Polymer 48 (2007) 530e541

www.elsevier.com/locate/polymer

Toughening mechanisms of nanoparticle-modified epoxy polymers

B.B. Johnsen a, A.J. Kinloch a, R.D. Mohammed a, A.C. Taylor a,*, S. Sprenger b
a
Department of Mechanical Engineering, Imperial College London, South Kensington Campus, London SW7 2AZ, UK
b
Nanoresins AG, Charlottenburger Strasse 9, 21502 Geesthacht, Germany
Received 11 August 2006; received in revised form 26 October 2006; accepted 16 November 2006
Available online 15 December 2006

Abstract

An epoxy resin, cured with an anhydride, has been modified by the addition of silica nanoparticles. The particles were introduced via a sole
gel technique which gave a very well-dispersed phase of nanosilica particles which were about 20 nm in diameter. Atomic force and electron
microscopies showed that the nanoparticles were well-dispersed throughout the epoxy matrix. The glass transition temperature was unchanged
by the addition of the nanoparticles, but both the modulus and toughness were increased. The measured modulus was compared to theoretical
models, and good agreement was found. The fracture energy increased from 100 J/m2 for the unmodified epoxy polymer to 460 J/m2 for the
epoxy polymer with 13 vol% of nanosilica. The fracture surfaces were inspected using scanning electron and atomic force microscopies, and
the results were compared to various toughening mechanisms proposed in the literature. The toughening mechanisms of crack pinning, crack
deflection and immobilised polymer were discounted. The microscopy showed evidence of debonding of the nanoparticles and subsequent plas-
tic void growth. A theoretical model of plastic void growth was used to confirm that this mechanism was indeed most likely to be responsible for
the increased toughness that was observed due to the presence of the nanoparticles.
Ó 2006 Elsevier Ltd. All rights reserved.

Keywords: Epoxy; Nanoparticles; Fracture

1. Introduction e.g. [6e8], into the epoxy polymer can increase their tough-
ness. Here the rubber or thermoplastic particles are typically
Epoxy polymers are widely used for the matrices of fibre- about 1e5 mm in diameter with a volume fraction of about
reinforced composite materials and as adhesives. When cured, 5e20%. However, the presence of the rubbery phase typically
epoxies are amorphous and highly-crosslinked (i.e. thermoset- increases the viscosity of the epoxy monomer mixture and
ting) polymers. This microstructure results in many useful reduces the modulus of the cured epoxy polymer.
properties for structural engineering applications, such as Hence rigid, inorganic particles have also been used, as
a high modulus and failure strength, low creep, and good these can increase the toughness without affecting the glass
performance at elevated temperatures. transition temperature of the epoxy polymer. Here glass beads
However, the structure of such thermosetting polymers also or ceramic (e.g. silica or alumina) particles with a diameter of
leads to a highly undesirable property in that they are rela- between 4 and 100 mm are typically used, e.g. [9e14]. How-
tively brittle materials, with a poor resistance to crack initia- ever, these relatively large particles also significantly increase
tion and growth. Nevertheless, it has been well established the viscosity of the resin, reducing the ease of processing. In
for many years that the incorporation of a second microphase addition, due to the size of these particles they are unsuitable
of a dispersed rubber, e.g. [1e5], or a thermoplastic polymer, for use with infusion processes for the production of fibre
composites as they are strained out by the small gaps between
the fibres.
* Corresponding author. Tel.: þ44 20 7594 7149; fax: þ44 20 7594 7017. More recently, a new technology has emerged which holds
E-mail address: [email protected] (A.C. Taylor). promise for increasing the mechanical performance of such

0032-3861/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.polymer.2006.11.038
 B.B. Johnsen et al. / Polymer 48 (2007) 530e541 531

thermosetting polymers. This is via the addition of a nanophase 2.2. Glass transition temperatures
structure in the polymer, where the nanophase consists of
small rigid particles of silica [15e18]. Such nanoparticle- The glass transition temperature, Tg, of the various poly-
modified epoxies have been shown to not only increase further mers was measured using differential scanning calorimetry.
the toughness of the epoxy polymer but also, due to the very The sample was heated to 175  C at a rate of 10  C/min,
small size of the silica particles, not to lead to a significant and then cooled to 0  C. The sample was then heated again
increase in the viscosity of the epoxy monomer. to 175  C, and the results quoted are from this second heating
The aims of the present work were to investigate the frac- run.
ture toughness of epoxy polymer modified with silica nanopar- Dynamic mechanical thermal analysis (DMTA) was per-
ticles, and to establish the structure/property relationships. The formed by testing bars 48  3  2 mm in size in the three-
toughening mechanisms which may be operating will be point-bending mode at 1 Hz. The storage modulus, loss
reviewed, and the mechanism most likely to be responsible modulus and loss factor, tan d, were calculated as a function of
will be identified. temperature, over a range from 40 to 175  C. The glass tran-
sition temperature, Tg, was also determined, and was taken to
be the temperature at which the peak value of tan d occurred.
2. Experimental
2.3. Mechanical properties
2.1. Materials
Tensile dumbbell specimens were machined from the
The materials were based upon a one-component hot-cured plates. These were tested at a displacement rate of 1 mm/
epoxy formulation. The epoxy resin was a standard diglycidyl min and a test temperature of 21  C, according to the ISO
ether of bis-phenol A (DGEBA) with an epoxy equivalent standard test method [21,22]. The strain in the gauge length
weight (EEW) of 185 g/mol, ‘Bakelite EPR 164’ supplied by was measured using a clip-on extensometer, and the Young’s
Hexion Speciality Chemicals, Duisburg, Germany. The silica modulus, E, was calculated.
(SiO2) nanoparticles were supplied as a colloidal silica sol in
the resin matrix, ‘Nanopox F400’, by Nanoresins, Geesthacht, 2.4. Fracture testing
Germany. The particles are synthesised from aqueous sodium
silicate solution [19,20]. They then undergo a process of sur- The single-edge notch bend (SENB) test was used to deter-
face modification with organosilane and matrix exchange, to mine the fracture toughness, KIc, at the onset of crack growth
produce a masterbatch of 40 wt% (26 vol%) silica in the epoxy of the polymers. Specimens were machined from the sheets,
resin. The nanosilica particles had a mean particle size of and the fracture toughness was determined according to the
about 20 nm, with a narrow range of particle-size distribution; relevant ISO standard [23], using a displacement rate of
laser light scattering shows that almost all particles are be- 1 mm/min and a test temperature of 21  C. Four replicate
tween 5 and 35 nm in diameter. The particle size and excellent specimens were tested for each blend composition. The ma-
dispersion of these silica particles remain unchanged during chined notch was sharpened by drawing a razor blade across
any further mixing and/or blending operations. Further, the notch tip before testing. All the specimens failed by unsta-
despite the relatively high silica content of 26 vol%, the nano- ble crack growth, and hence only a single initiation value of
filled epoxy resin still has a comparatively low viscosity due to the fracture toughness was obtained from each specimen.
the agglomerate-free colloidal dispersion of the nanoparticles The coefficient of variation in the values of KIc was 15%.
in the resin. The small diameter and good dispersion of the The value of the fracture energy, GIc, was calculated from
nanoparticles of silica have been previously reported and knowledge of the values of KIc and E, using the relationship:
shown [15,16]. The curing agent was an accelerated methyl-
hexahydrophthalic acid anhydride, namely ‘Albidur HE 600’ KIc2

