Carbon 114 (2017) 557e565

Contents lists available at ScienceDirect

Carbon
journal homepage: www.elsevier.com/locate/carbon

Reinforcing mechanism of graphene at atomic level: Friction, crack

surface adhesion and 2D geometry
Shu Jian Chen a, Chen Yang Li a, Quan Wang b, Wen Hui Duan a, *
a
Department of Civil Engineering, Monash University, Clayton, VIC 3800, Australia
b
Department of Architecture and Civil Engineering, City University of Hong Kong, Kowloon, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history: Owing to its superior mechanical properties, graphene has been used to reinforce and substantially
Received 5 July 2016 improve the strength of composite materials. Still lacking, however, is a clear understanding of gra-
Received in revised form phene's reinforcing mechanism at the atomic level, especially in relation to its pull-out behavior. By
2 December 2016
molecular dynamics (MD), it is found that pull-out of graphene, different from that of micro fibers, is not
Accepted 11 December 2016
Available online 22 December 2016
governed by friction only. Rather, the pull-out force is revealed to be governed by a “crack surface
adhesion” phenomenon due to unbalanced adhesion at the crack surface when graphene is not func-
tionalized and the crack opening rate is small. Crack surface adhesion produces a constant pull-out force
(about 0.2e1 N per meter width) regardless of the embedded length. There is a transition from crack
surface adhesion governed pull-out to friction governed pull-out when the crack opening speed, gra-
phene size and degree of functionalization increase. A new model is developed to integrate friction and
crack surface adhesion with the 2D geometry of graphene. The new model can be used to predict the
crack bridging stress for 2D graphene (or other 2D atomic thin reinforcements). The outcome of this
study benefits the understanding and design of new graphene composites.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction A major advantage of using nanosized reinforcements, such as

Owing to their superior strength and elastic modulus, graphene
and its derivatives have been among the best candidates for rein-
forcing biomaterials [1e3], polymer [4e10], metal [11e14], and
ceramic/cementitious materials [15e18]. It has been reported that
the elastic modulus of chitosan increased over 200% with the
addition of 0.1e0.3 wt% of graphene [1]; the inclusion of 0.7 wt%
graphene oxide (GO) increased the tensile strength and Young's
modulus of polyvinyl alcohol (PVA) by 76% and 62%, respectively
[10]; the tensile strength of aluminum composite reinforced by
0.3 wt% graphene was increased by over 62% [11]; with the addition
of 0.8 vol% graphene, the fracture toughness of an alumina ceramic
was increased by 40% [15]; the compressive strength and flexural
strength of cement increased by 15e33% and 41e59% respectively
with the introduction of 0.05 wt% GO [18]. Thus, graphene or
graphene-like atomically thin materials (ATMs) have been
acknowledged as promising reinforcing materials for generating
stronger composites.
* Corresponding author.
E-mail address: Wenhui.duan@monash.edu (W.H. Duan).
graphene, is that they can distribute more evenly in the matrices,
allowing them to arrest crack propagation at earlier stage
[17,19e22]. The pull-out of nanometer scale reinforcements during
fracture has been one of the major mechanisms for enhancing
mechanical properties [23,24]. During pull-out, the interaction
between the embedded reinforcement and the matrix, i.e., friction
and chemical bonding, consumes energy and resists the widening
and propagation of cracks [25,26].
The pull-out process of reinforcement, such as with a one-
dimensional (1D) fiber, has often been divided into two stages:
the debonding stage and the pull-out stage [25,27]. During
debonding, chemical bonds between the fiber and the matrix
break, allowing relative movement of the fiber. It has been widely
recognized that after debonding, the pull-out of the fiber is gov-
erned by friction between the matrix and the fiber [20,23,27e34].
Theoretical models based on friction have been developed to pre-
dict the reinforcing effect of microfiber reinforcements [27,28]. In
the previous studies of the present authors and colleagues [24,34],
it was also demonstrated that the friction-based pull-out model
could be extended to nanometer level for multiwall carbon nano-
tubes with diameter around 10e20 nm. In these friction governed
pull-out models, the pull-out force decreased with the embedded

