You are on page 1of 10

Composites Part B 174 (2019) 106958

Contents lists available at ScienceDirect

Composites Part B
journal homepage: www.elsevier.com/locate/compositesb

Fracture modeling of fiber reinforced concrete in a multiscale approach


Marcello Congro a, b, Eleazar Cristian Mejía Sanchez b, Deane Roehl a, b, *, Ederli Marangon c
a
Department of Civil and Environmental Engineering, Pontifical Catholic University of Rio de Janeiro, Brazil
b
Tecgraf Institute, Pontifical Catholic University of Rio de Janeiro, Brazil
c
Department of Civil Engineering, Federal University of Pampa, Rio Grande do Sul, Brazil

A R T I C L E I N F O A B S T R A C T

Keywords: This paper proposes numerical models for the study of fracture of fiber reinforced concrete. The composite
Multiscale analysis material behavior is described at the macroscale and at the mesoscale. The macroscale model considers homo­
Fracture geneous equivalent continuum properties taken from experimental curves of the composite. In order to repro­
Finite element analysis (FEA)
duce the effect of the random distribution of steel fibers within the cementitious matrix, random values of
Concrete
Fibers
equivalent elasticity modulus and equivalent tensile strength are assigned to solid and cohesive elements,
Numerical analysis respectively. The mesoscale model represents the fiber explicitly inside the concrete matrix through cohesive
interface elements with steel properties. In this case, fibers are located and oriented randomly in the matrix. In
addition, to allow matrix fracture, cohesive elements with softening constitutive behavior are placed at the edges
of the solid elements in the meso and macroscale models. The numerical models were applied to the simulation of
a direct tensile laboratory test for a steel fiber reinforced concrete. The macroscale model uses probability
functions to define the mechanical properties for each element. The predicted fracture paths and load capacity
present satisfactory results when compared to those obtained experimentally. In the mesoscale model, distinct
mechanical properties are applied for the steel fibers and for the cementitious matrix. The results from the
mesoscale approach reinforce the concept that fiber dispersion and orientation affect the structural load capacity
and matrix brittleness. In addition, cohesive interface elements proved to be an attractive approach to predict
fracture propagation in the composite material.

1. Introduction mechanical behavior of concrete. These methodologies include several


numerical models for fracture modeling in fiber-reinforced concrete
In recent decades, fiber reinforced composites have been widely used structures. Some examples of numerical models are cohesive interface
in engineering structures and in high technology areas, especially due to elements [10,11,42,48], smeared crack approaches [20,34,37] and
their excellent mechanical properties. The addition of fibers in the microplane models [7,40].
brittle matrix can significantly improve the brittleness of the composite Understanding the interaction between fibers and concrete in these
material. The effects of the individual fibers along the rupture plane, models becomes even more challenging, since nonlinear complex phe­
generating distinct stress-strain material behavior, define the macro nomena are involved. Interaction between fiber and matrix with load
hardening. transfer through adhesion, cementitious material damage and fiber
For structures involving composite materials, as fiber reinforced plastic deformation are some examples of these phenomena. Recent
concrete, probabilistic studies help to determine how fiber distribution applications include multiscale approaches [16,18,29,49], incorpo­
affects structural response. Load carrying capacity, ductility and fracture rating mesoscale effects to analyze the internal structure of the com­
patterns are some of the outputs of interest in these studies. According to posite. From the mesoscale material behavior, it is possible to explain
fiber arrangement, orientation, and geometry, fractures can propagate further what is observed at the macroscopic scale.
along different paths. Computational simulations predict fiber rein­ Recent studies on material behavior at the mesoscale have been
forced concrete behavior at distinct observation scales. published in literature with different numerical techniques. Liang & Wu
According to Zhan & Meschke [48], since the 1960s several [26] have proposed a 3D mesoscale model considering fiber and con­
computational models are employed in order to understand the crete separately, linking them with slide line contact. Ogura et al. [32]

* Corresponding author. Department of Civil and Environmental Engineering, Pontifical Catholic University of Rio de Janeiro, Brazil.
E-mail address: deane@tecgraf.puc-rio.br (D. Roehl).

https://doi.org/10.1016/j.compositesb.2019.106958
Received 11 March 2019; Received in revised form 16 May 2019; Accepted 27 May 2019
Available online 30 May 2019
1359-8368/© 2019 Elsevier Ltd. All rights reserved.
M. Congro et al. Composites Part B 174 (2019) 106958

