Engineering Fracture Mechanics 103 (2013) 103–114

Contents lists available at SciVerse ScienceDirect

Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech

Cracking behaviour of fibre-reinforced cementitious

composites: A comparison between a continuous and
a discrete computational approach
Roberto Brighenti ⇑, Andrea Carpinteri, Andrea Spagnoli, Daniela Scorza
Department of Civil-Environmental Engineering & Architecture, University of Parma, Viale Usberti 181/A, 43124 Parma, Italy

a r t i c l e i n f o a b s t r a c t
Keywords: In the present paper, the mechanical behaviour of fibre-reinforced brittle-matrix compos-
Brittle materials ites, with emphasis to cementitious composites, is examined by adopting both a discontin-
Micromechanical model
uous-like FE approach and a lattice model. The main phenomena involved, such as crack
Discontinuous FE
formation and propagation, crack fibre bridging, fibre debonding, fibre breaking, are taken
Lattice model
Fracture into account. The basic assumptions and theoretical background of such approaches are
Fibre-reinforced composite outlined, and some experimental data related to plain and fibre-reinforced concrete spec-
Fibre debonding imens under Mode I and Mode I + II loading are analysed. The comparison of the numerical
simulation results shows that the lattice model allows us a very detailed description of the
fracture pattern, whereas the discontinuous FE approach mainly gives us only global infor-
mation in terms of both crack path and stress–strain response curve. Nevertheless, the FE
approach is computationally convenient and a useful tool for studying problems which do
not require a detailed description of the fracture process.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction

As is well-known, brittle or quasi-brittle materials suffer from several drawbacks, such as low tensile strength and low
fracture and fatigue resistance, and poor wear resistance and durability under repeated loading. Since ancient ages, it has
been observed that the above shortcomings can be reduced by adding fibres to the (matrix) material so as to obtain the de-
sired mechanical properties of the fibre-reinforced composite (FRC) materials which are often economical and competitive in
comparison to more technologically advanced materials. The above advantages have produced an increasing interest in the
computational simulation of such a class of materials when designing composite structures.
Since FRC materials are multiphase in nature, the mechanical characteristics of the different phases as well as their interac-
tions must be taken into account in order to correctly describe their effective behaviour. Phenomena such as matrix cracking,
fibre crack bridging effects on the matrix material, fibre debonding, and fibre breaking have to be modelled. For design pur-
poses, the knowledge of the macroscopic mechanical behaviour of FRC materials generally allows us to assess the in-service
safety level of structural components. In order to examine such materials, various approaches can be used, such as microme-
chanical models (physically-based approach [1–3]), and homogenisation models (mathematically-based approach [4,5]).
Due to low fracture toughness of brittle materials, crack propagation up to failure can easily occur even if the fibre phase
has a beneficial effect in limiting such a phenomenon. From a mathematical point of view, cracking corresponds to a severe
strain localisation phenomenon which is not easily represented by numerical models: computational instabilities, diver-
gence or non-uniqueness of the solution due to the discontinuous displacement field, which develops in narrow highly

⇑ Corresponding author. Tel.: +39 0521 905910; fax: +39 0521 905924.
E-mail address: [email protected] (R. Brighenti).

0013-7944/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfracmech.2012.01.014
104 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114

strained zones, can take place. Several models to solve this class of mechanical problems can be found in the literature: clas-
sical smeared crack approaches (which are affected by a strong mesh-dependence [6]), specific strategies based on the
description of the evolving cracked geometry, such as remeshing or mesh adaptivity [7,8], finite element enrichment ap-
proaches [9–12], interface element approaches [13,14], meshless methods [15,16], and discontinuous formulations [17–
24]. Discrete models, such as the well-known lattice model [25–27], can also be employed to solve such problems.
When the fracture process occurs in a fibre-reinforced material, the description of the phase-interaction behaviour as well
as bridging effects produced by the reinforcing phase must be taken into account. As a matter of fact, even if each component
behaves in a linear elastic manner, the composite material can show a non-linear mechanical behaviour due to the imperfect
bonds between the constituents, leading to fibre debonding or breaking.
In the present paper, two different mechanical models are compared:

(i) a continuous model based both on a fracture energy approach for the brittle matrix [28] that simulates the cohesive
crack behaviour through an appropriate stress field relaxation and on a micromechanical approach to examine the
macroscopic fibre-reinforcing effects (for both random and unidirectional fibre distribution) [29];
(ii) a micromechanical discrete lattice model [27,30] that can be used to simulate heterogeneous materials and multi-
phase composites such as fibre-reinforced ones.

The basic assumptions and theoretical background of such approaches are firstly outlined and discussed and, finally,
experimental data related to both plain and fibre-reinforced cementitious composites under Mode I or Mode I + II monotonic
loading are analysed. The results provided by the two approaches are compared and some conclusions are drawn.

2. Fracture simulation in brittle or quasi-brittle materials

In this Section, the main features of the two theoretical models being compared are reviewed. Details of such models can
be found in Refs. [28] and [27,30] for the continuous model and the discrete lattice model, respectively.

