Three-Dimensional Modeling of Short Fiber-Reinforced

Composites with Extended Finite-Element Method

Matthew G. Pike 1 and Caglar Oskay 2

Abstract: This manuscript presents a modeling approach based on the extended finite-element method (XFEM) for modeling the mechanical
behavior of three-dimensional short fiber composites including interface debonding. Short fibers are incorporated into the XFEM framework
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

as deformable elastic two-dimensional rectangular planar inclusions. Enrichment functions account for both the presence of axial deformable
fibers within the composite domain and the progressive debonding along the fiber matrix interfaces. A modeling strategy is provided that
is particularly suitable for failure analysis of composites with high-aspect-ratio inclusions, in which direct numerical analysis is computa-
tionally intractable. The performance of the proposed XFEM model is numerically assessed by comparing model predictions to the direct
finite-element method. DOI: 10.1061/(ASCE)EM.1943-7889.0001149. © 2016 American Society of Civil Engineers.
Author keywords: Three-dimensional (3D) extended finite-element method; Short fiber composites; Interface debonding; Random short
fibers.

Introduction high-aspect-ratio fibers is computationally infeasible. This is

Incorporation of short fibers enhances the mechanical properties of
many engineered materials, including increase in elastic modulus,
load carrying capacity, flexural strength, and flexural toughness
(Won et al. 2012; Banthia 1995; Conner et al. 1998). The short fiber
composite materials also provide unique and favorable functional
properties, including crack control, electromagnetic field shielding,
and self sensing (e.g., Chung 2000; Reza et al. 2004; Fu and Chung
1996). This study presents an extended finite-element method
(XFEM) to model high-aspect-ratio short fiber composites in three
dimensions. In the proposed approach, fibers are represented as
two-dimensional (2D) elastic deformable planar rectangles with
progressive debonding capability on the fiber–matrix interface.
The elastic and strength properties of short fiber-reinforced
composites have been modeled using the micromechanical ap-
proach based on Eshelby’s solution of ellipsoidal inclusions com-
bined with the Mori-Tanaka scheme (Tandon and Weng 1986;
Huang 2001), and Hashin-Strichman bounds (Ponte-Castaneda
and Willis 1995), and others. While micromechanical-based analy-
sis provides good estimates of the homogenized properties of
the composite, localized failure mechanisms as well as debonding
between the constituents are not directly modeled. A fuller under-
standing of the idealized failure response can be gained based on
detailed numerical analysis of resolved representative volumes
(Bohm et al. 2002; Lusti et al. 2002). This approach is useful
for understanding fiber interactions with dilute concentrations
(Evans and Zok 1994; Mishnaevsky Jr. and Brndsted 2008), but
direct numerical analysis in the presence of a large number of
1
Dept. of Civil and Environmental Engineering, Vanderbilt Univ., VU
Station B#351831, 2301 Vanderbilt Place, Nashville, TN 37235. E-mail:
matthew.g.pike@vanderbilt.edu
2
Dept. of Civil and Environmental Engineering, Vanderbilt Univ.,
VU Station B#351831, 2301 Vanderbilt Place, Nashville, TN 37235
(corresponding author). E-mail: caglar.oskay@vanderbilt.edu
Note. This manuscript was submitted on December 18, 2015; approved
on June 3, 2016; published online on July 26, 2016. Discussion period open
until December 26, 2016; separate discussions must be submitted for in-
dividual papers. This paper is part of the Journal of Engineering Me-
chanics, © ASCE, ISSN 0733-9399.
because the discretization of the domain must utilize small ele-
ments to resolve the fiber and ensure mesh compatibility between
the fibers and the matrix for capturing the response accordingly.
Approaches to eliminate the need for direct resolution of indi-
vidual fibers in reinforced composites have been previously pro-
posed. Analytical models for short fiber composites have been
developed to capture the elastic response and estimate fiber stresses
(Pan 1996; Hsueh 1990). Numerical models have been also devel-
oped to account for embedded fibers by idealizing the fibers as truss
elements using their end coordinates, without explicitly discretizing
within the finite-element mesh (Elwi and Hrudey 1989; Lykidis
and Spiliopoulus 2008).
Applying the principles of XFEM is an alternative approach for
modeling the behavior of fiber-reinforced composites. With the
XFEM approach, the need to discretize individual fibers and
the need to enforce mesh compatibility between the fibers and the
matrix phase is eliminated. The approximation basis enrichment
strategy of XFEM furnishes the standard finite-element basis with
additional nodal enrichment functions capable of representing
inhomogeneities and discontinuities within the problem domain
without explicitly representing them through meshing (Moës et al.
1999; Belytschko and Black 1999). The enrichment functions are
known a priori to represent the response well within the whole
domain or subdomain of the problem, around strong or weak
discontinuities and incorporated such that the partition of unity
property for the enrichment is satisfied (Babuska and Melenk
1997). XFEM has been successfully used to model strong and weak
discontinuities (Belytschko et al. 2001; Sukumar and Prevost 2003;
Fries and Belytschko 2006) as well as multiple cracks and inclu-
sions (Budyn et al. 2004; Daux et al. 2000; Hiriyur et al. 2011).
Further investigations on XFEM and cohesive behavior were
performed by Zi and Belytschko (2003), Moës and Belytschko
(2002), Asferg et al. (2007), Grogan et al. (2015), Zamani et al.
(2012), Unger et al. (2007), and Wang and Waisman (2015), among
others.
XFEM modeling of cracks in three-dimensional (3D) problems
has been recently proposed for fatigue and dynamic propagation
(Sukumar et al. 2000, 2003; Duarte et al. 2001; Moes et al.
2002), among others. Duarte et al. (2000) accounted for the

© ASCE 04016087-1 J. Eng. Mech.

J. Eng. Mech., 04016087

 presence of internal boundaries and geometries in a 3D domain.
Three-dimensional woven fiber composites were modeled using Ωm
Γt x3
XFEM by Moës et al. (2003). The modeling of cylindrical fibers 2
has been proposed by Soghrati and Geubelle (2012). Oswald et al. x2
(2009) developed an approach to model carbon nanotube and thin
films using 3D XFEM. n
Similar in principals to XFEM modeling, methods to model in- 3
clusion discontinuities have been proposed (e.g., Simo et al. 1993;
Γf 1
Jirasek 2000; Hansbo and Hansbo 2004; Oskay 2009). Radtke et al.
l
(2010, 2011) were the first to model high-aspect-ratio fibers as zero-
measure elastic inclusions in the context of short fiber-reinforced
composites. The strong discontinuity present due to tangential de- x4
bonding at the fiber–matrix interface was modeled with a Heaviside Γu 4 x
1
enrichment function. Tangential slip was idealized based on a w
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

