delamination was further confirmed by the
amount of peaks observed on the force-time curves for both cases and by the measurement of the damage area
(see §3.2). In addition, this first load drop is related to matrix strength [35], again consistent with the fact that the
matrix in PCL(25) samples was of lower strength, as demonstrated in our previous study [17]. While Hertzian
failure was more difficult to detect in Plain specimens, PCL(25) ones demonstrated that the position of the first
load drop was constant for varying impact energies, as reported in earlier studies [36]. Still focusing on the load
curve, the peak load appeared later for the PCL(25) as compared to the Plain samples, indicating that more
damage was accumulated within the PCL(25) system, again in accordance with the composite intrinsic toughness
and its lower resistance to delamination. The level of impact force at which this peak load appeared was different
for the two assessed systems and did not follow the same trend between the three impact energies used. This
behaviour, observed previously elsewhere [34,37], indicated a difference in strain-rate sensitivity between the
two matrices.

Impact energy during test was calculated from the force-time curves following ASTM standard D7136 [30] and
is shown on Figure 2 (right axis) for the two assessed systems. Energy increases up to a peak value before being
partly restituted to the impactor and reaching a final value which can be considered as the amount of dissipated
energy by the impactor within the impacted sample. However, this last quantity does not consider the possibility
for other energy consumption mechanisms to take place, including friction with the linear guide during falling,
deformation of the grips, energy transfer to the impactor or the base machine [32]. We therefore quantified
energy dissipation by measuring the inbound and rebound velocities of the impactor with the help of light gates
placed right above the samples. The energy dissipated during the entire impact process is represented in Figure 3
for the three different energies and appeared to be invariant between the two assessed systems. Indeed, while the
delamination area between two samples made of two different matrices can be different because of their intrinsic
delamination resistance, the energy absorption mechanism during impact is mainly related to the capacity of the

6
fibres to absorb energy [35]. For two samples having different matrices, but similar lay-up, reinforcement
architecture and number of plies, the magnitude of energy dissipation will thus be the same [38]. It has been
demonstrated that fibre volume fraction has an influence on the impact response of composites [39], however, it
is important to remember that in a concern to create the same damage amount in each sample before the healing
process, the impact energy was adapted to the thickness of the tested sample (Eq. (1)) and therefore the variation
in impact properties resulting from a variation in fibre volume fraction should have been avoided.

Figure 4 presents the impact force as a function of the composite deflection response (integrated from the force-
time histories) for the two assessed systems and the three impact energies. Oscillations, as in Figure 2, were
characteristic of damage initiation and propagation within the samples and were found to be lower for lower
impact energy levels. This plot is of interest as its slope indicates the stiffness of the sample, which by evidence
is the same within each sample for the three different energies. However, stiffness differs between Plain and
PCL(25), indicating that it was different for the two assessed systems, as expected from our previous studies
[17,28]. Quantification of stiffness was however not performed here as many assumptions, which might bias the
results, were performed during the curve determination [32]: neglecting friction between the impactor and the
specimen, the damping effect in the structure, the gravity during the impact event, or assuming the impactor as a
rigid body.

In addition to the analysis of the force-time histories during the impact event, the observation of the damage
shape was performed through optical images (Figure 5). The impact showed a typical nut shape, both for Plain
and PCL(25) systems, in accordance with the stacking sequence used [40]. The damage area was higher for
PCL(25) because of its lower resistance to crack propagation, as discussed before. Also, the damage was larger
on the back face of the laminate, a combined effect between top and bottom surface where the impact event
induced respectively compressive and tensile stresses, as already discussed elsewhere [41]. As the optical images
shown in Figure 5 display translucent samples, the true observed damage area was not influenced by larger back
face damage. These images were taken before and after a thermal mending cycle at 150°C for 30 minutes and
showed the capacity of the modified composites (PCL(25) samples) to heal impact induced damage. However, as
quantification of the damage extent through these optical images was not conclusive due to the inability to
exactly determine matrix microcracking (i.e. the exact surrounding of the damage), C-scans were performed.