2.1. Continuum approach to fracture: a plasticity-like FE model

A crack process zone in a continuum material can mathematically be represented as a high strain localisation occurring in
a very narrow region. Let us assume the existence of a discontinuity of the displacement field along the line S, contained in a
solid which occupies the region X (Fig. 1).
The discontinuous displacement field can be expressed as follows [24]:

dðxÞ ¼ dðxÞ þ HðxÞ  ½½dðxÞ ¼ dðxÞ þ HðxÞ  wðxÞ ð1Þ

|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
dd ðxÞ

that is, d(x) is the sum of its continuous part, dðxÞ, and the discontinuous one, dd(x) = H(x)  [[d(x)]] = H(x)  w(x), with H(x)
being the Heaviside jump function placed across the crack line, i.e. H(x) = 0 if x e X, H(x) = 1 if x e X+.
The displacement jump vector across the line S, [[d(x)]], coincides with the displacement discontinuity vector w(x), i.e.
[[d(x)]] = w(x), and such a vector can be written as follows: w(x) = u(x) + v(x) = iu(x) + jv(x), where the versors i and j identify
the normal and tangential jump displacement components, respectively (Fig. 2).
In the following the above relationships are considered referred to a FE by using the subscript c for the related quantities.
The mechanical behaviour of a cracked body can conveniently be described by a cohesive-friction law for the cracked
zone and by an elastic or an elastic–plastic law for the uncracked (continuous) region. According to the cohesive crack model
[31], the crack faces are assumed to transmit a non-zero stress whose value rc(uc) can be represented by a decreasing func-
tion of the relative displacement, uc, normal to the crack face. In particular, such a stress is here assumed to be described by a
decreasing exponential continuous law [28]:

Fig. 1. Discontinuous displacement field along the line S in a 2-D solid. Definition of the versors i and j at point C on S.
 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114 105

Fig. 2. Embedded discontinuity of the displacement field in a finite element: (a) uncracked and (b) cracked four-node finite element.

2f t ðu0 uc Þ

rc ðuc Þ ¼ ft  e 2Gf ft u0 ; with rc ðuc Þ u ! 0þ ð2Þ

!1 c

where ft is the maximum tensile strength of the material, u0 is the lower crack opening limit at which the bridging process
starts, and Gf is the fracture energy of the material (energy for unit surface crack).
A friction shear stress described by the following expression [28]
8 h   i
< signðv Þ  ½r ðu Þ  b  1  uc m if 0 < uc < 2  r c and vc – 0
c c c
sc ðuc ; v c Þ ¼ 2rc ð3Þ
:
0 if uc > 2  r c or vc ¼ 0
is assumed to be present between the crack faces, where the crack surface roughness rc is the mean asperity size. A non-zero
shear stress exists only if the surface asperities are in contact (0 < uc < 2  rc), obeying a Coulomb-like law with a decreasing
friction coefficient b (according to the exponent m) by increasing the relative crack displacement uc.
The finite element (FE) formulation of the above problem can employ an appropriate stress field correction in the cracked
ðiÞ
element, in order to get the unbalanced nodal force vector f e;u at the generic iteration step i:
Z
ðiÞ ðiÞ
f e;u ¼ f e;ext  Bt  rrel ðwc ÞdX ð4Þ
Xe

which must be iteratively driven to very small values. In Eq. (4), rrel(wc) is the stress tensor fulfilling some stress relaxation
requirements according to Eqs. (2) and (3).
The FE displacement field d(x) can be written as follows: d(x) = N(x)  d + [H(x)  N+(x)]  wn, where the interpolated dis-
placement field d(x) is given by the sum of the continuous displacement field N(x)  d (d contains the standard nodal degrees
of freedom) and the discontinuous part of the displacements [H(x)  N+(x)]  wn, where wn is nodal counterpart of the dis-
placement discontinuity vector wc.
The stresses evaluated at the Gauss point are properly modified in order to take into account the fracture effects for the
next iteration. This corrected stress field, rrel(wc), can be obtained from the knowledge of rc(uc) and sc(uc, vc) that depend on
uc, vc.
In the simple case of a 2D plane stress problem, the relaxed stress tensor rrel,S in the crack co-ordinate system n, g can be
written as follows (Fig. 2b):
 
rc ðuc Þ sc ðuc Þ
rrel;S ðwc Þ ¼ ð5Þ
sc ðuc Þ rg
where rg is the stress in the material, acting parallel to the crack direction. Such a stress is assumed not to be influenced by
the crack. The corresponding relaxed tensor in the global co-ordinate system is given by:

rrel ðwc Þ ¼ Rt  rrel;S ðwc Þ  R ¼ r  rjump ðwc Þ ð6Þ

where R is the rotation matrix. Therefore, the relaxed stress tensor is equal to the difference between the current stress ten-
sor r and a correction stress tensor, rjump(wc) = r  rrel,S(wc), which represents the stress tensor difference between the pre-
existing stress tensor and that after the appearance of the crack.
Once the current maximum principal stress r1(C) at the centre point C of the finite element (Fig. 2b) reaches the material
tensile strength ft, a crack is assumed to appear lying on a straight line S normal to the above maximum principal stress. We
can evaluate the incremental strain tensor from the above stress tensor:

de ¼ C1  ½dr  drjump ðwc Þ ¼ B  dd  ½C1  drjump ðwc Þ ð7Þ

1 1
where C  drjump(wc) = C  (T  dwc), and T is an appropriate tensor that provides the stress jump as a function of the rel-
ative crack displacement jump vector wc. The displacement discontinuity vector wc = uc + vc must be evaluated through an
iterative process up to the fulfilling of proper convergence requirements.
106 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114