nonlinear cohesive law, while the normal fiber–matrix interface

separation was not considered. Previous work performed by the au-
thors developed a two-dimensional formulation of high aspect ratio
fiber composites using XFEM. An XFEM model for random rigid
short fibers in an elastic domain was proposed in Pike and Oskay
(2015a). Pike and Oskay (2015b) and Pike et al. (2015) formulated
a progressive failure model for random short fiber-reinforced
composite materials for elastic deformable fiber inclusions that ac-
counts for the presence of multiple fibers within close proximity.
This manuscript builds on the ideas of Radtke and co workers as
wells as Pike and Oskay (2015a) and Pike et al. (2015), and
presents a three-dimensional XFEM model for short fiber compo-
sites. In a three-dimensional setting, high-aspect-ratio fibers are
modeled as two-dimensional planar rectangular inclusions. The fi-
bers are taken to be elastic and deformable in the axial direction,
which is the dominant deformation mechanism for short fibers em-
bedded in a matrix domain. Fiber and debonding enrichment func-
tions account for the presence of the fibers within the composite
domain and idealize the progressive debonding on the fiber–matrix
interfaces, respectively. The capabilities of the proposed XFEM
model are verified against the direct finite-element method and
the performance of the model is assessed. The proposed formu-
lation includes the following key features:
• The high-aspect-ratio fiber is modeled as a two-dimensional rec-
tangular inclusion in three dimensions, where the resolved faces
are used to initiate debonding;
• Three-dimensional enrichment functions are used to account for
traction-separation behavior and the strain discontinuity at fiber-
matrix interfaces;
• Multiple zero-measure inclusions are incorporated in the same
element to allow the presence of fibers close to one another
without excessive localized meshing; and
• The present formulation accounts for decohesion in both tan-
gential and normal directions. The computational efficiency
gained by the proposed approach would also allow multiscale
analysis based on both scale separable and inseparable modeling
approaches (e.g., Oskay 2012, 2013; Zhang and Oskay 2015;
Feyel 2003).
Three-Dimensional Enrichment Functions
Fig. 1. Schematic representation of a short fiber composite domain
with two-dimensional fiber identification
et al. (1999), and Pigeon et al. (1996), among others. Other work
utilizing rectangular short fibers of different sizes and materials has
also been performed in e.g., Gopalaratnam and Shah (1987), Lange
et al. (1996), and Yi and Ostertag (2001). Considering no twisting,
a fiber is defined by three corner points of the thin rectangle, which
forms the plane of the fiber along with its width and length. Fibers
are modeled as two-dimensional inclusions due to their high
aspect ratio.
The displacement field discretization that accounts for the
planar fibers is expressed as (Pike et al. 2015)
X
nn n X
X nen α 
uðxÞ ¼ N a ðxÞûa þ N Iαb ðxÞψα ðxÞĉbα
a¼1 α¼1 b¼1
n X
X nαen 
þ N Iαc ðxÞϒα ðxÞd̂cα ð1Þ
α¼1 c¼1
where u = displacement field; x = space coordinate; nn = total num-
ber of mesh nodes in the finite-element discretization; n = number
of fibers; nαen = number of enriched nodes for fiber α; N a = standard
finite-element shape function associated with node a; ûa , ĉbα ,
and d̂cα = nodal coefficients of the standard, fiber enrichment,
and debonding enrichments for each fiber α, respectively; I α =
index set of enriched nodes for fiber α; I αa ∈ I α = index of an
enriched node, a; and ψα and ϒα = fiber enrichment and debonding
enrichment functions, respectively.
The first right-hand side term in Eq. (1) represents the standard
finite-element approximation. The second term represents the
presence of the fiber within the domain, in which the strain
discontinuity in the approximation space is a function of the fiber-
enrichment function, ψα . The third term corresponds to the sepa-
ration along the face of the fiber and the matrix due to the
progressive loss of cohesion between them.

In this manuscript, short fibers in a three-dimensional XFEM do-
main are modeled as two-dimensional objects. Fibers are taken to
have high aspect ratios with l ≫ w ≫ t, where l, w, and t denote
length, width, and thickness of the fiber, respectively, the domain of
the matrix is Ωm , the interface between the fiber and the matrix is
Γf , and the boundaries of the prescribed displacements and trac-
tions are Γu and Γt , respectively (Fig. 1). Fibers of such geometry
to reinforce a cement matrix are employed in Banthia (1995), Pierre
Fiber Enrichment Function
The enrichment function for the fiber is expressed in terms of a
distance function to the fiber body. For simplicity, α, which indi-
cates the fiber, is omitted in this section.
The reinforcing fiber is taken to be entirely embedded in the
open bounded domain of the three-dimensional composite body,
Ω ⊂ R3 . The distance function associated with the fiber body,
ϕc ðxÞ, is expressed as

© ASCE 04016087-2 J. Eng. Mech.

J. Eng. Mech., 04016087


Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.
Fig. 2. Short fiber inclusion enrichment function for three dimensions;
black represents values close to 0 and white represented the highest
value: (a) three-dimensional view with the black line representing
the fiber; (b) fiber enrichment values in the X-Y plane; (c) fiber enrich-
ment values in the Y-Z plane
ϕc ðx; xp Þ ¼ jjx − xp jj2 ; x ⊂ ΩΓ f ; xp ⊂ Γf ⊂ R3 ð2Þ
where xp = any point on the fiber domain, Γf ; x = position within
the matrix; and jj · jj2 ¼ L2 norm.
The distance function defined in Eq. (2) is used to express the
enrichment function for the fiber as the shortest distance between
any point within the matrix and the fiber domain:
ψðxÞ∶ ¼ min ½ϕc ðx; xp Þ
xp ∈Γf
The enrichment function is illustrated in Fig. 2(a) for a randomly
placed short fiber in a three-dimensional domain, where the value
of the enrichment function increases from black (zero) to white
(highest). Cross-sectional contours of the enrichment function
are shown in Figs. 2(b and c). The enrichment function, ψ, incor-
porates a strain discontinuity mode along the fiber position, and
displacements around the fiber can be captured without explicitly
discretizing the fiber domain. The form of Eq. (3) for the enrich-
ment function ensures that the approximation basis captures the
strain discontinuity but stays smooth otherwise around the sides
and face of the fiber.
Debonding Enrichment Function
The debonding enrichment function describes the debonding be-
tween the fiber and the matrix, and introduces a strong discontinu-
ity in the displacement field. To model the shape of the debonding
between the fiber and matrix interface in three dimensions, a
Heaviside function is used to represent the physical separation
in the normal and tangential directions. Since the fiber is idealized
as planar, the debonding is considered along the plane of the fiber.
The possibility of debonding along the sides of the fiber is not
considered in this work.
© ASCE
J. Eng. Mech., 04016087
The debonding enrichment function is expressed in terms of the
level set functions describing the fiber domain. The level set asso-
ciated with the plane of the fiber, ϕf ðxÞ, is expressed as
ϕf ðxÞ ¼ n̂ · ðx − xc Þ ð4Þ
where n̂ = normal vector to the fiber plane; and xc = reference point
on the plane of the fiber, set to the center of the fiber; ϕf divides the
domain along the plane of the fiber and is the signed distance func-
tion, which vanishes on the fiber plane. The level set functions for
the fiber edges, e, are expressed as
ϕe ðxÞ ¼ ðx − xxe Þ · te ; e ¼ 1,2; 3,4 ð5Þ
where e = fiber edge; te = tangent at the fiber edge; xx1 = end point
on the fiber edge, e (i.e., fiber corner point); and ϕe provides the
zero-level set along the plane normal to the fiber edge, e, with
positive values on one side of the domain cut by ϕe and negative
elsewhere in the domain.
The debonding enrichment function for the fiber is then ex-
pressed in terms of the level set functions as
Y
4 
ϒðxÞ ¼ Hðϕf Þ Hð−ϕe Þ ð6Þ
e¼1
The debonding enrichment function, ϒ, is employed to capture
debonding along both normal and tangential directions. Interfacial
damage in directions normal and tangential to the fibers has been
observed experimentally in fiber-reinforced composites (e.g., Pan
and Leung 2007).
Governing Equations and Computational
Formulation
The mechanical equilibrium constitutive laws and the boundary
conditions within the three-dimensional domain for the short
fiber-reinforced composite are
∇ · σðx; tÞ ¼ 0; x∈Ω ð7Þ
ð3Þ
σ ¼ LðxÞ∶ϵðx; tÞ ð8Þ
~
uðx; tÞ ¼ uðx; tÞ; x ∈ Γu ð9Þ
σ · n ¼ t~ðx; tÞ; x ∈ Γt ð10Þ
where σ = stress tensor; ∇ · ð·Þ = divergence operator; ϵ = strain
tensor, where the strain is taken to be the symmetric gradient of
the displacement field (ϵ ¼ ∇s u); and L = tensor of elastic moduli
of the matrix or of the fiber, as a function of position within the
domain, where L is taken to be symmetric and positive definite
for both materials. While in this study, the matrix and fibers are
taken to be elastic, it is straightforward to consider nonlinear or
damage behavior within the matrix as demonstrated in a 2D setting
in Pike and Oskay (2015b). u~ and t~ = prescribed boundary dis-
placements and tractions defined on boundaries Γu and Γt ,
respectively, such that Γu ∩ Γt ¼ ∅ and ∂Ω ¼ Γu ∪ Γt . The
domains of fiber, α and the matrix are denoted as Ωα and Ωm , re-
spectively. The short fibers do not intersect with each other
(i.e., Ω ¼ Ωm ∪∪nα¼1 Ωα ) and are taken to be embedded fully in
the matrix.
Across the fiber–matrix interface the traction continuity is ex-
pressed as
04016087-3 J. Eng. Mech.
 (a) (b) (c)

