Sie sind auf Seite 1von 30

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS

Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435


Published online 17 July 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/nag.2411

Comparison of continuous and discontinuous constitutive models to


simulate concrete behaviour under mixed-mode failure conditions

Jerzy Bobiński*,† and Jacek Tejchman


Faculty of Civil and Environmental Engineering, Gdańsk University of Technology, Narutowicza 11/12, Gdańsk 80-233,
Poland

SUMMARY
The paper presents detailed FE simulation results of concrete elements under mixed-mode failure conditions
according to the so-called shear-tension test by Nooru-Mohamed, characterized by curved cracks. A contin-
uous and discontinuous numerical two-dimensional approach was used. In order to describe the concrete’s
behaviour within continuum mechanics, two different constitutive models were used. First, an elasto-plastic
model with isotropic hardening and softening was assumed. In a compression regime, a Drucker–Prager
criterion with a non-associated flow rule was used. In turn, in a tensile regime, a Rankine criterion with
an associated flow rule was adopted. Second, an isotropic damage constitutive model was applied with a
single scalar damage parameter and different definitions of the equivalent strain. Both constitutive laws were
enriched by a characteristic length of micro-structure to capture properly strain localization. As an alterna-
tive approach, the extended finite element method was used. Our results were compared with the experimen-
tal ones and with results of other FE simulations reported in the literature. Copyright © 2015 John Wiley &
Sons, Ltd.

Received 21 April 2014; Revised 21 May 2015; Accepted 9 June 2015

KEY WORDS: elasto-plasticity; damage mechanics; XFEM; strain localization; crack; numerical simulations

1. INTRODUCTION

Concrete is a highly heterogeneous and non-linear material that belongs to quasi-brittle materials,
which exhibit a gradual decrease of strength with increasing strain after the peak. A choice of the
constitutive model and knowledge on its limitations and drawbacks is essential to obtain theoretical
results being in a satisfactory agreement with experimental outcomes. This problem is especially
important when the evolution of curved cracks is taken into account. The fracture mechanics in
plane simulations defines two cracking modes: I (opening mode) and II (shearing mode). While
modes I and II are rather simple to be validated, more complex stress states, which include, for
example a combination of two these modes, are still a demanding numerical challenge. Usually,
model parameters are calibrated from uniaxial (mode I) tests (tension, compression), and a response
of models under mixed-mode conditions is the consequence of the assumed formulation. Moreover,
not only the validation of constitutive laws has to be performed at the material point level but also,
its ability to simulate properly structural elements has to be checked.
Despite of a huge variety of constitutive laws intended to describe the concrete behaviour, there is a
limited number of laboratory experiments under mixed-mode conditions. Very often, single-edge or
double-edge notched beams under anti-symmetrical four-point shear loading were chosen as
benchmark tests. The outcomes from experiments performed by Arrea and Ingraffea [1], Schlangen

*Correspondence to: Jacek Tejchman, Faculty of Civil and Environmental Engineering, Gdańsk University of Technol-
ogy, Narutowicza 11/12, Gdańsk 80-233, Poland.

E-mail: tejchmk@pg.gda.pl

Copyright © 2015 John Wiley & Sons, Ltd.


CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 407

[2] or Carpinteri et al. [3] were directly compared with finite element method (FEM). In the context of
reinforced concrete, for example a dowel disk test was simulated [4]. However, the most commonly
used experiment to validate constitutive models for concrete was performed by Nooru-Mohamed [5]
at Delft University. A double-edge notched specimen was tested under different combinations of
horizontal and vertical loads. The main difficulty in simulating this problem is a continuous rotation
of principal directions of the stress state.
The intention of the paper is to numerically analyze in detail a quasi-static laboratory test by Nooru-
Mohamed [5] using a continuous and discontinuous numerical approach under plane stress conditions.
Two relatively simple continuum constitutive laws to check their effectiveness to describe a cracks’
geometry during a complex failure mode were applied. In elasto-plasticity, the Rankine criterion in
tension and the Drucker–Prager criterion in compression were used. A constitutive law with
isotropic degradation of the material was also assumed with the different equivalent strain measures.
Both models were equipped with a characteristic length of micro-structure by applying an integral
non-local theory to properly regularize boundary value problems. This is in contrast with many
formulations (Section 2), where a simple crack band approach was taken into account (unable to
describe the width of a localized zone). Besides a smeared description of cracks, the extended FEM
(XFEM) in its standard formulation was also used to simulate cracks in a discontinuous way. The
numerical results were compared with the original experiments and with the results reported by
other researchers with respect to the shape of cracks and load–displacement diagrams. Advantages
and disadvantages of models used were outlined. For a better comparison of the curvature of cracks,
a new geometric measure in the form of a crack height was introduced. To our knowledge, such a
comprehensive detailed comparison of different constitutive models (continuous and discrete)
describing the concrete behaviour in a complex failure mode has not been performed yet.
The paper is organized as follows. After introduction (Section 1) and literature review (Section 2),
elasto-plastic and isotropic continuum damage constitutive laws are described in Sections 3 and 4. For
damage mechanics, different definitions of equivalent strain are presented. Both approaches are
regularized with an integral non-local theory discussed in Section 5. In Section 6, the general
information about the XFEM is provided with a description of a discrete cohesive law and a crack
propagation algorithm. The laboratory tests are depicted in Section 7. The numerical results of the
Nooru-Mohamed test (enhanced continuum models and XFEM) are comprehensively analyzed in
Section 8. A discussion with other results reported in the research literature is also given. The final
conclusions are listed in Section 9.