http://dx.doi.org/10.1016/j.carbon.2016.12.034
0008-6223/© 2016 Elsevier Ltd. All rights reserved.
558 S.J. Chen et al. / Carbon 114 (2017) 557e565
length of the fiber/CNTs reinforcement, due to the reduced fiber-
matrix contact area [24,25,27].
Despite the widely reported and superior reinforcing effect of
graphene, investigation of their reinforcing mechanism, such as the
forces in pull-out and the effect of their 2D geometry, has still been
limited. Due to the sub-nanometer thickness of graphene, existing
models and theories describing the pull-out of micro- [27,28] and
nanoscale [20,23,29e34] reinforcing materials may not be valid for
graphene. It has been found that at nanometer scale, friction also
follows Amonton's First Law: being proportional to normal force
and being a result of surface roughness, or in other words, asper-
ities [35]. However, some key issues, such as whether the friction
governed pull-out mechanism can be directly applied to graphene
(with thickness around or below 1 nm and with correspondingly
low surface roughness), have rarely been investigated. Further-
more, existing pull-out models or theories used to predict the
reinforcing effect have been based on 1D fibers. The 2D geometry of
a micro-/nanoreinforcement has rarely been considered. For ap-
plications of graphene/ATM-reinforced composites, therefore, it is
imperative to clarify the pull-out mechanisms and effects of 2D
geometry.
In this study, three steps are taken to understanding the pull-out
mechanism of atomically thin graphene. The first step is to evaluate
the friction between atomic thin graphene and matrices. On the
basis of molecular dynamics (MD) simulation of friction between
graphene and various matrices, it is demonstrated that friction
between graphene and matrix can be significantly affected by
sliding speed and covalent functionalization. The second step is to
evaluate another source of pull-out resistance due to the surface
energy. Using MD, by coupling and quantifying the energy and
forces during the pull-out of graphene, it is suggested that gra-
phene pull-out from various matrices subjects to a constant resis-
tance force (R). The origin of this constant R is the unbalanced
adhesion force between the matrix and the graphene at the crack
surface. This unbalanced adhesion force is therefore referred as
“crack surface adhesion” in this paper.
The third step is to integrate the crack surface adhesion with the
2D geometry of graphene/ATMs. A new theoretical model is
developed to describe the pull-out of 2D circular graphene/ATMs
with random spatial distribution. Furthermore, a new formulation
of the crack bridging stress is established for estimating the rein-
forcing effect of 2D graphene/ATM. The reinforcing mechanism and
model presented here provides better understanding on the effects
of size, crack surface adhesion, friction, and surface properties of
atomically thin reinforcements, contributing to the optimal design
of future graphene/ATM-reinforced composites.
2. Simulation methods
Molecular dynamics (MD) simulations were conducted to
quantify the friction and pull-out interactions between graphene
sheet and various matrices. Three types of matrix were used, cal-
cium silicate hydrate (C-S-H), aluminum (Al), and polyethylene
(PE), that are typical representatives of ceramics/cementitious
materials, metals, and polymers, respectively. The atomic interac-
tion was modeled by the COMPASS force field (condensed-phased
optimized molecular potential for atomistic simulation studies)
[36,37], which is the first ab initio force field that was parameter-
ized and validated using condensed-phase properties. This force
field has been shown to be applicable in describing the mechanical
properties of sp2 carbon [38,39], calcium silicate hydrate (C-S-H)
[40e42], metal [43], and polymers [43e45]. The force field has also
been used to study interfacial properties and interaction between
graphene and polymers [46] or C-S-H [47]. Comparison between
first principle and force field computed graphene-aluminum
interaction energy shows less than 5% difference [48].
2.1. Friction
A periodic simulation box (49.7 Å
25.9 Å
30.8 Å for C-S-H,
88.8 Å
25.6 Å
42.5 Å for Al and 42.3 Å
24.4 Å
27.0 Å for PE)
was created with two layers of matrix materials sandwiching a
monolayer graphene (MLG). After geometry optimization, a 100 ps
NPT ensemble (a constant number of particles, constant pressure
and temperature dynamics simulation) was employed to optimize
the shape of the lattice and relax the system. The pressure was set
to be the atmospheric pressure. Then the sheet was assigned a
speed vx and a NVE ensemble (a constant number of particles,
constant volume, and energy dynamics simulation) was conducted.
The speed vx of all the atoms in the MLG was then extracted under
different sliding distances d. Based on vx, the loss of global kinetic
energy (DEkinetic) of the MLG during sliding was computed as
1

DEkinetic ¼ m v2x ðdÞ  v2x ðd0 Þ (1)
2
where vx ðd0 Þ is the average speed of all the atoms of the MLG after
being assigned an initial speed, vx ðdÞ is the average speed of all the
atoms of the MLG after sliding and m is the mass of the MLG.
The friction force f was then calculated as:
DEkinetic
F¼ (2)
d
The interfacial shear strength t could be given by:
F
t¼ (3)
A
where A is the area of MLG.
In addition, first principle MD was performed to evaluated
friction at adhesion sites where graphene and PE matrix were
covalently bonded via SiO4. Using SiO4 (or silane based compounds)
to bond graphene with polymer matrix such as PE is often seem in
creating functionalized graphene polymer composites [49]. A small
simulation box (9.1 Å
7.9 Å
10.7 Å) was created with two layers
of PE sandwiching a MLG while one side of graphene forms a co-