supplied by Nanoresins, Geesthacht, Germany. GIc ¼ 1  n2 ð1Þ
E
Bulk sheets of unmodified epoxy and nanosilica-modified
epoxy polymers were produced to determine the properties where E is the modulus of elasticity obtained from the tensile
of the polymers. Firstly, the simple DGEBA resin was mixed tests, and n is the Poisson’s ratio of the polymer, taken to be
together with given amounts of the nanosilica-containing ep- 0.35 [24].
oxy resin. The value of the EEW of the blend was then mea-
sured via titration. Secondly, the stoichiometric amount of 2.5. Microscopy
the curing agent was added to the mixture, which was poured
into release-coated moulds and pre-cured for 1 h at 90  C, The fracture surfaces of the specimens were investigated
followed by a cure of 2 h at 160  C. using scanning electron microscopy (SEM). A JEOL ‘JSM-
The densities of the plates were measured. An epoxy den- 5300’ scanning microscope was used, and all specimens
sity of 1100 kg/m3 and a silica density of 1800 kg/m3 were were coated with a thin layer of sputtered gold before analysis
calculated. The volume fraction of silica was calculated to prevent charging. An acceleration voltage of 20 kV was
from the known weight fractions using the measured densities. used.
532 B.B. Johnsen et al. / Polymer 48 (2007) 530e541

High resolution SEM was performed on the fracture sur-

faces with an electron microscope equipped with a field emis-
sion gun (FEG-SEM). The instrument used was a ‘Leo 1525’
from Carl Zeiss equipped with a ‘Gemini’ column. Typically,
the acceleration voltage was set at 5 kV. All specimens were
coated with an approximately 5 nm thick layer of gold or
platinum before analysis.
Thin sections, approximately 60e80 nm thick, of the
blends were cryo-microtomed (at 50  C) for subsequent
examination using transmission electron microscopy (TEM).
The TEM was performed using a JEOL ‘JEM-2000FX II’
electron microscope at an acceleration voltage of 200 kV.
Atomic force microscopy (AFM) studies were undertaken
using a ‘MultiMode’ scanning probe microscope from Veeco
equipped with a ‘J’ scanner and a ‘NanoScope IV’ controller.
A smooth surface was first prepared by cutting samples on
a cryo-ultramicrotome at temperatures down to 100  C.
The scans were performed in tapping mode using silicon
probes, and both height and phase images were recorded.

2.6. Surface roughness

Surface roughness measurements of the fracture surfaces

were performed using a ‘Form Talysurf Series 2’. A sharp sty-
lus with a tip radius of 2.5 mm, attached to a cantilever, was
drawn across the surface at a constant speed for a set distance.
Five measurements with a traverse length of 4 mm were per-
formed perpendicular to the direction of the crack growth.
The average roughness, Ra, of the line profiles was calculated.

2.7. Microstructure

Microscopy of the unmodified epoxy polymer showed that

a homogeneous thermoset was formed. For the nanosilica-
modified materials, atomic force and transmission electron mi- Fig. 1. Atomic force micrographs (height image) of the epoxy polymer
crographs both showed that there was no agglomeration of the containing 9.6 vol% nanosilica.
nanosilica particles, even at higher volume fractions of nano-
silica. Instead, the particles are well-dispersed throughout the
epoxy, as shown in Figs. 1 and 2. These images confirm that
the nanoparticles are approximately 20 nm in diameter. Note
that the TEM slice shown in Fig. 2 is approximately 100 nm
thick, and hence the apparent volume fraction of particles is
higher than the true volume fraction.

3. Results

3.1. Glass transition temperatures

Dynamic mechanical thermal analysis (DMTA) was used to

determine the storage modulus, G0 , and the glass transition
temperature, Tg, of the formulations used. A Tg of 153  C
was measured for the unmodified epoxy polymer, as shown
in Table 1 and Fig. 3. Addition of the nanosilica did not signif-
icantly affect the measured glass transition temperature, and
all values lay within the range 152  2  C. The glass transition
temperatures measured using differential scanning calorimetry Fig. 2. Transmission electron micrograph of the epoxy polymer containing
also showed no significant effect of the addition of nanosilica, 9.6 vol% nanosilica.
 B.B. Johnsen et al. / Polymer 48 (2007) 530e541 533

Table 1 5.0
Glass transition temperatures, modulus and fracture properties of the anhy-
dride-cured epoxy polymer containing nanosilica particles
Nanosilica Nanosilica Tg ( C) Tg ( C) E KIc GIc
content (wt%) content DSC DMTA (GPa) (MN m3/2) (J/m2) 4.5

Young's Modulus, E (GPa)

(vol%)
0 0 143 153 2.96 0.59 103
4.1 2.5 137 152 3.20 1.03 291 4.0
7.8 4.9 136 154 3.42 1.17 352
11.1 7.1 141 151 3.57 1.18 343
14.8 9.6 138 152 3.60 1.29 406
20.2 13.4 138 150 3.85 1.42 461 3.5

3 1.2
3.0 Experimental
Halpin-Tsai prediction
2.5 1 Nielsen no slip
Nielsen interfacial slip
Storage Modulus, G' (GPa)