2. LITERATURE REVIEW

The simulations of the Nooru-Mohamed test were first performed in the literature using different
smeared crack models within continuum damage mechanics. Unfortunately, improper crack patterns
were obtained with some definitions of the equivalent strain. To improve it, Grassl and Jirásek [6]
used a local isotropic constitutive law and anisotropic micro-plane model (coupled with the external
tracking algorithm to determine the direction of the crack propagation). The crack band approach
was used as a regularization method. Later, they extended this approach with the orthogonal fixed-
crack method [7]. The same idea was used also by Cervera and Chiumenti [8]. Slobbe et al. [9]
coupled this method with a sequentially linear analysis and quadratic finite elements. This approach
is opposite to a standard continuum formulation, where the evolution of cracks is a natural
consequence of equilibrium equations. Ožbolt and Reinhardt [10] simulated mixed mode in concrete
using a modified version of the micro-plane model M2 together with a crack band approach. Patzak
and Jirásek [11] used a micro-plane-based anisotropic damage model regularized by a non-local
integral theory. For a better capture of a crack trajectory, the adaptive remeshing technique was
used. Another anisotropic damage model was formulated by Desmorat et al. [12] based on the
Mazars’s definition of the equivalent strain and equipped with a characteristic length via an integral
non-local theory. Di Prisco et al. [13] compared the performance of different continuum constitutive
laws. They analyzed Ottosen’s model coupled with a crack band approach, a gradient plasticity
model with the Rankine criterion and an integral non-local scalar damage model. A similar

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
408 J. BOBIŃSKI AND J. TEJCHMAN

comparison was carried out by Pivonka et al. [14]. In this study, the so-called extended Leon’s model,
two multi-surface plasticity constitutive laws – the Drucker–Prager criterion with the tension cut-off
and the Drucker–Prager law with the Rankine criterion – and micro-plane models were investigated.
All formulations were regularized with a fracture energy approach. The Nooru-Mohamed test served
also as a benchmark for validating coupled damage-plasticity constitutive laws. Jefferson [15]
examined his plastic-damage-contact ‘Craft’ model with a crack band approach. Abu Al-Rub and
Kim [16] proposed a coupled model based on Lubliner’s yield criterion. Another definition of such
coupling was given by Ananiev and Ožbolt [17]. Wu and Xu [18] formulated a multi-crack elasto-
plastic damage model. In the last three examples, however, a regularization technique was not applied.
Not only some smeared descriptions of cracks were used to simulate the Nooru-Mohamed test.
Different descriptions of displacement jumps in continuum were also examined. Lensa et al. [19]
analyzed this problem using cohesive surfaces. Several models were formulated within the strong
discontinuity approach with embedded discontinuities in finite elements and cohesive softening laws
defined in damage [20–23]. Gasser and Holzapfel [24] performed three-dimensional simulations
using the XFEM. With XFEM, different crack propagation criteria were examined in the Nooru-
Mohamed experiment [25–27]. In turn, Dong et al. [28] and Zhang [29] adopted a meshless method
with a cohesive crack definition.
Several authors applied discrete models in order to simulate the Nooru-Mohamed test. Schlangen and
Garboczi [30, 31] used a two-dimensional (2D) lattice composed of small beam elements, in which after
exceeding the assumed threshold stress value, beam elements were removed. They simulated one-phase
shear (horizontal) load scenario only. The same problem with the aid of a rigid-body-spring network
model and a Mohr–Coulomb fracture criterion with the tension cut-off was analyzed by Bolander and
Saito [32]. Chang et al. [33] defined micro-polar microstructural model based on the underlying
regular lattice network. Delaplace and Ibrahimbegovic [34] adopted a Reissner beam link model
based on a random distribution of Voronoi cells. A confinement–shear lattice model with softening in
damage prescribed to lattice beams was used by Cusatis et al. [35]. To increase the ductility of lattice
models, Kozicki and Tejchman [36] assigned different properties to linear elastic perfectly brittle
beams simulating aggregates, cement matrix and interfacial transition zones.