Fig. 3. Three-dimensional direct finite-element simulation (cross section cut at the center of the fiber width): (a) initial representative volume element
(RVE) with the short vertical fiber before tensile loading; (b) amplified deformed RVE with the short vertical fiber after tensile loading; (c) amplified
deformed RVE with the short vertical fiber after shear loading
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

tractions along the length of the fiber. No significant shear stress or

⟦T⟧ ¼ ⟦σ · n⟧ ¼ 0 x ∈ Γα ≡ ∂Ωm ∩ ∂Ωα α ¼ 1,2,..,n
bending moment develops within the domain of the fiber and the
ð11Þ stress in the fiber is taken to be predominately axial. These consid-
erations are verified using direct finite-element simulations where
where the traction T is defined in the local coordinate system as a the fiber is resolved with highly resolved meshes in three dimen-
function of the normal and tangential components; n = outward unit sions. Fig. 3 displays a finite-element simulation of a cross section
vector to the boundary; ⟦ · ⟧ = jump operator; and Γα = interface cut of a vertical short fiber (l=w=t ¼ 1000=20=1) in a matrix do-
between the fiber, α, and the matrix. main subject to tensile and shear loading. Fig. 3(a) displays the
A cohesive law describing the relationship between surface trac- initial state while Fig. 3(b) shows the deformation state under ten-
tion and separation represents the physical deterioration that occurs sile loading. The displacements are amplified for visualization pur-
at the fiber–matrix interface. A bilinear cohesive law is considered: poses and the figures shows that fiber bending is insignificant.
8 Fig. 3(c) demonstrates the deformation profile under shear loading,
>
> ⟦ûi ⟧ max which also exhibits insignificant bending in the fiber.
>
> di ≥ ⟦ûi ⟧ ≥ 0
>
> ti With the high-aspect-ratio fiber and the assumption of uniform
>
< di stress across the width of the fiber, the third term in Eq. (13) ap-
T i ð⟦ûi ⟧Þ ¼ tmax ð12Þ proximated as
>
>
i
ðdcrit
n − ⟦ûi ⟧Þ di ≥ ⟦ûi ⟧ ≥ di
crit
>
> d crit
− d
>
>
i i Z Z
>
:0 ⟦ûi ⟧ ≥ dcrit
i σ∶δϵdΩ ≃ Afα σαf δϵαf dΩ ð14Þ
Ωα Ωα

where ⟦û⟧ = displacement jump vector (i.e., separation) defined

in the local coordinate system; di = cohesive characteristic separa- where Afα = cross-sectional area of the fiber. The axial stress
tion length for the normal and tangential directions (di ¼ in fiber α is taken to be proportional to the axial strain
dn ; dt1 ; dt2 ); tmax = ultimate traction; and dcrit (i.e., σαf ¼ Ef ϵαf ), where Ef = elastic modulus of the fiber.
i i = maximum displace-
ment jump along the normal and tangential directions. The domain of the matrix is taken to occupy the entire domain,
To develop the weak form of the model from Eqs. (7)–(11), the since the domains of the fibers are vanishingly small and modeled
standard finite-element procedure is used and the weak form is ex- as zero-measure inclusions (i.e., 2D). The weak form of the gov-
pressed as follows: erning equations for the fiber composites then becomes
Z Xn Z Z n Z
X X
n Z
σ∶δϵdΩ þ T · δ⟦u⟧dΓ σ∶δϵdΩ þ T · δ⟦u⟧dΓ þ Afα Ef ϵαf δϵαf dΩ
Ωm α¼1 Γα Ω α¼1 Γα α¼1 Ωα
n Z Z Z
X
þ σ∶δϵdΩ − t~ · δudΓ ¼ 0 ð13Þ − t~ · δudΓ ¼ 0 ð15Þ
Ωα Γt Γt
α¼1

where δu = test function; and δϵ = gradient of the test function.
With the condition that the short fibers have high aspect ratios, it
is assumed that tractions along the two opposing faces of a fiber in
the thickness direction are uniform. Debonding along the two faces
of a fiber would typically occur concurrently for a short fiber em-
bedded in a matrix under the traction conditions. However, the fiber–
matrix debonding is likely to initiate at a weak spot at one face of the
fiber. Upon complete debonding at the weak face, the tractions along
the opposing (unbonded) face relax. Since the width of the fiber is
significantly smaller than the length, lα =wα ≫ 1, tractions across the
width of the fiber are also taken to be uniform.
Fibers are assumed to be fully embedded in the domain,
have high aspect ratios, and be subjected to approximately uniform
XFEM is employed to discretize and evaluate the governing
equations [Eqs. (7)–(15)]. The standard Ritz-Galerkin procedure
is utilized and results in a nonlinear system, which is incrementally
evaluated using the Newton-Raphson method. The computational
complexity of the proposed method is related to the number of fi-
bers present in the system as the number of enrichment degrees of
freedom is proportional to fiber density. The nonlinear iterations to
convergence also increases as a function of fibers, since more
possible debonding sites (and hence more-pronounced cohesive
behavior) exist within the problem domain. The 3D formulation
associated with the extended finite-element discretization generally
follows the 2D counterpart, which is described in detail in Pike and
Oskay (2015b) and Pike et al. (2015). The following subsection

© ASCE 04016087-4 J. Eng. Mech.