3.2. C-scans

Figure 6 depicts the impact damage area as measured by C-scans on Plain and PCL(25) samples impacted at the
three different energies. As expected from the impact behaviour (see Figure 2), the damage extent, linked to the
system intrinsic toughness and strength, was higher in PCL(25) samples as compared to Plain ones. However,
the progression in damage area for the two samples did not follow the same trend: at low impact energies, the
slope of the curve was nearly the same for the two samples while it was different at higher energies. During the
impact event, many different modes of failures are found in FRPs, including [37,42]: matrix damage,
delamination and fibre damage. These three modes appear successively during the impact event and when
increasing the impact energy. While at low energies, matrix cracking occurs due to compression, tension or
shear, delamination is produced by interlaminar stresses. These two modes depend on the material's toughness

7
and do not give rise to the same damage progression, which explained the damage area evolution observed in
Figure 6.

C-scans of samples impacted at the three energies were performed before and after a thermal mending cycle at
150°C for 30 minutes. This procedure, realised on at least three samples per condition, allowed determining the
capacity for damage area recovery of PCL(25) samples following Eq. (2). Examples of C-scans before
(Damaged) and after the healing cycle (Healed) are depicted on Figure 7 for the three impact energies, colours
represent the sound intensity and damage was defined when 80% of the sound intensity was lost. Even though
the damage area could not be fully recovered, the efficiency of the thermal mending cycle increased as the
impact energy decreased. Indeed, for impact energies of 34, 17 and 8.5 Joules, the healing efficiency in terms of
damage area recovery was respectively of 39.9% ± 0.6%, 55.4% ± 2.7% and 94.8% ± 2.7%. At high impact
energy, two mechanisms are responsible for the incomplete recovery of damage: (i) fibre deformation and
damage and (ii) formation of large cracks. The repair system being efficient only in the matrix, fibre deformation
and damage will be permanent within such a structure after impact damage at high energy, showing the
limitation of this healing mechanism. However, at low energies, when the impact event induced only matrix
microcracking, incomplete healing was also observed. The expansion capacity of PCL (14% at 150°C) in such a
system varies between 35 and 53 microns (see §3.3 for calculations), therefore the impact event should have
induced some thicker cracks which could not be healed. At low damage extent, the echo signal from the C-scan
was reflected and gave information on the transverse crack thickness. In a large damage however, the echo signal
was blocked and could not provide information on the crack thickness distribution within the sample. Transverse
cuts of the damaged samples were thus performed (see §3.3) to provide understanding of the healing mechanism
of such system after impact damage.

3.3. Morphological characterization

Figure 8 shows reconstructed cross-sectional area taken from 200 optical microscopy images, for the three
assessed impact energies in PCL(25) samples and in the damaged state (i.e. unhealed). As for impact data and C-
scans, when the impact energy decreased, the amount of damage (i.e. cracks) decreased. This mapping also
revealed the type of damage mechanism (matrix damage, delamination and fibre damage [37,42]) occurring at
each impact energy: (i) at 8.5 Joules, only matrix microcracks could be observed, (ii) at 17 Joules, some
delamination was also present, (iii) at 34 Joules, higher amount of delamination was present and fibre damage
appeared. 34 Joules energy led to a pine tree pattern, typical in FRPs [40]. Notice also that the infusion quality of
the composites could be confirmed within the undamaged zones, showing no porosities, as expected from our
previous study [28].

From the maps in Figure 8, the crack thickness distribution due to the impact event was measured for the three
impact energies. Thickness distribution over each sample is depicted on Figure 9 and fitted with a standard
Gaussian procedure. As expected from optical observations, the amount and size of cracks increased when the
impact energy increased. Knowing the amount and size of cracks further allowed determining a theoretical
healing efficiency from the theoretical PCL volume expansion within each sample. Indeed, from the study of
Rodgers [29], PCL has a volume expansion capacity of 14% at 150°C (i.e. the healing temperature). In addition,
from C-scans data, it is known that the damage area was on average of 2.39, 3.94 and 9.08 cm2 for impact

8
energies of respectively 8.5, 17 and 34 Joules. Assuming that these damage areas were present on the entire
thickness of the laminate (i.e. 15 interfaces) and knowing sample dimensions, the PCL expansion capacity over
the sample free surfaces, and thus its crack filling capacity for impacts of 8.5, 17 and 34 Joules could be
determined and was respectively of 53, 47 and 35 µm. As the energy of impact increased, the PCL filling
capacity was therefore theoretically decreased because of a higher amount of free surfaces for the PCL to expand
and thus less through thickness expansion capacity. Using the crack thickness distribution of Figure 9, the PCL
could thus fill 99.1%, 69.9% and 32.3% of the cracks created by impact events at energies respectively of 8.5, 17
and 34 Joules. These data correlated well with damage area recovery observed by C-scans.