valent bond with SiO4 which was connected with carbon atoms in
PE chains. Either LDA or PBE functional was used to calculate the
force on atoms based on density functional theory (DFT). The re-
sults from the two functionals show minimum difference (see
Fig. S2 in supplementary information) in estimated friction pa-
rameters indicating low level of error in the DFT calculation [50].
The molecular structure was optimized and relaxed in a NVT until
the temperature is stable. Then a velocity of vx was applied to the
graphene and the movement of the atoms is calculated using
newton's second law of motion. To prevent overheating of the
system, the temperature of matrix is control at 298 K (via tem-
perature scaling). The friction F and t was then calculated using Eqs.
(2) and (3).
2.2. The pull-out
To describe the pull-out, a sandwiched periodic structure was
built with similar style to that for the simulation of friction. Periodic
boundary condition was also used. But an isolated MLG sheet
(IMLG) was used instead of a continuous MLG sheet. Geometry
optimization and a 100 ps NPT ensemble at 298 K with atmospheric
pressure were employed to optimize the parameter of the simu-
lation box (lattice) and relax the system. To leave space for the pull-
out of the isolated MLG, a vacuum gap was introduced in the pull-
 S.J. Chen et al. / Carbon 114 (2017) 557e565
out direction, while keeping the optimized lattice parameter in the
other two directions unchanged. The introduction of the vacuum
gap creates exposed crack surface in the matrix. No covalent bond
was broken when creating these crack surfaces. For C-S-H surface,
extra OH ions and Ca2þ ions were introduced to balance charge on
the crack surface [40]. Then the pull-out displacement d was
increased step by step and in each step (including when d ¼ 0), a
100 ps NVT ensemble (a constant number of particles, constant
volume and temperature dynamics simulation) at 298 K was
employed to relax the system. On the basis of the relaxed structure,
the forces acting on the atoms in the IMLG in the pull-out direction
direction (Fx) applied by the matrices were computed. The inter-
action energy EI between the matrix and the sheet could be
calculated as:
EI ¼ Etotal  ðEmatrix þ Esheet Þ (4)
where Etotal is the total energy of the composite system; Ematrix and
Esheet are the energies of the matrix and sheet, respectively. By
calculating the difference between EI in the initial state (d ¼ 0) and
in the pulled out state (d > 0), the corresponding pull-out energy for
each step was calculated.
3. Friction between graphene and matrix
Pull-outs of a micro- or nanofibre, with diameters from tens of
microns down to tens of nanometers, were considered as friction
Fig. 1. Schematics of (a) friction governed micro-/nanofibre pull-out (red areas indi-
cate friction-damaged materials during pull-out), and (b) crack surface adhesion
resistance for ATM during pull-out (circles/arcs indicate different force equilibriums at
locations I, II, and III). (A colour version of this figure can be viewed online.)
559
governed processes [24,27]. As shown in Fig. 1a, when those fibres
are pulled out from the matrix, the abrasion or wear of the surface
micro-convexities or asperities (red in Fig. 1a) causes a frictional
resistance [51,52]. The energy conservation in this case is mainly
recognized as a balance between the work done by the pull-out
force and the energy consumed in bond-breaking/reformation
[53] and agitation of the atom movement at the friction interface.
However, when the material is atomically thin with very low
surface roughness, such as the graphene shown in Fig. 1b, the
friction caused by abrasion and wear is substantially reduced. MD
simulations were conducted to evaluate the friction level between
MLG and various matrices. In the example of the simulated system
shown in Fig. 2a, the MLG has a sliding speed vx and a sliding dis-
tance d. Fig. 2bed show the decrease of vx during the sliding in
different matrices from d ¼ 0 to 10 Å due to friction. From Fig. 2bed
it can be seen that there is an initial sharper drop in vx within 2 Å for
each of the three different matrices and after that, the rate of
decrease of vx becomes relatively stable.
By measuring the decrease in the speed of the MLGs in the
stable stage, the friction force and the interfacial shear strength, t,
can be calculated, as shown in Table 1. It is seen that the frictional
interfacial shear strength t can ranges from 0.3 to 233 MPa when
there is no covalent bonding between graphene and the matrix. The
computed t is found to increase significantly to the increase sliding
speed. This effect can be explained by the increasing rate of entropy
which increase with the temperature difference in the system.
Higher speed of graphene means higher temperature difference in
the system. Therefore, more ordered kinetic energy (oriented
speed) from the graphene turns into disordered kinetic energy
(temperature) in the system at higher vx. C-S-H matrix is found to
have the least t. This is likely due to the water molecules in the C-S-
H matrix and the spaces between graphene and C-S-H acting as
lubricant. With sliding speed less than 100 m/s, the t for the three
matrices are found below 1.6 MPa.
As the results show that the value of t can vary significantly with
pull-out speed, value of t should be determined based on the strain
rate and crack propagation speed for each specific application. The
Fig. 2. (a) Snapshot of a sliding graphene in a periodic simulation box with PE matrix
and the decrease of speed of due to friction in C-S-H (△), metal (B) and polymer (,)
matrices with initial speed about (b) 10 (c) 5 (d) and 1 nm/ps respectively. (A colour
version of this figure can be viewed online.)
560 S.J. Chen et al. / Carbon 114 (2017) 557e565