introduced a mesoscale model of a fiber reinforced concrete under propagation in fiber reinforced concrete has intensified considerably in
mixed-mode fracture, based on a pullout behavior including the pullout the last decades, with the adoption of several methodologies to represent
angle of fibers to consider the fiber resistance against shear deformation material behavior at the macroscale. One of these methodologies uses
across the crack plane. Alternatively, Huschke & Unger [19] proposed a the so-called cohesive fracture model in association with interface ele­
2D and 3D mesoscale finite element model by employing an implicit ments [8]. Modeling of fractures follows with the insertion of interface
gradient enhanced damage model, initially studied by Peerlings et al. elements between solid finite elements. It is important to emphasize that
[33]. Previous research of Eligehausen et al. [14] proposes a local bond these special elements have distinct constitutive laws from those used in
stress-slip model of deformed bars under generalized excitations for the the solids model.
interaction between concrete and fibers. It is also important to point out A constitutive damage law describes the material softening behavior
that many formulations are not related just to cementitious matrix, but and consists of three phases. Fig. 1 shows a typical traction (σ ) – sepa­
also to inorganic-matrix composites, including polymeric or ceramic ration (Δ) model for normal and tangential relative displacements of
composites, for example [13,15,28,36]. interface elements, identified by subscripts n and t, respectively. The
Since the 1970s, several mathematical formulations were employed first phase describes material loading, characterized by linear-elastic
to compute the critical fiber volume in reinforced concrete. These for­ behavior up to peak load σ max to which correspond the critical relative
mulations consider the fibers continuous and perfectly aligned to the displacements δnc and δtc , respectively; the second phase describes
main stress axis. However, these premises are not present under prac­ material softening, where the material begins to suffer damage, which
tical conditions. In practice, fibers are discontinuous and randomly evolves until the cohesive interfaces are entirely separate at final rela­
distributed in the concrete matrix. Thus, the actual composite behavior tive displacement values δn and δt . Under unloading, damage is irre­
is different from predictions of the classical formulations of Aveston, versible and the traction-separation curve decreases to the origin [42].
Cooper and Kelly [2]. This constitutive damage model, initially proposed by Da �vila et al.
The variability in the cross-section of the fibers, for example, will [12] considers a parameter λ related to the norm of the displacement
provide deviations in the force that each fiber can withstand. The var­ jump tensor. The equivalent displacement jump is positive and a
iations of the fiber stiffness in the concrete matrix also influence the continuous function, as indicated in Equation (1).
strength properties of the material, providing different behaviors for the qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
composite according to the random dispersion of the fibers. The elas­ λ ¼ Δ23 þ ðΔshear Þ2 (1)
ticity modulus is variable for not only steel fibers, but also can be
extended to any other fiber type, in particular to natural fibers, such as where : is the MacAuley bracket, which sets any negative values to zero;
sisal and bamboo [9,23,38,39]. Δ3 refers to the displacement jump in mode I, normal to the mid-plane;
In this sense, probabilistic analyses are also performed in order to and Δshear is the tangential displacement considering the displacement
simulate the random dispersion of fibers in the concrete matrix, veri­ jump in modes II and III, as shown in Equation (2).
fying how this dispersion can affect the structural load capacity. The qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
prediction of the composite mean strength and the surface energy as a Δshear ¼ Δ21 þ ðΔ2 Þ2 (2)
function of two major components and their mutual interaction have
been started since the classical studies of Naaman [31]. In recent years, In this way, mixed mode behavior is considered, where normal and
many probabilistic approaches using the Monte Carlo method are tangential effects are coupled. Notice that, in the exponential softening
developed for fiber reinforced concrete structures [1,22,43], analyzing behavior, α is the parameter that controls the decay rate of the curve, D
the spatial variability and geometrical properties within structural is the damage variable, K0 is the stiffness coefficient, and δ is the asso­
elements. ciated displacement.
8
This paper proposes two formulations in order to represent the <
fracture propagation patterns in a fiber reinforced concrete, considering ð1 DÞK0 * Δδ; δ0 < Δδ < δf
T¼ (3)
0; δf < Δδ
a multiscale approach. The numerical simulations aim to reproduce the :
material behavior from experimental tests. In a macroscale approach,
probabilistic functions are considered to represent the stiffness and
distinct fiber dispersions in the cementitious matrix, avoiding the
explicit representation of the fibers in the finite element model. In
addition, this approach reduces the computational effort required for
performing analyses. However, it is an approximate approach, since
probabilistic functions describe the fiber random dispersion behavior.
On the other hand, the mesoscale approach consists of an innovative
way of representing steel fibers by using the cohesive interface element
formulation, allied to a damage constitutive model. The inclusion of
discrete fibers in the finite element mesh, assigning them distinct me­
chanical properties from cementitious matrix, becomes an interesting
alternative to simulate the heterogeneous structure of the composite
material. Thereby, it is possible to acquire the real fracture behavior of
fiber reinforced concrete in direct tensile tests.
Consequently, through repeated simulations, it is possible to obtain a
measure of sensitivity of system response to changes in the input pa­
rameters. Finally, after these simulations, it is possible to analyze the
load-displacement curve obtained when these mathematical models and
approaches are employed in order to represent fracture propagation of
the experimental tests.

2. Cohesive interface fracture model

The development of numerical models to simulate fracture Fig. 1. Cohesive constitutive model for interface elements.