3.4. Compression After Impact (CAI)

CAI strength has been recognized since many years as the property suffering the largest from impact damage as
compared to undamaged materials [40], quantification of damage recovery through this technique is thus of great
interest. Figure 10 depicts the compressive residual strength of Plain, PCL(25) samples in the damaged state (i.e.
after the impact event) and PCL(25) samples in the healed state (i.e. after the impact event and a healing cycle at
150°C during 30 minutes) as a function of the impact energy level. Compression properties of the samples in the
undamaged state (i.e. at 0 Joules impact energy) were also tested as reference. Each condition was tested with at
least three samples. For both systems, as the impact energy increased, the compressive residual strength
decreased due to the higher amount of damage provided during the impact event. PCL(25) samples exhibited for
each impact energy, a lower compressive residual strength as compared to Plain. CAI strength is related to two
phenomena [43]: buckling of the impact damage area and crack propagation next to the point of impact before
fibre and sample failure. Thermoplastic based composites demonstrated elsewhere [44] higher CAI strength as
compared to thermoset based composites, the higher toughness (i.e. resistance to crack propagation) of
thermoplastic composites being the driving force for this improved behaviour. As demonstrated in our previous
studies [17,28], resistance to crack propagation in epoxy-PCL matrices (neat as well as when integrated to FRPs)
is lower as compared to epoxy ones because the PCL stays as a confined layer in between the epoxy particles
within the structure which prevents matrix strain. The difference in compressive behaviour between the two
systems was constant up to 17 Joules energy and then decreased because at higher impact energies, fibre failure
induced by impact influenced the compressive residual properties. Nevertheless, CAI strength values observed
for the assessed systems were in the range of conventional values in FRPs [43–45] thanks mainly to the woven
reinforcement and the high volume fraction of fibers. Notice that FCAI for the Plain samples at 8.5 Joules energy
is depicted in grey on Figure 10 because failure occurred by end brooming, thus leading to an invalid
measurement according to the ASTM standard.

Recovery in CAI strength was assessed on some PCL(25) samples, damaged and successively healed by a
thermal mending cycle at 150°C for 30 minutes. On Figure 10, the effect of this healing cycle (PCL(25) healed)
is superimposed with damaged Plain and PCL(25) samples. The healing efficiency could then be calculated by
Eq. (3) and was of 95.9% ± 6.0%, 73.0% ± 10.4%, 22.4% ± 0.5% for impact energies respectively of 8.5, 17 and
34 Joules. As expected from C-scan tests and optical microscopy, at low impact energies, PCL could fill the
impact induced cracks whereas at high impact energies it could not due to its limited expansion capacity. High
recovery of CAI strength at low impact energy was further confirmed by the fact that on all samples in the
damaged state, compressive failure occurred through the impact damage (see Figure 11 (a)) while on all samples

9
in the healed state, failure occurred by end brooming (see Figure 11 (b)). However, healing efficiency values did
not reach 100% due to the destructive nature of CAI testing which requires different samples for each condition,
thus creating an additional standard deviation. Also, as demonstrated by Slattery et al. [45], test geometry and
buckling in CAI tests are really sensitive to scaling effects, thus if repair has to be validated for a real structure,
dimensions of the CAI specimens have to be as close as possible to those of the targeted structure.

3.5. Healing efficiency comparison

Impact damage healing efficiency has been quantified by means of C-scans through recovery in damage area as
well as CAI strength recovery, but also theoretically estimated knowing the expansion capacity of PCL and the
crack thickness distribution within the impact damage. Healing efficiency values for these three methods are
superimposed on Figure 12 as a function of the impact energy. At low impact energy, thanks to the expansion
capacity of PCL to fully bridge the crack gap, the overlap between the three methods is almost perfect. At higher
impact energies however, the difference between the three methods was higher. Owing to the different nature of
these three characterization methods, the overlap could be considered as sufficient to confirm the observed trend,
as well as to validate the theoretical crack filling estimation.

3. Conclusion
Glass fibre reinforced epoxy-PCL composites with fibre volume fractions around 50vol% were successfully
processed through Vacuum Assisted Resin Infusion Moulding at elevated temperature. Samples were used to
assess the impact response of such systems, characterize their damage area recovery as well as their potential for
recovery of compressive residual strength. As expected from our previous study [28], resistance to crack
propagation was lower for epoxy-PCL composites as compared to pure epoxy composite because Hertzian
failure appeared earlier and the amount of peaks in the load-time histories (representing delamination events)
was higher. Also, damage accumulation was higher in Epoxy-PCL systems as the peak load appeared later in the
load-time histories.