Table 1 4. Crack surface adhesion during pull-out

Decreasing rate of velocity and computed friction coefficient.

Matrix Initial sliding speed (km/s) Interfacial shear strength, t (MPa) As shown in Fig. 1b, a monolayer sheet is pulled out from the
C-S-H 10 16.9 (±0.2a)
matrix. For atoms that are distant from the exit, such as atom I,
C-S-H 5 12.7 (±0.6) adhesion forces apply to them from all directions. For atoms that are
C-S-H 1 1.27(±0.00) close to the crack plane (or crack/fracture surface) of the matrix,
C-S-H 0.1 0.63(±0.00) however, such as atoms II and III, due to the absence of matrix atoms
Al 10 187.1(±1.7)
beyond the crack plane, the adhesion forces in those directions
Al 5 133.0(±1.5)
Al 1 34.7(±0.3) disappear. As shown in Fig. 1b, the resultant adhesion near the exit is
Al 0.1 0.285(±0.00) in the opposite direction to the pull-out. Since this adhesion applies
PE-Si-b 10 3407(±93)c only to atoms near the crack surface, it is here referred to as crack
PE 10 232.78(±1.9)
surface adhesion. This generates the resistance force R. For ATMs
PE 5 86.8(±0.9)
PE 1 17.0(±0.00)
with unit width, this R is theoretically a constant during the pull-out,
PE 0.1 1.60(±0.00) regardless of the embedded length of the ATM in the matrix.
a To demonstrate and validate the crack surface adhesion concept,
Standard deviation.
b
PE matrix and graphene covalently bonded with SiO4. analyses were conducted to investigate the forces on an IMLG
c
The average value calculated by LDD and GGA. applied by the matrix. The forces on each atom of the IMLG in the x
direction (Fx) were computed and the distributions of Fx for
different matrices and different pull-out distances are shown in
cracks can propagate at sound speed [54], typically 1e12 km/s in
Fig. 3c, f, h, and j. It is seen that, in the embedded part of the IMLG
various solids. For high strain rate applications, the materials may
that is distant from the crack plane/surface, the positive and
be subjected to impact with speed up to 4e12 km/s [55] which may
negative forces usually appear in pairs and show a patterned dis-
also influence the pull-out speed.
tribution, thereby balancing each other. In the crack plane/surface
Fig. 3aed show snapshots of the sliding process where the
region, however, most atoms have a negative force, forming an
graphene and the PE-Si- matrix is bonded covalently. It can be seem
“adhesion band” (ADB). Fig. 3a and d show the accumulation of Fx
that the bond between the SiO4 and graphene can be broken and
over x. It can be seen that, for the part outside the ADB, the sum-
reformed during the sliding process. Fig. 3e shows that sliding
mation of force is close to zero. Only in the ADB does the force
speed decreased much more significantly compared with the cases
summation begin to decrease to negative. Such calculations were
(in Fig. 2b) without covalent bond. In Fig. 3e, the resistant force for
likewise conducted for IMLGs with Al (Fig. 3g) and PE matrix
the sliding increased significantly, during the stretching of the co-
(Fig. 3i), where similar phenomena were found. The magnitude of
valent bond (a and c) and then sharply decreased during bond
the crack surface adhesion force (0.27 N/m for C-S-H, 0.88 N/m for
break/reformation (b and d). Slight increases in sliding speed and
Al, and 0.50 N/m for PE) is shown to be greater than the friction at
negative resistance at b and d in Fig. 3e indicate release of kinetic
low speed when the IMLG is around 1 by 1 mm in size. The results
energy during bond break/reformation. However, this release of
here show that the constant R for ATM is dominated by the crack
kinetic energy is much less than the energy consumed during the
surface adhesion within the ADB. It should be noted that several
stretching of the bond. This difference may be due to that the
water molecules vaporized from the C-S-H surface and entered into
released kinetic energy is in random orientation and therefore have
the vacuum region. These molecules are found to have negligible
limited contribution to the sliding speed of the graphene. As shown
effect on the graphene-matrix interaction.
in Table 1, at the adhesion site on a graphene where bond break and
By coupling the force and energy during pull-out, factors that
reformation occurs, the frictional interfacial shear strength can
affect the magnitude of R were also investigated, including matrix
increase significantly (see Fig. 4).
material and different sizes of the IMLG (C-S-H: Fig. 5a and d, Al:
Fig. 5b, and PE: Fig. 5c) and surface decoration (hydroxide and
epoxide) of IMLG (Fig. 5d). Fig. 5(eel) shows the crack-surface-
adhesion-governed pull-out energy (Ep, ad) and differential of en-
ergy (dEp, ad/dd) against d during different pull-out conditions. Note
that dEp, ad/dd represents the average pull-out force. As indicated in
Fig. 5e and f, a change in the length of the IMLG has no effect on Ep,
ad, that increases in an approximately linear manner (constant dEp,
ad/dd) except at the very beginning and the end of the pull-out.
Fig. 5g indicates that doubling the width of the IMLG also doubles
Ep, ad and dEp, ad/dd. The type of matrix produces the most signifi-
cance differences in pull-out energy (Ep, ad) and force (dEp, ad/dd),
with Al matrix producing the highest pull-out force, followed by PE
and C-S-H (Fig. 5e, i and j). Moreover, surface decoration of the
IMLG, which introduces charges and electrostatic attractions be-
tween the IMLG and the matrix, increases the magnitude of Ep, ad
and dEp, ad/dd, as shown in Fig. 5k and l. The size of the unit cell (or
in other words, the uniformity of atom distribution) in the matrix is
shown to affect the variation of dEp, ad/dd as the C-S-H (with the
most heterogeneous atom distribution) gives the greatest variation
of dEp, ad/dd, whereas Al (with the least heterogeneous atom dis-
tribution) gives the smallest. This is because the adhesion forces as
Fig. 3. (a)e(d) Snapshots during the sliding of graphene in a PE-Si- matrix and (e) the
illustrated in Fig. 1b may not be identical from different directions
variation of sliding speed and resistance during sliding. The green shading in (a)e(d) at atomic scale due to the non-uniform and non-continuous spatial
indicates electron density. (A colour version of this figure can be viewed online.) distribution of atoms in the matrix.
 S.J. Chen et al. / Carbon 114 (2017) 557e565 561

Fig. 4. Forces on atoms in IMLG in pull-out direction (x-direction) Fx (c, f, h, and j) and cumulative Fx (a, d, g, and i) applied by the matrix when the IMLG is (a and c) half pull-out and
(b and f) three quarters embedded in a C-S-H matrix (△) and half embedded in (g and h) Al (B) and (i and j) PE (,) matrix. (b) and (e) show snapshots of the MD model at the
corresponding pull-out distances. (A colour version of this figure can be viewed online.)

Overall, the increase of pull-out energy in Fig. 5e, g, i and k is
mostly linear except for the first and last 1e2 nm of the pullout
process. The effect of the first and last 1e2 nm can be minimal
when the sheet is greater than 50e100 nm length in reinforcing
applications [11]. Therefore, when calculating the reinforcing effect,
it is reasonable to assume that the surface adhesion induced pull-
out force is constant without introducing significant error.
5. A new model for the reinforcing effect of 2D ATM/graphene
Since results in Section 4 show that the crack surface adhesion
plays an important role during pull-out of graphene, a new model is
developed here to describe pull-out of 2D ATMs considering crack
surface adhesion, friction and initial chemical bonding. First, the
relations between the pull-out force P and the pull-out displace-
ment d for ATMs is established in analogy to the micromechanical
model of Li [26] for 1D fibres. As shown in Fig. 6a, a section of
atomically thin sheet with infinitesimal width dw, thickness t, and
initial embedded length Le is considered. The pull-out process can
be described by two stages: a debonding stage and a debonded
stage. During the debonding stage, only for the part within L, the
chemical bond between the sheet and the matrix breaks and
the part is under tension. dP is the integral of forces along x from
0 to L. The relationship between the pull-out force dP and the
pull-out displacement d during the debonding stage can be given as
(see Appendix A in supplementary information for detailed
derivations):
dP ¼ f ðd0 jt; C; Gd ; E; h; Le ; tÞ$dw; d < d0 (5)
R is the stress as a
where E is the elastic modulus of the sheet, C ¼ tdw
result of the constant R, Gd is the chemical bond energy, h is related
to the young's modulus and volume fraction of the sheets and
matrix (Eq. (A7) in Appendix A in supplementary). As the pull-out
displacement d increases, the sheet and the matrix keep debonding
until eventually all the chemical bonds between the sheet and the
matrix are broken, leading to the debonded stage:
dP ¼ f ðd0 jt; C; Le ; tÞ$dw; d  d0 : (6)
The relationship between dP and d is used to derive a new model
for the crack bridging stress (sB) of ATM embedded in a matrix. sB,
562 S.J. Chen et al. / Carbon 114 (2017) 557e565