J. Eng. Mech., 04016087

 describes the numerical integration of various enriched elements
necessary to evaluate Eq. (15) in the XFEM context.
Numerical Integration
In XFEM, the extent of the subdomain around an inclusion that is
enriched is chosen based on the geometry or the discretization. A
geometry-based approach is considered for modeling cracks using
a specified radius around the crack tip since the stress fields around
the crack tip vary as a function of the distance from the tip. In this
work, the enrichment domain is chosen based on the discretization
since the enrichments functions accurately represent the local
behavior around the inclusion. Various methods to integrate enrich-
ment functions in three dimensions have been investigated by
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.
Laborde et al. (2005), Bechet et al. (2005), and Park et al. (2009).
In this study, the 3D domain is comprised of a standard mesh of
hexahedral elements with four different element types: (1) far-field
elements; (2) partially enriched elements; (3) fully enriched
elements; and (4) fully enriched elements that also have partial
enrichment.
The far-field elements have no enrichment and use standard
quadrature rules with eight integration points in a trilinear hexahe-
dral element. Partially enriched elements include some nodal en-
richment from one or multiple fiber inclusions but are not fully
enriched (i.e., one or more fibers lie in elements adjacent to the
current element). Since no displacement discontinuity exists, stan-
dard quadrature rule is employed with eight integration points in a
hexahedral element. Fully enriched elements are those crossed by
one or multiple fibers. Considering the position of each fiber, the
element domain is split into tetrahedral subdomains (Fig. 4) using
Delaunay triangulation. The tetrahedral subdomains are placed
such that no subdomain crosses the domain of the fiber and sub-
domains line up with the fiber interface, to ensure that the compo-
nents of the enrichment function for the fiber edges and domain
level sets are integrated separately. Each tetrahedral subelement
contains 11 integration points creating high-order integration in
the enriched element. Fully enriched elements that also have partial
enrichment use the same element-splitting rules as in the fully en-
riched elements, and use higher-order integration rules for each
subelement.
The cohesive interface integration is performed based on two-
dimensional Gauss quadrature on the two-dimensional domain of
1
0 counterparts. The maximum cohesive separation lengths (dcrit
-1
1 64 elements up to approximately 33,000 elements with correspond-
0 1
0 the fiber domain, the average element edge length is approximately
-1 -1
Fig. 4. Three-dimensional visualization of tetrahedral subelements in
an hexahedral enrichment element; thick black line represents the fiber
inclusion
© ASCE 04016087-5
J. Eng. Mech., 04016087
the fiber. The interface integration is independent of the subdomain
discretization. The fiber–matrix interface Γα is divided into multi-
ple rectangular domains, based on the length of the fiber, lα . Each
rectangular domain uses a four-point quadrature rule to perform the
cohesive integration on the fiber domain.
Numerical Examples
In this section, the performance of the proposed model in capturing
the response of short fiber-reinforced composites is assessed.
The first example investigates the accuracy characteristics of a
randomly oriented single fiber embedded in a domain with elastic
matrix perfectly bonded with the fiber, as well as in the presence
progressive debonding along the fiber–matrix interface. The second
example investigates the fiber–matrix debonding and localized
stresses for a case with two randomly oriented fibers in close prox-
imity to each other. The effects of the orientation of the fiber inter-
face within the domain in relation to the applied loading, the
influence of the aspect ratio of the fiber with respect to the interface
debonding are presented next. The study then evaluates the re-
sponse of a domain reinforced by multiple fibers.
Single-Fiber Inclusion
The schematic representations of the model problems are shown in
Fig. 5. Two cases are investigated: a cubic domain subjected to uni-
form displacement-controlled tensile loading [Fig. 5(a)]; and a do-
main subject to uniform displacement-controlled shear loading
[Fig. 5(b)]. In both cases, the amplitude of loading is 3% bulk
strain. Symmetric boundary conditions are imposed on the left
(x ¼ 0), bottom (z ¼ 0), and back (y ¼ 0) faces for the tensile case.
The shear-loading case utilizes fixed boundary conditions in all
three directions at the left bottom vertex (x ¼ 0, y ¼ 0, and
z ¼ 0) and fixed z-direction boundary conditions on the back face
(z ¼ 0). The edge length of the domain is 4 mm. The matrix
material is taken to be concrete with the Young’s modulus and Pois-
son’s ratio set as 14 GPa and 0.3, respectively. The fiber is taken to
have a Young’s modulus of 207 GPa, and width and thickness of 20
and 1 μm, respectively. The fiber is placed such that its face is par-
allel to the x-direction. The coordinates of the end points of the
fiber body are (2.02, 2.37, 1.73) and (2.34, 2.14, 2.38), with a fiber
length of 1 mm.
When included, progressive debonding at the fiber–matrix in-
terfaces is modeled using the bilinear cohesive zone law [defined
in Eq. (12)] for both the proposed model and reference simulations.
The peak normal traction (tmaxn ) and normal cohesive characteristic
separation length (dn ) are set as 10 MPa and 1 nm respectively.
The peak shear traction (tmax max
t1 , tt2 ) and shear cohesive character-
istic separation lengths (dt1 , dt2 ) are set identical to their normal
n ,
dcrit crit
t1 , and dt2 ) are taken as 8 nm both for normal and shear direc-
tions (Nicholas et al. 2013).
In the proposed XFEM model, the domain is discretized with a
uniform grid (hexahedra) of various mesh densities ranging from
ing element sizes of h ¼ 1 mm and h ¼ 0.125 mm, respectively.
The reference model consists of explicitly modeled, fully resolved
3D solid fibers and a very fine nonuniform discretization. Within
0.33 μm. In order to ensure that the reference model itself is ac-
curate, a mesh sensitivity analysis for the direct finite-element
method has also been performed. A very fine and convergent mesh
has been employed as the reference simulation. The reference
model discretization consists of approximately 550,000 elements.
J. Eng. Mech.
 E E

F F
B1
B1

A1 A1
C C

Z D Z D

z
Y X y Y X
x
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)

Fig. 5. Geometry and boundary conditions of the single-fiber inclusion examples: (a) tensile loading case with loading at the right Y-Z face (shaded
area); (b) shear loading case with loading at the right Y-Z face and top X-Y face (shaded area)

Fig. 6 shows the pointwise displacement errors of the proposed The pointwise error for each case is presented as a function of
XFEM model with respect to the reference simulation for the element size in the XFEM discretizations. The six locations at which
tensile [Figs. 6(a and b)] and shear loading [Figs. 6(c and d)] the pointwise errors are computed are shown in Figs. 5(a and b).
cases. The pointwise error is computed using the discrete norm The errors from the elastic tensile loading example (i.e., no de-
expressed as bonding is considered) are shown in Fig. 6(a). The errors show
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P3 monotonic convergence as a function of mesh size. The left fiber
Δui ðx̂Þ2 end (Point A1) resulted in the largest error in all cases, but was
Error ¼ p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pi¼1 ð16Þ
3 2
i¼1 ui ðx̂Þ
below 1% for the smallest mesh size. Points C, D, E, and F all
had minimal error for all mesh sizes. Fig. 6(b) illustrates the point-
where Δu = difference in displacement predicted by the proposed wise errors corresponding to the case where the progressive de-
model and reference simulation; u = value of the reference simu- bonding between the fiber and the matrix is considered. Similar
lation; and x̂ = position of the point where the error is assessed. to the elastic case, the errors showed monotonic mesh convergence,

6 5
Point A1 Point A1
5 Point B1 Point B1
Point C 4 Point C
Absolute Error [%]

Absolute Error [%]

Point D Point D
4
Point E Point E
3
Point F Point F
3
2
2

1 1

0 0
0 0.2 0.4 0.6 0.8 1 0 0.5 1
(a) Mesh Size, h[mm] (b) Mesh Size, h[mm]
8 8
Point A1 Point A1
Point B1 Point B1
6 Point C 6 Point C
Absolute Error [%]

Absolute Error [%]

Point D Point D
Point E Point E
Point F Point F
4 4

2 2

0 0
0 0.5 1 0 0.5 1
(c) Mesh Size, h[mm] (d) Mesh Size, h[mm]

Fig. 6. Pointwise error as a function of mesh density: (a) elastic domain with tensile loading; (b) domain with progressive interface debonding with
tensile loading; (c) elastic domain with shear loading; (d) domain with progressive interface debonding with shear loading loading

© ASCE 04016087-6 J. Eng. Mech.