This lower resistance to crack propagation was further confirmed by C-scan analysis where epoxy-PCL
composites showed a damage area 2.5 times larger at 34 Joules impact energy. C-scans were performed before
and after a thermal mending cycle at 150°C for 30 minutes, and demonstrated a damage area recovery from 40%
up to 90%. Incomplete recovery has been attributed to fibre damage, which could not be healed, and formation
of rather thick cracks, which could not be healed by PCL due to its limited expansion capacity. Transverse cuts
of samples impacted at the three energy levels showed typical damage shape in FRPs and the determination of
transverse crack distribution within epoxy-PCL composites revealed crack thicknesses ranging from few µm up
to 300 µm. The theoretical PCL expansion capacity of PCL further emphasized that, depending on the energy
level, 32 to 99% of the cracks could be bridged, thus confirming C-scans analysis. In addition, epoxy-PCL
composites were able to recover between 22 and 96% of compressive residual strength after low-velocity impact
damage, in accordance with C-scan and optical microscopy observations, even though these showed lower
resistance to crack propagation as compared to pure epoxy composites (a CAI strength decreased by 29% in
undamaged samples).

10
Although epoxy-PCL composites demonstrated lower compressive residual strength as compared to pure epoxy
composites, the used woven reinforcement and the high fibre volume fraction provided CAI strength values
close to those encountered in conventional FRPs. In addition, the developed system allowed full healing of an
impact damage of 8 Joules, corresponding to the main concern of maintenance activities in the composite
industry (e.g. tools dropped from 1 m height). Epoxy-PCL composites, providing their manufacturing feasibility
through conventional industrial processes, their acceptable impact damage resistance and their ability to fully
heal low-velocity impact damage, can thus be of interest in composite structures that are subjected to moderate
loads and not easily accessible to repair.

4. Acknowledgments
This work is funded by the Swiss National Science Foundation (SNF 200020-150007-1). We would like to thank
Perstorp for providing the PCL, COMATEC institute at HEIG-VD for C-scan tests and the Laboratory of
Mechanical Metallurgy (LMM) for CAI tests.

5. References
[1] Schoeppner GA, Abrate S. Delamination threshold loads for low velocity impact on composite
laminates. Compos Part A Appl Sci Manuf 2000;31:903–15.
[2] ASTMD7137. Standard Test Method for Compressive Residual Strength Properties of Damaged
Polymer Matrix Composite Plates. ASTM Stand 2012:1–16.
[3] Myhre SH, Labor JD. Repair of advanced composite structures. J Aircr 1981;18:546–52.
[4] Blaiszik BJ, Kramer SLB, Olugebefola SC, Moore JS, Sottos NR, White SR. Self-Healing Polymers and
Composites. Annu Rev Mater Res 2010;40:179–211.
[5] Wu DY, Meure S, Solomon D. Self-healing polymeric materials: A review of recent developments.
Prog Polym Sci 2008;33:479–522.
[6] Garcia SJ. Effect of polymer architecture on the intrinsic self-healing character of polymers. Eur Polym
J 2014;53:118–25.
[7] Kessler M., White S. Self-activated healing of delamination damage in woven composites. Compos Part
A Appl Sci Manuf 2001;32:683–99.
[8] Manfredi E, Cohades A, Richard I, Michaud V. Assessment of solvent capsule-based self-healing for
woven E-glass fibre-reinforced polymers. Smart Mater Struct 2015;24:1–11.
[9] Patel AJ, Sottos NR, Wetzel ED, White SR. Autonomic healing of low-velocity impact damage in fiber-
reinforced composites. Compos Part A Appl Sci Manuf 2010;41:360–8.
[10] Yin T, Rong MZ, Wu J, Chen H, Zhang MQ. Healing of impact damage in woven glass fabric
reinforced epoxy composites. 2Composites Part A Appl Sci Manuf 2008;39:1479–87.
[11] Williams GJ, Bond IP, Trask RS. Compression after impact assessment of self-healing CFRP. Compos
Part A Appl Sci Manuf 2009;40:1399–406.
[12] Norris CJ, Bond IP, Trask RS. Healing of low-velocity impact damage in vascularised composites.
Compos Part A Appl Sci Manuf 2013;44:78–85.
[13] Zhong N, Post W. Self-repair of structural and functional composites with intrinsically self-healing
polymer matrices: A review. Compos Part A Appl Sci Manuf 2015;69:226–39.
[14] van der Zwaag S, Grande a. M, Post W, Garcia SJ, Bor TC. Review of current strategies to induce self-
healing behaviour in fibre reinforced polymer based composites. Mater Sci Technol 2014;30:1633–41.
[15] Hayes S a, Zhang W, Branthwaite M, Jones FR. Self-healing of damage in fibre-reinforced polymer-
matrix composites. J R Soc Interface 2007;4:381–7.
[16] Hayes S a., Jones FR, Marshiya K, Zhang W. A self-healing thermosetting composite material. Compos
Part A Appl Sci Manuf 2007;38:1116–20.