2.2. Discrete approach to fracture: a lattice model

The domain occupied by the material is described in a discretized manner by a two-dimensional triangular lattice, having
hexagonal unit cells with truss elements of length l (Fig. 3). One of the main advantage of the lattice models is to replace the
tensorial quantities (stresses, strains, etc., related to the continuum occupied by the material) with vectorial quantities.
The Young modulus of the truss elements in the lattice model determines the stiffness of the material. The relationship
between the Young modulus of the truss ðEÞ and that of the material (E) can be evaluated by equating the elastic strain en-
ergy of the material occupying an hexagonal unit cell (having unit thickness) with that of the lattice occupying the same
region (Fig. 3) [25,26]:
pffiffiffi
3l
E¼ E ð8Þ
2A
where A is the cross-sectional area of the truss elements. From now onwards we adopt the following notation: a bar above
the symbol means that the quantity is related to truss elements of the lattice model, whereas the plain symbol means that
the quantity is related to the material. The adopted lattice of truss elements, in contrast with that of beam elements, is com-
putationally less expensive (2 degrees of freedom per node instead of 3), but it has the limitation of enforcing a Poisson ratio
of the material equal to 1/3 [25].
In order to describe the stress–strain curve of the truss elements on the basis of that of the material, a transformation rule
for stresses has to be used. We consider here a plane stress field acting in the solid (with 3 components rx, ry, sxy in the x–y
frame of Fig. 3). Assuming that the lattice unit cell is small enough by size to be regarded as embedded in a uniform strain
field, we can write in a compact form (using Einstein summation rule) [30]:

r ðtÞ ¼ EniðtÞ nðtÞ

j eij ð9Þ

where the index t = 1, 2, 3 identifies the truss orientation with respect to the x–y frame (Fig. 3), the index i, j = 1, 2 identifies
ðtÞ ðtÞ
the coordinate axes (1 is for x-axis, 2 for y-axis) so that e11 = ex, e22 = ey, e12 = e21 = cxy/2. Further, ni ; nj arepthe
ffiffiffi direction co-
ð1Þ ð1Þ ð2Þ ð2Þ ð3Þ
sines pofffiffiffi the truss element (t) with respect to the x–y frame ðn1 ¼ 1; n2 ¼ 0; n1 ¼ 1=2; n2 ¼ 3=2; n1 ¼ 1=2;
ð3Þ
n2 ¼ 3=2Þ. The strain components appearing in Eq. (9) are related to the stress components in the material according
to the classical generalised Hooke’s law, namely:

1 dij krkk
eij ¼ rij  ð10Þ
2G 3k þ 2G
pffiffi
where r11 = rx, r22 = ry, r12 = sxy, and the Lame’s constants are expressed as a function of Eðk ¼ G ¼ 4l3A EÞ.
By inserting Eq. (10) into Eq. (9), the relationship between the stress r
 ðtÞ in the truss (t) and the general plane stress tensor
in the material can be written as follows [27,30]:
8 9 2 pffiffiffi pffiffiffi 38 9
<r
>  ð1Þ >
= l 3=2  3=6 0
p ffiffiffi < rx >
> =
6 7
r ð2Þ ¼ 4 0 3=3 1=2 5 ry ð11Þ
>
: ð3Þ > A pffiffiffi > >
r ; 0 3=3 1=2
:
sxy ;
where r  ð1Þ , r
 ð2Þ and r
 ð3Þ are the axial stresses acting in the trusses.
The tensile behaviour of a quasi-brittle material can be described according to the cohesive crack approach. Hence, the
stress–strain curve is the result of the contribution of two constitutive laws: that of the bulk material, here assumed to
be linear with Young modulus in tension equal to that in compression, and the crack bridging law of the cracked material
(Fig. 4). The resulting stress–strain curve is characterised by a perfectly-elastic behaviour in compression; the tensile behav-
iour is elastic up to the first cracking stress, and then a linear postcracking curve with a softening branch follows.

(2)
(3)
(1) l
(1) 3
y
(3)
(2)

l
x

Fig. 3. The unit cell of a regular triangular lattice.

 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114 107

(a) (b) (c)

(d) (e) (f)

Fig. 4. The constitutive curves (not drawn to scale) for the continuum model (a–c) and for the truss elements of the lattice model (d–f): (a) stress–strain
(re) curve of solid concrete (bulk material); (b) stress-crack opening (r–w) curve of plain concrete; (c) r–w crack bridging curve due to fibres; (d) r–e
curve of solid concrete (bulk material); (e) stress-cracking strain ðr ecr Þ curve of plain concrete; (f) r  ecr curve due to fibres.
 

Now let us note that, having Eq. (11) in mindpand ffiffiffi examining a uniaxial stress condition, the stress in the truss of the lat-
tice parallel to the loading axis is equal to r
 ¼ ðl 3=ð2AÞÞr (e.g. see in Eq. (11) the stress in the truss (1) when x is the loading
pffiffiffi
axis). Accordingly, the first cracking stress f t of the truss is equal to ðl 3=ð2AÞÞft , where ft is the first cracking stress of the
material. The strain eel of the truss at the elastic limit (strain at the first cracking stress) is equal to f t =E.
In line with a cohesive crack approach, the area under the stress r against crack opening w curve (characterised by a first
cracking stress ft and an ultimate crack opening wu) is equal to the Mode I fracture energy Gf (hence, for a linear curve r
against w, we have wu = 2Gf/ft). This concept can be translated to the truss elements of the lattice model, if one assumes
to smear the crack opening along the length of the truss. Hence, the ultimate cracking strain eu is given by [27,30]:

G
eu ¼ 2 f ð12Þ
lf t

where Gf is the Mode I fracture energy of the truss, that can be determined from the material counterpart following an en-
ergy conservation argument (that is, the energy dissipation at the surface of the crack
pffiffiffi in the material is lumped at the cross-
sectional area of the truss). Hence, by considering the influence area (equal to l= 3) assigned to a truss submitted to a pure
Mode I loading (e.g. see the truss (1) in Fig. 3, submitted to the uniaxial stress rx), we have [27,30]:

l
Gf ¼ pffiffiffi Gf ð13Þ
A 3

At this point, it is worth noticing that a regular triangular lattice with spatially homogeneous properties produces an overall
isotropic behaviour. However, we should bear in mind that some influence of the regular lattice on the resulting crack pat-
terns has been observed in the literature [25]. In order to reduce the bias of the crack trajectory, the triangular lattice can be
made irregular by randomly perturbing the nodal coordinates, e.g. shifting each node of every triangle by a random amount
within a square box, following a uniform probability distribution. In this way, the resulting lattice is constituted by truss
elements which are slightly skewed with respect to the original directions (1), (2) and (3) of the regular triangular lattice
shown in Fig. 3.
108 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114

3. Mechanics of fibre-reinforced materials

In this Section, the modelling of the fibre reinforcing effects within the framework of the two theoretical approaches
being compared is described. Unless otherwise specified, details of such a modelling can be found in Refs. [29] and
[30,34] for the continuous model and the discrete lattice model, respectively.

3.1. Mesomechanical model in the continuum approach

The heterogeneous (composite) material is supposed to consist of a matrix phase (denoted by the subscript m) and a fibre
phase (denoted by the subscript f) embedded in the matrix (Fig. 5). The assumption that the composite material (having
characteristic size D) has macroscopically the same mechanical characteristics of a small Representative Volume Element
(RVE, having characteristic size d << D) is made.
In the present paper, the model developed in Ref. [29] is extended. The tangent elastic tensor of the macroscopically
homogeneous composite, obtained through an energy balance between the composite and the equivalent macroscopically
homogeneous material, can be written as follows [4,29]:
! Z
drf
C0eq ¼ lm  C0m þ gf   F  FdU ¼ lm  C0m þ gf  C0f ð14Þ
dem
f U

where C0f ; C0eq are the tangent elastic tensor of the matrix, of the fibres and of the equivalent material, respectively; U is the
solid angle of the integral domain; F = k  k is a second-order tensor, where k is the unit vector parallel to the fibre axis,
k ¼ fk1 k2 k3 g ¼ fsin h  cos u sin h  sin u cos hg (Fig. 5). In Eq. (14), em f and rf indicate the strain in the matrix, mea-
sured in the fibre direction, and the actual fibre stress, respectively, whereas lm ¼ V m =V; gf ¼ V f =V respectively represent
the matrix volume fraction and fibre volume fraction in the composite RVE and, in turn, in the material.
Preferential orientation of the fibres in one particular space direction can be taken into account through suitable proba-
bility distribution density functions of the orientation angles, pu(u), ph(h), to be introduced in Eq. (14):
Z Z pZ p
pu ðuÞ  ph ðhÞ  F  Fdu dh ¼ pu ðuÞ  ph ðhÞ  ðF  FÞdu dh ð15Þ
SolidAngle 0 0

where
pa ðaÞ ¼ Aa ðaÞ þ Ba ðaÞ þ C a ðaÞ ð16Þ
with
1 1 al 2 1 1 apl 2 1 1 aþpl 2
Aa ðaÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  e2ð da Þ ; Ba ðaÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  e2ð da Þ ; C a ðaÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  e2ð da Þ ð17Þ
2 2 2
2  p  da 2  p  da 2  p  da

and a = u, h, 0 6 a 6 p.
The above function pa(a), defined in the intervals 0 6 a 6 p, is theoretically formulated by assuming a Gaussian-like
expression, which attains the maximum value at a ¼ u; h, and the minimum value at a ¼ u  p=2; h  p=2. Further, the
Rp Rp
cumulated probability over the function domain is equal to one, i.e. 0 pu ðuÞdu ¼ 1 and 0 ph ðhÞdh ¼ 1.

Fig. 5. Scheme of the RVE (Representative Volume Element) in a fibre-reinforced composite material, and identification of the fibre orientation in the 3D
space.
 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114 109

 or l ¼ 
The parameter l represents the mean value of the probability distribution density functions (i.e. l ¼ u h), whereas
cu and dh are the corresponding variances. The case of randomly oriented fibres can be obtained by setting pu ðuÞ ¼ 1=2p and
ph(h) = sin (h); in such a case, both the macroscopic fibre elastic tensor C0f and the elastic tensor of the composite C0eq are
isotropic.
In order to take into account the fibre–matrix debonding, Eq. (14) can be rewritten as follows [29]:
! Z " # Z
drf dsðem f Þ
C0eq ¼ l  C0m þ gf   F  F dU ¼ l  C0m þ gf  E0f sðem
f Þ þ em
f   F  F dU ð18Þ
dem
f U demf U

where E0f is the tangent Young modulus of the fibres, em f ¼ ðk  kÞ : e (with e the matrix strain tensor) is the matrix strain
measured in the fibre direction. Further, sðem
f Þ is the sliding function that allows us to write the fibre strain ef as a function
of the matrix one emf :

ef ¼ emf  ½½ef m  ¼ sðemf Þ  ½ðk  kÞ : e ð19Þ

|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
em
f

The sliding function quantifies the effectiveness of the stress transfer between the matrix and the fibres [29], and can explic-
itly be written if the shear stress distribution along the fibre is known.
The fibre failure phenomenon can be taken into account by observing that the maximum tensile stress along a fibre is
always reached at its centre [29]; when such a maximum stress, rf(0), reaches the fibre tensile strength, i.e. rf(0) P ft,f,
the fibre is assumed to break in two parts having the same length. It is worth noting that the local perturbation of the stress
field in the neighbourhood of the fibre breakage point (this perturbation induces the appearance of a shear stress acting be-
tween the broken parts of the fibre and the matrix) is not accounted by the present model which is based on a homogeni-
sation procedure of the composite material.