2
M. Congro et al. Composites Part B 174 (2019) 106958

� � ��1
0 displacement data.
1 exp α* Δδ δ0
�δ � B δf δ0 C Fig. 4 shows typical plots obtained during the tensile tests for the
*B1 (4)
0 C
D¼1
Δδ @ 1 expð αÞ A cement matrix and the composites. The observed tensile strength was
4.63 � 0.39 MPa and 4.66 � 0.42 MPa for the matrix and the composite,
respectively. The elastic modulus obtained under tensile load was
Therefore, due to the nonlinearity of Equation (4), it is necessary to
33.1 � 2.1 GPa for the matrix and 35.8 � 2.0 GPa for the composite.
apply numerical methods to obtain iteratively the softening parameters
α and the total displacement at failure Δδ. This iteration was calculated
4. Fiber modeling in a macroscale approach
by using the Generalized Minimum Square Method, with the assistance
of the computational software MATLAB ®. The experimental points from
Macroscale computational simulations for concrete structures
Marangon’s test [27] were fitted to the exponential softening model,
generally consider homogeneity and, therefore, disregard the intrinsic
obtaining α ¼ 1.3 and Δδ ¼ 4.62 mm.
heterogeneous nature of a composite material. At this observation scale,
continuum equivalent properties represent the concrete mechanical
3. Experimental procedure
behavior.
In this paper, random values of stiffness and tensile strength assigned
The analysis developed in this paper is based on the direct tensile test
to the finite elements represent the presence of fibers in the concrete
of a self-compacting concrete reinforced with 1% hooked-end steel fi­
matrix. These arrangements gather random variables with known or
bers [27]. The matrix, without fibers, was also tested. The steel fibers
assumed probability distributions. It is possible, in this way, to carry out
had length of 60 mm, aspect ratio of 80, tensile strength of
repeated simulations, using in each of them a particular set of values of
1238 � 25 MPa and elasticity modulus of 202 � 4 GPa.
the random variables. Thus, the results of the simulations are statisti­
Table 1 presents the compositions of matrix and fiber composite. The
cally variable, converting response distribution rules into response
mixture was produced in a planetary mixer, in a temperature-controlled
variables.
room at 22 � 2 � C. Firstly, the interior of the mixer was wetted and
excess water was removed. Dry aggregates (fine and coarse) were placed
4.1. 2D macroscale analysis
into the mixer and blended for 60 s. Next, the cementitious materials
were added and the materials were mixed for other 60 s. After mixing
The composite material was discretized with 2D quadrilateral linear
the dry materials, 80% of the total water was added. At this point, the
elements (9.5 � 7.5 mm) under the assumption of linear elasticity. At
superplasticizer and a viscosity modifier agent were also included, fol­
each continuum element edge, an interface element with exponential
lowed by the remaining water, and the mixture was blended for 8 min.
softening law is inserted. The arc-length method controls the incre­
Finally, for the fiber reinforced concrete, the steel fibers were added and
mental analysis. Fig. 5 summarizes the finite element mesh and the
mixed for another 4 min.
boundary conditions of the numerical test.
The concrete samples were cast with dimensions of 100x100 � 400
To take into account random material property, typical for fiber
mm with reduction of section height in the center to 75 mm. Fig. 2 shows
composites, a random generation procedure defines the properties of
a scheme regarding the dimensions of the specimens. Curing happened
each finite element. For comparison, three probability density functions
in a fog room, under controlled temperature 22 � C � 2 � C and humidity
were tested: normal distribution, lognormal distribution and logistics
of above 95%. After 20 days of curing, each specimen was cut in half in
distribution. The direct tensile test provides the mean and standard
longitudinal direction, generating two samples 50 mm thick. The tests
deviation values for the elasticity modulus and tensile strength of the
were carried out after 28 days of curing. Four samples of the matrix and
samples.
four of the composite were tested.
The main idea of the macroscale approach is to reproduce the direct
In order to perform the direct tensile tests, a rigid system with fixed
tensile test in a finite element analysis. A sub-routine developed in
ends was introduced. Fig. 3 brings a scheme of the experimental setup.
software MATLAB ® generates random values of the equivalent com­
This setup avoids the contact between machine and specimen, pre­
posite elasticity modulus E and tensile strength σ t , according to the
venting crushing of the latter. Two steel plates are glued to the specimen
probability function. These values are assigned to each element in the
with high strength epoxy adhesive. They are connected to a U-shaped
central region of the mesh, given the expectation of fracture occurrence
steel apparatus by three steep pins. The testing machine clamps the
in this region. Fig. 6 exhibits the computational model that replicates the
apparatus at the fixing region. In addition, a laser level in the test ma­
direct tensile test. In this sense, cohesive interface elements are placed at
chine is responsible for assuring specimen alignment.
all edges of continuum elements in the central region of the specimen.
The direct tensile tests were conducted in a Shimadzu servo-
Three distinct simulation sets were developed: (i) random generation
controlled hydraulic testing machine with load capacity of 1000 kN.
of Young’s modulus for each continuum element using a quadrangular
The actuator displacement controls the test at a constant rate of 0.1 mm/
structured mesh, (ii) random generation of tensile strength for each
min. Two linear variable displacement transducers (LVDTs), placed in
cohesive element using a quadrangular structured mesh and (iii) random
the middle of the specimen according to Fig. 3b, register the
generation of tensile strength for each cohesive element using an un­
structured mesh. The third simulation set was included in the analyses
Table 1 because non-structured meshes reproduce better the fracture path and
Composition of matrix and composite. the structure behavior.
Composition (kg/m3) Matrix Composite