3. ELASTO-PLASTIC CONSTITUTIVE LAW

3.1. Standard formulation


The elasto-plasticity model was defined within a standard plasticity theory. It is based on a
decomposition of the small total strain rates ε_ ij into the elastic and plastic strain rates ε_ ije and ε_ ijp . The
elastic behaviour (the calculation of stresses σ ij) is governed by the linear elastic stiffness tensor Cijkl
with Young modulus E and Poisson ratio v. The plastic strain rates are calculated on the basis of the
plastic multiplier λ_ and assumed potential function g

∂g
ε_ ij ¼ ε_ ije þ ε_ ijp ; σ_ ij ¼ Cijkl ε_ kle ; ε_ ijp ¼ λ_ (1)
∂σ ij
with the following loading–unloading conditions:

λ_ ≥ 0; f ≤ 0; λ_ f ¼ 0: (2)
In order to simulate the behaviour of concrete elements, a multi-surface constitutive law was
formulated (Figure 1a). In a tensile regime, the classical Rankine criterion was used. The failure
function FR was defined as

F R ¼ f R  σ t ðκR Þ ¼ maxfσ 1 ; σ 2 ; σ 3 g  σ t ðκR Þ; (3)


where σ 1, σ 2 and σ 3 are the principal stresses, σ t the current uniaxial tensile failure stress and κR the
(hardening) softening parameter. The softening parameter κR was defined as the cumulative value of

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 409

Figure 1. Plasticity yield functions under plane stress conditions: following (a) Rankine–Drucker–Prager
(Eqs 3 and 5) and (b) modified Rankine (Eq. 9) (ticks are marked for values 1 and 10).

the plastic multiplier λ_ from the failure plane with the maximum principal stress. When more surfaces were
active, a multi-surface plasticity algorithm with independent plastic multipliers λ_ i (its number depended on
the number of active planes) applying Koiter’s rule was used. The same formula was used to calculate the
softening parameter κR; that is, the additional positive plastic multipliers were not taken into account. In a
uniaxial load case, the softening parameter κR was equal to the maximum principal plastic strain ε p1 . An
associated flow rule was assumed; the potential function from Eq. 1 (denoted for the Rankine criterion
as gR) was equal to fR. To describe softening under tension, an exponential curve was chosen:
 
κR
σ t ðκR Þ ¼ f t exp  ; (4)
κu
where ft is the concrete strength in uniaxial tension (ultimate failure stress) and the parameter κu controls
the slope of the softening curve. It can be interpreted as an intersection point of a tangent line from the
peak point with a horizontal axis. This curve was chosen to obtain a better agreement with the XFEM
formulation for a one-dimensional case (Section 6). It should be noted that a formulation with this
simple softening curve produces very similar results to ones obtained with more sophisticated curves,
for example the Hordijk’s non-linear curve [37], which was calibrated based on tensile experiments.
In a compression regime, a simple yield surface based on the linear Drucker–Prager criterion and
isotropic hardening and softening was used (following the formulation proposed in ABAQUS/
Standard [38]). The yield surface was assumed as
 
1
f P ¼ q þ p tanφ  1  tanφ σ c ðκP Þ; (5)
3
where q is the Mises equivalent stress, p the mean stress, φ the friction angle, σ c the current uniaxial
compression failure stress and κP the hardening (softening) parameter (equal to plastic strain in
uniaxial compression ε11p ). The invariants p and q were defined as
rffiffiffiffiffiffiffiffiffiffiffiffi
1 3
p ¼ σ kk and q¼ sij sji ; (6)
3 2
where sij denotes the deviatoric stress tensor. The flow potential gp for the Drucker–Prager criterion
was assumed as
gP ¼ q þ p tanψ (7)

with the dilatancy angle ψ. The plastic strains rates ε_ ijp were calculated as
κ_ P ∂gP
ε_ ijp ¼ : (8)
1  ð tanψ Þ=3 ∂σ ij

