Computers and Structures 185 (2017) 27–46

Contents lists available at ScienceDirect

Computers and Structures

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

An effective discrete model for strain hardening cementitious

composites: Model and concept
Tiansheng Shi ⇑, Christopher K.Y. Leung
Depart. of Civil and Environmental Engineering, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong

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

Article history: Strain hardening cementitious composite (SHCC) are materials exhibiting high deformation capacity and
Received 31 October 2016 excellent crack control. In applications where SHCC is employed to enhance durability, information on
Accepted 2 March 2017 the crack width and spacing under loading is important. In conventional finite element analysis, the
material is commonly modeled as a continuum with tri-linear tensile behavior, which cannot capture
the crack pattern. Here an efficient discrete model for SHCC is proposed to address such an issue, with
Keywords: the use of continuum element for matrix damage/cracking, truss element for fiber bridging effect and
Discrete model
interface element for matrix-fiber interaction. Appropriate constitutive laws are assumed for these ele-
Multiple cracking
Strain hardening cementitious composites
ments and the parameters are calibrated from direct tensile test. The validity of the model is shown
(SHCC) by analyzing a tensile specimen and the realistic multiple cracking process of SHCC is captured.
Nonlinear fracture Through a systematic parametric study, the effects of important model parameters on the tensile behav-
Structural application ior of the composites are assessed. The proposed model is further improved to accurately reproduce the
evolution of crack pattern, including average and maximum crack width, crack density and crack width
distribution. Efficient and accurate, the model can be used for analysis of SHCC members under bending,
restrained shrinkage or subject to reflective cracking.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Through proper matrix design, fiber selection and interfacial
treatment, strain hardening cementitious composites (SHCC) with
tensile failure strain up to several percent (often 3–5%) can be
designed. In such materials, strain hardening is accompanied by
the formation of multiple cracks with opening of 50 lm or below.
With high ductility and excellent crack control, SHCC can be
employed to enhance the deformation capacity, energy dissipation,
damage tolerance as well as durability of structural components.
The ductility of the material is important for deformation capacity
and damage tolerance. While for durability enhancement, an
important design parameter is the crack opening on the member
surface, which governs the penetration of water and other chemi-
cals such as chlorides.
When tensile cracking occurs in the SHCC matrix, the stress
released by the cracked matrix is taken up by the fiber. The fiber
bridging law guarantees the multiple cracking behavior of SHCC
if it satisfies both strength and energy criteria. The strength crite-
rion is that the maximum bridging stress should be higher than
⇑ Corresponding author.
E-mail address: tshiaa@connect.ust.hk (T. Shi).
the first cracking strength of matrix. The energy criterion is that
the complementary energy of fiber bridging law is larger than
the fracture energy of matrix, which would lead to steady state
crack propagation with a large part of the crack profile staying at
constant (and small) opening. The details of these two criteria
can be found in Li and Leung [9] and Li [8]. To find the multiple
cracking pattern, earlier work (e.g., [28]) considers the random dis-
tribution of material property, especially the matrix strength, so
cracking occurs subsequently at different locations. However with
this approach, the distribution of the simulated crack spacing is
different from that observed in experiments. As shown by the
pioneering work of Aveston and Kelly [1], the spacing of multiple
cracking is governed by the stress transfer distance from a crack
beyond which the matrix strength is reached again. To find this
distance, the interaction between fiber and matrix needs to be ana-
lyzed. Conventional finite element analyses based on continuum
models do not take such interactions into consideration. Instead,
the composite is modeled with continuum element following var-
ious kinds of tensile constitutive laws. For example, the homoge-
nization based constitutive law ([14,3]) follows a tri-linear
tensile behavior (linear elastic stage, strain hardening stage and
strain softening stage, as shown in Fig. 1(a)) idealized from direct
tensile tests. This approach can predict the load vs. deformation
relation as long as each element covers a few cracks under

http://dx.doi.org/10.1016/j.compstruc.2017.03.002
0045-7949/Ó 2017 Elsevier Ltd. All rights reserved.
28 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46
(a) Homogenization based model
Fig. 1. Constitutive law of conventional model for SHCC.
relatively low stress gradient, so the ‘average’ composite behavior
is properly represented. However, the analysis cannot provide any
information on crack spacing or crack opening. Moreover, when
the stress gradient is high and thus deformation is not homoge-
neous, e.g., when SHCC is employed to resist concentrated stress
or shear, the accuracy of the continuum approach is doubtful.
Kabele [15], Kabele [16] and Yang and Fischer [29] proposed an
individual crack based constitutive law for SHCC, which is as
shown in Fig. 1(b). In this approach, the bridging stress vs. crack
opening relation is first determined from theoretical derivation
or experimental testing of a notched tensile specimen. For a partic-
ular element, the post-cracking stress vs. strain relation can then
be obtained by simply dividing the crack opening by the element
dimension perpendicular to the crack (Note: in this case, as shown
in Fig. 1(b), there is a post-peak stress drop due to matrix cracking
followed by stress increase with further crack opening as loading is
picked up by the fibers). According to Kabele [15], this model can
improve the result for non-proportional loading (e.g., shear in the
case studied). However, with the model of Kabele [15] one crack
will form in each element as long as the tensile strength is reached,
so crack spacing is determined by the mesh size. For correct anal-
ysis of the crack pattern and crack development process, a priori
information on the crack spacing is therefore required to come
up with the proper mesh size. The concepts of Kabele [15], i.e., a
priori determined crack spacing and with bridging force exerted
along crack surface, is similar to smeared model proposed by Bel-
letti et al. [2] for reinforced concrete. As an alternative approach,
Yang and Fischer [29] and Kabele [16] considered heterogeneity
of material property (i.e., variation of strength and crack bridging
behavior among different sections) based on experimental results
and imposed a minimum crack spacing condition so cracks are
sequentially formed at random location under increased loading.
Recently Huang and Zhang [33] implemented a similar technique,
i.e., randomness of matrix and fiber bridging law as well as mini-
mum crack spacing condition, in the framework of eXtended Finite
Element Method or XFEM [25], with cohesive zone model adap-
tively embedded in the model.
There are also some advanced models for SHCC which attempts
to model the random fibers directly and consider the post-cracking
interaction of each fiber with the matrix explicitly for multiple
cracking analysis. These include the lattice model [34,11] and rigid
body spring model [19,17]. While these models have the potential
to capture how the actual stress transfer mechanism between fiber
and matrix affects the multiple cracking behavior, there are many
degrees of freedom in the model which makes the computation
very time-consuming. These approaches are therefore inappropri-
ate for analysis of structural members.
The multiscale framework proposed by Kabele [35] and lattice
model proposed by Spagnoli [23] are two other approaches for
the modeling of SHCC. While they are computationally more effi-
cient than the advanced models described above, the stress trans-
fer between fiber and matrix is not explicitly considered as fiber
bridging stress is applied directly on the crack surface.
(b) Individual based model
Based on the above discussion, there is a need for a new
model, which can not only calculate crack spacing and width
in SHCC, but with sufficient computational efficiency for analysis
of structural members. Thus in this paper, a new discrete model
for SHCC material is proposed to fill this gap. After a crack forms
in the matrix, the crack spacing (and opening) between this
crack and the nearby possible crack is governed by the stress
transfer between fiber and matrix. To consider such stress trans-
fer explicitly in numerical analysis, we model matrix, fiber and
fiber/matrix interface separately with three different types of
elements, including continuum element (which can undergo
cracking) for matrix, truss element for bridging fiber and inter-
face element for stress transfer. In the following, the modeling
concept is first described in detail. In reality, there are many
fibers bridging the crack and interfacial stress distribution is very
complicated. Instead of modeling the details of every individual
fiber (length, orientation and location) and the fiber/matrix
interaction (bond and friction), the current study aims at devel-
oping equivalent truss and interface elements that can represent
the overall effect of these fibers, which thus reduces the compu-
tational effort. Corresponding to each matrix element, only one
truss element and one interface element will be employed. The
model will then be simple and efficient enough for the analysis
of structural members. A major challenge in this work is to
determine the constitutive behavior of the equivalent elements,
which cannot be directly measured. In view of the complex
stress transfer mechanisms, purely theoretical derivation is also
very difficult. A physically based phenomenological approach
will hence be adopted. Based on previous theories, the form of
the constitutive behavior is first obtained. Material parameters
for the constitutive relation are then determined from results
of direct tensile test of SHCC. The proposed discrete model is
validated by analyzing a direct tensile specimen and the result
is compared with that of test. Then a systematic parametric
study is performed to study the sensitivity of SHCC’s perfor-
mance to the material parameters. Based on the finding of the
parametric study, the proposed model is further improved to
enable more accurate prediction of experimental results on crack
pattern and crack width.
2. Physical process of multiple cracking and basic modeling
concept
According to Li and Leung [9], brittle matrix reinforced with
short random fibers could show multiple cracking if properly
designed. In the last two decades, ECC (Engineered Cementitious
Composite) [27] and UHPFRC (Ultra High Performance Fiber Rein-
forced Concrete) materials demonstrating strain hardening and
multiple cracking behavior have been studied by many researchers
and applied in civil engineering projects. To facilitate their wider
application, a proper model for their multiple cracking behavior
is desired.
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46
2.1. Physical process of multiple cracking
To model multiple cracking process of SHCC, the physical mech-
anisms involved in the process as well as their controlling factors
are first reviewed. The cracking process which is schematically
illustrated in Fig. 2, is summarized as follows:

Matrix cracking occurs when its strain (or stress) reaches a cer-
tain level.

Once a crack is formed, stress at the crack surface of matrix is
reduced and the released load is taken up by the fiber.

Stress is transferred from the fiber back to matrix through bond
or friction.

The stress transferred to the matrix increases with distance
from the crack.

Beyond a certain distance from the crack, the matrix stress can
reach the cracking strength again, so it is possible for another
crack to form.
From the above physical process, it can be seen that multiple
cracking process is strongly affected by the matrix cracking behav-
ior, fiber bridging action and fiber-matrix interaction. These factors
are studied in detail below. Since most SHCCs are made with poly-
meric fibers, our discussions will focus on flexible fibers, in which
case the bending stiffness can be neglected.
2.2. Matrix cracking and inclined fiber reinforcement
At a crack in the matrix, a free body diagram showing the dif-
ferent stresses acting on the matrix is given in Fig. 3. After matrix
cracking, cohesive stress due to aggregate bridging or interlocking
acts on the cracked surface. The dotted line in the figure repre-
sents an inclined fiber which interacts with the matrix in two dif-
ferent ways. At the crack surface, bending of fiber introduces a
local force (referred to as the pulley force) that is transferred
immediately to the matrix. Along the fiber, there is a distributed
interfacial stress (i.e., bond and friction stress between fiber and
matrix), which causes the matrix stress to increase with distance
from the crack.
If the bond/friction stress along the fiber is constant, the max-
imum bridging stress is governed by the shorter embedded
length of fiber on the sides of the crack. Once this stress is
reached, the fiber will be pulled out from the short side. How-
ever, for the PVA fiber that is commonly used in SHCC, the inter-
facial friction is found to increase with fiber sliding, so it is
possible for the load acting on the short side to exceed the max-
imum load carrying capacity of the long side. In such a case,
Fig. 2. Schematics for multiple cracking process.
29
Cohesive stress of
matrix
Interfacial
Pulley force
stress of fiber
of fiber
Fig. 3. Force/stress acting on the matrix at/around a crack.
two-way pull-out [31] can occur, and this will have a significant
effect on the bridging stress vs. crack opening relation. Other
mechanisms affecting the fiber bridging force include fiber rup-
ture and matrix spalling.
2.3. Effect of multiple fibers
In SHCC material, there are many fibers with different embed-
ded length and oriented at different angles. To develop a simplified
model, these randomly distributed fibers are represented by one
equivalent fiber (Fig. 4), and their integrated crack bridging effects
as well as interaction with the matrix are described by the consti-
tutive laws of the equivalent fiber and the interface between the
equivalent fiber and matrix, respectively.
2.4. Interaction between nearby cracks
The interaction between nearby cracks may occur in two differ-
ent ways, due to the bridging of several cracks by a single fiber, or
the presence of many fibers between two adjacent cracks.
When fiber length is larger than the crack spacing, there is a
high probability for multiple cracks to form along one fiber
(Fig. 5). The interaction of such cracks has been analyzed in Přinosil
and Kabele [21].
Also, between two adjacent cracks, there are many fibers and a
part of the stress they transfer to the matrix cancel each other so
formation of a new crack between these two cracks is only
resulted from the net stress transferred to the matrix as shown
in Fig. 6.
Fig. 4. Representation of many fibers by one equivalent fiber (or truss).
Fig. 5. Stress distribution along one fiber bridging several cracks.
30 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

= +

Fig. 6. Cancellation of stresses exerted on the matrix by two fibers pulled at adjacent cracks.

Truss element for fiber

Continuum element for
matrix

Interface element

Fig. 7. Schematic illustration of the proposed new discrete model.

2.5. A discrete model for SHCC fiber bridging law proposed by Li and Leung [9] for flexible fibers
with purely frictional interface, except that the pulley force acting
Considering the physical processes described above, a new on the crack surface is not considered. The governing equations are
finite element model is proposed to properly reflect these three given below.
important factors. The proposed FE model consists of three differ- "
0:5 #
ent types of elements, i.e., a continuum element, a truss element d d Vm
and a bond/frictional interface element (Fig. 7). Formulated with
rtr ¼ r0 2  ð1Þ
dm dm V f
smeared crack model, the continuum element can describe the
cracking behavior of matrix, including formation of small dis- (
tributed micro cracks and localized cracks. After cracking, the ele- etr Spf etr < dm =Spf
d¼ ð2Þ
ment carries a cohesive stress due to aggregates bridging and dm þ ðetr  dm =Spf Þle etr > dm =Spf
interlocking action. In addition, as the pulley force from inclined
fibers also acts directly on the crack surface of matrix, it should
also be included in the post-cracking constitutive law of the con-
sL2f
dm ¼ ð3Þ
tinuum element. The constitutive behavior of the truss element ð1 þ gÞEf df
will then represent the fiber bridging force subtracting the pulley
force. The interface element is employed to model the net stress V f Ef
transferred between fiber and matrix, which controls crack spac-
g¼ ð4Þ
V m Em
ing. As the continuum element, truss element and interface ele-
ment are already available in most general purpose finite rtr is the stress in the truss element. r0 is a normalization factor
element softwares (e.g., ABAQUS, ANSYS, ADINA or DIANA) as well defined as Vf s(Lf/df)/2 in Li and Leung [9]. In this work, however,
as special purpose finite element softwares (e.g., ATENA), the pro- we take it as an empirical parameter to be obtained by fitting of
posed discrete element model could thus be easily implemented in experimental data (ru). d is the crack opening displacement. etr is
common softwares for the analysis of SHCC members. The details strain of the truss (i.e., equivalent fiber); Vf, Vm are volumes of fiber
of the constitutive law for each element are discussed in the fol- and matrix; Ef, Em are modulus of fiber and matrix; df, Lf are diam-
lowing section. eter and length of fiber. s is the bond strength between individual
fiber and matrix. Of special attention is the parameter Spf, which
is a fictitious crack spacing considering the non-uniformity of strain
3. Material constitutive behavior of various elements in the distribution along the truss and it is calibrated from the measured
discrete model for SHCC value of eu, as discussed later. It should also be noted that when
etr > dm =Spf , deformation of the equivalent fiber will localize in
The material law for each element of the proposed discrete ele- one truss element, so the expression of d is different from that when
ment is proposed as follows: etr < dm =Spf . In Li and Leung [9], Eq. (1) is only valid when
etr < dm =Spf , but here it is considered applicable to all situation
3.1. Truss element for simplicity. This assumption does not affect the multiple cracking
behavior of SHCC. For the truss element, the area of its cross
The constitutive law of truss element which represents the section (Atr) is obtained by Vf/Vm times the area of matrix (Am),
bridging behavior of all the fibers is assumed to follow a simple i.e., Atr/Am = Vf/Vm.
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46 31