Coarse Aggregate ϕmax ¼ 19 mm 354.9 323.6


4.2. First simulation set
Coarse Aggregate ϕmax ¼ 9.5 mm 152.3 136.8
Fine Aggregate ϕmax ¼ 2.4 mm 949.6 949.6 The first simulation set used random values of the elasticity modulus
Silica Mesh 325 70.0 70.0 E, following normal, lognormal and logistic distributions. These nu­
Brazilian Cement CPIII-40 360.0 360.0
merical values were assigned to the elements in the central region of the
Fly Ash 185.3 185.3
Silica Fume 48.1 48.1 specimen. For each property distribution function, at least five random
Superplasticizer (Glenium 51 ®) 17.3 28.8 generations were analyzed. Table 2 summarizes the mechanical prop­
Viscosity Modifier Agent – UW410-Basf 0.36 0.36 erties of the fiber reinforced concrete used throughout these simulations.
Water 158.0 159.0 Fig. 7 presents the best results for the final fracture path for each dis­
Steel Fiber (Dramix ®) 0.00 78.0
tribution function. The results are in reasonable agreement with the

3
M. Congro et al. Composites Part B 174 (2019) 106958

Fig. 2. (a) Scheme with the dimensions of the specimens for the direct tensile tests (in millimeters) (b) examples of some reinforced concrete specimens used for the
direct tensile test.

Fig. 3. (a) Scheme of the setup for the direct tensile test. The apparatus consists of two steel accessories – one clamped by the testing machine and the other attached
to the specimen; (b) LVDT positions in the middle of the sample.

Fig. 5. Prismatic specimen discretization for the macroscale analysis.

Fig. 4. Typical plot for the direct tensile tests for the cement matrix and for the
tested composites.

experimental main fractures.

4.3. Second simulation set Fig. 6. Computational model and detail of the central region of the specimen
with random equivalent composite properties.
The second simulation set generated random values for the tensile

4
M. Congro et al. Composites Part B 174 (2019) 106958

Table 2 each distribution function.


Mechanical properties of the fiber reinforced concrete for probabilistic simula­ It is important to emphasize that several simulations were carried out
tions of Simulation Set I. considering distinct probability distributions for each simulation set. In
Fiber Reinforced Concrete this approach, according to the adoption of distinct probability density
E (GPa) Variable
functions representing fibers dispersion, several load-displacement be­
haviors might be observed. Consequently, the structural load capacity of
v( ) 0.2
the part will increase or decrease. Hence, the probability distributions
σt (MPa) 4.65
can be employed in order to capture the deviations in the load-
Kn ; Ks (MPa/m) 33.1 e3
displacement curves of the specimen.
Fig. 10 compares these curves for all three-simulation sets of the
macroscale models. Each plot exhibits one of the probability distribu­
strength σt , following normal, lognormal and logistic distributions.
tions of the structure behavior. In all three sets and rounds, the normal
These numerical values were associated with the elements in the central
and logistic distributions produced load-displacement curves in very
region of the specimen. The analyses were repeated for at least five
close agreement with the experiments.
random generations for each distribution. Table 3 summarizes the me­
chanical properties of the fiber reinforced concrete used throughout
these simulations. Fig. 8 presents the fracture patterns obtained 5. Fiber modeling in a 2D-MESOSCALE approach
numerically in comparison to the laboratory results. These are the best
matches for each distribution, although more rounds were performed. In the mesoscale, the numerical models explicitly represent the in­
It is important to point out that the numerical simulations using the dividual steel fibers in the concrete matrix. This representation con­
cohesive interface elements formulation consider a mixed-mode siders the orientation and distribution of the fibers in the cement matrix.
behavior of fracture. In Figs. 7 and 8, cracks propagate horizontally, Distinct material models can describe each component of the mesoscale.
once the stresses in y-direction are much higher than the stresses in x- This methodology requires the discretization of the internal structure of
direction. In addition, the structured mesh with interface elements along the material, which represents a considerable increase in computational
the edges restricts fracture propagation to horizontal or vertical. Since effort. Due to this complexity, mesoscale simulations are limited to
the tensile stresses on the horizontal cohesive elements are much higher smaller models.
than the tensile stresses on vertical planes no fracture deviation is Consequently, some simplifying assumptions assist the 2D numerical
observed. Therefore, the fracture paths are distinct from the modeling of fiber reinforced concrete at the mesoscale, as detailed in the
experiments. following sections.