In general, the defined model shares the choice of two criteria: the Rankine and Drucker–Prager with
many similar formulations [14, 39, 40]. It reflects the main features of the concrete behaviour.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
410 J. BOBIŃSKI AND J. TEJCHMAN

However, the choice of the Drucker–Prager criterion raises some controversies. Although it is able to
simulate, for example the different strengths in biaxial and uniaxial compression (by calibrating the
internal friction angle), it has several shortcomings. The tensile and compression meridians are
straight, and the shape of the yield curve in a deviatoric plane is circular (the shape does not depend
on the similarity angle). In reality, these meridians are curved, and the shape of the yield surface in a
deviatoric plane is non-circular. These disadvantages may be overcome by adopting more advanced
constitutive laws (e.g. by Lee and Fenves [41] and by Menetrey and Willam [42]). In the formulation
of the model, the cases with two or three active yield planes of the Rankine criterion were taken into
account by applying a standard multi-surface plasticity model with the Koiter’s rule in order to
calculate increments of plastic strains. The only difference is that the hardening parameter κR is the
same for all active yield planes (the plastic multipliers λ_ i are, however, independent for each active
surface ‘i’). A simultaneous activity of the Drucker–Prager and any of the Rankine planes was treated
in the same way.

3.2. Modified formulation


In order to obtain a better agreement with experimental results in a tension–compression regime, the
original Rankine criterion was modified to reproduce a decrease of the material tensile strength with
increasing perpendicular compressive stress. The following failure function FQ was postulated for
3D cases:

F Q ¼ σ max  k f ðσ med þ σ min Þ  σ t ðκQ Þ; (9)

where σ max, σ med and σ min are the maximum, intermediate and minimum principal stresses,
respectively. The coefficient kf describes the strength reduction, and κQ denotes the softening
parameter. Under plane stress conditions by taking kf = ft/fc (fc – uniaxial compression strength), two
extra characteristic points in the principal plane stress space (σ 1 and σ 2) were obtained: (ft, 0) for
uniaxial tension and (0, fc) for uniaxial compression. By introducing this modification in a
tension–compression regime only and keeping the original Rankine yield definition in a tension–
tension regime, one extra yield line and one corner point were added in 2D simulations with the
symmetry consideration. In order to simplify the numerical implementation, the original Rankine
failure function (Eq. 3) was also replaced by the modified formulation of Eq. 9 in a tension–tension
regime (Figure 1b). The drawback of the modification was the fact that the tensile biaxial strength
ftb was slightly higher because the intersection point of two yield lines (in 2D problems) along a
hydrostatic line (given by Eq. 9) was equal to ftb = ft/(1  kf). However, the strength’s overestimation
was small (if the coefficient kf was small), and therefore, this simplified modification was used in
our numerical simulations. The flow potential gQ was also extended to the following format:

gQ ¼ σ max  kg ðσ med þ σ min Þ; (10)

where kg denotes the coefficient describing the change of a flow direction as compared with a standard
approach. When kf = kg, the associated constitutive law was obtained. The standard model was
recovered (Eqs 1–3) by assuming kf = kg = 0. The modification was not coupled with the Drucker–
Prager criterion, so it could be solely used in tension-dominated problems. The coefficient kf has to
be positive. Otherwise, a linear increase of the concrete strength in a tension–compression regime
occurs (which is not confirmed in biaxial experiments).

4. CONTINUUM DAMAGE MECHANICS LAW

The second constitutive law was defined within continuum isotropic damage mechanics. The initiation
and propagation of micro-cracks is described by a scalar damage parameter D. The virgin state with the
elastic stiffness corresponds to the parameter D = 0, while a completely cracked material is described
by D = 1. The damage variable D can be interpreted as an indicator of the mechanical degradation of
the material. It acts as a stiffness reduction factor:

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 411

σ ij ¼ ð1  DÞC eijkl εkl : (11)

Please note that isotropic damage causes a degradation of Young modulus only, while the Poisson
ratio remains not affected. The growth of the variable D is controlled by a monotonically increasing
damage history parameter κ, which is defined as the maximum of the equivalent strain measure eε
reached during the entire load history:

κðt Þ ¼ maxfeεðτ Þg: (12)


τ≤t

During loading, the parameter κ grows, and during unloading and reloading, it remains constant. The
equivalent strain eε is a scalar that takes into account the sensitivity of a degradation process to different
principal strain components. The chosen model reflects a tensile-dominated failure mechanism in
concrete-like materials. We used three different definitions of the equivalent strain eε : the modified
von Mises [43], the Rankine [44] and the one proposed by Häuβler-Combe and Pröchtel [45]. In the
literature, the definitions based on the energy release rate [46] or those proposed by Mazars [47] are
also frequently applied. It is worth mentioning that all these relationships in uniaxial tension reduce
to ε ¼ eε. Therefore, to describe the evolution of the damage variable D, an exponential softening law
was used for all definitions of the equivalent strain eε:

κ0  
D¼1 1  α þ αeβðκκ0 Þ ; (13)
κ
wherein κ0 is the initial value of the damage threshold parameter κ and α and β are the material
parameters. The parameters α and β are usually calibrated based on a force–displacement curve in
softening from uniaxial tension.

4.1. Modified von Mises equivalent strain


The modified von Mises definition was originally proposed by de Vree et al. [43]:
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s  ffi
k1 1 k1 ε 2 12k
eε ¼ Iε þ I þ J 2ε ; (14)
2kð1  2νÞ 1 2k 1  2ν 1 ð1 þ νÞ2

where k is the parameter and I ε1 and J ε2 are the first invariant of the strain tensor and second invariant of
the deviatoric strain tensor:

1 1  2
I 1ε ¼ ε11 þ ε22 þ ε33 ; J 2ε ¼ εij εij  I 1ε : (15)
2 6
The equivalent strain depends not only on the second invariant J ε2 (as in the original von Mises criterion
for elasto-plastic materials) but also on the first invariant I ε1 . Thus, it becomes closer to the Drucker–
Prager criterion. For uniaxial loading cases, the compressive stress kσ has the same influence on a
damage growth as the tensile stress σ. Therefore, k is usually defined as the ratio of the compressive
to tensile strength of concrete (Figure 2a). The model requires six parameters: Young modulus E,
Poisson ratio v, coefficient k, threshold value κ0 and constants α and β to describe a softening curve.
The threshold value is determined with the aid of the tensile strength ft. Note that in a biaxial
deformation state, an elliptical failure envelope is obtained with the overestimation of the concrete
strength under the biaxial tensile/compressive stress. For k = 10 and under the proportional loading
σ 1 > 0, σ 2 = 2σ 1 and σ 3 = 0, the maximum tensile strength is twice as high as the maximum
uniaxial tensile strength ft (after [11]). As a consequence, the shear resistance is non-physically
augmented. Although this definition does not reflect the failure envelope of concrete in the three-
dimensional context, it is used to simulate concrete beams during, for example point bending or
anti-symmetric four-point loading (as it was shown by Rodriguez-Ferran and Huerta [48] and
Simone et al. [49]).

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
412 J. BOBIŃSKI AND J. TEJCHMAN

Figure 2. Equivalent strain definition under plane stress conditions: (a) by modified von Mises (Eq. 14), (b)
by Rankine (Eq. 16) and (c) by Häuβler-Combe and Pröchtel (Eq. 18) (ticks are marked for values 1 and 10).

4.2. Rankine equivalent strain


Jirásek [44] proposed a definition equivalent to the Rankine criterion in plasticity (Figure 2b):
n o
max σ eff
i
eε ¼ ; (16)
E
eff
where σ i are the principal values of the effective stress tensor defined as

σ eff
ij ¼ C ijkl εkl :
e
(17)

A modified version with a smooth round-off in the region of multi-axial tension was also formulated
[44, 50].

4.3. Häuβler-Combe–Pröchtel equivalent strain


The most sophisticated definition used in the simulations was originally proposed by Häuβler-Combe
and Pröchtel [45]. It is based on the failure criterion by Hsieh, Ting and Chen [51] for concrete, which
describes fairly adequately the yield surface under plane stress conditions (Figure 2c). Later, it was
slightly reformulated by Jirásek [52]. The formula can be written in the following format:

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 413

 ffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 pffiffiffiffiffi  pffiffiffiffiεffi 2
eε ¼ c2 J 2ε þ c3 ε1 þ c4 I 1ε þ c2 J 2 þ c3 ε1 þ c4 I 1ε þ 4c1 J 2ε ; (18)
2
where ε1 is the maximum principal total strain and c1, c2, c3 and c4 are the coefficients depending on
the following parameters: α1 (ratio between uniaxial tensile strength ft and uniaxial compressive
strength fc), α2 (ratio between biaxial compressive strength fbc and uniaxial compressive strength fc)
and multipliers α3 and γ in triaxial compression (σ 1 = σ 2 = γ fc, σ 3 = α3 fc with α3 > γ). The
definition includes the modified von Mises and Rankine formulae as special cases. The model
requires nine parameters: Young modulus E, Poisson ratio v, coefficients α1, α2, α3 and γ to define
the equivalent strain, threshold value κ0 and constants α and β to describe a softening curve.

5. INTEGRAL NON-LOCAL THEORY

Standard constitutive laws are not able to describe properly the strain softening of the material when
using FEM. Because these models contain no information about the size and spacing of localized
zones, their enrichment by a characteristic length of micro-structure is necessary. A characteristic
length restores also the well-posedness of boundary value problems and makes the FE results mesh-
independent.
In the continuum constitutive laws described in the paper, this enrichment was performed with the
help of a non-local theory in the integral format (Pijaudier-Cabot and Bažant [46]), which is based on a
spatial weighted averaging of a local tensor or scalar state variables. The non-local averaging can be
applied only to internal variables coupled with a dissipative process and softening of materials. The
influence of the material neighbouring points in determining the state of the analyzed point is taken
into account. As a consequence, the principle of local action is not valid any more.
In all FE simulations based upon an isotropic damage model, the equivalent strain measure eε was
replaced by its non-local counterpart ε [46] in Eq. 12:

∫V ωðkx  ξ kÞeεðξ Þdξ


εðxÞ ¼ ; (19)
∫V ωðkx  ξ kÞdξ
where V denotes the body volume, x are the coordinates of the considered point, ξ are the coordinates
of the surrounding points and ω denotes the weighting function.
In the calculations within elasto-plasticity, the softening parameters κi (i = R, P for softening parameters
of the standard Rankine and Drucker–Prager criteria, Eqs 3 and 5 or i = Q for the modified Rankine
formulation, Eq. 9) were assumed to be non-local [53] with the so-called over-non-local formulation:

∫V ωðkx  ξ kÞκi ðξ Þdξ


κi ðxÞ ¼ ð1  mÞκi ðxÞ þ m ; (20)
∫V ωðkx  ξ kÞdξ
where κi(x) are the non-local softening parameters and m is the additional non-locality parameter. These
non-local parameters are used to define the tensile σ t or compression failure stress σ c. This averaging is
performed independently for both yield surfaces in the Rankine–Drucker–Prager model. For m = 0, a
local formula is retrieved, and for m = 1, a classical non-local model is recovered. For the condition
0 ≤ m ≤ 1, the model does not fully regularize a boundary value problem, and plastic strains localize in
a zero thickness zone. To create a plastic region with a finite size and to obtain mesh-independent
results, the parameter m has to be greater than 1 [53–55]. It was equal to m = 2 [55] in all numerical
simulations.
In the both approaches, a Gaussian function was chosen to calculate the weighing function:
   
1 r 2
ωðr Þ ¼ pffiffiffi exp  ; (21)
l π l
where r is the distance between two considered material points and l is the characteristic length of
micro-structure. The characteristic length l determines the size of localized zones (in plasticity, the

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
414 J. BOBIŃSKI AND J. TEJCHMAN

geometry of localized zones depends also on the non-locality parameter m). Although the Gaussian
function has an unbounded support (for any distance r, it returns a non-zero value), in practice, the
averaging is restricted to a small area around each material point (the influence of points at the
distance of r = 3l is only of 0.01%). The softening non-local parameters near boundaries were
calculated also with Eqs 19–21 (which satisfy the normalizing condition).
With a non-local definition of the softening parameter in elasto-plasticity, the consistency condition
in a Gauss point is no longer local. The non-local softening parameter depends on the unknown values
of local softening parameters (related to unknown plastic strains) in neighbouring points. To solve this
problem, extra sub-iterations are required [56]. The simplified approach determines the influence of
surrounding points by estimated values of local softening parameters from known total strains [53].
We used another option in order to calculate the influence of neighbouring points by taking the
values of local softening parameters from the previous global iteration [57]. The increment of the
non-local softening parameter can be calculated as
  ðk1Þ !
ðk Þ ðk Þ ∑j ω r ij Δκj Vj ðk1Þ
Δ κi ¼ Δ κi þm    Δκi ; (22)
∑j ω rij V j