2.5
2 0.8 0 2 4 6 8 10 12 14
Nanosilica Content (vol%)
Tan Delta

1.5 0.6 Fig. 4. Tensile modulus versus nanosilica content. Points are experimental
data, and solid line is the prediction using the HalpineTsai model. The dashed
lines are predictions using the McGee and McCullough form of the Lewise
Control (Modulus)
1 0.4 Nielsen model for no slippage at the particle-matrix interface (kE ¼ 2.167)
2.5 vol% nanosilica (Modulus)
Control (Tan Delta)
and for interfacial slippage (kE ¼ 0.867) for non-agglomerated spheres in a
2.5 vol% nanosilica (Tan Delta) matrix with n ¼ 0.35.
0.5 0.2

models are summarised in review papers by Ahmed and Jones

0 0
0 20 40 60 80 100 120 140 160 180 [31] and by Nielsen and Landel [32]. Note that these models
Temperature (°C) make a number of assumptions. They assume that the matrix
Fig. 3. Dynamic mechanical thermal analysis data, showing flexural modulus
and the particles are linear-elastic, isotropic and that there is
and loss factor (tan d), measured at 1 Hz, versus temperature for unmodified perfect bonding between the particles and the matrix. They
epoxy polymer and epoxy polymer with 2.5 vol% nanosilica. also assume that the particles are (where applicable) uniform
in their aspect ratio and perfectly aligned with the applied
load. Many models also ignore any agglomeration and
with Tg values of 140  3  C being measured. (Note that particleeparticle interactions.
DMTA gave somewhat higher Tg values than those obtained The rule of mixtures and the modified rule of mixtures have
using differential scanning calorimetry, as has been observed generally been found to give poor predictions [33], and they
previously when comparing these techniques, e.g. by Johnsen also give poor predictions when compared to the present
et al. [25].) The modulus determined from the DMTA tests experimental data. They will not be considered further. The
gave a similar trend to that observed using the tensile tests MorieTanaka model is also frequently used for predicting
discussed below. the modulus of particle-modified polymers [34,35]. Here the
particles are treated as ellipsoidal (either fibre-like or plate-
3.2. Tensile modulus like) with a constant aspect ratio, and are assumed to be
aligned. However, this model works best for relatively high
A tensile modulus of 2.96 GPa was measured for the un- aspect ratios, as when the aspect ratio approaches unity the
modified epoxy polymer. The measured modulus was found model predicts a negligible reinforcing effect, and hence is
to increase with the nanosilica content, see Fig. 4. The increase not suitable for use in the present work.
in modulus is expected because the modulus of silica, The HalpineTsai model comes from work by Halpin and
E ¼ 70 GPa [26,27], is much greater than that of the epoxy co-workers [30,36], and gives the modulus of the material as
matrix. A maximum modulus of 3.85 GPa was measured, for a function of the modulus of the matrix polymer, Em, and of
the epoxy polymer with 13.4 vol% of nanosilica. The mea- the filler particles, Ef, and also as a function of the aspect ratio
sured moduli can be compared to theoretical predictions. by the inclusion of a shape factor. The predicted composite
There are many theoretical models that may be used to pre- modulus is
dict the moduli of particle-modified polymers, e.g. [28e30].
The most commonly used models are the rule of mixtures, 1 þ zhVf
Ec ¼ Em ð2Þ
the HalpineTsai and the MorieTanaka relationships. Other 1  hVf
534 B.B. Johnsen et al. / Polymer 48 (2007) 530e541

where z is the shape factor, Vf is the volume fraction of parti- interface, and kE ¼ 1.0 if there is slippage [32]. Nielsen [41]
cles, and has shown that the value of kE is reduced when the Poisson’s
   ratio of the matrix is lower than 0.5. In the present work
Ef Ef n ¼ 0.35, so values of kE are reduced by a factor of 0.867.
h¼ 1 þz ð3Þ
Em Em Hence, kE ¼ 2.167 if there is no slippage and kE ¼ 0.867 if
there is slippage. The predictions for these two cases are
Halpin and Tsai noted that the value of z must lie between
shown in Fig. 4. It can be seen that reducing the adhesion,
zero and infinity. Indeed, if z ¼ N then Eq. (2) reduces to the
and hence the value of kE, reduces the predicted modulus.
rule of mixtures, and if z ¼ 0 then it reduces to the transverse
The LewiseNielsen predictions bracket the experimental
rule of mixtures. They suggested that the value of z correlated
data, and indicate that there is likely to be imperfect adhesion
with the geometry of the reinforcing phase, especially with the
between the particles and the matrix.
aspect ratio (w/t) of the particles, where w is the length of the
An alternative explanation for the relatively poor agreement
particle and t is its thickness. By comparison of the predictions
being observed between the HalpineTsai model and the ex-
with the results of a finite-element analysis, Halpin and Kardos
perimental data is the agglomeration of the particles. The
[37] suggested that a shape factor of z ¼ 2w/t is used for cal-
values of the maximum volume fraction of particles and the
culating the modulus of a polymer with the particles aligned
Einstein coefficient in the LewiseNielsen model take into ac-
with the loading direction. They recommended using z ¼ 2
count the degree of dispersion of particles. For example, for
for the modulus perpendicular to the loading direction. For
random close-packed agglomerated particles, Vmax ¼ 0.37,
the spherical particles used in the present work the aspect ratio
and kE ¼ 6.76  0.867 ¼ 5.86 [32]. Substitution of these
is unity, and hence z ¼ 2 will be used.
values into the equations above gives predictions which are
The predictions are shown with the experimental data in
higher than the LewiseNielsen prediction with no slippage,
Fig. 4. This figure shows that the modulus is expected to in-
as shown in Fig. 4, and hence the agreement between the
crease with the volume fraction of nanosilica [38], and that
experimental data and the modified model becomes worse.
the agreement between the predictions and experimental data
In addition, the micrographs of the materials used to show no
is reasonable. However, the measured moduli generally lie
agglomeration of the silica nanoparticles, and hence it seems
slightly below the predicted relationship. This is a common
unlikely that agglomeration is responsible for the relatively
observation, as the model assumes that there is perfect bond-
poor agreement between the HalpineTsai model and the
ing between the particles and the matrix, which is normally
experimental data.
not the case. This effect can be considered further using the
LewiseNielsen model [39] and the work of McGee and
3.3. Fracture toughness and fracture energy
McCullough [40]. The authors showed that the composite
modulus can be predicted using:
A mean fracture toughness, KIc, of 0.59 MN m3/2 was
1 þ ðkE  1ÞbVf measured for the unmodified epoxy polymer. The measured
Ec ¼ Em ð4Þ toughness was increased by the addition of nanosilica, and
1  bmVf
a maximum KIc of 1.42 MN m3/2 was measured for the poly-
where kE is the generalised Einstein coefficient, and b and m mer with 13.4 vol% nanosilica, see Table 1. The fracture en-
are constants. The constant b takes into account the relative ergy, GIc, was calculated for each formulation via Eq. (1),
modulus of the particles and the matrix, and is given by using the modulus measured by tensile testing. These data
   are shown in Fig. 5 and Table 1. The fracture energy increases
Ef Ef steadily with the nanosilica content, see Fig. 5.
b¼ 1 þ ðkE  1Þ ð5Þ
Em Em
3.4. Fracture surfaces
The value of m depends on the maximum volume fraction
of particles, Vmax, and can be calculated from
The fracture surface of the unmodified epoxy polymer is