J. Eng. Mech., 04016087

 in approximately 33,000 elements. The reference simulation,
using the direct finite-element method, consists of approximately
740,000 elements. Similar to the previous example, the fiber do-
mains are completely resolved with a fine grid of solid elements,
with the average element edge length approximately 0.33 μm.
Fig. 8 displays the magnitude of the displacement jump along
Z the length of the fibers computed by the proposed model and the
reference simulation. For both fibers, the proposed model captures
2 the general shape and magnitude of the displacement jump. The
1
displacement jump for Fiber 1 [Fig. 8(a)], displays approximately
the same maximum amplitude over the length of the fiber for both
models. In both simulations, the ends of the fiber display a dis-
Y placement jump. The debonding computed by the proposed model
X
at each of the fiber tips was slightly greater than that of the refer-
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

ence simulation. Fig. 8(b) shows the displacement jump along the
Fig. 7. Geometry and boundary conditions of the two-fiber inclusion fiber length with respect to Fiber 2. The shape of the debonding in
example the proposed model does not match exactly to the reference sim-
ulation, but does display similar magnitude and shape. In both fi-
bers, the magnitude of the displacement jump is controlled by the
normal displacement jump, since in these examples, the face of the
with all errors below 0.25% at the smallest mesh. The largest error
fiber is parallel to the direction of loading. The displacement jump
occurred at Point A1 for all mesh sizes and the errors at the edges of
in both Fibers 1 and 2 display partial separation across their
the domain (Points C, D, E, and F) were negligible.
Figs. 6(c and d) display the pointwise errors for the elastic respective lengths under the applied load amplitude. A three-
and debonding shear loading cases, respectively. The errors dimensional representation of the progressive debonding in the
from both elastic and debonding cases show a decrease in error with x-direction, y-direction, and z-direction is illustrated in Fig. 9,
mesh size. The largest errors occurred at the left and right fiber for a planar view in the X-Y plane [Fig. 9(a)], a three-dimensional
tips (Points A1 and B1) for all mesh sizes studied. Similar to view of the debonding in the domain [Fig. 9(b)], and a magnified
the case of tensile loading, the error at the smallest mesh size of view of the three-dimensional debonding [Fig. 9(c)]. The debond-
h ¼ 0.125 mm was approximately below 1.5% for all points ing is amplified (1 × 105 times) to show the magnitude and shape
considered. of displacement jumps. The slight discrepancy between the dis-
placement jump profiles computed by the proposed and reference
models is attributed to the discretization error introduced by the
Two-Fiber Inclusion coarser XFEM mesh, especially in capturing the high-displacement
In this section, a two-fiber inclusion example is studied and com- jump gradients at the fiber tips.
pared against the direct finite-element method to assess the accu- Figs. 10(a and b) illustrate the local variation of the normal stress
racy of the proposed approach in capturing interfacial debonding around Fiber 1 as computed by the reference model [Fig. 10(a)]
and stress. Two fibers are placed neighboring to each other in the and proposed model [Fig. 10(b)], at the time step when tractions
domain subjected to displacement-controlled tensile loading with along the interface are near the peak value. The stress contours
an amplitude of 3% bulk strain. Fig. 7 displays the positions of are from the cross section parallel to the X-Y plane at Z ¼
the two fibers within the domain. The domain size, loading and 1.75 mm. In both the reference model and proposed model, the
boundary conditions, and constituent interface properties are taken stress field away form the fibers is uniform. Around the fiber,
to be the same as in the single-fiber inclusion example. Both of the the stress distributions computed by the proposed and reference
fibers in the domain are 1 mm in length. The coordinates of the end models are similar. Figs. 10(c and d) compare the stress distributions
points of Fiber 1 are (1.33, 1.10, 1.53) and (2.20, 2.38, 1.55), of the proposed and reference models at a time step when the inter-
whereas the coordinates of the end points of Fiber 2 are (1.77, face separations along the fibers are near maximum separation
1.84, 1.56) and (1.36, 2.74, 2.19). Both fibers are oriented parallel (i.e., separation beyond which tractions vanish). Higher stress con-
to the direction of the load. In this example, the fibers are modeled centrations are captured by both models at each of the fiber tips.
with progressive debonding along the fiber–matrix interfaces. Along the fiber edges, relatively low stresses occur as the tractions
The two-fiber composite domain with the proposed model unload within the softening regime of the cohesive behavior, past
is discretized using a mesh size of h ¼ 0.125 mm, resulting the peak tractions.
Displacement Jump [mm]
Displacement Jump [mm]

x 10-6 x 10-6
10 10
Reference Simulation Reference Simulation
8 XFEM 8 XFEM
6 6
4 4
2 2

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

(a) Position Along Fiber [mm] (b) Position Along Fiber [mm]

Fig. 8. Displacement jump along the fiber length for the two-fiber inclusion case: (a) Fiber 1; (b) Fiber 2

© ASCE 04016087-7 J. Eng. Mech.

J. Eng. Mech., 04016087

 4 4

2 2
Y

0
Z0 4
2 0 2 4
4 2 Reference Simulation
0 2 4 0 0 XFEM
X Y X
(a) (b) (c)

Fig. 9. Debonding in a 3D domain for the two-fiber inclusion case: (a) perspective view; (b) 3D view for debonding (magnitude amplified for
visualization); (c) magnified view; solid straight black line represents each fiber, solid gray line displays results of the reference model, and dotted
line is the results of the proposed model
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 10. Cross-sectional plot of the normal stress in the x-direction of Fiber 1; plots taken in the X-Y plane: (a) reference simulation at time step of
peak traction; (b) proposed model at time step of peak traction; (c) reference simulation at time step maximum debonding separation; (d) proposed
model at time step of maximum debonding separation

The stress variations around Fiber 2 near the peak traction
and near maximum separation are displayed in Fig. 11.
Figs. 11(a and b) illustrate the local variations in the normal stress
around Fiber 2 as computed by the reference model [Fig. 11(a)]
and proposed model [Fig. 11(b)], at the time step when tractions
along the interface are near the peak value. The stress contours are
from the cross section parallel to the X-Y plane at Z ¼ 1.55 mm.
There is a uniform stress field away from the fibers, in both the
reference and proposed models. The computed stress distributions
for the proposed model and reference models around the fiber are
similar along its length. The stress distribution at a time step when
the interface separations along the fibers are near maximum sep-
aration are compared in Figs. 11(c and d). Along the length of the
fiber, slightly lower stresses occur on either side of the fiber as
compared with the uniform stress in the domain, similar to
Fiber 1.
Effect of Fiber Orientation
The orientation of the fiber relative to the loading direction signifi-
cantly affects the response of the composite. In this section, the
effect of the orientation of the fiber with respect to the direction
of the load is investigated.
The single-fiber case discussed earlier is modeled with
progressive debonding for various fiber orientations. The same
material parameters, cohesive law, and boundary conditions are
used in the earlier example are employed. The mesh size is set
at h ¼ 0.125 mm for all simulations. The fiber is rotated about

© ASCE 04016087-8 J. Eng. Mech.