where superscript (k) in the global iteration counter, i and j denote the integration point numbers, rij is
the distance between integration points i and j and Vj is the volume associated with the integration point
j. This approach ‘freezes’ the influence of neighbouring points and restores a local character of the
plasticity algorithm at the integration point level.
In damage mechanics, the implementation is simpler, because values of local equivalent strains
(quantities to be averaged) are calculated based on known total strains (at the beginning of the
iteration). The only extra cost is the calculation of these equivalent strains in all integration points in
the system before starting a loop over finite elements to determine stresses. In order to avoid this
extra loop over all elements (and integration points), the similar procedure as in elasto-plasticity was
used. The non-local equivalent strain was calculated as

  ðk1Þ
ðk Þ ωðrii ÞV i ðkÞ ∑j≠i ω rij eεj Vj
εi ¼   eεi þ   : (23)
∑j ω r ij V j ∑j ω rij V j

The 2D non-local models were implemented into the commercial finite element code ABAQUS
[38]. The user elements with non-local constitutive laws were defined with the aid of user element
subroutine. The independent FORTRAN module was written to handle the non-local data. For
the solution of a non-linear equation of motion governing the response of a system of finite
elements, the calculations were performed with a symmetric elastic global stiffness matrix. For
non-local elasto-plastic constitutive laws, the determination of a tangent stiffness matrix within
a non-local theory is virtually impossible. The use of a local tangent stiffness matrix resulted in
a poor convergence in a softening regime; therefore, the elastic matrix was preferred [53]. In
non-local damage mechanics, the derivation of a fully consistent non-local tangent stiffness
matrix was solely presented for models with a linear dependence between state variables and
local strains [50, 58]. However, the implementation of such a matrix is not a straightforward
task. When using commercial codes, it may be even impossible. Alternatively, the secant or
local tangent stiffness matrices may be also used [59]. The following convergence criteria were
assumed [38]:

rmax ≤ 0:01e
q and cmax ≤ 0:01Δumax ; (24)

where rmax is the largest residual out-of-balance force, e


q the average value of all elements and
externally applied forces in the model (so-called the spatial average force), cmax the largest
correction of the displacement and Δumax the largest change of the displacements in the
increment.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 415

6. EXTENDED FINITE ELEMENT METHOD

6.1. Displacement field


The method assumes that a displacement field is discontinuous across the crack just after its initiation.
It is based on the local partition of unity concept by Melenk and Babuška [60]. It enables adding ‘ad
hoc’ extra terms to a standard FE displacement field interpolation. These extra functions are
responsible for capturing displacement jumps. In XFEM, cracks do not have to be placed along
finite element edges; they may pass through elements without any need of remeshing. Because of
the local nature of enrichment, extra displacement degrees of freedom are introduced only in finite
elements cut by a discontinuity. They are gradually added to subsequent finite elements when a
crack propagates.
The formulation used in the paper follows (with some improvements and modifications) the original
concept presented by Wells and Sluys [61]. To describe jumps in the displacement field, the so-called
shifted-basis enrichment was used [62]. In a finite element, the displacement field u in the point x was
interpolated as

uðxÞ ¼ NðxÞa þ NðxÞΨðxÞb; (25)

where N contains the shape functions, a and b are the standard and enriched displacements in nodes,
respectively, and Ψ is the diagonal matrix defined as:
2 3
Ψ11 ðxÞ 0 …
6 7
ΨðxÞ ¼ 4 0 Ψ22 ðxÞ … 5 with ΨII ðxÞ ¼ ðΨ ðxÞ  Ψ ðxI ÞÞI; (26)
⋮ ⋮ ⋱

where xI are the node I coordinates, ψ is the generalized step function and I is the unit matrix (its size is
equal to the geometrical dimension of the problem, e.g. for plane problems is equal to 2). The function
ψ (also called a sign function) was defined as
(
1 x ∈ Ωþ
Ψ ðxÞ ¼ ; (27)
1 x ∈ Ω

where Ω+ and Ω– are the ‘positive’ and ‘negative’ parts of the body (finite element) separated by a
crack.
The shifted-basis enrichment simplifies the implementation of XFEM (two types of finite elements
exist only); moreover, in nodes, total displacements are equal to standard displacements a. The weak
form of equilibrium equations can be defined for XFEM with the shifted-basis enrichment. After
several transformations, the matrix format of discretized equations with unknowns displacements
(increments) a and b is obtained (the details of this derivation can be found in the paper by Zi and
Belytschko [62] and Tejchman and Bobiński [63]).