4.4. Third simulation set 5.1. Simplifying assumptions

The third simulation set generated random values of tensile strength In this paper, cohesive interface elements model the contact zone
σr , following normal, lognormal and logistic distributions for a non-
structured finite element mesh. These numerical values were associ­ Table 3
ated with the elements in the central region of the specimen. The ana­ Fracture mechanical properties for probabilistic simulations of Set 2.
lyses were repeated at least five times for each distribution. The Fiber Reinforced Concrete
mechanical properties of the fracture are the same as in the second σt (MPa) Variable
simulation set.
Fig. 9 presents the fracture pattern obtained with the different dis­ α 1.3
tribution functions, comparing to the results obtained experimentally in Kn ; Ks (MPa/m) 33.1 e3
the central area. This figure shows the best results among all rounds for

Fig. 7. Comparative analysis of fracture propagation for the best simulation rounds in Simulation Set 1 with the experiments.

5
M. Congro et al. Composites Part B 174 (2019) 106958

Fig. 8. Comparative analysis of fracture propagation for the best matches to experimental data in Simulation Set 2.

Fig. 9. Fracture propagation for the best simulation rounds in the third sets.

between solid elements and segments representing the fibers. Tensile of the test specimen. Generating a 3D finite element mesh for such a high
strength (σ n ) and normal stiffness (Kn ) of these elements correspond to number of fiber segments results in a complex finite element mesh,
fiber mechanical properties. This methodology also reproduces the demanding greater computational effort to analyze the fracture propa­
sliding of the fiber with respect to the cementitious matrix, controlled by gation at this scale.
the strength (σt ) and stiffness (Kt ) properties of the fiber-matrix inter­ In this way, this paper adopts a 2D plane stress representation of the
action. Additionally, fiber debonding from cement matrix is represented specimen. Therefore, the central region of the specimen is a plate con­
by normal and tangential stiffness (Kn and Kt , respectively), associated taining 71 fiber segments with random orientation and distribution.
to cohesive interface element formulation, as reported by Da �vila et al. Each segment has a cross section equivalent to two fibers. Fig. 12 il­
[12]. lustrates schematically the simplifying hypothesis adopted in the 2D
The number of fibers included in the mixture is a function of the simulations at the mesoscale.
volumetric content. The fibers are considered as cylindrical geometric Determination of the equivalent properties for the fibers embedded
entities with diameter equal to 0.75 mm and length L ¼ 60 mm, as shows in the concrete follows from the law of mixtures for composite materials.
Fig. 11. Equations (6) and (7) give the equivalent elastic modulus and tensile
The number of fibers per unit volume is calculated by Equation (5), strength. In these two expressions, Em ; σ tm refer to the Young modulus
where Vuf refers to the fiber unit volume assuming a cylindrical shape; Vt and the tensile strength of the concrete matrix; Am is the area of the
is the total volume of the prismatic specimen and φ ¼ 1% is the fiber cement matrix, Ef; σ tf indicate the Young modulus and tensile strength
volumetric content. of the steel fiber, respectively; Af is the cross section area of one steel
fiber and nf is the number of fibers in the specimen cross section (in this
Vt
nfibers 3D ¼ *φ (5) case 2).
Vuf
Em Am þ Ef Af nf
Equation (1) gives a value of 142 fibers located in the central region Eeq ¼ : (6)
Am þ Af n f

6
M. Congro et al. Composites Part B 174 (2019) 106958

Fig. 10. Load-displacement curves for each studied sets.

Fig. 11. Simplified fiber geometry in 3D space.

σtm Am þ σtf Af nf
σ eq ¼ (7)
Am þ Af nf

5.2. Random fiber generation

From the simplifying assumptions mentioned above, a finite element


model was developed to analyze the load-displacement behavior of the
steel fiber reinforced concrete specimen in a tensile test. In this step,
Fig. 12. Simplified 2D fiber distribution.
straight segments of equal length were randomly inserted in the model.
Each of these geometric entities represents two steel fibers. Their