1  Vf 
 shown in Fig. 6a, where the direction of crack propagation
m¼1þ Vmax Vf þ ð1  Vmax Þ 1  Vf ð6Þ is from left to right. The fracture surface is relatively smooth
Vmax
and glassy, which is typical of a brittle thermosetting polymer
Values of Vmax have been tabulated by Nielsen and Landel [42], and shows that no large-scale plastic deformation has
[32] for a range of particle shapes and types of packing. The occurred during fracture. These observations agree well with
micrographs shown in the present work indicate that the parti- the low measured toughness of the material, where KIc ¼
cles appear non-agglomerated and randomly arranged. For 0.59 MN m3/2. In addition, there are apparent steps and
random close packing, non-agglomerated spheres, Nielsen changes of the level of the crack which can be observed in
and Landel quote a value of Vmax ¼ 0.632. This value will Fig. 6a. These features are feather markings, which are caused
be used in the modulus predictions. by the crack forking due to the excess of energy associated
The value of kE varies with the degree of matrix to particle with the relatively fast crack growth. This repeated forking
adhesion. For a matrix with a Poisson’s ratio of 0.5 containing and the multi-planar nature of the surface are ways of absorb-
dispersed spheres, kE ¼ 2.5 if there is no slippage at the ing this excess energy in a very brittle material [43].
 B.B. Johnsen et al. / Polymer 48 (2007) 530e541 535

500 is proportional to its length, additional energy input is there-

fore required. Reducing the spacing between the particles
450
and their size will increase the strengthening (toughening)
400 effect [27].
Crack pinning is identified by the presence of bowing lines
Fracture Energy, GIc (J/m2)

350 on the fracture surface. Lange [48] identified pinning by the

bowed shape of the crack front in perfectly brittle materials,
300
e.g. glass and magnesium oxide, and Green et al. [49] mea-
250 sured the shape of the bowed crack at breakaway from the pin-
ning particles in glass containing 132 mm diameter nickel
200 particles. Kinloch et al. [45] also observed bowing lines using
an epoxy matrix containing 50 mm diameter spherical glass
150
particles.
100 In these cases, crack pinning occurred when the particles
were larger than the crack-opening displacement. To see if
50 a pinning argument is reasonable for nanoparticles, the
crack-opening displacement can be calculated and compared
0
0 2 4 6 8 10 12 14 to the particle size. The plastic zone ahead of the crack tip
Nanosilica Content (vol%) can be modelled as a line-zone using the Irwin analysis. Under
plane-strain conditions, the crack-opening displacement, dtc,
Fig. 5. Fracture energy, calculated using Eq. (1), versus nanosilica content.
can be calculated using the relationship [50]:

The fracture surface of the nanosilica-modified materials KIc2
GIc

dtc ¼ 1  n2 ¼ ð8Þ
showed similar features to those of the unmodified epoxy Esy sy
polymer, as shown in Fig. 6b and c. Crack forking and feather
markings are observed, and the fracture surfaces have a brittle where sy is the yield stress of the matrix.
appearance. However, the addition of nanosilica did not give For the unmodified epoxy polymer, using the data in Tables
an apparent increase in the roughness observed by scanning 1 and 2, the value of the crack-opening displacement is
electron microscopy, unlike for micrometre-sized particles, 1.7 mm. For the maximum toughness measured in the present
e.g. [33]. This aspect is discussed in more detail below. work, using the 13.4 vol% nanosilica formulation, a value of
dtc ¼ 7.5 mm is calculated, as shown in Table 3. The crack
3.5. Toughening micromechanisms tip is shown schematically in Fig. 7, which shows the relative
size of the nanoparticles and the crack-opening displacement.
Many authors have observed that the addition of particles to From this figure, it is clear that particles which are so much
a brittle material increases the measured fracture energy. smaller than the crack-opening displacement are unlikely to
Huang and Kinloch postulated that the fracture energy of a cause crack pinning. Indeed, the crack pinning and bowing
particle-modified polymer may be expressed [44] as: analysis by Green et al. [51] require that the particles are
much larger than the plastic zone size. This is not the case
GIc ¼ GIcu þ J ð7Þ in the present work, as the nanosilica particles have a mean
particle size of about 20 nm, whereas the diameter of the plas-
where GIcu represents the fracture energy of the unmodified tic zone varies from 5 to 29 mm, as shown in Table 3.
epoxy, and J represents the overall toughening contributions. By comparison, where crack pinning has been observed,
These toughening contributions have been ascribed to the particle diameter is much larger than the crack-opening
a range of processes. They can be broadly categorised as on- displacement. Kinloch et al. [45] quoted dtc ¼ 1.8 mm at
plane processes (such as crack pinning or bowing and crack room temperature while the mean particle diameter was
deflection) or off-plane processes (such as debonding and plas- 50 mm, using an epoxy matrix with glass particles. In this
tic void growth). The following discussion considers these case the diameter of the smallest plastic zone was about
processes in more detail, and uses analytical models to predict 35 mm. Similarly, Green et al. [49] demonstrated pinning
the toughening contributions from these processes. with 132 mm diameter particles, but dtc ¼ 0.3 mm and the di-
ameter of the plastic zone was less than 60 mm (conservative
3.5.1. Crack pinning estimates calculated assuming that the yield stress is equal
A crack pinning mechanism has been invoked for to the quoted fracture stress). Others [10,13,45,46] have in-
micrometre-sized glass particles [10,45,46], and also for nano- voked pinning though the lack of microscopic evidence makes
particles [18]. This draws parallels to the impediment of the it difficult to establish whether pinning really occurred.
movement of dislocations through metals by particles (disper- Norman and Robertson [52] observed no bowing lines on
sion hardening), e.g. [47], where the dislocation bows between the fracture surfaces and discounted crack pinning, finding
particles, increasing its length. As the energy of a dislocation that off-plane processes dominated the toughening effect.
536 B.B. Johnsen et al. / Polymer 48 (2007) 530e541

Fig. 6. Scanning electron micrographs of fracture surfaces, showing the precrack towards the left of the image, for (a) unmodified epoxy polymer, (b) epoxy
polymer with 2.5 vol% nanosilica, and (c) epoxy polymer with 13.4 vol% nanosilica. (Crack propagation is from left to right.)