J. Eng. Mech., 04016087

Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 11. Cross-sectional plot of the normal stress in the x-direction of Fiber 2; plots taken in the X-Y plane: (a) reference simulation at time step of
peak traction; (b) proposed model at time step of peak traction; (c) reference simulation at time step maximum debonding separation; (d) proposed
model at time step of maximum debonding separation

its center axis for 13 different angles. The angle is measured at θf ¼
0° for fiber face completely parallel to the direction of the load in
the x-direction [Fig. 12(b)]. θf is rotated every 15° until the fiber
face is perpendicular to the direction of the load in both directions.
The maximum normal separation along the fiber interface is re-
ported in Fig. 12(a), for angles ranging from −90° to 90°. Naturally,
the maximum normal displacement jump occurs when the fiber
face is parallel to the direction of the load. When the fiber is rotated
15° in either direction, there is a steep drop in maximum normal
separation at the same load amplitude. The maximum normal
x 10-6
8
Max Normal Separation [mm]
separation decreases when the fiber face continues to rotate toward
being perpendicular to the load. When the fiber face is completely
perpendicular to the load, there is negligible normal separation.
It can be seen that the direction of the fiber face influences the
response of the fiber–matrix debonding. A slight asymmetry is
observed in Fig. 12(a) since the nonsymmetric position of the fiber
within the domain leads to different interface tractions under pos-
itive or negative rotations of equal amplitude.
Effect of Fiber Aspect Ratio
Next, the effect of the fiber’s aspect ratio (l=w) on the interface
debonding properties is investigated. A composite with a single fi-
ber, with the orientation of the fiber parallel to the direction of the

load, is modeled with various aspect ratios using the proposed

XFEM model.
6 Fig. 13 shows the maximum of the normal separations along the
fiber length plotted against the fiber aspect ratio at identical load
4 amplitudes. Seven different fiber aspect ratios are modeled ranging
from 5 to 200. The largest maximum normal separation occurs with
2
the lowest fiber aspect ratio, which is 5. The maximum normal sep-
aration drops as a function of the aspect ratio. There appears to be
an exponential decrease of separation with the increase in aspect
0 ratio, with minimal change in separation between the larger-aspect
-100 -50 0 50 100
ratio fibers. As illustrated, the aspect ratio of the fiber significantly
Fiber Face Rotation [ ]
effects the response of the debonding.
(a) (b)

Fig. 12. Effect of fiber rotation response: (a) maximum normal separa-
tion as a function of the fiber orientation; (b) fiber orientation with
respect to the center of the fiber
Dense Fiber Domain
This section investigates a three-dimensional dense fiber domain
with the proposed XFEM approach for the progressive debonding

© ASCE 04016087-9 J. Eng. Mech.

J. Eng. Mech., 04016087

 response. A domain of 4 × 4 × 4 mm is taken to have 250 ran- within the domain. The random dispersion of fibers results in multi-
domly oriented fibers in its domain with a mean length of ple elements having multiple fiber enrichments. Fibers are placed
1 mm (0.1 mm), and a fixed width of 0.02 mm, as displayed such that they do not touch and do not intersect each other. The
in Fig. 14. All 250 fibers are randomly positioned and oriented mesh size used in this example is h ¼ 0.25 mm. Fig. 14 does
not display the mesh size used in this example, but uses a much
larger grid size for visualization purposes. The boundary condi-
x 10-5 tions, matrix properties, fiber properties, and cohesive law are taken
2
to be the same as in the previous examples.
Fig. 15 summarizes the maximum separation at the fiber–matrix
Max Normal Separation [mm]

interface for each of the fibers. The magnitude range of the maxi-
1.5 mum separation are categorized in Fig. 15(a). There were approx-
imately 100 fibers that resulted in a maximum interface separation
of less than 2.5×10−6 mm. A total of 75 fibers had a maximum
1 separation between 2.5×10−6 and 5×10−6 mm, 30 fibers had a
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.

maximum separation between 5×10−6 and 7.5×10−6 mm, and

45 fibers resulted in a maximum separation over 7.5×10−6 mm.
0.5 Fig. 15(b) displays the severity of debonding. The majority of
the fibers in the domain displayed partial separation at their
fiber–matrix interface (175 of the 250 fibers); however, 30 fibers
0 showed no separation and 45 fibers had complete separation at the
0 50 100 150 200
Fiber Aspect Ratio
fiber–matrix interface. The varied separations of the fibers occur
due to the fiber position and face orientation.
Fig. 13. Maximum normal separation as a function of fiber aspect ratio

Conclusions

The formulation and implementation of an extended finite-element

4 method (XFEM) approach for modeling short fiber composites in a
three-dimensional domain was presented. Short fibers were incor-
porated into the XFEM framework as elastic two-dimensional rec-
tangular planar inclusions that were able to deform in the axial
2
Z

direction. A fiber enrichment function to account for the fiber pres-

ence and a debonding enrichment function to idealize the
progressive debonding on the fiber–matrix interfaces were devel-
oped. The approach provided a modeling strategy for very high
0
4 4
aspect ratio inclusions with fiber–matrix debonding capabilities
in three dimensions. The performance of the XFEM approach
2 2
Y X
was assessed against the direct finite-element method for single-
0 0 fiber and two-fiber domain cases and displayed high accuracy. The
XFEM model also demonstrated its capabilities in modeling the
Fig. 14. Three-dimensional dense fiber domain
response of dense short fiber microstructures.

100 200

80
Number of Fibers

Number of Fibers

150

60
100
40

50
20

0 0
m

mm

ion
ion
mm

ion
m
-6m

-6m

rat
rat

rat
e-6
e-6

epa
.5e

epa

epa
.5e
7.5
-5
<2

>7

lS
S

ll S
2.5

5-

No

rtia

Fu
Pa

(a) (b)

Fig. 15. Summary of displacement jumps for the fibers in the dense fiber domain: (a) total number of fibers with their respective maximum separation
of each displacement jump; (b) total number of fibers with no separation, partial separation, and full separation

© ASCE 04016087-10 J. Eng. Mech.