Micro damage stage

Matrix cohesive + pulley forceof fiber
ft ft ft

Em
unload
unload

md

(a) (b) (c)

Fig. 8. The tensile behavior for continuum element: (a) stress-strain relationship before matrix crack; (b) stress-crack opening width relationship after matrix cracking
(considering pulley force); (c) micro damage stage of matrix.

where g ¼ 4þf pf =2 Þ and f is the snubbing coefficient of the

2 ð1 þ e
3.2. Continuum element 2

fiber in mortar; em is the matrix strain corresponding to maximum

Before cracking, the properties (elastic modulus, Poisson’s ratio, pulley force. Le is the length of the finite element perpendicular to
and compressive constitutive law) of continuum element are the crack direction.
assumed to be the same as that of the composite, which can be Under compression, the behavior of continuum element is
obtained directly from tensile and compressive tests. After crack- taken as that of the composite, with the contribution of fiber to
ing, the tensile stress carried by the continuum element is com- compressive behavior considered. As multiple cracking under ten-
posed of two parts: (i) cohesive stress of the matrix, and (ii) sion is the major concern in this paper, compressive behavior will
pulley force of the fiber. Modeling of these two components is not be discussed in detail.
described separately below. For the cementitious matrix, the
simplest model commonly used is linear elastic before cracking
3.3. Interface element
and linear softening after cracking. The pulley force provided by
fiber upon matrix cracking surface is calculated through microme-
The interface element is employed to model the distributed
chanical approach, by integrating the pulley force of every fiber
stress transferred between fiber and matrix along the length of
crossing the crack. The full constitutive law of the continuum ele-
fiber, with appropriate constitutive law to describe the relation-
ment under tension is illustrated in Fig. 8(a) and (b).
ship between bond stress and interfacial slip, as shown in Fig. 9.
According to Van Mier [26], there are 4 stages for the fracture of
In the simplest case, a constant bond stress-slip law is first
concrete: elastic stage, micro-cracking stage, onset of macro-
assumed for the interface. In the later part of this paper, linear
cracking stage and bridging stage. A similar concept is applied to
hardening law or piecewise linear hardening law will also be
the matrix phase (i.e., mortar) of SHCC in this work. After the elas-
employed. It should be noted that softening law can also be
tic stage, a ‘‘micro-damage” stage is assumed to account for the for-
applied to the interface element but this will not be covered
mation of very small distributed micro-cracks in the mortar that
in this paper. To facilitate computation, a short elastic stage
may even be invisible. Due to the ability of micro-fibers in SHCC
(with steep slope) is added.
to arrest micro-cracking, the ‘‘micro-damage” stage of mortar
In Yang and Fischer [29], Spagnoli et al. [24], Spagnoli [23]
may be quite extended. During this stage, the stress in mortar is
and Kabele [16], multiple cracking of SHCC was simulated by
assumed to stay constant with the increase of strain and unload
considering the random distribution of material properties. As
towards the origin. The behavior of mortar during micro-damage
a first study on the new discrete model of SHCC, we would
stage (emd) is shown in Fig. 8(c).
focus on the formation of multiple cracks due to stress trans-
Following the same functional form for the bridging stress of
fer, which will occur even with uniform material properties.
truss elements, the pulley stress acted on the cracked element is
Additional effects of material property variation will be consid-
given by:
ered in future work. Despite the use of uniform material prop-
"
0:5 #
e e erties in the computation, multiple cracking will not occur
rp ¼ ðg  1Þr0 2  ð5Þ simultaneously in a direct tensile specimen according to our
em em
calculation, as very small numerical errors (such as those due
to truncation or convergence criterion of Newton-Raphson
dm
em ¼ ð6Þ method to solve the nonlinear problem) always exist and the
Le resulting numerical randomness will also lead to a sequential
formation of multiple cracks, as shown in the results presented
in later sections.
hardening
4. Determination of model parameters from the direct tensile
constant bond stress
n test

softening
4.1. The process for determining model parameters

Direct tensile test is commonly used to characterize the

mechanical behavior of SHCC. A typical tensile curve of SHCC is
shown in Fig. 10. In this figure, three points, including the turning
sy s point (etp, rtp) indicating deviation from linear elastic behavior,
first cracking point (efr, rfr) where there is a drop of load indicating
Fig. 9. Constitutive law for interface element. formation of macro crack and ultimate strength point (eu, ru), can
32 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46
First cracking
Ultimate
( fr fr )
strength ( u u)
Turning point
( tp tp )
Fig. 10. Stress-strain relationship from direct tensile test of SHCC.
be identified. The crack density at various load levels can also be
counted. Specifically, the average crack spacing at ultimate load
is denoted as Sp. In this section, the determination of material
parameters for the proposed discrete model from these test results
will be discussed.
4.2. Continuum element
For the continuum element, its elastic behavior follows the
experimental tensile curve before the turning point, and its
strength is taken to be the tensile strength of matrix, which can
be obtained from the turning point (i.e., ft = rtp). To analyze post
cracking behavior of continuum element, fracture energy of matrix
can be assumed to be around 5 J/m2, a value commonly used for
the mortar in SHCC [30]; the pulley force of fiber is obtained
together with property of the truss element based on the measured
strength of composite ru, as discussed in the following sub-section.
The snubbing coefficient f of pulley force depends on the mortar
and fiber used. For PVA fiber and mortar, the snubbing coefficient
is 0.5 [10].
4.3. Truss element
The fiber properties (diameter, length and modulus) are deter-
mined from measurements on the fiber or data provided by man-
ufacturer. Assuming the matrix has softened to zero stress at
ultimate load, the strength of tensile specimen is contributed by
the fiber bridging force carried by the truss element and pulley
force at the crack surface, i.e.:
ru Am ¼ rtru Atr þ rpu Am ¼ g r0 Am
ru
r0 ¼
g
where rtru and rpu are strength of truss element and ‘‘pulley force”
acting on crack surface.
The fictitious crack spacing Spf is obtained as follows. When the
ultimate strength of composite is reached, the crack pattern is
assumed to be saturated in the specimen. Due to the assumed uni-
form interfacial bond/friction stress along truss, the stress of the
trusses decreases linearly from the crack surface, while the stress
in the matrix increases linearly. The maximum crack spacing is
that where the stress in matrix at the middle is equal to the
strength of matrix (ft). The minimum crack spacing is equal to half
of the maximum crack spacing. The crack opening can then be
obtained by integrating the strain of truss between two cracks
(at a spacing of Sp). By equating the calculated ultimate strain
(which is obtained by integrating the strain along the truss, or by
dividing crack opening by crack spacing) to the measured value
(eu), an expression for Spf is derived in Appendix A as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi


ffi !
d 3k g rtp k g rtp
Spf ¼  ðg  1Þ  8  ðg  1Þ þ 6
6eu 2 ru 2 ru
where k is a coefficient considering the randomly distribution of
crack spacing, which is taken to be 1.337 following Kimber and Keer
[18].
4.4. Interface element
To have a steep slope for the bond-slip law before slipping
occurs at s = sy, a relatively small value of 1  106 m is assigned
to sy. Using the same assumption as above for the calculation of
Spf, the interfacial bond stress (sn) is calculated from the
experimentally measured averaged crack spacing at ultimate load
(sp) by:
kV m df ðgrtp  ðg  1Þru Þ
sn ¼ ð10Þ
4gV f Spav g
Note that the underlying assumption for Eq. (10) is that the
state of ‘‘saturated” crack pattern is reached in the specimen,
i.e., all possible cracks are formed, or the largest crack spacing
is no larger than 2 times of the minimum possible crack
spacing.
As a special case, if the snubbing effect of fiber is assumed to be
0, i.e., f = 0 (or g = 1), the pulley force is not considered in constitu-
tive law of matrix, and the simplified constitutive laws for truss
and interface element are:
r0 ¼ ru ð11Þ
0 sffiffiffiffiffiffiffiffiffi 1
d @3krtp krtp
Spf ¼ 8 þ 6A ð12Þ
6eu 2ru 2ru
kV m rtp df
sn ¼ ð13Þ
4V f Spav g
5. Experimental validation
A simple example is given to illustrate the calculation process
and also to verify that the proposed model can reproduce the ten-
sile behavior as well as multiple cracking behavior of SHCC
properly.
ð7Þ
5.1. General model information
ð8Þ
The tensile test ‘‘FINE” in Magalhães et al. [12] is chosen for
analysis, as the full set of data on multiple cracking, including
average crack spacing, both maximum and average crack width
evolution as well as the evolution of crack width distribution
are reported in the paper or can be found in Zijl et al. [36].
The test is labeled ‘‘FINE” in the paper because fine sand is
used to make the SHCC tensile specimen. Moreover compres-
sive, tensile and flexural tests on SHCC have all been conducted
and reported in the paper. For one of their tensile tests with
very detailed data (FINE), the key points of the tensile curve
are captured from their picture for comparison with the calcu-
lated results. The PVA fiber content is Vf/Vm = 2%. For other
related data (e.g. mix design of SHCC), readers can refer to
the original paper.
From the test, ru = 3.384 MPa, rtp = 2 MPa and eu = 3.6%. At
ultimate load, the number of cracks per meter is reported to
be around 300, so the average crack spacing Sp is
1000/300 = 3.33 mm. The important micro-mechanical parameter
s (real bond strength between PVA fiber and mortar) is not given
in Magalhães et al. [12], and here it is chosen to be 2 MPa fol-
ð9Þ lowing Zhang et al. [32]. The elastic modulus of material E is
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46 33

20 GPa according to the compressive stress vs strain curve of
SHCC. Because of PVA fiber used, the snubbing factor f is 0.5,
and thus g is 1.5.
According to the process described in section Part 3, for FE
model considering pulley force, the set of parameters are deter-
mined as: ft = 2 MPa; Gf = 5 J/m2 (assumed); rtru = 112.8 MPa;
dm = 0.173 mm; Spf = 2.170 mm; sn = 0.175 MPa; rpu = 1.128 MPa.
While for model not considering pulley force, those parameters
are: ft = 2 MPa; Gf = 5 J/m2 (assumed); rtru = 169.2 MPa; dm =
0.173 mm; Spf = 1.727 mm; sn = 0.4 MPa.
5.2. FE model and result
With these parameters, the proposed finite element model is
implemented in the Finite Element software ATENA (Version
5.0.3d) to simulate the direct tensile test. The geometry of the
specimen is 50  0.5  100 mm (length  width  thickness).
The small value of specimen width is chosen to save computa-
tional time as preliminary analysis indicates that specimen
width has little effect on the results. It is a plane stress prob-
lem, for which the thickness of specimen does not have any
effect on the result. The mesh size for continuum element is
0.5 mm  0.5 mm, and one truss element (with associated inter-
Fig. 11. Finite element model for analysis of SHCC.
face elements) is embedded in each continuum element along
the loading direction. The left side of the tensile specimen is
fixed in x-direction, while the right side is loaded by displace-
ment control and the total displacement is 2.0 mm (correspond-
ing to a total strain of 4.0%, this value may vary from case to
case). The displacement and reaction force of nodes at right
side are monitored during loading. The detailed finite element
model in ATENA [4] is shown in Fig. 11.
From Fig. 12, it is found that the stress-strain curves predicted
by the FE model considering pulley force (‘‘FE pulley” in Fig. 12
(a)) and not considering pulley force (‘‘FE nopulley” in Fig. 12(a))
both agree well with the test (test ‘‘FINE” in [12], though the lines
are lower than the test results when the strain is 0.5–1.5%. The
computed average crack spacing of 3.33 mm for both models is
exactly the same as the experimental value of average crack spac-
ing. The fluctuation of stress-strain curve indicates sequential for-
mation of crack, which is consistent with experimental
observation. However, according to the calculation, most of the
cracks are formed before the strain reaches 2.0%, while cracks were
formed over the whole loading stage in the experiments. Generally,
the result predicted by model considering pulley force is almost
the same as that by model not considering pulley force. However,
from the crack pattern at ultimate stage in Fig. 12(b) and (c), it is
found that the maximum crack width (136.7 lm) at ultimate stage
predicted by model considering pulley force is significantly smaller
than that (225.1 lm) predicted by model not considering pulley
force.
Using model considering pulley force, the evolution of crack
pattern and crack width of tensile specimen is shown in Fig. 13
(b). To draw these figures, only cracks with width over 2 lm
are included. ‘‘Cracks” whose width is smaller than 2 lm do
exist and occur in almost every element, but they are

(a) stress-strain curve

(b) crack pattern at ultimate stage strain = 3.6% (pulley, average crack spacing = 3.33 mm)

(c) crack pattern at ultimate stage strain = 4.0% (no pulley, average crack spacing = 3.33mm)

Fig. 12. Result calculated by proposed discrete model (Note: the values on crack pattern figures are crack width in unit of meter, same in the following; ‘‘test” refers to the
experiment of ‘‘FINE” in [12], same in the following).
34 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

(a) Diffused “micro crack” (damage) at strain 0.4%

Strain = 0.4%

Strain = 0.8%

Strain = 1.2%

Strain = 1.6%

Strain = 2.0%

Strain = 2.8%

Strain = 3.6%

(b) Calculated crack pattern at different loading stage

Fig. 13. Evolution of crack pattern (by FE model considering pulley).