As the nanosilica particles are so much smaller than the 3.5.2. Crack deflection
crack-opening displacement they are unlikely to cause crack The toughening effect could be due to crack deflection oc-
pinning, and this conclusion is supported by the theory. The curring, where the crack front tilts and twists when it encoun-
absence of bowed crack front markings on the fracture ters the particles and hence passes around them [42,53]. This
surfaces in the present work, and that by Zhang et al. [18], causes an increase in the total fracture surface area and also
indicates that crack pinning is unlikely to be responsible for causes the crack to grow locally under mixed-mode I/II condi-
the observed increase in toughness. tions. The former toughening effect can be evaluated by (i)
comparing the measured fracture toughness with the surface
Table 2
Material properties of the unmodified epoxy polymer, and constants used for
calculation of the fracture energy, GIc, and for the contribution to the increase Table 3
in fracture energy from the plastic void growth mechanism, DGv Theoretical values of the crack-opening displacement and the plastic zone size
using the Irwin model, calculated using Eqs. (8) and (11)
Property Symbol Ref. Units Value
3/2 Nanosilica Nanosilica Crack-opening Radius of a circular
Fracture toughness KIc 1 MN m 0.59
content (wt%) content (vol%) displacement dtc (mm) plastic zone ry (mm)
Modulus Em 1 GPa 2.96
Poisson’s ratio n [24] e 0.35 0 0 1.7 4.9
Tensile yield stress sy 1 MPa 61.1 4.1 2.5 4.8 15
Compressive yield stress syc 1 MPa 77.3 7.8 4.9 5.7 19
von Mises pressure sensitivity mm [65] e 0.2 11.1 7.1 5.6 20
Max. stress concentration factor Kvm [65] e 2.22 14.8 9.6 6.6 24
20.2 13.4 7.5 29
1: Measured in the present work.
 B.B. Johnsen et al. / Polymer 48 (2007) 530e541 537

increases in the true (i.e. local) fracture surface area are not
20 nm solely responsible for the increases in the toughness.
particles
The measured fracture energies may also be compared to
predictions by Faber and Evans [53], which consider that crack
deflection causes the crack to grow locally under mixed-mode I/
II conditions. This analysis uses the shape of the particles,
4.8µm which is spherical in this case, and their volume fraction. For ex-
ample, the Faber and Evans model predicts that the fracture en-
ergy for a volume fraction of 2.5% of silica spheres will be 1.2
times that of the unmodified epoxy. The GIc of the pure epoxy is
103 J/m2 and thus the predicted value of GIc from employing the
Faber and Evans model is 120 J/m2. However, the measured
fracture energy of 291 J/m2 is significantly higher than this pre-
Fig. 7. Schematic showing crack tip, with calculated crack-opening displace- dicted value. The Faber and Evans model consistently underpre-
ment (from Eq. (8)) and nanoparticles for the epoxy polymer containing
dicts the fracture energy, and this suggests that this mechanism
approximately 2.5 vol% nanosilica.
is unlikely to fully explain the increased toughness.
In addition, Fig. 7 shows that the nanoparticles are so much
roughness, and (ii) the latter toughening effect can be evalu-
smaller than the crack-opening displacement that they are un-
ated using the analysis by Faber and Evans [53].
likely to cause the crack to deflect. Indeed, Evans states that
Work by Arakawa and Takahashi [54], as reported by Hull
the equations can only be used provided that the obstacle di-
[55,56], showed that the toughening effect due to an increase
mension is larger than the plastic zone [57], which is not the
in the true fracture surface area gives a linear relationship be-
case in the present work, as discussed above. This and the
tween the surface roughness and the overall toughening contri-
relatively poor agreement with the crack deflection models
bution, J [54]. In the present work, the average surface
indicate that crack deflection is unlikely to fully explain the
roughness, Ra, of the fracture surfaces was measured. The av-
observed toughening effect.
erage roughness of the unmodified epoxy polymer was rela-
tively low, an Ra value of 0.04 mm being measured, which is
3.5.3. Immobilised polymer
typical for an unmodified epoxy polymer, e.g. [33]. The rough-
Another mechanism that has been proposed [18,58] is the
ness generally increased with an increasing concentration of
formation of an interphase or immobilised layer of polymer
nanosilica, from a minimum of Ra ¼ 0.03 mm at 2.5 vol%
around the particles. Because the interparticle distance for
nanosilica to a maximum of Ra ¼ 0.15 mm at a concentration
nanocomposites is so small, it is suggested that due to the
of 13.4 vol% nanosilica. These data are shown in Fig. 8, where
proximity of the particles this interphase may be present
the measured roughness is plotted against the overall toughen-
throughout the entire polymer matrix. The formation of an im-
ing contribution, J, from the presence of the nanosilica. These
mobilised layer of polymer around particles has been observed
do not show a linear relationship, and hence it appears that
for various systems. Using absorption studies, the bound layer
thickness is typically 1e5 nm for carbon black-elastomer sys-
0.25 tems [59], and values of 10e20 nm have been reported for ru-
tile (TiO2) filled polyethylene [60]. Viscometric studies
(13.4)
generally give larger thicknesses [59], in the order of tens of
0.2 nanometres, and values as high as 1.4 mm have been reported
Average Roughness, Ra ( m)

for PMMA with glass particles [61].

Thermal analysis has also been used. Work by Gerard et al.
0.15 [62] comparing treated and untreated glass particles in an ep-
oxy matrix, as reported by Rothon and Hancock [59], showed
(9.6) that uncoated particles reduce the mobility of the polymer.
(4.9)
0.1 However, when the particles were coated with a rubbery inter-
layer the effect was largely absent. In addition, immobilisation
Unmodified of the polymer would be expected to increase Tg of the epoxy
epoxy (2.5)
0.05 [62], and/or to broaden the dynamic mechanical peak, see
Iisaka and Shibayama [63] for example. However, Fig. 3
shows that both the width and position of the tan d peak mea-
(7.1)
0 sured using DMTA are unaffected. The same is true for the
0 50 100 150 200 250 300 350 400 other DMTA data, as confirmed by the Tg values which are
Toughening Increment, (J/m2) summarised in Table 1.
Fig. 8. Mean roughness of the fracture surfaces versus measured toughening Hence it can be concluded that there is no evidence for
increment, J. The volume fractions of nanosilica are shown in parentheses, a toughening mechanism which involves the formation of an
error bars are 1 standard deviation. immobilised layer of polymer around the nanosilica particles.
538 B.B. Johnsen et al. / Polymer 48 (2007) 530e541

3.5.4. Plastic void growth nanosilica, see Fig. 9, showed the presence of voids around
The toughening mechanisms associated with micrometre- several of the nanoparticles. This shows that plastic void
sized particles have frequently been shown to be due to de- growth of the epoxy matrix, initiated by debonding of the
bonding of the particles followed by plastic void growth, nanoparticles, has occurred. The diameter of these voids is
e.g. [12,13,42]. Indeed, Kinloch and Taylor [42] have also typically 30 nm. These voids were also observed in the frac-
demonstrated that the voids around particles closed-up when ture surfaces of samples with different contents of nanosilica.
the epoxy polymer was heated above its Tg and allowed to re- Although the samples are coated to prevent charging in the
lax. The debonding process is generally considered to absorb electron microscope, the voids are not an artefact of the coat-
little energy compared to the plastic deformation of the matrix. ing as they could not be observed on a coated fracture surface
However, debonding is essential because this reduces the con- of the pure epoxy polymer, see Fig. 10. Also, the nanosilica-
straint at the crack tip and hence allows the matrix to deform modified samples appeared similar whether they were coated
plastically via a void growth mechanism. with platinum or gold.
High resolution scanning electron microscopy (FEG-SEM) In addition, similar voids were observed by AFM of
of a fracture surface of the polymer containing 9.6 vol% uncoated fracture surfaces, see Fig. 11. However, the apparent
diameter of the nanoparticle in the void highlighted in Fig. 11
is 30 nm, as shown in Fig. 12, whereas transmission electron
microscopy has shown that the mean particle size is actually
around 20 nm. This discrepancy is due to the tip-broadening
effect when the AFM is used to identify such small features.
As the tip radius of the AFM probe is about 10 nm, this makes

Fig. 10. Scanning electron micrograph (FEG-SEM) of the fracture surface of

the unmodified epoxy polymer.