3.2. Fibre crack bridging in the lattice model

The tensile behaviour of a fibre-reinforced composite material can be described according to the cohesive crack approach.
Hence, the stress–strain curve for the cracked matrix described in Section 2.2 is combined with the crack bridging law due to
fibres (Fig. 4). The resulting stress–strain curve is characterised by a perfectly-elastic behaviour in compression; the tensile
behaviour is elastic up to the first cracking stress, and then a linear piecewise postcracking curve with softening branches
follows (Fig. 6).
The crack bridging law due to fibres is based on the results coming from an isolated fibre loaded at its end with a pull-out
force P acting along the fibre axis. The resulting P–d function (d = pull-out displacement) is given by [32]:
p qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
PðdÞ ¼ pffiffiffi ð1 þ qÞEf D3f s0 d ð20Þ
2
where Ef is the Young modulus of the fibre. Further, s0 is the frictional bond stress, Df is the fibre diameter, and
q ¼ ðEf V f Þ=½Eð1  V f Þ (E = Young modulus of the matrix).

Fig. 6. Example of the resulting stress–strain curve (not drawn to scale) for the truss elements of the lattice model in the FRC material. EðiÞ is the secant
Young modulus in the cracking stage at the ith load step.
110 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114

Eq. (20) is valid until the sliding zone reaches the embedded end of the fibre (full debonding of the fibre), which corre-
sponds to a pull-out displacement dP and a force PP given by:
PP ¼ pDf ls0 ð21Þ
where l is the embedded length of fibre. By replacing P(d) with PP and d with dP in Eq. (20), we obtain:
2
dP ¼ ð2l s0 Þ=½ð1 þ qÞEf Df  ð22Þ
After full debonding, the fibre pull-out continues, and the pull-out force decreases as a result of the decreasing embedded
length of the fibre. Assuming a constant frictional bond and ignoring the elastic stretching of the fibre and the fibre–matrix
unit cell at this stage, the following relationship holds [32]:

d  dP
PðdÞ ¼ pDf ls0 1  ð23Þ
dco  dP
where dco is the cut-off displacement (of the loaded end of the fibre) corresponding to a complete pull-out of the fibre of
initial embedded length l, that is: dco = l. Within the framework of the proposed model, the non-linear expression (20) is
approximated by the secant passing through the point of coordinates (dP, PP) and the origin.
For a non-aligned fibre (that is, when there is an inclination angle a between the fibre and the loading axis, coincident
with the normal to the crack Mode I plane), which is the case of randomly distributed fibre composites, various studies have
indicated an increase of the peak load PP with increasing a. As was originally proposed by Morton and Groves [33], we can
assume:

PP ðaÞ ¼ pDf ls0 ef a ð24Þ

(instead of PP given in Eq. (21)), where f is the so-called snubbing coefficient whose value usually ranges from 0.7 to 0.9. For
each single fibre intersecting the crack, the relationship between the crack bridging force P and the crack opening displace-
ment w might be described by Eqs. (20)–(24), where we should consider that the crack opening displacement is equal to
twice the pull-out displacement (w = 2d), and the embedded length l of the fibre is equal to Lf  z (z = distance of the fibre
centroid from the crack plane, measured along the fibre axis). In a continuous approximation (which is reasonable provided
that the fibre mean spacing is much smaller than the crack length), a relationship between the crack bridging stress r, under-
stood as a bridging force per unit crack surface, and the crack opening displacement w can be considered, and the related
peak crack bridging stress rP can be determined. Assuming a uniform distribution for the fibres, corresponding to the follow-
ing probability density functions: PrðaÞ ¼ sin2 a and Pr (z) = 1/(2Lf), we have [32]:
Z p=2 Z Lf
16 1 f a sin a
rP ¼ Vf pDf ðLf  zÞs0 e dz da ð25Þ
p D2f 0 0 2Lf 2

Integration of Eq. (25) yields [34]:

!
f p=2
V f 2Lf s0 1 þ fe
rP ¼ ð26Þ
Df 1 þ f2

If unidirectional fibres are examined, PrðaÞ ¼ dða

^ Þ is the Dirac delta function, and rP depends on the angle a
^ between fibre
direction and normal to crack Mode I plane, that is:
V f 2Lf s0 f a^
rP ¼ e ð27Þ
Df
Now the peak pffiffiffistress due to the fibre crack bridging in the truss of the lattice normal to a putative Mode I crack plane is
equal to r
 0 ¼ ðl 3=ð2AÞÞr0 (e.g. see, in Eq. (11), the stress in the truss (1) when x is the loading axis), where r0 is the peak
stress due to fibres. If random fibres are considered, r0 ¼ rP where rP is given by Eq. (26). In the case of unidirectional fibres,
it might be conjectured that r0 ¼ r2P ð1 þ cos 2a^ Þ, where rP is given by Eq. (27) and the angle a
^ is the angle between the fibre
direction and truss direction in the lattice, being the crack Mode I plane assumed to be normal to the truss direction. The r0
expression for unidirectional fibres is based on the transformation rule of the normal stress component due to a rotation of
the angle a^.
The characteristic cracking strain values of the crack bridging curve due to fibres (Fig. 6) can be determined by smearing
the crack opening along the length of the truss, namely:
w0 wu;f
e0 ¼ ; eu;f ¼ ð28Þ
l l
where w0 = crack opening at the peak stress of the crack bridging law due to fibres; wu,f = ultimate crack opening of the bridg-
ing law due to fibres. Note that wu,f is typically assumed to be equal to a half of the fibre length [32]. Finally, the resulting
stress–strain curve in the truss elements of the lattice model (see Fig. 6) can be obtained once the following values of stress/
strain are computed [30]:
 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114 111