7
M. Congro et al. Composites Part B 174 (2019) 106958

random dispersion is defined by a subroutine created in MATLAB ®, Table 4


which exports the geometric configuration of the fibers to the finite Mechanical properties of the concrete matrix for simulations at the mesoscale.
element mesh pre-processor. Next, the meshing procedure generates Concrete Matrix
interface elements along the fibers and solid elements with edges
Em (GPa) 33.10
matching the fibers. In this way, fibers are represented explicitly and
v( ) 0.2
fiber lengths and orientations are preserved in the mesh.
σtm (MPa) 4.63
In simulations at the mesoscale, the explicit representation of fibers
Kn ; Ks (MPa/m) 33.1 e3
in the model demands a higher refinement in order to capture the correct
mechanical response of the composite material. According to Refs. [44,
46,47], mesoscale models with poor mesh refinement may result in
fracture propagation patterns which differ from those observed in Table 5
experimental tests. Cracks propagate along cohesive elements so crack Mechanical properties of steel fibers for the simulations at the mesoscale.
path is bias. Good predictions of crack trajectory requires unstructured Steel Fibers
mesh topology with high refinement. In this application, due to the a(GPa) 202
presence of several fibers and aiming at better identification of the v( ) 0.2
fracture pattern, the model was discretized with higher refinement in σtf (MPa) 1238
the central region.
Cohesive interface elements with composite equivalent properties
represent the mechanical behavior of the fibers. In addition, at mesh
Table 6
edges where no fiber is present, cohesive interface elements with con­
Equivalent properties for each interface element at fiber location, considering
crete properties are introduced, allowing fracture propagation. Fig. 13 law of mixtures.
shows the summarized workflow for the numerical simulations.
Equivalent Properties
Tables 4–6 combine the mechanical properties relevant to the nu­
merical model. In these tables, Em , Ef and Eeq refer to the Young Eeq (GPa) 38.167
modulus of concrete matrix, fibers and equivalent continuum, respec­ v( ) 0.2
tively; v is the Poisson coefficient; σtm ; σtf and σ eq are the tensile σeq (MPa) 41.63
strength from matrix, fibers and equivalent continuum, respectively; Kn , Kn (MPa/m) 41.63e6
Kt are the normal and tangential stiffness related to the interface ele­ Kt (MPa/m) 41.63e6
ments; and δ is the total displacement at failure. Fig. 14 presents two δ (mm) 4.62
finite element models employed in the numerical simulations consid­
ering discrete fiber modeling at the mesoscale. The fracture propagation
Fig. 16 presents a comparison between results for macro- and
paths for each model validate the numerical analysis against the
mesoscale models using unstructured meshes against experimental data.
experimental results.
As expected, load capacity behavior increases or decreases according to
Fig. 15 compares the load-displacement curves considering the
dispersion of fibers. Both models were able to predict the experimental
mesoscale analysis and the experimental results. The inclusion of
mechanical response.
random steel fibers in the concrete matrix reinforces the idea of
Hence, the methodology using cohesive interface elements for nu­
increasing the load capacity of the part. In this way, fibers will be
merical modeling of fiber reinforced concrete at the mesoscale proved to
responsible for controlling material cracking, enhancing strength and
be valid to represent concrete behavior in a tensile test. The results
durability of the structural element.
obtained were adequate when compared to the experimental results.
Fiber distribution and orientation affect the global behavior of the
specimen. In this study, although positioning of fibers was random, no
fiber tangling or clustering is considered. These events are present in
some mixtures and can modify the global composite response even more
strongly.

6. Conclusions

This study proposed two finite element models to represent rein­


forced steel fiber concrete behavior, one at the macroscale and the other
at the mesoscale. In order to consider the inhomogeneity due to random
fiber distribution, a probabilistic approach was developed. Results from
tensile experimental tests performed on specimens of a cementitious
mixture with steel fibers provide data for the probabilistic definition of
model parameters and reference results. Three distribution functions
define the parameter values, randomly modifying each element of the
central region of the model.
The model at the macroscale is easy to generate and adopts equiv­
alent continuum properties for the composite material, which is
randomly distributed in the central region of the specimen. Three dis­
tribution functions were tested. The random properties are the Young’s
modulus and tensile strength of the composite material. Finally, we
compare results from simulations using structured and unstructured
meshes. The many possibilities of orientation of steel fibers produced
distinct behaviors in the load-displacement curves of the simulated
specimens. All three-distribution functions lead to results in the
Fig. 13. Workflow for the mesoscale modeling.

8
M. Congro et al. Composites Part B 174 (2019) 106958

Fig. 14. Finite element mesh in the central region of the fiber reinforced specimen. Random fiber distribution and fracture path (a) model 1; (b) model 2.

Fig. 16. Load-displacement curves comparison for the experimental and nu­
merical unstructured macro and meso-scale models.

mesoscale. Finally, multiscale numerical simulations using cohesive el­


ements at macro- and mesoscale are an attractive approach to predict
Fig. 15. Load-displacement curves for the experimental and numerical models fracture patterns in fiber reinforced concrete.
with the 2D mesoscale approach.

Acknowledgements
reference experimental range. However, the lognormal distribution
produced results slightly deviated from the experimental tension curve.
This study was financed in part by the Coordenaça ~o de Aperfeiçoa­
The results confirm that fiber dispersion contributes to increase the
mento de Pessoal de Nivel Superior - Brasil (CAPES) - Finance Code 001.
structural load capacity. The unstructured mesh was capable of repro­
The authors would like to thank CAPES, Tecgraf Institute at Pontifical
ducing the crack propagation pattern in the reinforced concrete
Catholic University of Rio de Janeiro for the financial support and
material.
assistance during the execution of this project.
In the mesoscale analysis, a numerical model was developed where
the fibers were introduced explicitly in the finite element mesh. In this
References
case, the positions of the fibers are generated randomly and fiber
behavior is introduced in the model through cohesive interface ele­ [1] Alann A, Johnsson H. Flax fiber-reinforced glued-laminated timber in tension
ments. To allow fracture propagation in the matrix, interface elements perpendicular to the grain: experimental study and probabilistic analysis. J Mater
with cement properties were inserted along the edges of the continuum Civ Engng 2010;22:827–35. https://doi.org/10.1061/(ASCE)MT.1943-5533
.0000070.
elements. Simulations at this scale demand higher computational effort [2] Aveston J, Cooper GA, Kelly A. The properties of fibre composites. Surrey, England:
since they require fine meshes. The numerical model at this observation IPC Science and Technology Press; 1971. p. 15–26. https://scholar.google.com/
level also reached good accuracy when compared to the experimental scholar_lookup?title¼Single%20and%20Multiple%20Fracture&publication_ye
ar¼1971&author¼J.%20Aveston&author¼G.A.%20Cooper&author¼A.%20Kelly.
results, validating the cohesive interface elements as a valuable alter­
[7] Caner FC, Bazant ZP, Wendner R. Microplane model M7f for fiber reinforced
native to simulate the fiber reinforced concrete behavior at the concrete. Eng Fract Mech 2013;105:41–57. https://doi.org/10.1016/j.engf
racmech.2013.03.029.