Fig. 9. Scanning electron micrographs (FEG-SEM) of the fracture surface of

the epoxy polymer containing 9.6 vol% nanosilica. (Voids with nanoparticles Fig. 11. Atomic force micrograph (height image) of a fracture surface of the
are circled in the central image.) epoxy polymer containing 9.6 vol% nanosilica.
 B.B. Johnsen et al. / Polymer 48 (2007) 530e541 539

40 simplicity, we will initially assume that the adhesion between

30 the silica and epoxy polymer phases is relatively poor, and
hence debonding will not contribute to the toughening effect.
20
Particle ~ 30 nm
The contribution to the increase in fracture energy from the
Height (nm)

10 plastic void growth mechanism, DGv, is given [44] by

0


-10
DGv ¼ 1  m2m =3 Vv  Vf syc ryu Kvm
2
ð10Þ

-20
where mm is a material constant, Vv is the volume fraction of
-30 voids, Vf is the volume fraction of particles, syc is the com-
-40 pressive yield stress of the unmodified epoxy polymer, ryu is
0 100 200 300 400 500 the radius of the plastic zone of the unmodified epoxy poly-
Surface distance (nm) mer, and Kvm is the maximum stress concentration factor of
Fig. 12. Surface profile of the line drawn across the nanosilica particle and the von Mises stress in the plastic matrix. The value of mm,
void in Fig. 11. which describes the pressure sensitivity of the material in
the von Mises yield criterion, has been reported [64] to be be-
features that are protruding out of a surface appear larger than tween 0.175 and 0.225, and is normally taken to be 0.2, as re-
their true size in the micrographs. The void diameter in this ported in [44]. The material properties for the calculation of
case is about 70 nm, and it would appear that AFM can only DGv are summarised in Table 2. The maximum stress concen-
be reliably used to detect the largest voids. tration factor, Kvm, was found from a finite-element analysis
Voids with no nanoparticles were also observed with FEG- [65] to be 2.22 around a void in an epoxy matrix.
SEM. Here the particles associated with these voids will be sit- The size of the plastic zone ahead of the crack tip can
uated in the opposite fracture surface, or have fallen out of the be calculated assuming linear-elastic fracture-mechanics
surface completely during fracture, as is commonly observed (LEFM) behaviour. Under plane-strain conditions and assum-
with micrometre-sized particles [42]. (It should be noted that ing that the zone is circular [50] as proposed by Irwin, the
the diameters of most of these holes are less than those dis- radius of the plastic zone, ry, is given by
cussed above, as the matrix is unlikely to fail across the widest
point of the void. Further, the coating, which is 5 nm thick,  2
1 KIc
will partially fill the voids, and hence the observed size may ry ¼ ð11Þ
6p sy
be smaller than the true (uncoated) diameter.)
In the present work, for the unmodified epoxy polymer KIc ¼
3.5.5. Modelling of the contribution from the plastic
0.59 MN m3/2, sy ¼ 61.1 MPa and hence ryu ¼ 5 mm.
void growth mechanism
Now, Eq. (10) requires knowledge of the volume fraction of
To confirm whether the observed debonding and plastic
voids and it is difficult to measure this parameter accurately
void growth which occur for the nanoparticle-modified epoxy
from the micrographs, as it is difficult to identify all the voids
could be responsible for the toughening effect, the increase in
due to the resolution of the microscopes used. Nevertheless,
toughness can be compared to a theoretical model. A suitable
a mean void diameter can be calculated from the features
model for this is by Huang and Kinloch [44]. This model as-
which are clearly voids, see Fig. 9 for example. Analysis of
sumes that the fracture specimens behave in a bulk linear-elas-
micrographs of the polymer with 9.6 vol% nanosilica gave
tic manner, and that the energy dissipation is localised to
a mean void diameter of 30 nm. Micrographs of the formula-
a small plastic zone at the crack tip, as observed in the present
tion containing 2.5 vol% nanosilica indicated an average void
work.
diameter of just below 30 nm. However, some voids up to
Huang and Kinloch suggested that the overall toughening
50 nm in diameter are observed in all formulations where
contribution, J, was composed of contributions from particle
micrographs from FEG-SEM analysis are available.
bridging, localised shear banding in the epoxy matrix, and
The predicted toughening increment, J, was calculated us-
plastic void growth of the epoxy matrix which is initiated by
ing Eqs. (7), (10) and (11), assuming (a) that the measured
cavitation or debonding of the particulate phase. However,
toughening increment is entirely caused by plastic void growth
for rigid spherical particles as used in the present work, mi-
(i.e. J ¼ DGv), and (b) that all of the particles initiate the
croscopy has shown that bridging does not occur. Hence the
growth of voids which have a final diameter of 30 nm. For
plastic void growth mechanism seems to be by far the domi-
the epoxy with 9.6 vol% of nanosilica, the measured toughen-
nant toughening micromechanism, and thus the relationship
ing increment is 303 J/m2 and the predicted toughening incre-
in Eq. (7) may be written as:
ment is 394 J/m2. The toughening increment can also be
GIc zGIcu þ DGv ð9Þ predicted for the other formulations, for which FEG-SEM im-
ages are not available, assuming that the average void diameter
where DGv is the contribution to the increase in fracture is constant and that all of the particles initiate void growth.
energy from plastic void growth of the epoxy matrix. For The results are shown in Table 4.
540 B.B. Johnsen et al. / Polymer 48 (2007) 530e541