considered as tiny and diffused ‘‘damage” as shown in Fig. 13
(a). In the calculation results, there is no crack with width in
the range of 2–10 lm, indicating that the cracks are formed
abruptly, with their width ‘‘jumping” to a value larger than
10 lm. This sudden formation of crack is also observed in the
test. The crack pattern evolution shows that the new cracks
form sequentially at different strains, as also indicated in the
stress-strain curve. When the strain reaches 2.0%, the number
of cracks in the specimen does not change, which is also consis-
tent with previous finding in the calculated stress-strain curve
(see Fig. 12(a)).
The calculated average crack width (Fig. 14) is increasing
during the whole loading stage, which is different from the test
result of Magalhães et al. [12]. However, in other literatures
(e.g., [5,20]), the tensile test of ECC shows a trend of increasing
average crack width similar to that obtained by the proposed
model.
For FE model considering pulley force, additional results of
interest are shown in Fig. 15. The maximum crack width is found
to increase with strain, which is similar to the trend of the average
crack width, except that when the specimen is approaching failure,
the maximum crack width increases more rapidly. As the opening
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46
Fig. 14. Evolution of averaged crack width.
of cracks can significantly increase the composite strain, the
contribution of crack opening to the total strain of the composite
is studied here. The term ‘‘nominal strain” is defined as average
crack width/average crack spacing (or sum of the width of all
cracks/length of specimen), and the ratio of nominal strain to mea-
sured strain is plotted in Fig. 15(d). From the figure, it is found the
nominal strain ratio is 0.49 at strain 0.2%, with the remaining part
(0.51) of total deformation coming from elastic deformation and
diffused damage; the ratio is increasing quickly when the strain
increases from 0.2% to 1.2%; At the strain of 1.2%, it increases to
0.98 (very close to 1.0) and remains stable, which means the SHCC
deformation is mostly contributed by crack opening. The
calculated ratio may be slightly higher than 1.0 due to the
computation error of ATENA.
(a)Maximum crack width evolution
(b) Minimum crack width evolution
Fig. 15. Result calculated by FE model considering pulley force.
35
5.3. Parametric study
A Parametric study is performed to identify the effects of the
important parameters on the behavior of SHCC, which would pro-
vide guidelines for further improving the model to obtain better
agreement with experimental results. From Eqs. (9)–(12), ru/ft or
ru/rtp is found to determine the ultimate strain of the tensile spec-
imen, while sn affects crack spacing and crack width. Thus the
mortar strength ft (or rtp) and interfacial strength sn are selected
as parameters to be studied. The length of micro damage stage of
mortar is discussed here as well, considering that the micro dam-
age stage of mortar may be elongated by the arresting of micro-
cracks by the fibers. In the parametric study, the model considering
pulley force presented in the above section is used as the standard
case, the constitutive law and material parameters of other cases
are the same as that of the standard case, unless stated otherwise.
The mesh convergence behavior and unloading property of the
proposed model are important issues. While they are not the focus
of the present work, a preliminary study has been conducted and
the results are summarized in Appendices B and C respectively.
According to the simulations in Appendix B with the same material
parameters but different mesh sizes, similar results can be
obtained for mesh size below 0.8 mm. However, as crack spacing
in SHCC members can be as small as 2 mm, a mesh size of
0.5 mm or below is recommended. For the study of unloading
behavior in Appendix C, the unloading moduli of truss and inter-
face elements are taken to be the same as the initial elastic moduli
of the corresponding elements (indicating plastic or sliding behav-
ior without damage), while the continuum elements can undergo
damage through smear cracking The composite behavior of the
(c) Nominal strain/strain ratio
(d) Evolution of Crack density
36 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

SHCC then shows an unloading modulus below the initial modulus
of the composite, which is similar to the results from the model
proposed by Gencturk and Elnashai [6]. For accurate simulation
of cyclic behavior of SHCC, the unloading behavior should be fur-
ther refined. More in-depth experimental and theoretical studies
will then be required to derive accurate degradation models for
the continuum, truss and interface elements.
Fig. 16 shows the effect of matrix tensile strength on the stress
vs. strain behavior. With a higher tensile strength of 2.5 MPa, the
calculated result (ft = 2.5) agrees well with test data when strain
is 0.5–1.2%. This implies that the variation of material properties
such as mortar strength at different locations of a real specimen
may have an effect on its cracking behavior during the multiple
cracking stage. It is also found that higher tensile strength of
matrix leads to low ductility and also larger crack spacing at ulti-
mate stage, which agrees with experimental observations as well
as theoretical concept.
Fig. 17 shows the effect of interfacial strength. Within the range
of interfacial strength studied i.e., sn = 0.14, 0.175 and 0.2 MPa, the
stress-strain curves are almost the same. Also for model with
higher interfacial strength, more cracks form when the strain is
large. The ultimate crack spacing and crack width are then also
smaller. It is interesting to note that the ultimate strain is almost
the same for these models with different interfacial strength. This
trend can be also seen from the previous Eqs. (9) and (10).
Fig. 18 shows how the micro damage stage affects the results.
emd is the strain increase due to micro damage of mortar, which
is assumed to take place at constant stress ft. Simulation was car-
ried out with different values of emd (0, 1e3, 2e3 and 3e3),
and the larger value of emd results in lower ductility as shown in
the stress-strain curve, though such effect is smaller than that of
ft. Larger value of emd also leads to higher first cracking strength
as well as higher stress at early multiple cracking stage. From the
crack pattern at ultimate stage shown in Fig. 18(b), it is found that

(a) Stress-strain curve

Crack pattern at ultimate stage strain = 3.6% (f t = 2.0 MPa)

Crack pattern at ultimate stage strain = 2.6% (f t = 2.5 MPa)

(b) Crack pattern at ultimate stage

(c) Average crack width evolution

Fig. 16. Parametric study of mortar strength (with pulley, ft = 2.0 and 2.5 MPa).
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46 37

(a) Stress-strain curve

( n = 0.14 MPa, strain = 3.6%)

( n = 0.175 MPa, strain = 3.6%)

( n = 0.2 MPa, strain = 3.6%)

(b) Crack pattern at ultimate stage

(c) Average crack width evolution

Fig. 17. Parametric study of interfacial bond stress (with pulley, sn = 0.14, 0.175 and 0.2 MPa).

larger value of emd does not necessary lead to smaller maximum
crack width, though few cracks may form very close to each other
(even smaller than minimum crack spacing).
5.4. Further study on crack width
Crack width is an important design parameter in the practical
application of SHCC as it governs transport properties and durabil-
ity. However, the parametric study above shows that the proposed
model with simple material constitutive law cannot capture the
trend of experimental data (from [12]) on crack width vs. strain.
Specifically, even with the change of various parameters, the crack
width is found to increase monotonically with strain, but the test
results indicate the reaching of a plateau at relative low strain.
In Zijl et al. [36], tensile test results of SHCC specimens made
with PVA fibers by different groups around the world are reported.
The various test results reveal a general trend of average crack
width evolution as shown in Fig. 19. In the figure, there are three
stages of crack evolution: initial stage, steady stage and increasing
stage. In the real application of SHCC, we are most interested in the
crack width during the serviceability state, when the strain is
lower than 1–2%. Beyond this range, the ductility of SHCC is more
important for control of ultimate failure and resilience of the
structure.
38 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

(a) Stress-strain curve

md = 0 (strain = 3.6%, max crack

md = 1e-

md = 2e-

md = 3e-

(b) Crack pattern at ultimate stage for different value of md

(c) Average crack width evolution

Fig. 18. Parametric study of different emd (micro damage stage = 0, 1e3, 2e3 and 3e3).