11
[17] Cohades A, Manfredi E, Plummer J-C, Michaud V. Thermal mending in immiscible poly(ε-
caprolactone)/epoxy blends. Eur Polym J 2016;81:114–28.
[18] Luo X, Ou R, Eberly DE, Singhal A, Viratyaporn W, Mather PT. A thermoplastic/thermoset blend
exhibiting thermal mending and reversible adhesion. ACS Appl Mater Interfaces 2009;1:612–20.
[19] Meure S, Wu DY, Furman S. Polyethylene-co-methacrylic acid healing agents for mendable epoxy
resins. Acta Mater 2009;57:4312–20.
[20] Pingkarawat K, Bhat T, Craze D a., Wang CH, Varley RJ, Mouritz a. P. Healing of carbon fibre–epoxy
composites using thermoplastic additives. Polym Chem 2013;4:5007.
[21] Chen J, Chang F. Phase Separation Process in Poly (ε -caprolactone ) - Epoxy Blends. Macromolecules
1999;32:5348–56.
[22] Yamanaka K, Inoue T. Structure development in epoxy resin modified with poly(ether sulphone).
Polymer (Guildf) 1989;30:662–7.
[23] Girard-reydet E, Sautereau H, Pascault JP, Keates P, Navard P, Thollet G, et al. Reaction-induced phase
separation mechanisms in modified thermosets. Polymer (Guildf) 1998;39:2269–79.
[24] Cabanelas JC, Serrano B, Baselga J. Development of Cocontinuous Morphologies in Initially
Heterogeneous Thermosets Blended with Poly(methyl methacrylate). Macromolecules 2005;38:961–70.
[25] Wang CH, Sidhu K, Yang T, Zhang J, Shanks R. Interlayer self-healing and toughening of carbon
fibre/epoxy composites using copolymer films. Compos Part A Appl Sci Manuf 2012;43:512–8.
[26] Pingkarawat K, Mouritz a. P. Stitched mendable composites: Balancing healing performance against
mechanical performance. Compos Struct 2015;123:54–64.
[27] Meure S, Wu D-Y, Furman S a. FTIR study of bonding between a thermoplastic healing agent and a
mendable epoxy resin. Vib Spectrosc 2010;52:10–5.
[28] Cohades A, Michaud V. Thermal mending in E-glass reinforced Poly(ε-caprolactone)/epoxy blends, In
Press, Accepted Manuscript. Compos Part A Appl Sci Manuf 2016.
[29] Rodgers PA. Pressure-Volume-Temperature Relationships for Polymeric Liquids: A review of
Equations of State and Their Characterisitc Parameters for 56 Polymers. J Appl Polym Sci
1993;48:1031–80.
[30] ASTMD7136. Standard Test Method for Measuring the Damage Resistance of a Fiber-Reinforced
Polymer Matrix Composite to a Drop-Weight Impact Event. ASTM Stand 2015:1–16.
[31] Shyr T-W, Pan Y-H. Impact resistance and damage characteristics of composite laminates. Compos
Struct 2003;62:193–203.
[32] Uyaner M, Kara M. Dynamic Response of Laminated Composites Subjected to Low-velocity Impact. J
Compos Mater 2007;41:2877–96.
[33] Hosur M V., Karim MR, Jeelani S. Experimental investigations on the response of stitched/unstitched
woven S2-glass/SC15 epoxy composites under single and repeated low velocity impact loading. Compos
Struct 2003;61:89–102.
[34] Evci C, Gülgeç M. An experimental investigation on the impact response of composite materials. Int J
Impact Eng 2012;43:40–51.
[35] Cantwell WJ, Morton J. The impact resistance of composite materials - a review. Composites
1991;22:347–62.
[36] Belingardi G, Vadori R. Influence of the laminate thickness in low velocity impact behavior of
composite material plate. Compos Struct 2003;61:27–38.
[37] Richardson MOW, Wisheart MJ. Review of low-velocity impact properties of composite materials.
Compos Part A Appl Sci Manuf 1996;27:1123–33.
[38] Quaresimin M, Ricotta M, Martello L, Mian S. Energy absorption in composite laminates under impact
loading. Compos Part B Eng 2013;44:133–40.
[39] Reghunath R, Lakshmanan M, Mini KM. Low velocity impact analysis on glass fiber reinforced
composites with varied volume fractions. IOP Conf Ser Mater Sci Eng 2015;73:12067.
[40] Abrate S. Impact on composite structures. Cambridge: Cambridge University Press; 1998.
[41] Namala KK, Mahajan P, Bhatnagar N. Digital Image Correlation of Low Velocity Impact on