9
M. Congro et al. Composites Part B 174 (2019) 106958

[8] Chandra N, Li H, Shet C, Ghonem H. Some issues in the application of cohesive [29] Montero-Chac� on F, Cifuentes H, Medina F. ‘Mesoscale characterization of fracture
zone models for metal-ceramic interfaces. Int J Solids Struct 2002;39:2827–55. properties of steel fiber-reinforced concrete using a lattice–particle. Model. Mater.
https://doi.org/10.1016/S0020-7683(02)00149-X. 2017;10:207. https://doi.org/10.3390/ma10020207.
[9] Chandramohan D, Marimuthu K. A review on natural fibers. Int J Res Rev Appl Sci [31] Naaman AE, Moavenzadeh F, McGarry FJ. Probabilistic analysis of fiber reinforced
2011;8:194–206. https://www.arpapress.com/Volumes/Vol8Issue2/IJRRAS_8 concrete. J Eng Mech 1974;100:397–413. ISSN: 0733-9399, https://trid.trb.org/vi
_2_09.pdf. ew/100515.
[10] Congro M, Mejia C, Roehl D. ‘XFEM and interface element models applied to [32] Ogura H, Kunieda M, Ueda N, Nakamura H. MESO-scale modeling for fiber
fracture of fiber reinforced concrete beams’. XXXVIII CILAMCE – ibero Latin- reinforced concrete under mixed mode fracture. In: Proceedings of the 8th
American congress of computational methods in engineering. Brazil: international conference on fracture mechanics of concrete and concrete
Florian�opolis; 2017. https://doi.org/10.20906/CPS/CILAMCE2017-0195. structures, FraMCoS; 2013. p. 1422–9. https://framcos.org/FraMCoS-8/p428.pdf.
[11] D’Antino T, Colombi P, Carloni C, Sneed L. Estimation of a matrix-fiber interface [33] Peerlings RHJ, De Borst R, Brekelmans WAM, De Vree JHP. Gradient enhanced
cohesive material law in FRCM-concrete joints. Compos Struct 2018;193:103–12. damage for quasi-brittle materials. Int J Numer Methods Eng 1996;39:3391–403.
https://doi.org/10.1016/j.compstruct.2018.03.005. https://doi.org/10.1002/(SICI)1097-0207(19961015)39:19%3c3391::
[12] D�avila CG, Camanho PP, Turon A. NASA/TP-2007-214869 cohesive elements for AID-NME7%3e3.0.CO;2-D.
shells. 2007. https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/200700 [34] Rafi MM, Nadjai A. Comparison of numerical behaviors of FRP reinforced concrete
18344.pdf. beams using three smeared crack models. Mater Struct 2012;45:93–106. htt
[13] Dmitriev AI, Osterle
€ W, Wetzel B, Zhang G. Mesoscale modeling of the mechanical ps://doi.org/10.1617/s11527-011-9753-6.
and tribological behavior of a polymer matrix composite based on epoxy and [36] Sadowski T, Marsavina L. Multiscale modelling of two-phase ceramic matrix
6vol.% silica nanoparticles. Comput Mater Sci 2015;10:204–14. https://doi.org/ composites. Comput Mater Sci 2011;50:1336–46. https://doi.org/10.1016/j.co
10.1016/j.commatsci.2015.08.029. mmatsci.2010.04.011.
[14] Eligehausen R, Popov EP, Bertero VV. Local bond stress-slip relationships of [37] Sato Y, Naganuma K. Discrete-like crack simulation by smeared crack-based FEM
deformed bars under generalized excitations. Proceedings of the 7th european for reinforced concrete. Earthq Eng Struct Dyn 2007;36:2137–52. https://doi.
conference on earthquake engineering, 4. Athens: Techn. Chamber of Greece; org/10.1002/eqe.720.
1982. p. 69–80. Universit€at Stuttgart, 1982, https://doi.org/10.18419/opus-8473. [38] Silva FA, Chawla N, Filho RDT. Tensile behavior of high performance natural
[15] Fagiano C, Genet M, Baranger E, Ladav� eze P. Computational geometrical and (sisal) fibers. Compos Sci Technol 2008;68:3438–43. https://doi.org/10.1016/j.co
mechanical modeling of woven ceramic composites at the mesoscale. Compos mpscitech.2008.10.001.
Struct 2014;12:146–56. https://doi.org/10.1016/j.compstruct.2014.01.045. [39] Silva FA, Mobasher B, Soranakom C, Filho RDT. Effect of fiber shape and
[16] Ghosh A, Chaudhuri P. Computational modeling of fracture in concrete using a morphology on interfacial bond and cracking behaviors of sisal fiber cement based