( r 1 (
E
þ e0 if e0 < eu r 0 þ f t eueue0 if e0 < eu
e1 ¼
r 1 r 1 ¼ ð29Þ
E
þ eu if e0 P eu r 0 eeu0 if e0 P eu
( r 2 (
r 0 eeu;fu;f eeu0 if e0 < eu
 
E
þ eu if e0 < eu
e2 ¼
r 2
r 2 ¼ ð30Þ
E
þ e0 if e0 P eu r 0 if e0 P eu
where the total strains in the above expressions are the sum of elastic and cracking strains. If, at a certain load step, the ten-
sile strain e in the truss is higher than eel , an iterative procedure up to convergence is performed using a secant stiffness ap-
proach (EðiÞ is the secant Young modulus after convergence at the ith load step, see Fig. 6).

4. Comparison between numerical and experimental results

The fracture behaviour of plain and fibre-reinforced concrete beams under Mode I or Mode I + II loading conditions is
examined herein by means of the two computational models being compared. The first model, based on a continuous ap-
proach, has been implemented in an in-house code, whilst the second model, based on a discrete approach, has been imple-
mented in the subroutine UMAT of the commercial FE code ABAQUS.
Firstly, an experimental campaign related to prismatic single-edge notched concrete specimens with a three-point bend-
ing scheme is considered [35]. The specimens have a span of 500 mm and a cross section of 150  150 mm; the notch depth
is equal to 25 mm (Fig. 7a) and its width is equal to 4 mm. Plain and steel fibre-reinforced (with random fibre distribution)
concretes are analysed. The mechanical parameters of the beam material are the following: Young modulus E = 32 GPa, ulti-
mate tensile strength ft = 2.6 MPa, fracture energy Gf = 94 N/m. Due to lack of experimental data, the latter two values have
been estimated according to the empirical formulas of the CEB-FIP Model Code 90 [36], which depend on the compressive
strength of concrete and the maximum size of aggregates. Given the inherent approximation of the CEB-FIB formulas and the
remarkable influence of the parameters ft and Gf on the two models here discussed, the ensuing results should be considered
with caution, bearing in mind that a better estimation capacity would be possible if such parameters were directly deter-
mined by means of experimental tests. The relevant parameters for steel fibres are: gf = 0.45%, Ef = 210 GPa, 2Lf = 50 mm,
Df = 1 mm, s0 = 1 MPa. In the present analysis, the aspect ratios of fibres are assumed as such as to exclude fibre tensile fail-
ure (only fibre debonding is herein taken into account). The analysis is performed under displacement control by imposing a
progressive vertical displacement at the central loaded point. A plane stress condition is assumed. In Fig. 7 the discontinuous
FE as well as the adopted lattice model discretization are represented.
Vertical load against crack mouth opening displacement (CMOD) for plain concrete and for steel fibre-reinforced concrete
specimens are illustrated in Fig. 8a and b, respectively; literature numerical and experimental results are also reported [35].
In the case of plain concrete the peak load is well represented by the present FE model while it is overestimated by the lattice
model, on the other hand the softening branch of the P – CMOD curve is well represented by both approaches. The overes-
timated values of the peak load obtained by the lattice model might be attributed to the estimated value of the tensile
strength, which apparently plays a more important role in the lattice model in comparison to that in the discontinuous-like
FE approach. However, further investigations are needed to fully understand this trend.
The fibre reinforced beam presents a peak load slightly higher than that of the unreinforced case, whereas the post peak
curve indicates a ductile behaviour due to the bridging effect of the fibres crossing the main crack developed in the mid span
cross section of the beam.
The (mixed mode) fracture behaviour of a four-point shear loaded single-edge notched beam is then examined. Such a
beam configuration has been used by several authors as a benchmark test for numerical analyses [22,23]. The geometrical

Fig. 7. Three-point bending (plain and fibre-reinforced) concrete beam [35]: FE model (with 344 nodes and 308 four node elements) with geometrical
dimensions expressed in (mm). Overall view (a) of the lattice model (the side parts are discretized with elastic 4-node plane elements) with a detail of the
notch zone (b).
112 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114

20000 22500
Numerical (ref. [35]) Numerical (ref. [35])
Experimental (ref. [35]) 20000 Experimental (ref. [35])
17500
Lattice model Lattice model
Discontinuous FE 17500 Discontinuous FE
15000
Vertical load, P (N)

15000
12500
12500
10000
10000
7500
7500

5000
5000
f=0% f = 0.45%
2500 2500
(a) (b)
0 0
0E+000 1E-004 2E-004 3E-004 4E-004 0E+000 1E-004 2E-004 3E-004 4E-004 5E-004
Crack Mouth Opening displac., CMOD (m) Crack Mouth Opening displac., CMOD (m)