Table 4 these effects would tend to lead to the experimentally mea-

Measured and predicted toughening increments sured values of GIc being greater than the predicted values.
Nanosilica Nanosilica Toughening increment, J (J/m2) Hence, these effects would offset the above effect of not all
content (wt%) content (vol%) Measured Predicted of the particles debonding. Nevertheless, there is no direct
4.1 2.5 188 107 evidence for these effects being of significance and, for the
7.8 4.9 249 209 present, will not be considered further.
11.1 7.1 240 297 In summary, considering the simplifying assumptions made,
14.8 9.6 303 394 the agreement between the measured and the predicted tough-
20.2 13.4 358 540
ening increments is very reasonable. The predicted values are
Predicted values calculated using Eqs. (7) and (10), assuming that J ¼ DGv. somewhat larger than the measured values for the high nanosil-
(Note GIcu ¼ 103 J/m2.)
ica contents, but the predicted values agree very well with the
experimental data for the lower nanosilica contents. This con-
Considering the assumptions made in the above quantita- firms that plastic void growth of the epoxy matrix, initiated by
tive modelling calculations, then firstly, it is highly unlikely a void being formed by debonding of the nanosilica, can indeed
that all of the particles will debond and initiate void growth. produce the major increases in the values of the fracture en-
Indeed, from the FEG-SEM images of the fracture surfaces ergy, GIc, recorded in the present work.
in Fig. 9 it can be seen that not all of the nanosilica particles
seem to initiate void growth. The difficulty of quantifying the 4. Conclusions
number of debonded particles, when considering the present
nanosilica particles, has been discussed above and has been re- An epoxy resin cured with an anhydride has been used.
ported by van Hartingsveldt and van Aartsen [66] who studied This was modified by the addition of silica nanoparticles, man-
the debonding of glass particles with a diameter of 20e50 mm ufactured using a solegel process, which were 20 nm in diam-
in a polyamide matrix. They clearly demonstrated that only eter. These particles were well-dispersed through the epoxy
a fraction of the particles debonded, but that this fraction could matrix with no agglomeration observed using transmission
be as high as about 0.8e0.9. Further, such a value was attained electron and atomic force microscopies. The addition of the
at levels of applied strain which are relatively low compared to nanoparticles did not affect the glass transition temperature;
those which exist in the region of the crack tip, as is being con- the Tgs of the unmodified and nanoparticle-modified epoxy
sidered in the present work. Also, the present nanosilica parti- polymers were measured to be in the range of 140  4  C us-
cles do possess a very narrow distribution of particle size. ing differential scanning calorimetry. Dynamic mechanical
Thus, they would be expected to possess an almost constant thermal analysis confirmed this observation. The addition of
value of the applied stress needed to debond them [67,68], nanoparticles increased the modulus of the epoxy polymer
and hence might be expected to exhibit a relatively high frac- as expected. The measured moduli were compared to theoret-
tion of particles which show a similar behaviour. (Although, ical values, calculated using the HalpineTsai and Lewise
previous work [67,68] has suggested that, for a given level Nielsen models, and good agreement was found.
of interfacial adhesion, the nanosilica particles should require The fracture toughness of the polymers was measured, and
a relatively high level of debonding stress compared to glass a KIc of 0.59 MN m3/2 was recorded for the unmodified ep-
particles which are many micrometres in diameter.) Thus, on oxy. Addition of the nanoparticles increased the fracture
balance, it appears not unreasonable to assume that most of toughness, a maximum value of 1.42 MN m3/2 being mea-
the nanosilica particles will debond and initiate plastic void sured for the epoxy polymer with 13.4 vol% of nanoparticles.
growth in the epoxy matrix. However, clearly, if only a fraction These values were converted to fracture energies, GIc, using
of the particles initiates void growth, or if only partial debond- the measured modulus. The unmodified epoxy polymer gave
ing occurs, then the predicted toughening increments would be GIc ¼ 103 J/m2, and a maximum fracture energy of 460 J/m2
reduced compared to the values shown in Table 4. was calculated. Hence there is a significant toughening effect
Secondly, for simplicity, it has been assumed that the adhe- due to the addition of the silica nanoparticles.
sion between the silica and epoxy phases was poor. If this The toughening mechanisms for particle-modified epoxy
is not the case, then the energy absorbed by the debonding polymers reported in the literature were reviewed. The con-
process will also contribute to the toughening effect. Thirdly, stant Tg values discount the immobilised polymer mechanism,
the presence of nanoparticles in the ligaments between the and the lack of pinning marks on the fracture surfaces indi-
debonded nanoparticles will increase the stiffness of the cates that crack pinning is not occurring. Crack deflection
plastically-deforming ligaments, and hence may be expected was discounted by a lack of correlation between the measured
to increase the strain-energy density absorbed by the ligaments roughness of the fracture surfaces and the toughening incre-
during their plastic deformation. Indeed, the data from tensile ment. As the silica nanoparticles are so much smaller than
tests show that the strain to failure is unaffected by the addi- the crack-opening displacement, it is unlikely that these on-
tion of nanoparticles, but the modulus and the strength are plane mechanisms are responsible for the toughening effect.
increased. Hence, the strain-energy density at fracture (in a Indeed, these theories assume that the particle diameter is
tensile test) of the nanosilica-modified epoxy polymer is much greater than the plastic zone size, which is not the
greater than that of the pure epoxy polymer. Clearly, both of case with nanoparticles.
 B.B. Johnsen et al. / Polymer 48 (2007) 530e541 541