12
 Glass/Epoxy Composite. Int J Comput Methods Eng Sci Mech 2014;2287:140214124149004.
[42] Tan W, Falzon BG, Chiu LNS, Price M. Predicting low velocity impact damage and Compression-
After-Impact (CAI) behaviour of composite laminates. Compos Part A Appl Sci Manuf 2015;71:212–26.
[43] Rivallant S, Bouvet C, Hongkarnjanakul N. Failure analysis of CFRP laminates subjected to
compression after impact: FE simulation using discrete interface elements. Compos Part A Appl Sci
Manuf 2013;55:83–93.
[44] Vieille B, Casado VM, Bouvet C. Influence of matrix toughness and ductility on the compression-after-
impact behavior of woven-ply thermoplastic- and thermosetting-composites: A comparative study.
Compos Struct 2014;110:207–18.
[45] Slattery PG, McCarthy CT, O’Higgins RM. Assessment of residual strength of repaired solid laminate
composite materials through mechanical testing. Compos Struct 2016;147:122–30.

13
Figure 1: Schematic top view of the VARIM lay-out, including the fibre preform (grey), the flow medium (grey mesh), the
inlet and outlet lines. Dimensions are in mm.

Figure 2: Impact force and energy as a function of time for three different impact energies, for (a) Plain system and (b)
PCL(25) system.

14
Figure 3: Dissipated energy during impact as a function of the impact energy, for Plain as well as PCL(25) samples.

Figure 4: Impact force as a function of the deflection response for Plain and PCL(25) systems as well as three impact
energies: 34, 17 and 8.5 Joules.

15
Figure 5: Front face optical image of PCL(25) samples after impact (Damaged) and after a thermal mending cycle of 30
minutes at 150°C (Healed) for three impact energies: 34, 17 and 8.5 Joules.

Figure 6: Damage area measured by C-scan analysis as a function of the impact energy, for Plain as well as PCL(25)
samples.

16
Figure 7: C-scan images of PCL(25) samples after impact (Damaged) and after a thermal mending cycle of 30 minutes at
150°C (Healed) for three impact energies: 34, 17 and 8.5 Joules.

Figure 8: Transverse cut optical microscopy mapping of PCL(25) system impacted at energies of 8.5, 17 and 34 Joules.

17
 Figure 9: Crack thickness distribution corresponding to the optical microscopy mapping of Figure 8.

Figure 10: Compressive residual strength as a function of the impact energy for Plain, PCL(25) in the damaged state and
PCL(25) in the healed state.

18
Figure 11: Back face optical image of CAI samples from the PCL(25) system impacted at 8.5 Joules (i.e. E/3) (a) not healed
and (b) healed.

Figure 12: Healing efficiency as a function of the impact energy, for three characterisation techniques: (i) damage area
determined by C-scan; (ii) compression after impact tests; (iii) crack thickness distribution by optical microscopy
observations.

Table 1: Mechanical properties of E-glass fibre-reinforced polymer composites (Vf=50%) with an epoxy matrix and a blend
of epoxy and 25 vol% of PCL, both cured with DDS [28]. Notice that 150°C is the system healing temperature. Numbers in
brackets show standard deviations.

Drop in E’ at Flexural modulus Flexural strength

Material E’ [GPa] at 25°C GI [J/m2] at 25 °C
150 °C [%] [GPa] at 25°C [MPa] at 25°C
E-glass/epoxy 6.61 (0.50) 16.19 15.99 (1.24) 389.68 (31.75) 947.40 (109.99)
E-glass/epoxy-PCL 6.29 (0.54) 39.90 15.30 (0.63) 349.89 (26.74) 620.32 (67.51)

19