Fig. 5. Snapshot of pull-out of a monolayer graphene from (b) metallic (Al), (c) polymeric (PE), and (a) and (d) cementitious (C-S-H) matrices and (e, g, i, and k) Ep,ad and (f, h, j, and l)
changing rate of Ep,ad (pull-out force) against d for 20
20* Å (>), 40
20 Å(✲), 60
20 Å(△), 40
40 Å(8) graphene and 60
20 Å surface decorated graphene (*) pull-out from
C-S-H matrix and 60
20 Å graphene pull-out from Al (B) and PE (,) matrix. Parts of the matrices in (a) and (d) are made invisible to show the embedded portion of the graphene
sheet. *the dimensions of the sheet are indicated as length
width, with length being in the pull-out direction. (A colour version of this figure can be viewed online.)

which represents the stress that helps close the crack during crack out and the part to the right of plane II remains embedded. Based
propagation, is a measure of the reinforcing effect and reflects the on force equilibrium, it is usually assumed that the smaller side of
post-cracking strength improvement in the macro sense. Assuming the reinforcement is pull-out due to the smaller debonding force
the volume fraction of the ATM is Vre, the sheet is a circular flat disk they can withstand [26]. sB can be computed as the integration of
with a radius r and the ATM sheets are 3D randomly located and PðdÞ over different 4 and z:
oriented in the matrix. The new model for sB of 2D ATMs is
developed (see Appendix B in supplementary information for ðrÞcos
Zp=2 Z 4
detailed derivation). 2V
As shown in Fig. 5b, the x’ is perpendicular to the opening crack
sB ðdÞ ¼ re PðdÞpð4ÞpðzÞdzd4 (7)
prt
planes I and II. The angle between the ATM/graphene and x’ is 4. As 4¼0 z¼0
d increases, the part of ATM/graphene to the left of plane II is pulled where pð4Þ and pðzÞ are the probability density functions of 4 and z
 S.J. Chen et al. / Carbon 114 (2017) 557e565 563

Fig. 6. a) pull-out model of a section of an ATM sheet with infinitesimal width, b) randomly oriented 2D disk pull-out model in 3D space, c) demonstration of debonding and
debonded parts during pull-out, and d) sB vs d where the red, blue, and green curves and arrows indicate the effect of increasing Gd, C and r respectively. (A colour version of this
figure can be viewed online.)

respectively. The model presented here adopts a random distribu-
tion of the orientation (4) and position (z) of graphene. For specific
applications where the distribution of graphene can be controlled,
Eqs. A20 and A21 in Supplementary Information can be altered to
adopt specific spatial distribution of the graphene.
Results in Section 3 and 4 provided estimation of typical values
for two key parameters during graphene pull-out: the frictional
interfacial shear strength, t, ranging from 0.3 to 1.6 MPa with low
crack opening speed but can increase to over 200 MPa with high
crack opening speed or large number of bond break/reformation
site on the graphene (Table 1); the crack adhesion introduce stress
in graphene, C, varies with the type of matrices from 0.71 to
2.39 GPa. Since the R is due to vdW, electrostatic and chemical
bonding forces between the matrix and the outermost 1e2 layers of
atoms on a reinforcement [56], it can be seen as proportional to the
perimeter of the cross-section of a reinforcement. The C for CNTs
and graphene can then be formulated as 4R/d and R/t, where d is the
diameter of CNT. Therefore, for common MWCNT with d (>10 nm)
far greater than t, C of these MWCNTs can be much less than
graphene.
Another key parameter for pull-out is the chemical bond energy,
Gd, which is the energy consumed to permanently break some of
the bonds between graphene and matrix. Once these bonds are
broken the ATM/graphene starts to slide against the matrix. For
simplicity, the value of Gd in this study is estimated as the number
of bonds per unit area of graphene times the energy to break each
bond. For example, if 1 covalent bond is broken on every
0.5e100 nm2 of graphene/ATM, each of which consumes energy