Fig. 8. Vertical load vs crack mouth opening displacement (CMOD) for a three-point bending concrete beam [35] without (a) and with steel fibres (b).
Literature numerical and experimental results [35] and results by the present lattice and FE discontinuous models are plotted.

parameters of the structure (beam thickness equal to 0.1 m and notch width equal to 5 mm) and of both the FE and lattice
discretisations are displayed in Fig. 9. The mechanical parameters of the beam material are the following: Young modulus
E = 35 GPa, ultimate tensile strength ft = 2.8 MPa, fracture energy Gf = 100 N/m (conversely to the previous example, the lat-
ter two values have experimentally been determined). Both plain and fibre-reinforced cases are examined. In particular, two
FRCs are adopted: the first one with PVA fibres (gf = 5%, Ef = 60 GPa, 2Lf = 6 mm, fibres diameter Df = 14 lm, s0 = 3 MPa), and
the second one with steel fibres (gf = 5%, Ef = 200 GPa, 2Lf = 24 mm, Df = 0.5 mm, s0 = 1.5 MPa). Similar to the case of the
three-point bending tests described above, the aspect ratios of fibres are assumed as such as to exclude fibre tensile failure.
The analysis is performed under displacement control by imposing a progressive vertical displacement at the two bottom
loaded points (Fig. 9a), and keeping the ratio between the displacements of the above two points equal to that obtained from
a load-controlled linear-elastic analysis with P and P/10, respectively. A plane stress condition is assumed.
The vertical applied load P against the crack mouth sliding displacement d (CMSD, [22,23]) is graphically represented in
Fig. 10a together with literature results [22,23]. All the results are in satisfactorily agreement. It can be noted that the lattice
model slightly overestimates the initial stiffness of the structure, while the peak load is well evaluated for the plain concrete
beam. Some differences can be appreciated in the softening branch of the numerical FEM curves for the case of plain con-
crete. Moreover, it can be observed that both the presented models well describe the increase of load-carrying capacity
and of the ductility in the material due to the presence of fibres, showing a better behaviour in the case of PVA fibres with
respect to steel ones. Finally, the crack patterns obtained from both the two models are displayed in Fig. 10c–f. For plain con-
crete and the same load level, an extended and diffused fracture zone is provided by the FEM model, while the lattice ap-
proach provides a narrow fracture zone.
In conclusion, by juxtaposing the outcomes of the numerical simulations performed with the two models being com-
pared, a general good agreement can be observed despite the fact that the two models are based on remarkably different
computational approaches. In more details, the crack paths are well described by both models, the lattice model allowing
a more detailed description of the phenomenon. The discontinuous FE model, which is characterised by a smaller number
of degrees of freedom in comparison to that of the lattice model, appears to be capable of well describing the overall

(a) (b)

Fig. 9. Single-edge notched beam under four-point shear: (a) FE discretisation with 436 four-node bilinear elements and 479 nodes. (b) Detail of the lattice
discretization with truss length l = 500 lm.
 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114 113

1.2E+005
Exp. (Alfaiate et al. [22, 23], plain concrete)
FEM (Alfaiate et al. [22, 23], plain concrete) (a)
Lattice (plain concrete)
1.0E+005 FEM (plain concrete)

8.0E+004
Load, P (N)

6.0E+004

4.0E+004
(c) (d)

2.0E+004

0.0E+000
1.2E+005
Lattice (5% PVA fibres)
relative displace
Lattice (5% steel fibres) (b)
FEM (5% PVA fibres)
1.0E+005 FEM (5% steel fibres)

8.0E+004
Load, P (N)

6.0E+004

4.0E+004
(e) (f)
2.0E+004

0.0E+000
0E+000 2E-005 4E-005 6E-005
relative displacement, d (m)

Fig. 10. Vertical bottom applied load against crack mouth sliding displacement, d, for (a) plain and (b) fibre-reinforced materials. Crack patterns obtained
through the lattice ((c) plain and (d) with PVA fibres) and FEM ((e) plain and (f) with PVA fibres) at P ffi 16 KN.

response in terms of load against deflection curves for both plain cementitious composites and fibre-reinforced cementitious
composites.

5. Conclusions

In the present paper, the crack formation and propagation in plain and fibre-reinforced quasi-brittle materials have been
analysed by using a discontinuous-like FE approach and a microstructural lattice model. The main theoretical aspects to de-
scribe the matrix fracture and the fibre bridging effects have been outlined, and experimental results related to plain and
fibre-reinforced concrete specimens under Mode I and Mode I + II fracture conditions have been examined.
The comparison of the numerical simulation results shows that the lattice model allows us a very detailed description of
the fracture pattern, whereas the discontinuous FE approach mainly gives us only global information in terms of both crack
path and stress–strain response curve also for fibre-reinforced brittle materials. Nevertheless, the FE approach is computa-
tionally convenient and a useful tool for studying problems which do not require a detailed description of the fracture
process.