meshfree meso-macro multiscale method. Comput Mater Sci 2013;69:204–2015. composites. Cement Concr Compos 2011;33:814–23. https://doi.org/10.1016/j.
https://doi.org/10.1016/j.commatsci.2012.11.025. cemconcomp.2011.05.003.
[18] Guan X, Liu X, Jia X, Yuan Y, Cui J, Mang HA. A stochastic multiscale model for [40] Smolcic Z, Ozbolt J. Meso scale model for fiber-reinforced-concrete: microplane
predicting mechanical properties of fiber reinforced concrete. Int J Solids Struct based approach. Comput Concr 2017;19:375–85. https://doi.org/10.1298
2015;56–57:280–9. https://doi.org/10.1016/j.ijsolstr.2014.10.008. 9/cac.2017.19.4.375.
[19] Huschke P, Unger JF. A mesoscale finite element model for fibre reinforced [42] Tailhan T, Rossi P, Daviau-Desnoyers D. Probabilistic numerical modelling of
concrete. In: Proceedings of european congress on computational methods in cracking in steel fibre reinforced (SRFC) structures. Cement Concr Compos 2015;
applied sciences and engineering – ECCOMAS 2016. Greece: Crete Island; 2016. htt 55:315–21. https://doi.org/10.1016/j.cemconcomp.2014.09.017.
ps://www.eccomas2016.org/proceedings/pdf/5328.pdf. [43] Thomos G, Trezos CG. Examination of the probabilistic response of reinforced
[20] Hwang J-H, Lee DH, Kim KS, Ju H, Seo S-Y. Evaluation of shear performance of concrete structures under static non-linear analysis. Eng Struct 2006;28:120–33.
steel fibre reinforced concrete beams using a modified smeared-truss model. Mag. https://doi.org/10.1016/j.engstruct.2005.08.003.
Concr. Res. 2013;65:283–96. https://doi.org/10.1680/macr.12.00009. [44] Vo�rechovský M, Sadílek V. Computational modeling of size effects in concrete
[22] Lee T-H, Mosalam KM. Probabilistic fiber element modeling of reinforced concrete specimens under uniaxial tension. Int J Fract 2008;154:27–49. https://doi.org/10.
structures. Comput Struct 2004;82:2285–99. https://doi.org/10.1016/j.co 1007/s10704-009-9316-9.
mpstruc.2004.05.013. [46] Wang X, Zhang M, Jivkov A. Computational technology for analysis of 3D meso-
[23] Li D, Yang Q-S, Liu X, Shang JJ. Experimental investigation on tensile properties of structure effects on damage and failure of concrete. Int J Solids Struct 2016;80:
carbon nanotubes wires. Mech Mater 2017;105:42–8. https://doi.org/10.1016/j. 310–33. https://doi.org/10.1016/j.ijsolstr.2015.11.018.
mechmat.2016.11.010. [47] Yaghoobi A, Chorzepa M, Kim S, Durham S. Mesoscale fracture analysis of
[26] Liang X, Wu C. Meso-scale modelling of steel fibre reinforced concrete with high multiphase cementitious composites using peridynamics. Materials 2017;10. 21pp.
strength. Constr Build Mater 2018;165:187–98. https://doi.org/10.1016/j.conb https://doi.org/10.3390/ma10020162.
uildmat.2018.01.028. [48] Zhan Y, Meschke G. Multilevel computational model for failure analysis of steel-
[27] Marangon E. Caracterizaç~ ao material e estrutural de concretos autoadens�aveis fiber–reinforced concrete structures. J Eng Mech 2016;142. https://doi.
reforçados com fibras de aço. PhD Thesis. Rio de Janeiro, Brazil: Federal University org/10.1061/(ASCE)EM.1943-7889.0001154.
of Rio de Janeiro; 2011. http://objdig.ufrj.br/60/teses/coppe_d/EderliMarangon. [49] Zhang Y, Meschke G. A multiscale oriented concept for the analyses of steel fiber
pdf. reinforced concrete materials and structures. In: Proceedings of the 8th
[28] Mazzuco G, D’Antino T, Pellegrino C, Salomoni V. Three-dimensional finite international conference on fracture mechanics of concrete and concrete
element modeling of inorganic-matrix composite materials using a mesoscale structures, FraMCoS 2013; 2013. p. 1352–9. Bochum, Germany, https://framcos.
approach. Compos B Eng 2018;143:75–85. https://doi.org/10.1016/j.compo org/FraMCoS-8/p418.pdf.
sitesb.2017.12.057.

10

You might also like