Fig. 7. (a) Effect of Gd on bridging stress, (b) prediction of graphene rupture using the model, (c) effect of C and t on bridging stress sB and (d) ratio between friction introduced and
total pull-out energy (Efriction/Epull-out) at various t and r. (A colour version of this figure can be viewed online.)
564 S.J. Chen et al. / Carbon 114 (2017) 557e565
about 300e600 kcal/mol [57], Gd can be estimates as 0.0045e2 J/
m2. It is also suggested that detailed evaluation of Gd in a future
study may contribute to better model prediction.
As shown in Fig. 7, the new model can be used to investigate the
influence of these parameters on the reinforcing mechanism and
reinforcing efficiency of graphene (See Table S1 in Supplementary
Information for the detailed list of parameters used in Fig. 7). Fig. 7a
and b demonstrated that Gd has a dominant influence on the
maximum sB and the maximum stress in graphene smax before
debonding. Increasing Gd can rise the potential strength improve-
ment of the composite and pull-out energy during the debonding
phrase. However, as shown in Fig. 7a, when Gd ¼ 0.2, there is a huge
drop in sB when the sheets are fully debonded. Such drop in sB can
contribute to brittleness in the post failure behavior which may
need to be avoided when ductility of the composite is concerned.
In addition, as shown in Fig. 7b, higher value of Gd leads to
higher smax in graphene and higher risk of graphene rupture at
smax > ~130 GPa [11]. Similarly, excessively high t, for example,
under high strain rate loading or having large number of
bond break/reformation sites (Table 1), will also increase risk of
graphene rupture in the debonding phrase (based Eq. (A22) in
Supplementary Information). The rupture of graphene during
debonding should be avoid as Fig. 7c suggest that crack bridging
R þ∞
energy ( 0 sB dd) at the debonded region (d > ~10 nm) is often
significantly larger than the debonding region (d < ~10 nm). Fig. 7c
also indicates that both C and t can contributes significantly to the
sB during debonded phrase. For instance, when C ¼ 2.39 GPa and
t ¼ 2.00 MPa, C and t have similar contribution to sB.
Fig. 7d shows the contribution of friction to the total pull-out (or
crack bridging) energy during the debonding phrase. Fig. 7d and
the data in Table 1 suggest that crack surface adhesion is the gov-
erning forces of the pull-out process under the conditions of low
crack opening speeds (<100 m/s) and minimum covalent bond
between matrices and graphene. As the size of the graphene and t
increases, the friction start to have more contribution. When t is
greater than 100 MPa and r is greater than 100 nm, friction will
become the governing force during pull-out. The increase of t can
be a result of increased pull-out speed and adhesion sites. One can
adopt different t value to take into account the strain rate effect and
adhesion effect using the model.
Literature on pull-out of inner tubes from perfectly nested
MWCNTs can show a constant pull-out force [58,59]. However, the
results in this study indicate that graphene reinforcements can
have a constant pull-out force from the matrices at certain cir-
cumstances, which has not been reported for MWCNT re-
inforcements. In addition, the 2D geometry of graphene leads to a
2D development of the debonded zone (Fig. S1 in supplementary
information). The recognition of this 2D debonding behavior is
important for estimating the reinforcing effect of graphene.
The model may be used to provide insight into experimental
observations. For example, a study has reported Al matrix with a
weight fraction of 0.3 wt % graphene oxide addition increase the
strength by 95 MPa [11]. Adopting a typical E ¼ 0.20 TPa [60] for
graphene oxide and C ¼ 2.3 GPa (from results in Section 4),
t ¼ 0.285 MPa (from Table 1), r ¼ 0.6 mm [11], the model gives a
maximum sB about 90 MPa and a value of Gd ¼ 0.7 J/m2. This value
of Gd indicates a fairly strong initial chemical bonding between
graphene oxide and the matrix. As discussed above, high Gd may
decrease ductility which is in agreement with the observations of
decreased ductility in Ref. [11].
6. Conclusions
MD simulations were conducted to evaluate the friction level
between graphene and different matrices. It was found that the
interfacial shear strength for graphene, which has atomic level
asperities, ranges from 0.3 to 1.6 MPa with low crack opening speed
but can increase to over 200 MPa with high crack opening speed or
large number of bond break/reformation sites.
The unbalanced adhesion around the crack surface (or crack
plane) caused a constant resistance force, R, during pull-out. Ana-
lyses of the forces on an IMLG applied by the matrix validated the
constant R theory and an ADB was found within which the constant
R originated. MD simulations were conducted to investigate the
factors affecting the magnitude of R. It was found that the magni-
tude of R is irrelevant to the embedded length but proportional to
the embedded width. Surface decoration of the IMLG also increased
the magnitude of R.
Formulization of the relations between the pull-out force and
the pull-out displacement for ATMs with infinitesimal width was
undertaken. The derived relations could be separated into a
debonding stage and a debonded stage. A new theoretical model
was developed that could be used to estimate the crack bridging
stress (sB) for 2D ATM reinforced composites. The ATM sheets were
modeled as a circular flat disk and all the sheets were 3D randomly