J. Eng. Mech., 04016087

 While the XFEM model in this manuscript works well for the
problems analyzed, several challenges remain in more realistic and
complicated applications. A key feature that needs to be addressed
is the modeling of realistic fiber volume fractions (e.g., 0.5 and 1%)
of high-aspect-ratio fibers in order to compare and reproduce ex-
perimental results. Using high-aspect-ratio fibers to achieve the
realistic fiber volume fractions produces a significant amount of
fibers in the representative volume element (RVE), which results
in numerous fibers in close proximity to each other. This may affect
the approximation of the composite, and this issue requires further
investigation. The interfacial properties can alter various mechani-
cal properties of the composites, including flexural strength and
ductility. In this work, the interface properties were considered
along the entire length of the fiber–matrix interface; in reality, their
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.
bond strength may differ at different locations along the length.
Therefore, spatial variability should be considered to accurately as-
sess the interfacial response. Furthermore, many of the fibers em-
ployed in short fiber-reinforced composites have circular cross
sections, wherein the debonding orientation is dictated by loading
direction rather than cross-sectional properties. The proposed strat-
egy will be generalized to account for fibers with circular cross sec-
tions. In order to account for matrix cracking as well as the
interactions between matrix cracking and fiber–matrix debonding,
matrix failure models based on regularized continuum damage me-
chanics or discrete crack modeling could be incorporated in future
studies.
Acknowledgments
The authors gratefully acknowledge the financial support from
Vanderbilt University and the Vanderbilt Institute of Software
Integrated Systems.
References
Asferg, J., Poulsen, P., and Nielsen, L. (2007). “A consistent partly cracked
XFEM element for cohesive crack growth.” Int. J. Numer. Methods Eng.,
72(4), 464–485.
Babuska, I., and Melenk, J. M. (1997). “The partition of unity method.”
Int. J. Numer. Methods Eng., 40(4), 727–758.
Banthia, N. (1995). “Uniaxial tensile response of microfibre reinforced
cement composites.” Mater. Struct., 28(9), 507–517.
Bechet, E., Minnebo, H., Moes, N., and Burgardt, B. (2005). “Improved
implementation and robustness study of the X-FEM for stress analysis
around cracks.” Int. J. Numer. Methods Eng., 64(8), 1033–1056.
Belytschko, T., and Black, T. (1999). “Elastic crack growth in finite ele-
ments with minimal remeshing.” Int. J. Numer. Methods Eng., 45(5),
601–620.
Belytschko, T., Moes, N., Usui, S., and Parimi, C. (2001). “Arbitrary dis-
continuities in finite elements.” Int. J. Numer. Methods Eng., 50(4),
993–1013.
Bohm, H. J., Eckschlager, A., and Han, W. (2002). “Multi-inclusion unit
cell models for metal matrix composites with randomly oriented discon-
tinuous reinforcements.” Comput. Mater. Sci., 25(1–2), 42–53.
Budyn, E., Zi, G., Moes, N., and Belytschko, T. (2004). “A method for
multiple crack growth in brittle materials without remeshing.” Int. J.
Numer. Methods Eng., 61(10), 1741–1770.
Chung, D. (2000). “Cement reinforced with short carbon fibers: A multi-
functional material.” Compos. Part B: Eng., 31(6–7), 511–526.
Conner, R., Dandliker, R., and Johnson, W. (1998). “Mechanical properties
of tungsten and steel fiber composites Zr41.25 Ti13.75 CU12.5 Ni10 Be22.5
metallic glass matrix composites.” Acta Mater., 46(17), 6089–6102.
Daux, C., Moes, N., Doblow, J., Sukumar, N., and Belytschko, T. (2000).
“Arbitrary branched and intersecting cracks with the extended finite
element method.” Int. J. Numer. Methods Eng., 48(12), 1741–1760.
© ASCE 04016087-11
J. Eng. Mech., 04016087
Duarte, C., Hamzeh, O., Liszka, T., and Tworzydlo, W. (2001). “A gener-
alized finite element method for the simulation of three dimensional
dynamic crack propagation.” Comput. Methods Appl. Mech. Eng.,
190(15–17), 2227–2262.
Duarte, C. A., Babuska, I., and Oden, J. (2000). “Generalized finite element
methods for three-dimensional structural mechanics problems.” Comp.
Struct., 77(2), 215–232.
Elwi, A., and Hrudey, T. (1989). “Finite element model for curved em-
bedded reinforcement.” J. Eng. Mech., 10.1061/(ASCE)0733-9399
(1989)115:4(740), 740–754.
Evans, A., and Zok, F. (1994). “The physics and mechanics of fibre-reinforced
brittle matrix composites.” J. Mater. Sci., 29(15), 3857–3896.
Feyel, F. (2003). “A multilevel finite element method (FE2 ) to describe the
response of highly non-linear structures using generalized continua.”
Comput. Methods Appl. Mech. Eng., 192(28–30), 3233–3244.
Fries, T., and Belytschko, T. (2006). “The intrinsic XFEM: A method for
arbitrary discontinuities without additional unknowns.” Int. J. Numer.
Methods Eng., 68(13), 1358–1385.
Fu, X., and Chung, D. D. L. (1996). “Submicron carbon filament cement-
matrix composites for electromagnetic interference shielding.” Cem.
Concr. Res., 26(10), 1467–1472.
Gopalaratnam, V., and Shah, S. (1987). “Tensile failure of steel fiber re-
inforced mortar.” J. Eng. Mech., 10.1061/(ASCE)0733-9399(1987)
113:5(635), 635–652.
Grogan, D., Bradaigh, C. O., and Leen, S. (2015). “A combined XFEM and
cohesive zone model for composite laminate microcracking and per-
meability.” Compos. Struct., 120, 246–261.
Hansbo, A., and Hansbo, P. (2004). “A finite element method for the sim-
ulation of strong and weak discontinuities in solid mechanics.” Comput.
Methods Appl. Mech. Eng., 193(33–35), 3523–3540.
Hiriyur, B., Waisman, H., and Deodatis, G. (2011). “Uncertainty quantifi-
cation in homogenization of heterogeneous microstructures modeled by
XFEM.” Int. J. Numer. Methods Eng., 88(3), 257–278.
Hsueh, C. (1990). “Interfacial debonding and fibre pull-out stresses of
fibre-reinforced composites.” Mater. Sci. Eng. A, 123(1), 1–11.
Huang, J. H. (2001). “Some closed-form solutions for effective moduli of
composites containing randomly oriented short fibers.” Mater. Sci. Eng.
A, 315(1–2), 11–20.
Jirasek, M. (2000). “Comparative study on finite elements with embedded
discontinuities.” Comput. Meth. Appl. Mech. Eng., 188(1–3), 307–330.
Laborde, P., Pommier, J., Renard, Y., and Salaun, M. (2005). “High-order
extended finite element method for cracked domains.” Int. J. Numer.
Methods Eng., 64(3), 354–381.
Lange, D., Ouyang, C., and Shah, S. (1996). “Behavior of cement based
matrices reinforced by randomly dispersed microfibers.” Adv. Cem.
Based Mater., 3(1), 20–30.
Lusti, H. R., Hine, P., and Gusev, A. (2002). “Direct numerical predictions
for the elastic and thermoelastic properties of short fibre composites.”
Compos. Sci. Technol., 62(15), 1927–1934.