Close examination of the data summarized in Zijl et al. [36]
as well as those reported by Lepech and Li [7] and Magalhães
et al. [12] reveals that the ratio of nominal strain (= average
crack width/average crack spacing) to measured strain is usu-
ally not equal to unity as shown in Fig. 20. This ratio in one
particular test is almost a constant value within the range of
strain from 0.5% to 4%. For different tests, the value of this
ratio is different. This may be caused by insufficient accuracy
of crack measurement method adopted (e.g., DIC or DIP, with
reported resolution of 10 lm) as well as complicated nature
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46 39

Fig. 19. Evolution of average crack width.

Leipzig fs1
Leipzig fs2
2 Qingdao formed
Qingdao finished
Rio fine
SU fs1
Nominal strain /strain Ratio

1.5 SU fs2
Leipzig cs
Rio cs
Lepech&Li09
1 Magalhães10 FINE

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Strain (%)

Fig. 20. Nominal strain/strain ratio for different tensile test of SHCC (most data from Zijl et al. [36] with the legends following the original paper; additional data from [7],
labeled as ‘‘Lepech&Li09” and [12], labeled as ‘‘Magalhães10 FINE”).

of the micro-cracking process (crack width around 10–50 lm, assumed to be 2e3 and ft is 2 Mpa. It is found that the
rough crack surface through thickness of specimen as well as piecewise linear hardening bond-slip law shown in Fig. 21
skewness of crack) in the SHCC specimen. While this issue can produce a crack width vs. strain curve comparable to test
requires further detailed study, the reported crack width is results for strain below 2%, as shown in Fig. 22(b). Also the
used in this paper as reference for comparison with the crack stress-strain curve as shown in Fig. 22(a) predicted by the
width evolution generated by the proposed discrete model.
Note because of the difference in nominal strain/strain ratio,
the crack density found in experiment and calculated by FE 1.2
will be different.
In the following, the constitutive laws and parameters are care- 1
fully calibrated to generate similar trend of crack width evolution 0.8
(MPa)

to that of the experiments, especially at strain below 2%. The refer-

ence crack evolution curves are from the tests labeled as ‘‘FINE” in 0.6
Magalhães et al. [12] and ‘‘M05” in Magalhães et al. [13], which are
n

0.4
denoted as ‘‘FINE_Mag_2010” and ‘‘M05_Mag_2014” respectively
here. These two specimens are of a very similar composition and 0.2
also show a similar tensile behavior.
0
From the parametric study presented above, the crack width 0 0.05 0.1 0.15 0.2
is found to be closely related to the bond-slip law of the inter- Slip (mm)
face element. As further study, different types of bond-slip law
are tried. For the other important parameters, the emd is Fig. 21. Hardening bond-slip law for the interface element.
40 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

(a) Stress-strain curve

(b) Average crack width-strain curve

(c) Crack pattern at ultimate stage (strain = 3.4%)

Fig. 22. Result of model considering hardening bond-slip law (‘‘FINE_Mag_2010” refers to test ‘‘FINE” in [12]; ‘‘M05_Mag_2014” refers to ‘‘M05” in [13]).

proposed FE model is very close to that of the test, except that
the load is very unstable during the ultimate stage, leading to
premature failure at a lower strain, i.e., 3.4% as shown FE
curve in Fig. 22(b).
To further improve the agreement between predicted and mea-
sured crack width vs. strain relation, the fiber bridging law is mod-
ified as shown in Fig. 23(a) where the original bridging law ‘‘fiber
bridging 1” (considering frictional bond-slip) is replaced by ‘‘fiber
bridging 2” to account for the pullout behavior of PVA fiber with
chemical bond and hardening bond-slip relationship as shown in
Fig. 23(b) [22]. The difference in simulated bridging law (in terms
of normalized stress vs. normalized crack opening) between the
pure friction case and the case considering chemical bond and
slip-hardening is shown in Fig. 23(c). From the figure, the differ-
ence increases with crack opening initially, but would then
decrease until there is very little difference near the peak. The
modification of fiber bridging law in Fig. 23(a) can be considered
a simple way to account for this effect.
The simulated results with the improved model are shown in
Fig. 24. The experimental stress-strain curve and almost all per-
spectives of the measured crack pattern can be well reproduced.
Specifically, the average crack width vs. strain curve up to 2% strain
is very close to the corresponding experimental curve. The maxi-
mum crack width at a certain strain, which is an important param-
eter for durability design, is also very accurately predicted.
Moreover the crack width distributions at different strain levels
are also very similar to that of experiment reported in Fig. 6 (L3-
fine) of Zijl et al. [36].
However, the ratio of nominal strain to strain of close to 0.6
for experiment of FINE_Mag_2010, and close to 0.45 for experi-
ment of M05_Mag_2014, while it is close to 1 according to our
FE analysis. Also, the simulated and experimental crack density
vs. strain curves are different from each other. This is as
expected because the nominal strain/measured strain ratio is
not equal to unity for the experimental data. (Please refer to
Fig. 21 and the corresponding discussions.) Interestingly, if the
experimentally determined nominal strain curve and crack den-
sity are modified by dividing with a factor of 0.6 for FINE_-
Mag_2010 and 0.45 for M05_Mag_2014, denoted as ‘‘FINE_
Mag_2010/0.6” and ‘‘M05_Mag_2014/0.45” in Fig. 24(g), very
good agreement between experimental and model results can
be achieved.
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46 41

120

100

80

Stress (MPa)
60
fiber bridging 1
40 fiber bridging 2

20

0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
Strain

(a) Bridging law for the truss element used for analysis

(b) Pullout of PVA fiber from mortar matrix

(c) Normalized bridging law considering chemical bond and hardening

fiber-matrix interaction
Fig. 23. Pullout of PVA fiber from mortar matrix and bridging law considering chemical bond and hardening bond-slip law (Note: ‘‘fiber bridging 1” is bridging law only
considering frictional bond-slip, while ‘‘fiber bridging 2” is a modified version to account for the pullout behavior of PVA fiber with chemical bond and hardening bond-slip
relationship; ‘‘hardening” curve is actually the curve ‘‘Vf = 2.0%” in Fig. 13 of [31], which considers the hardening bond-slip law; ‘‘friction” is calculated based on [9], which
only considers the fictional bond).

6. Conclusion by fiber on crack surface, (ii) truss element to represent

bridging effect of fibers, and (iii) interface element to
In this study, a new discrete model for SHCC material is account for stress transfer between fiber and matrix, the
developed. With the use of (i) continuum element to con- proposed model can capture the multiple cracking behavior,
sider the matrix cracking as well as pulley force provided i.e., accurately predict the stress-strain curve and crack pat-
42 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

tern of SHCC material under tension, with relatively few
degrees of freedom.
Different parameters of the proposed discrete model are studied
and it is found that: (1) strength of matrix affects the ductility of
specimen; (2) the length of mortar damage stage affects the first
peak load after matrix cracking; (3) bond strength of the interface
element affects the crack pattern.
As discrete model with simple material constitutive law cannot
predict the experimental trend of crack width evolution very well,
some improvement is made to the proposed model. It is found that
by adopting hardening bond-slip law for interface element and
modifying the behavior of truss element to account for the effect
of slip-hardening fiber-matrix interaction, the evolution of crack
pattern (including, maximum crack width, average crack width
and distribution of crack width) are well predicted by the proposed
discrete model. Generally speaking, the proposed model is applica-
ble to cases with tensile crack (or mode I fracture), usually found
in situations with tension along a certain direction, such as the bot-
tom of a bridge deck or girder. Though only uniaxial stress state is
studied here, the concept of the new discrete model can be easily
extended to multi-axial stress state (e.g., SHCC overlay repair
under restrained shrinkage from two directions) by adding equiv-
alent truss in other directions to bridge other possible cracks. How-
ever, for mode II or mixed mode fracture (e.g., shear crack or shear-
flexural crack), additional work need to be performed to derive an
accurate shear constitutive law for the composite.