distributed and oriented. The model shows that crack surface
adhesion is the governing forces of the pull-out process under the
conditions of low crack opening speeds (<100 m/s) and minimum
covalent bond between matrices and graphene while friction
become the governing force when t is greater than 100 MPa and r is
greater than 100 nm.
Acknowledgment
The authors are grateful for the financial support of the
Australian Research Council in conducting this study.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.carbon.2016.12.034.
References
[1] H. Fan, L. Wang, K. Zhao, N. Li, Z. Shi, Z. Ge, Z. Jin, Fabrication, mechanical
properties, and biocompatibility of graphene-reinforced chitosan composites,
Biomacromolecules 11 (9) (2010) 2345e2351.
[2] S. Baradaran, E. Moghaddam, W. Basirun, M. Mehrali, M. Sookhakian,
M. Hamdi, M.N. Moghaddam, Y. Alias, Mechanical properties and biomedical
applications of a nanotube hydroxyapatite-reduced graphene oxide compos-
ite, Carbon 69 (2014) 32e45.
[3] P.-P. Zuo, H.-F. Feng, Z.-Z. Xu, L.-F. Zhang, Y.-L. Zhang, W. Xia, W.-Q. Zhang,
Fabrication of biocompatible and mechanically reinforced graphene oxide-
chitosan nanocomposite films, Chem. Cent. J. 7 (1) (2013), 39e39.
[4] Z. Xu, C. Gao, In situ polymerization approach to graphene-reinforced nylon-6
composites, Macromolecules 43 (16) (2010) 6716e6723.
[5] Y. Xu, W. Hong, H. Bai, C. Li, G. Shi, Strong and ductile poly (vinyl alcohol)/
graphene oxide composite films with a layered structure, Carbon 47 (15)
(2009) 3538e3543.
[6] L.-C. Tang, Y.-J. Wan, D. Yan, Y.-B. Pei, L. Zhao, Y.-B. Li, L.-B. Wu, J.-X. Jiang, G.-
Q. Lai, The effect of graphene dispersion on the mechanical properties of
graphene/epoxy composites, Carbon 60 (2013) 16e27.
[7] S. Chatterjee, J. Wang, W. Kuo, N. Tai, C. Salzmann, W. Li, R. Hollertz, F. Nüesch,
B. Chu, Mechanical reinforcement and thermal conductivity in expanded
graphene nanoplatelets reinforced epoxy composites, Chem. Phys. Lett. 531
(2012) 6e10.
[8] G. Mittal, V. Dhand, K.Y. Rhee, S.-J. Park, W.R. Lee, A review on carbon
nanotubes and graphene as fillers in reinforced polymer nanocomposites,
J. Industrial Eng. Chem. 21 (2015) 11e25.
[9] O.M. Istrate, K.R. Paton, U. Khan, A. O'Neill, A.P. Bell, J.N. Coleman, Rein-
forcement in melt-processed polymeregraphene composites at extremely
low graphene loading level, Carbon 78 (2014) 243e249.
[10] J. Liang, Y. Huang, L. Zhang, Y. Wang, Y. Ma, T. Guo, Y. Chen, Molecular-level
dispersion of graphene into poly (vinyl alcohol) and effective reinforcement of
their nanocomposites, Adv. Funct. Mater. 19 (14) (2009) 2297e2302.
[11] J. Wang, Z. Li, G. Fan, H. Pan, Z. Chen, D. Zhang, Reinforcement with graphene
 S.J. Chen et al. / Carbon 114 (2017) 557e565
nanosheets in aluminum matrix composites, Scr. Mater. 66 (8) (2012)
594e597.
[12] S. Shin, H. Choi, J. Shin, D. Bae, Strengthening behavior of few-layered gra-
phene/aluminum composites, Carbon 82 (2015) 143e151.
[13] J. Hwang, T. Yoon, S.H. Jin, J. Lee, T.S. Kim, S.H. Hong, S. Jeon, Enhanced me-
chanical properties of graphene/copper nanocomposites using a molecular-
level mixing process, Adv. Mater. 25 (46) (2013) 6724e6729.
[14] W. Kim, T. Lee, S. Han, Multi-layer graphene/copper composites: preparation
using high-ratio differential speed rolling, microstructure and mechanical
properties, Carbon 69 (2014) 55e65.
[15] H. Porwal, P. Tatarko, S. Grasso, J. Khaliq, I. Dlouhý, M.J. Reece, Graphene
reinforced alumina nano-composites, Carbon 64 (2013) 359e369.
[16] L.S. Walker, V.R. Marotto, M.A. Rafiee, N. Koratkar, E.L. Corral, Toughening in
graphene ceramic composites, Acs Nano 5 (4) (2011) 3182e3190.
[17] S. Chuah, Z. Pan, J.G. Sanjayan, C.M. Wang, W.H. Duan, Nano reinforced
cement and concrete composites and new perspective from graphene oxide,
Constr. Build. Mater. 73 (2014) 113e124.
[18] Z. Pan, L. He, L. Qiu, A.H. Korayem, G. Li, J.W. Zhu, F. Collins, D. Li, W.H. Duan,
M.C. Wang, Mechanical properties and microstructure of a graphene oxi-
deecement composite, Cem. Concr. Compos. 58 (2015) 140e147.
[19] D. Qian, E.C. Dickey, R. Andrews, T. Rantell, Load transfer and deformation
mechanisms in carbon nanotube-polystyrene composites, Appl. Phys. Lett. 76
(20) (2000) 2868e2870.
[20] P.M. Ajayan, L.S. Schadler, C. Giannaris, A. Rubio, Single-walled carbon
nanotubeepolymer composites: strength and weakness, Adv. Mater. 12 (10)
(2000) 750e753.
[21] Z. Xia, L. Riester, W. Curtin, H. Li, B. Sheldon, J. Liang, B. Chang, J. Xu, Direct
observation of toughening mechanisms in carbon nanotube ceramic matrix
composites, Acta Mater. 52 (4) (2004) 931e944.
[22] G.Y. Li, P.M. Wang, X. Zhao, Mechanical behavior and microstructure of
cement composites incorporating surface-treated multi-walled carbon
nanotubes, Carbon 43 (6) (2005) 1239e1245.
[23] J.N. Coleman, U. Khan, W.J. Blau, Y.K. Gun’ko, Small but strong: a review of the
mechanical properties of carbon nanotubeepolymer composites, Carbon 44
(9) (2006) 1624e1652.
[24] S.J. Chen, B. Zou, F. Collins, X.L. Zhao, M. Majumber, W.H. Duan, Predicting the
influence of ultrasonication energy on the reinforcing efficiency of carbon
nanotubes, Carbon 77 (2014) 1e10.
[25] V.C. Li, H. Stang, H. Krenchel, Micromechanics of crack bridging in fibre-
reinforced concrete, Mater. Struct. 26 (8) (1993) 486e494.
[26] Z. Lin, T. Kanda, V.C. Li, On interface property characterization and perfor-
mance of fiber reinforced cementitious composites, Concr. Sci. Eng. 1 (3)