References

[1] Balendran B, Nemat-Nasser S. Bounds on elastic moduli of composites. J Mech Phys Solids 1995;43:1825–53.
[2] Kalamkarov AL, Kolpakov AG. Design problems for the fibre-reinforced composite materials. Composites Part B 1996;27B:485–92.
[3] Ngolle‘ A, Pe´ ra J. Microstructural based modelling of the elastic modulus of fibre reinforced cement composites. Adv Cem Based Mater 1997;6:130–7.
[4] Kalamkarov AL, Liu HQ. A new model for the multiphase fibre-matrix composite materials. Composites Part B 1998;29B:643–53.
[5] Hori M, Nemat-Nasser S. On two micromechanics theories for determining micro-macro relations in heterogeneous solids. Mech Mater
1999;31:667–82.
[6] Tin-Loi F, Tseng P. Efficient computation of multiple solutions in quasibrittle fracture analysis. Comput Methods Appl Mech Engng 2003;192:1377–88.
114 R. Brighenti et al. / Engineering Fracture Mechanics 103 (2013) 103–114

[7] Rashid MM. The arbitrary local mesh refinement method: an alternative to remeshing for crack propagation analysis. Comput Methods Appl Mech
Engng 1998;154:133–50.
[8] Bouchard PO, Bay F, Chastel Y, Tovena I. Crack propagation modelling using an advanced remeshing technique. Comput Methods Appl Mech Engng
2000;189:723–42.
[9] Belytschko T, Fish J, Engelmann BE. A finite element with embedded localization zones. Comput Methods Appl Mech Engng 1988;70:59–89.
[10] Belytschko T, Black T. Elastic crack growth in finite elements with minimal remeshing. Int J Numer Methods Engng 1999;45:601–20.
[11] Möes N, Dolbow J, Belytschko T. A finite element method for crack growth without remeshing. Int J Numer Methods Engng 1999;46:131–50.
[12] Jirásek M, Belytschko T. Computational resolution of strong discontinuities. In: Proceedings of the fifth world congress of computational mechanics.
WCCM V, 2002; Vienna.
[13] Alfaiate J, Pires EB, Martins JAC. A finite element analysis of non-prescribed crack propagation in concrete. Comput Struct 1997;63:17–26.
[14] Xie De, Waas AM. Discrete cohesive zone model for mixed-mode fracture using finite element analysis. Engng Fract Mech 2006;73:1783–96.
[15] Belytschko T, Lu YY, Gu L. Element-free Galerkin methods. Int J Numer Methods Engng 1994;37:229–56.
[16] Belytschko T, Lu YY, Gu L, Tabbara M. Element-free Galerkin methods for static and dynamic fracture. Int J Solids Struct 1995;32:2547–70.
[17] Oliver J. Modelling strong discontinuities in solid mechanics via strain softening constitutive equations. Part 1: fundamentals. Part 2: numerical
simulations. Int J Numer Methods Engng 1996;39:3575–623.
[18] Oliver J, Cervera M, Manzoli O. Strong discontinuities and continuum plasticity models: the strong discontinuity approach. Int J Plast 1999;15:319–51.
[19] Oliver J. On the discrete constitutive models induced by strong discontinuity kinematics and continuum constitutive equations. Int J Solids Struct
2000;37:7207–29.
[20] Wells GN, Sluys LJ. Three-dimensional embedded discontinuity model for brittle fracture. Int J Solids Struct 2001;38:897–913.
[21] Oliver J, Huespe AE, Pulido MDG, Chaves E. From continuum mechanics to fracture mechanics: the strong discontinuity approach. Engng Fract Mech
2002;69:113–36.
[22] Alfaiate J, Wells GN, Sluys LJ. On the use of embedded discontinuity elements with crack path continuity for mode-I and mixed-mode fracture. Engng
Fract Mech 2002;69:661–86.
[23] Alfaiate J, Simone A, Sluys LJ. Non-homogeneous displacement jumps in strong embedded discontinuities. Int J Solids Struct 2003;40:5799–817.
[24] Sancho JM, Planas J, Cendón DA, Reyes E, Gálvez JC. An embedded crack model for finite element analysis of concrete fracture. Engng Fract Mech
2007;74:75–86.
[25] Schlangen E, van Mier JGM. Simple lattice model for numerical-simulation of fracture of concrete. Mater Struct 1992;25:534–42.
[26] Schlangen E, Garboczi EJ. Fracture simulations of concrete using lattice models. Engng Fract Mech 1997;57:319–32.
[27] Guo L-P, Carpinteri A, Roncella R, Spagnoli A, Sun W, Vantadori S. Fatigue damage of high performance concrete through a 2D mesoscopic lattice model.
Comput Mater Sci 2009;44:1098–106.
[28] Carpinteri A, Brighenti R. A new continuum FE approach for fracture mechanics discontinuous problems. Comput Mater Sci 2009;45:367–77.
[29] Brighenti R. A mechanical model for fibre reinforced composite materials with elasto-plastic matrix and interface debonding. Int J Comput Mater Sci
2004;29:475–93.
[30] Spagnoli A. A micromechanical lattice model to describe the fracture behaviour of engineered cementitious composites. Comput Mater Sci
2009;46:7–14.
[31] Hillerborg A, Modéer M, Peterson PE. Analysis of crack formation and crack growth in concrete by means of fracture mechanics and finite elements.
Cem Concr Res 1976;6:773–82.
[32] Li VC. Postcrack scaling relations for fibre reinforced cementitious composites. J Mater Civil Engng ASCE 1992;4:41–57.
[33] Morton J, Groves GW. The effect of metal wires on the fracture of a brittle matrix composite. J Mater Sci 1976;11:617–22.
[34] Carpinteri A, Spagnoli A, Vantadori S. An elastic–plastic crack bridging model for brittle-matrix fibrous composite beams under cyclic loading. Int J
Solids Struct 2006;43:4917–36.
[35] Buratti N, Mazzotti C, Savoia M. Post-cracking behaviour of steel and macro-synthetic fibre-reinforced concretes. Constr Build Mater
2011;25:2713–22.
[36] CEB-FIP Model Code 90; 1993.