3.5
3.0
2.5
Stress (MPa)

2.0
1.5 FE

1.0
test
0.5
0.0
0.0 1.0 2.0 3.0 4.0 5.0
Strain (%)

(a) Stress-strain curve

(b) Crack pattern at ultimate stage (strain = 3.6%)

(c) Average crack width evolution

(d) Maximum crack width evolution (“Test” data are for test “L3” in
Table 4 of Zijl et al., 2016)
Fig. 24. Calculated result with proposed model (pulley, consider hardening bond and realistic bridging law).
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46 43

0.5%
6

Crack number (/50mm)

1.0%
1.5%
4 2.0%
2.5%
3.0%
2

Crack width ( m)

(e) Crack width distribution predicted by FE simulation (similar to that test)

(f) Nominal strain/strain ratio

(g) Crack density evolution

Fig. 24 (continued)

Being capable to efficiently and accurately capture the multiple in SHCC structures or SHCC repair system where tensile cracking is
cracking behavior of SHCC material, the proposed discrete model dominant. This capability of the model is important for durability
should be applicable to the prediction of crack spacing and opening design.
44 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

Acknowledgement Constitutive law of truss (fiber), adapted from bridging law

 qffiffiffiffi 
r ¼ f ðdÞ ¼ rtru 2 dd  dd , is
The authors gratefully acknowledge the financial support
received from The Research Grants Council of Hong Kong (Project rffiffiffiffiffiffiffiffiffi !
No. 16205414).
eSpf eSpf
r ¼ f ðdÞ ¼ rtru 2 
d d
Appendix A. Derivation of Spf
Equilibrium of truss, r ¼ rtru  2pArtrsn  y

qffiffiffiffiffiffiffi 
For a crack spacing of Sp (one real crack spacing at ultimate eSpf eSpf
state of tensile specimen), perform mechanical analysis (stress,
Then rtru  2pArtrsn  y ¼ rtru 2 d   d  , i.e.,
strain) at the ultimate state. sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!2
d 2pr sn
e¼ 1 y
Spf rtru Atr
The total strain at ultimate state, can be calculated by integrat-
ing the strain along truss

R Sp2  qffiffiffiffiffiffiffiffiffiffiffiffiffi2
R Sp2 d prsn
1  r2tru y dy
edy 0 Spf Atr
eu ¼ 0
Sp
¼ Sp
2 2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
A.1. For discrete model not considering pulley force d prsn Sp 4 2prsn Sp
¼  þ1
Spf rtru Atr 2 3 rtru Atr 2
For composite:

and

prsn Sp prsn V m rm r V m rm rm
¼ ¼ ¼
rtru Atr 2 rtru Atr V f sn 2 rtru V tr 2 2ru
Then

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
rffiffiffiffiffiffiffi 
d rm 4 rm d 3rm rm
eu ¼  2 þ1 ¼ 8 þ6
Spf 2ru 3 2ru 6Spf ru ru
For truss (fiber): Discussion:

(1) If (2x is max crack spacing), rm = ft;

Sp = 2x,
 qffiffiffiffi 
 ft 3f t
eu ¼ 6Sp
d
ru
f
 8 ru þ 6
(2) If Sp = x, (x is min crack spacing), rm = ft/2;
 qffiffiffiffiffiffiffiffiffiffiffiffi   qffiffiffiffiffiffi 
d 3f t ft d 3f t ft
eu ¼ 12Sp ru  8 2  ru þ 12 ¼ 6Sp 2ru  8 2ru þ 6
f f

(3) If Sp = kx, (kx is avg. crack spacing), rm = kft/2; then avg.

 qffiffiffiffiffi 
d 3rm rm
eu ¼ 6Sp ru  8 ru þ 6 (because the approximately linear
f

For matrix relationship between eu and rm ), i.e.,

0
sffiffiffiffiffiffiffiffiffi 1
d @3kft kft
eu ¼ 8 þ 6A
6Spf 2ru 2r u

So,
0 sffiffiffiffiffiffiffiffiffi 1
d @3kft kft
Spf ¼ 8 þ 6A
6eu 2ru 2r u
From equilibrium of composite: rtru Atr ¼ ru Am
Or
ru Am ru V m 0 sffiffiffiffiffiffiffiffiffi 1
rtru ¼ ¼
d @3krtp krtp
Atr Vf
Spf ¼ 8 þ 6A
6eu 2ru 2ru
Sp
From the equilibrium of matrix: 2
prsn ¼ Am rm ,
2
Am r m r V m r m r A.2. For model considering pulley force
Sp ¼ ¼
Atr sn V f sn
For truss element, define Spf so that the strain of truss is, Similarly,

 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ffi !
d d 3k g rtp k g rtp
e¼ Spf ¼  ðg  1Þ  8  ðg  1Þ þ 6
Spf 6eu 2 ru 2 ru
 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46 45

Appendix B. Mesh convergence study
To study the effect of mesh size on the results, the model con-
sidering pulley force is employed, with the following parameters:
constant bond for interface sn = 0.175 MPa; for bridging law of
fiber only friction is considered following Li and Leung [9] and
ru = 3.384 MPa. The mesh sizes of 0.8 mm, 0.5 mm, 0.35 mm and
0.25 mm are chosen for study and the results are summarized in
Fig. 25. It is found the various mesh sizes give comparable results.
In particular, if one is interesting in predicting crack width at the
serviceability state (with strain below 1%), which is important for
durability, the results for various crack sizes are very close.
However, as crack spacing in SHCC members under tension can
be as small as 2 mm, a mesh size smaller than 0.5 mm is
recommended.
Appendix C. Unloading behavior of the proposed model
To study the effect of unloading on the tensile stress vs strain
curve, the loading/unloading behavior of truss and interface ele-
ments in ATENA is adopted. For the truss, it is plastic and follows
isotropic hardening law. No damage occurs in the material during
loading, so the unloading modulus is equal to the initial elastic
modulus. For the interface, there is also no damage during loading

(a) Calculated stress-strain relationship

18
16
14
Crack number

12
10 mesh.25
8 mesh.35
mesh.5
6
mesh.8
4
2
0
0 1 2 3 4 5
Strain (%)

(b) Crack number evolution during loading

160 mesh.25
Avg. crack width (μm)

mesh.35

120 mesh.5
mesh.8

80

40

0
0 0.5 1 1.5 2 2.5 3 3.5 4

Strain (%)
(c) crack width evolution
Fig. 25. Mesh convergence study.
46 T. Shi, C.K.Y. Leung / Computers and Structures 185 (2017) 27–46

3.5

3.0

2.5

Stess (Mpa)
2.0 unloading
1.5
standard
1.0
test
0.5

0.0
0.0 1.0 2.0 3.0 4.0 5.0
Strain (%)

Fig. 26. Unloading behavior of the proposed discrete model for SHCC.