Observation of the fracture surfaces using scanning electron [22] ISO-527-2. Plastics e determination of tensile properties e part 2: test
and atomic force microscopies showed nanoparticles sur- conditions for moulding and extrusion plastics. Geneva: ISO; 1996.
[23] ISO-13586. Plastics e determination of fracture toughness (GIc and
rounded by voids, providing evidence of debonding of the KIc) e linear elastic fracture mechanics (LEFM) approach. Geneva:
nanoparticles and subsequent plastic void growth. An analyti- ISO; 2000.
cal model of plastic void growth was used to confirm whether [24] Kinloch AJ. Adhesion and adhesives: science and technology. London:
this mechanism could be responsible for the increased tough- Chapman & Hall; 1987.
ness. The mean void diameter was measured from the micro- [25] Johnsen BB, Kinloch AJ, Taylor AC. Polymer 2005;46(18):7352e69.
[26] Goodfellow product catalogue. Huntingdon: Goodfellow; 2005.
graphs, and the model was used to predict the toughening [27] Pascoe KJ. An introduction to the properties of engineering materials.
increment (compared to the fracture energy of the unmodified 3rd ed. London: Van Nostrand Reinhold; 1978.
epoxy polymer). The predicted values agreed well with the [28] Kerner EH. Proc Phys Soc B 1956;69:808e13.
measured values, indicating that debonding of the nanopar- [29] Nielsen LE. J Appl Polym Sci 1966;10:97e103.
ticles and subsequent plastic void growth were most likely [30] Halpin JC, Pagano NJ. J Compos Mater 1969;3:720e4.
[31] Ahmed S, Jones FR. J Mater Sci 1990;25(12):4933e42.
to be responsible for the increase in toughness that was [32] Nielsen LE, Landel RF. Mechanical properties of polymers and compos-
observed due to the presence of the nanosilica particles. ites. 2nd ed. New York: Marcel Dekker; 1994.
[33] Kinloch AJ, Taylor AC. J Mater Sci 2006;41(11):3271e97.
Acknowledgements [34] Fornes TD, Paul DR. Polymer 2003;44(17):4993e5013.
[35] Luo J-J, Daniel IM. Compos Sci Technol 2003;63(11):1607e16.
[36] Halpin JC. J Compos Mater 1969;3:732e4.
The authors gratefully acknowledge the financial support [37] Halpin JC, Kardos JL. Polym Eng Sci 1976;186(5):344e52.
provided by the Government of Trinidad and Tobago via a [38] Vollenberg PHT, Heikens D. Polymer 1989;30(9):1656e62.
studentship for Mr. R.D. Mohammed. Also, they wish to ac- [39] Lewis TB, Nielsen LE. J Appl Polym Sci 1970;14(6):1449e71.
knowledge the general support from the US Army European [40] McGee S, McCullough RL. Polym Compos 1981;2(4):149e61.
Research Office and Nanoresins (formerly Hanse Chemie). [41] Nielsen LE. J Compos Mater 1968;2(1):120e3.
[42] Kinloch AJ, Taylor AC. J Mater Sci 2002;37(3):433e60.
They would also like to thank Dr. B.R.K. Blackman, Mr. [43] Andrews EH. Fracture in polymers. 1st ed. Edinburgh: Oliver & Boyd;
M.Z. Sanat, Dr. M.G. Ardakani (Imperial College London) 1968.
and Dr. I.A. Kinloch (University of Manchester) for their [44] Huang Y, Kinloch AJ. J Mater Sci 1992;27(10):2763e9.
assistance with the microscopy. [45] Kinloch AJ, Maxwell DL, Young RJ. J Mater Sci 1985;20(11):4169e84.
[46] Lange FF, Radford KC. J Mater Sci 1971;6(9):1197e203.
[47] Caddell RM. Deformation and fracture of solids. Englewood Cliffs, New
References Jersey: Prentice-Hall; 1980.
[48] Lange FF. Philos Mag 1970;22(179):983e92.
[1] Drake RS, Siebert AR. SAMPE Quart 1975;6(4):11e21. [49] Green DJ, Nicholson PS, Embury JD. J Mater Sci 1979;14(6):1413e20.
[2] Kinloch AJ, Shaw SJ, Tod DA, Hunston DL. Polymer 1983;24(Octo- [50] Kinloch AJ, Young RJ. Fracture behaviour of polymers. London: Applied
ber):1341e54. Science Publishers; 1983.
[3] Kinloch AJ. MRS Bull 2003;28(6):445e8. [51] Green DJ, Nicholson PS, Embury JD. J Mater Sci 1979;14(7):
[4] Rowe EH, Siebert AR, Drake RS. Mod Plast 1970;47(August):110e7. 1657e61.
[5] Pearson RA, Yee AF. J Mater Sci 1986;21(7):2475e88. [52] Norman DA, Robertson RE. Polymer 2003;44(8):2351e62.
[6] Pascault JP, Williams RJJ. Formulation and characterization of thermo- [53] Faber KT, Evans AG. Acta Metall 1983;31(4):565e76.
setethermoplastic blends. In: Paul DR, Bucknall CB, editors. Polymer [54] Arakawa K, Takahashi K. Int J Frac 1995;48:103e14.
blends volume 1: formulation. New York, USA: John Wiley & Sons; [55] Hull D. Fractography: observing, measuring, and interpreting fracture
1999. p. 379e415. surface topography. Cambridge: Cambridge University Press; 1999.
[7] Bucknall CB, Partridge IK. Polymer 1983;24(May):639e44. [56] Hull D. J Mater Sci 1996;31(17):4483e92.
[8] Kinloch AJ, Yuen ML, Jenkins SD. J Mater Sci 1994;29(14):3781e90. [57] Evans AG. Philos Mag 1972;26(6):1327e44.
[9] Young RJ, Beaumont PWR. J Mater Sci 1975;10:1343e50. [58] Odegard GM, Clancy TC, Gates TS. Polymer 2005;46(2):553e62.
[10] Spanoudakis J, Young RJ. J Mater Sci 1984;19(2):473e86. [59] Rothon RN, Hancock M. General principles guiding selection and use of
[11] Amdouni N, Sautereau H, Gerard JF. J Appl Polym Sci 1992;46(10): particulate materials. In: Rothon RN, editor. Particulate-filled polymer
1723e35. composites. Harlow: Longman Scientific & Technical; 1995. p. 1e45.
[12] Lee J, Yee AF. Polymer 2000;41:8363e73. [60] Boluk MY, Schreiber HP. Polym Compos 1986;7(5):295e301.
[13] Kawaguchi T, Pearson RA. Polymer 2003;44(15):4239e47. [61] Iisaka K, Shibayama K. J Appl Polym Sci 1978;22(11):3135e43.
[14] Kitey R, Tippur HV. Acta Mater 2005;53(4):1167e78. [62] Gerard JF, Amdouni N, Sautereau H, Pascault JP. Introduction of a
[15] Kinloch AJ, Taylor AC, Lee JH, Sprenger S, Eger C, Egan D. J Adhes rubbery interphase in glass/epoxy composite materials: influence of the
2003;79(8e9):867e73. interlayer thickness on the viscoelastic and mechanical properties of
[16] Kinloch AJ, Mohammed RD, Taylor AC, Eger C, Sprenger S, Egan D. particulate and unidirectional composites. In: Ishida H, editor. Controlled
J Mater Sci 2005;40(18):5083e6. interphases in composite materials. New York: Elsevier; 1990. p. 441e8.
[17] Ragosta G, Abbate M, Musto P, Scarinzi G, Mascia L. Polymer [63] Iisaka K, Shibayama K. J Appl Polym Sci 1978;22:1845e52.
2005;46(23):10506e16. [64] Sultan JN, McGarry FJ. Polym Eng Sci 1973;13:29e34.
[18] Zhang H, Zhang Z, Friedrich K, Eger C. Acta Mater 2006;54(7): [65] Huang Y, Kinloch AJ. J Mater Sci 1992;27(10):2753e62.
1833e42. [66] van Hartingsveldt EAA, van Aartsen JJ. Polymer 1989;30(11):
[19] Sprenger S, Eger C, Kinloch AJ, Taylor AC, Lee JH, Egan D. Adhaesion 1984e91.
Kleben Dichten 2003;2003(3):24e8. [67] Vollenberg P, Heikens D, Ladan HCB. Polym Compos 1988;9(6):
[20] Hanse Chemie. Patent Application. WO 02/083776 A1, 2002. 382e8.
[21] ISO-527-1. Plastics e determination of tensile properties e part 1: [68] Pukanszky B, VanEs M, Maurer FHJ, Voros G. J Mater Sci 1994;
general principles. Geneva: ISO; 1993. 29:2350e8.