Lykidis, G., and Spiliopoulus, K. (2008). “3D solid finite element analysis
of cyclically loaded RC structures allowing embedded reinforcement
slippage.” J. Struct. Eng., 10.1061/(ASCE)0733-9445(2008)134:
4(629), 629–638.
Mishnaevsky, L., Jr., and Brndsted, P. (2008). “Three-dimensional numeri-
cal modelling of damage initiation in unidirectional fiber-reinforced
composites with ductile matrix.” Mater. Sci. Eng. A, 498(1–2), 81–86.
Moes, N., Gravouil, A., and Belytschko, T. (2002). “Non-planar 3D crack
growth by the extended finite element and level sets. Part I: Mechanical
model.” Int. J. Numer. Methods Eng., 53(11), 2549–2568.
Moës, N., and Belytschko, T. (2002). “Extended finite element method for
cohesive crack growth.” Eng. Fract. Mech., 69(7), 813–833.
Moës, N., Cloirec, M., Cartraud, P., and Remacle, J.-F. (2003). “A computa-
tional approach to handle complex microstructure geometries.” Comput.
Methods. Appl. Mech. Eng., 192(28–30), 3163–3177.
Moës, N., Dolbow, J., and Belytschko, T. (1999). “A finite element method
for crack growth without remeshing.” Int. J. Numer. Methods Eng.,
46(1), 131–150.
Nicholas, T., Boyajian, D., Chen, S., and Zhou, A. (2013). “Finite element
modeling of mode I failure of the single contoured cantilever CFRP-
reinforced concrete beam.” J. Struct., 2013, 1–8.
J. Eng. Mech.
 Oskay, C. (2009). “Two-level multiscale enrichment methodology for mod-
eling of heterogeneous plates.” Int. J. Numer. Methods Eng., 80(9),
1143–1170.
Oskay, C. (2012). “Variational multiscale enrichment for modeling coupled
mechano-diffusion problems.” Int. J. Numer. Methods Eng., 89(6),
686–705.
Oskay, C. (2013). “Variational multiscale enrichment method with mixed
boundary conditions for modeling diffusion and deformation prob-
lems.” Comput. Meth. Appl. Mech. Eng., 264, 178–190.
Oswald, J., Gracie, R., Khare, R., and Belytschko, T. (2009). “An extended
finite element method for dislocations in complex geometries: Thin
films and nanotubes.” Comput. Meth. Appl. Mech. Eng., 198(21–26),
1872–1886.
Pan, J., and Leung, C. K. Y. (2007). “Debonding along the FRP—Concrete
interface under combined pulling/peeling effects.” Eng. Fract. Mech.,
Downloaded from ascelibrary.org by La Trobe University on 07/29/16. Copyright ASCE. For personal use only; all rights reserved.
74(1–2), 132–150.
Pan, N. (1996). “The elastic constants of randomly oriented fiber comp-
sotes: A new approach to prediction.” Sci. Eng. Compos. Mater.,
5(2), 63–72.
Park, K., Pereira, J., Duarte, C. A., and Paulino, G. (2009). “Integration of
singular enrichment functions in the generalized/extended finite
element method for three-dimensional problems.” Int. J. Numer. Meth-
ods Eng., 78(10), 1220–1257.
Pierre, P., Pleau, R., and Pigeon, M. (1999). “Mechanical properties of steel
microfiber reinforced cement pastes and mortars.” J. Mater. Civ. Eng.,
10.1061/(ASCE)0899-1561(1999)11:4(317), 317–324.
Pigeon, M., Azzabi, R. P. P., and Banthia, N. (1996). “Durability of micro-
fiber-reinforced mortars.” Cem. Concr. Res., 26(4), 601–609.
Pike, M., Hickman, M., and Oskay, C. (2015). “Interactions between multi-
ple enrichments in extended finite element analysis of short fiber rein-
forced composites.” Int. J. Mult. Comput. Eng., 13(6), 507–531.
Pike, M., and Oskay, C. (2015a). “Modeling random short nanofiber- and
microfiber-reinforced composites using the extended finite-element
method.” J. Nanomech. Micromech., 10.1061/(ASCE)NM.2153-5477
.0000092, A4014005–11.
Pike, M., and Oskay, C. (2015b). “XFEM modeling of short microfiber
reinforced composites with cohesive interfaces.” Finite Elem. Anal.
Des., 106, 16–31.
Ponte-Castaneda, P., and Willis, J. (1995). “The effect of spatial distribution
on the effective behavior of composite materials and cracked media.”
J. Mech. Phys. Solids, 43(12), 1919–1951.
Radtke, F., Simone, A., and Sluys, L. (2010). “A partition of unity finite
element method for obtaining elastic properties of continua with em-
bedded thin fibres.” Int. J. Numer. Methods Eng., 84(6), 708–732.
Radtke, F., Simone, A., and Sluys, L. (2011). “A partition of unity finite
element method for simulating non-linear debonding and matrix
© ASCE 04016087-12
J. Eng. Mech., 04016087
failure in thin fibre composites.” Int. J. Numer. Methods Eng.,
86(4–5), 453–476.
Reza, F., Yamamuro, J. A., and Batson, G. B. (2004). “Electrical resistance
change in compact tension specimens of carbon fiber cement compo-
sites.” Cem. Concr. Compos., 26(7), 873–881.
Simo, J., Oliver, J., and Armero, F. (1993). “An analysis of strong disconti-
nuities induced by strain-softening in rate-independent inelastic solids.”
Comput. Mech., 12(5), 277–296.
Soghrati, S., and Geubelle, P. H. (2012). “A 3D interface-enriched gener-
alized finite element method for weakly discontinuous problems
with complex internal geometries.” Comput. Meth. Appl. Mech.
Eng., 217–220, 46–57.
Sukumar, N., Chopp, D., and Moran, B. (2003). “Extended finite element
method and fast marching method for three-dimensional fatigue crack
propagation.” Eng. Fract. Mech., 70(1), 29–48.
Sukumar, N., Moes, N., Moran, B., and Belytschko, T. (2000). “Extended
finite element method for three-dimensional crack modelling.” Int. J.
Numer. Methods Eng., 48(11), 1549–1570.
Sukumar, N., and Prevost, J.-H. (2003). “Modeling quasi-static crack
growth with the extended finite element method. Part 1: Computer im-
plementation.” Int. J. Solids Struct., 40(26), 7513–7537.
Tandon, G. P., and Weng, G. J. (1986). “Average stress in the matrix and
effective moduli of randomly oriented composites.” Compos. Sci. Tech-
nol., 27(2), 111–132.
Unger, J., Eckardt, S., and Konke, C. (2007). “Modelling of cohesive crack
growth in concrete structures with the extended finite element method.”
Comput. Meth. Appl. Mech. Eng., 196(41–44), 4087–4100.
Wang, Y., and Waisman, H. (2015). “Progressive delamination analysis of
composite materials using XFEM and a discrete damage zone model.”
Comput. Mech., 55(1), 1–26.
Won, J., Hong, B., Choi, T., Lee, S., and Kang, J. (2012). “Flexural behav-
iour of amorphous micro-steel fibre-reinforced cement composites.”
Compos. Struct., 94(4), 1443–1449.
Yi, C., and Ostertag, C. (2001). “Strengthening and toughening mecha-
nisms in microfiber reinforced cementitious composites.” J. Mater.
Sci., 36(6), 1513–1522.
Zamani, A., Gracie, R., and Eslami, M. (2012). “Cohesive and non-
cohesive fracture by higher-order enrichment of XFEM.” Int. J. Numer.
Methods Eng., 90(4), 452–483.
Zhang, S., and Oskay, C. (2015). “Variational multiscale enrichment
method with mixed boundary conditions for elasto-viscoplastic prob-
lems.” Comput. Mech., 55(4), 771–787.
Zi, G., and Belytschko, T. (2003). “New crack-tip elements for XFEM and
applications to cohesive cracks.” Int. J. Numer. Methods Eng., 57(15),
2221–2240.
J. Eng. Mech.