so the unloading modulus and initial elastic modulus are identical.
For the continuum element representing the matrix, smear crack-
ing can occur, and the unloading line from any point run towards
the origin.
To simulate unloading, the specimen is loaded to a strain of 2%
and then unloaded to a strain of 1.6%. It is then reloaded to 2.2%,
unloaded to 1.2% and reloaded again to 2.4%. (Note: The loading
scheme is chosen for the sake of easy computation.)
The results in Fig. 26 show that the unloading slope of the mem-
ber is much lower than the initial elastic slope (which is very high
as indicated by the closeness of the initial loading line to the y-axis
in the figure). As the truss and interface elements are assumed not
damaged, the damage in the continuum elements is responsible for
the decrease in slope. A similar behavior has been obtained in the
modeling work of Gencturk and Elnashai [6].
References
[1] Aveston J, Kelly A. Theory of multiple fracture of fibrous composites. J Mater Sci
1973;8(3):352–62.
[2] Belletti B, Cerioni R, Iori I. Physical approach for reinforced-concrete (PARC)
membrane elements. J Struct Eng 2001;127(12):1412–26.
[3] Boshoff WP, Van Zijl GPAG. A computational model for strain-hardening fiber-
reinforced cement-based composites: technical paper. J S Afr Inst Civ Eng
2007;49(2):24–31.
[4] Cervenka V, Cervenka J. User’s manual for ATENA 2D. Prague (Czech
Republic): Cervenka Consulting; 2002.
[5] Fischer G, Stang H, Dick-Nielsen L. Initiation and development of cracking in
ECC materials: experimental observations and modeling. Proceedings of the
6th international conference on fracture mechanics of concrete and concrete
structures, vol. 3. p. 1517–22.
[6] Gencturk B, Elnashai AS. Numerical modeling and analysis of ECC structures.
Mater Struct 2013;46(4):663–82.
[7] Lepech MD, Li VC. Water permeability of engineered cementitious composites.
Cement Concr Compos 2009;31(10):744–53.
[8] Li VC. On engineered cementitious composites (ECC). J Adv Concr Technol
2003;1(3):215–30.
[9] Li VC, Leung CKY. Steady-state and multiple cracking of short random fiber
composites. J Eng Mech 1992;118(11):2246–64.
[10] Li VC, Kanda T, Lin ZC. Influence of Fiber/Matrix interface properties on
complementary energy and composite damage tolerance. Key engineering
materials, vol. 145. Trans Tech Publications; 1998. p. 465–72.
[11] Lukovic M, Savija B, Schlangen E, Ye G, van Breugel K. A modelling study of
drying shrinkage damage in concrete repair systems. In: Structural faults and
repair conference 2014, 8–10 July 2014, London, UK. The Electrochemical
Society; 2014.
[12] Magalhães MS, Toledo filho RD, Fairbairn EMR. Fracture processes of strain-
hardening cement-based composites (SHCC) under direct tension and bending
loads Retrieved from <http://www.gruppofrattura.it/ocs/index.php/esis/
ECF18/paper/viewFile/6173/2048>; 2010.
[13] Magalhães MDS, Toledo Filho RD, Fairbairn EDMR. Influence of local raw
materials on the mechanical behaviour and fracture process of PVA-SHCC.
Mater Res 2014;17(1):146–56.
[14] Kabele P, Horii H. Analytical model for fracture behaviors of pseudo strain-
hardening cementitious composites. In: Proceedings-Japan society of civil
engineers. Dotoku Gakkai; 1996. p. 209–20.
[15] Kabele P. Finite element fracture analysis of reinforced SHCC members. In: van
Zijl GPAG, Boshoff WP, editors. Advances in cement-based
materials. Stellenbosch South Africa: CRC Press, Taylor & Francis Group;
2009. p. 237–44.
[16] Kabele P. Stochastic finite element modeling of multiple cracking in fiber
reinforced cementitious composites. Fract Damage Adv Fibre-Reinforced Cem-
Based Mater 2010:155–63.
[17] Kang J, Bolander JE. Spatial representation of fiber bridging forces in strain-
hardening cementitious composites. In: Proceedings of FraMCoS-9, 9th
international conference on fracture mechanics of concrete and concrete
structures. Berkeley, US.
[18] Kimber AC, Keer JG. On the theoretical average crack spacing in brittle matrix
composites containing continuous aligned fibres. J Mater Sci Lett 1982;1
(8):353–4.
[19] Kunieda M, Ogura H, Ueda N, Nakamura H. Tensile fracture process of Strain
Hardening Cementitious Composites by means of three-dimensional meso-
scale analysis. Cement Concr Compos 2011;33(9):956–65.
[20] Paegle I, Fischer G. Evaluation of test methods used to characterize fiber
reinforced cementitious composites. In: The international scientific
conference: innovative materials, structures and technologies. p. 121–6.
[21] Přinosil M, Kabele P. Modeling of fiber bridging in multiply-cracking mortar.
Eng Mech 2012;37.
[22] Redon C, Li VC, Wu C, Hoshiro H, Saito T, Ogawa A. Measuring and modifying
interface properties of PVA fibers in ECC matrix. J Mater Civ Eng 2001;13
(6):399–406.
[23] Spagnoli A. A micromechanical lattice model to describe the fracture
behaviour of engineered cementitious composites. Comput Mater Sci
2009;46(1):7–14.
[24] Spagnoli A, Yang E-H, Li VC, Micromechanical modelling of multiple fracture in
engineered cementitious composites, In: Proceedings of the 17th Biennial
European Conference on Fracture (ECF 17); 2008, p. 2407–2414.
[25] Sukumar N, Chopp DL, Moës N, Belytschko T. Modeling holes and inclusions by
level sets in the extended finite-element method. Comput Methods Appl Mech
Eng 2001;190(46):6183–200.
[26] Van Mier JG. Concrete fracture: a multiscale approach. CRC Press; 2012.
[27] Wang S, Li Victor C. Tailoring of pre-existing flaws in ECC matrix for saturated
strain hardening. In: The proceedings of FRAMCOS-5, Vail, Colorado, USA, April
2004. p. 1005–12.
[28] Wu HC, Li VC. Stochastic process of multiple cracking in discontinuous random
fiber reinforced brittle matrix composites. Int J Damage Mech 1995;4
(1):83–102.
[29] Yang J, Fischer G. Simulation of the tensile stress-strain behaviour of strain
hardening cementitious composites. In: Measuring, monitoring and modeling
concrete properties. Netherlands: Springer; 2006. p. 25–31.
[30] Yang EH, Li VC. Numerical study on steady-state cracking of composites.
Compos Sci Technol 2007;67(2):151–6.
[31] Yang EH, Wang S, Yang Y, Li VC. Fiber-bridging constitutive law of engineered
cementitious composites. J Adv Concr Technol 2008;6(1):181–93.
[32] Zhang J, Leung CK, Gao Y. Simulation of crack propagation of fiber reinforced
cementitious composite under direct tension. Eng Fract Mech 2011;78
(12):2439–54.
[33] Huang T, Zhang YX. Numerical modelling of mechanical behaviour of
engineered cementitious composites under axial tension. Comput Struct
2016;173:95–108.
[34] Prabowo H. Fiber Reinforced cementitious composite tailoring through 3D
lattice fracture simulations; 2009.
[35] Kabele P. Multiscale framework for modeling of fracture in high performance
fiber reinforced cementitious composites. Eng Fract Mech 2007;74
(1):194–209.
[36] van Zijl Gideon PAG, Slowik V, Filho RDT, Wittmann FH, Mihashi H.
Comparative testing of crack formation in strain-hardening cement-based
composites (SHCC). Mater Struct 2016;49(4):1175–89.