(1999) 173e184.
[27] V.C. Li, Postcrack scaling relations for fiber reinforced cementitious compos-
ites, J. Mater. Civ. Eng. 4 (1) (1992) 41e57.
[28] C.-H. Hsueh, Interfacial debonding and fiber pull-out stresses of fiber-
reinforced composites, Mater. Sci. Eng. A 123 (1) (1990) 1e11.
[29] B.S. Harrison, A. Atala, Carbon nanotube applications for tissue engineering,
Biomaterials 28 (2) (2007) 344e353.
[30] S. Bakshi, D. Lahiri, A. Agarwal, Carbon nanotube reinforced metal matrix
composites-a review, Int. Mater. Rev. 55 (1) (2010) 41e64.
[31] H. Kwon, M. Estili, K. Takagi, T. Miyazaki, A. Kawasaki, Combination of hot
extrusion and spark plasma sintering for producing carbon nanotube rein-
forced aluminum matrix composites, Carbon 47 (3) (2009) 570e577.
[32] S.S. Samal, S. Bal, Carbon nanotube reinforced ceramic matrix composites-a
review, J. Minerals Mater. Charact. Eng. 7 (04) (2008) 355.
[33] S. Chen, F. Collins, A. Macleod, Z. Pan, W. Duan, C. Wang, Carbon nano-
tubeecement composites: a retrospect, IES J. Part A Civ. Struct. Eng. 4 (4)
(2011) 254e265.
[34] B. Zou, S.J. Chen, A.H. Korayem, F. Collins, C. Wang, W.H. Duan, Effect of
ultrasonication energy on engineering properties of carbon nanotube rein-
forced cement pastes, Carbon 85 (2015) 212e220.
[35] Y. Mo, K.T. Turner, I. Szlufarska, Friction laws at the nanoscale, Nature 457
(7233) (2009) 1116e1119.
[36] D. Rigby, H. Sun, B. Eichinger, Computer simulations of poly (ethylene oxide):
565
force field, PVT diagram and cyclization behaviour, Polym. Int. 44 (3) (1997)
311e330.
[37] H. Sun, COMPASS: an ab initio force-field optimized for condensed-phase
applications overview with details on alkane and benzene compounds,
J. Phys. Chem. B 102 (38) (1998) 7338e7364.
[38] W. Duan, Q. Wang, K.M. Liew, X. He, Molecular mechanics modeling of carbon
nanotube fracture, Carbon 45 (9) (2007) 1769e1776.
[39] Q. Wang, W. Duan, K. Liew, X. He, Inelastic buckling of carbon nanotubes,
Appl. Phys. Lett. 90 (3) (2007) 033110.
[40] D. Hou, Z. Li, Molecular dynamics study of water and ions transported during
the nanopore calcium silicate phase: case study of jennite, J. Mater. Civ. Eng.
26 (5) (2013) 930e940.
[41] D. Hou, Y. Zhu, Y. Lu, Z. Li, Mechanical properties of calcium silicate hydrate
(CeSeH) at nano-scale: a molecular dynamics study, Mater. Chem. Phys. 146
(3) (2014) 503e511.
[42] C.Y. Li, S.J. Chen, Y. Lu, W.H. Duan, Molecular dynamics simulations of gra-
phene pull-out from calcium silicate hydrate, in: International Conference on
Mechanics and Physics of Creep, Shrinkage, and Durability of Concrete and
Concrete Structures (Christian Hellmich, Johann Kollegger and Bernhard
Pichler 21 September 2015 to 23 September 2015), American Society of Civil
Engineers, 2015, pp. 913e918.
[43] S.W. Bunte, H. Sun, Molecular modeling of energetic materials: the parame-
terization and validation of nitrate esters in the COMPASS force field, J. Phys.
Chem. B 104 (11) (2000) 2477e2489.
[44] F. Liu, N. Hu, H. Ning, Y. Liu, Y. Li, L. Wu, Molecular dynamics simulation on
interfacial mechanical properties of polymer nanocomposites with wrinkled
graphene, Comput. Mater. Sci. 108 (2015) 160e167. Part A.
[45] C. Wu, W. Xu, Atomistic molecular modelling of crosslinked epoxy resin,
Polymer 47 (16) (2006) 6004e6009.
[46] C. Lv, Q. Xue, D. Xia, M. Ma, J. Xie, H. Chen, Effect of chemisorption on the
interfacial bonding characteristics of graphene polymer composites, J. Phys.
Chem. C 114 (14) (2010) 6588e6594.
[47] H. Alkhateb, A. Al-Ostaz, A.H.-D. Cheng, X. Li, Materials genome for graphene-
cement nanocomposites, J. Nanomechanics Micromechanics 3 (3) (2013)
67e77.
[48] W. Lee, S. Jang, M.J. Kim, J.-M. Myoung, Interfacial interactions and dispersion
relations in carbonealuminium nanocomposite systems, Nanotechnology 19
(28) (2008) 285701.
[49] Y.-J. Wan, L.-X. Gong, L.-C. Tang, L.-B. Wu, J.-X. Jiang, Mechanical properties of
epoxy composites filled with silane-functionalized graphene oxide, Compos.
Part A Appl. Sci. Manuf. 64 (2014) 79e89.
[50] D. Rappoport, N.R. Crawford, F. Furche, K. Burke, Which Functional Should I
Choose?, 2008.
[51] N.-H. Sung, N.P. Suh, Effect of fiber orientation on friction and wear of fiber
reinforced polymeric composites, Wear 53 (1) (1979) 129e141.
[52] X. Dangsheng, Friction and wear properties of UHMWPE composites rein-
forced with carbon fiber, Mater. Lett. 59 (2) (2005) 175e179.
[53] S. Zhandarov, E. Ma €der, Characterization of fiber/matrix interface strength:
applicability of different tests, approaches and parameters, Compos. Sci.
Technol. 65 (1) (2005) 149e160.
[54] F.F. Abraham, H. Gao, How fast can cracks propagate? Phys. Rev. Lett. 84 (14)
(2000) 3113.
[55] C. Doolan, A Two-Stage Light Gas Gun for the Study of High Speed Impact in
Propellants, DTIC Document, 2001.
[56] J.N. Israelachvili, Intermolecular and Surface Forces, revised third ed., Aca-
demic press, 2011.
[57] M.S. Silberberg, R. Duran, C.G. Haas, A.D. Norman, Chemistry: the Molecular
Nature of Matter and Change, McGraw-Hill, New York, 2006.
[58] Z. Xia, W. Curtin, Pullout forces and friction in multiwall carbon nanotubes,
Phys. Rev. B 69 (23) (2004) 233408.
[59] M.-F. Yu, O. Lourie, M.J. Dyer, K. Moloni, T.F. Kelly, R.S. Ruoff, Strength and
breaking mechanism of multiwalled carbon nanotubes under tensile load,
Science 287 (5453) (2000) 637e640.
[60] J.W. Suk, R.D. Piner, J. An, R.S. Ruoff, Mechanical properties of monolayer
graphene oxide, ACS Nano 4 (11) (2010) 6557e6564.