6.2. Constitutive relationships


Two constitutive laws were defined to describe the behaviour of a solid body with a cohesive crack. In
a bulk (un-cracked) continuum, a linear elastic relationship between stresses and strains was assumed.
On the discontinuity, a softening cohesive law was formulated, which links tractions t with
displacement jumps [[u]] and governs the crack behaviour. The following format of the loading
function was chosen:
f ð½½un ; κn Þ ¼ ½½un -κn ; (28)

with the history parameter κn equal to the maximum value of the normal displacement jump [[un]]
achieved during a loading history. The softening of the normal component of the traction vector tn
was described with

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
416 J. BOBIŃSKI AND J. TEJCHMAN

   
f df f t
t n ¼ Df f f exp κn t with Df ¼ 1  exp κn ; (29)
Gf Gf
where Gf is the fracture energy and df denotes the drop factor [64]. It is an additional parameter that
improves the convergence in a tension–compression transition of the traction tn. With increasing df,
the value of Df approaches 1, and the standard exponential softening formula is recovered. In a
compressive regime, the penalty stiffness in a normal direction was used. This penalty stiffness was
defined as d f f 2t =Gf (the derivative of tn in Eq. 29 with κn = 0). During unloading, the secant stiffness
was used with a return to the origin (damage format). In a tangential direction, for the sake of
simplicity, a linear relationship between the displacement jump [[us]] and traction ts was defined
with the stiffness TS. Alternatively, a decrease of the shear stiffness upon the normal crack opening
was assumed (linear dependence on the current tensile failure traction), which was close to the
exponential softening postulated by Wells in Sluys [61]:

tn
ts ¼ T S : (30)
ft

Summing up, the description of the bulk material requires two constants: Young modulus E and
Poisson ratio v. The discrete crack definition needs four parameters: tensile strength ft, fracture
energy Gf, drop factor df (numerical constant) and tangential stiffness TS. Usually, a uniaxial tension
test serves to calibrate the parameters ft and Gf. The introduction of a discontinuity line into a
cracked element requires a new integration scheme for calculating the strains, stresses, internal force
vector and stiffness matrix. In this paper, three integration points for three-node triangles were
defined [65]. In quadrilaterals, a total of 12 integration points were defined [66]. For both finite
elements, two integration points on a discontinuity line were also defined. A finite element with
displacement jumps was defined using the user element subroutine. The enriched displacement
(jumps) in nodes (vector b in Eq. 25) was defined by activating rotational degrees available in
ABAQUS. One-node user elements with no stiffness and zero force were defined at nodes with
imposed boundary conditions. They transmitted the information from the input file to the algorithm
about the defined displacement boundary conditions and prescribed displacements. The element
forces on these constrained degrees of freedom were not taken into account while calculating the
largest residual out-of-balance force rmax (Eq. 24), and displacement jump degrees of freedom were
not activated there.
The data about residuum forces and displacement corrections were gathered independently of
ABAQUS to calculate all quantities in the convergence criteria (Eq. 24). These values were taken
from user elements. It allowed for an independent control of the convergence process and detection
of converged iterations. To drive the convergence process in ABAQUS and to enable extending
cracks without starting a new increment, another user element with no stiffness was added. In the
iterations in which a new crack segment was added (after finding the converged solution), to not
allow ABAQUS from starting the new increment, a very large force vector was defined in order to
increase the largest residual out-of-balance force rmax. The first inequality of the convergence criteria
(Eq. 24-1) was not satisfied, and ABAQUS continued the iteration process within the same
increment with the updated configuration of the crack. In other iterations, this element did not
modify the value of rmax. An extra module was written to visualize crack patterns. The XFEM
simulations were four times faster on average than non-local calculations (for the same mesh).

6.3. Crack initiation and propagation


In order to activate (create or extend) a crack, the Rankine condition has to be fulfilled at least in one
integration point at the element at the front of the crack tip:

maxfσ 1 ; σ 2 ; σ 3 g > f t : (31)

It was assumed that a crack tip might be located at element edges only. The crack tip placement in a
finite element node was also forbidden (to simplify the numerical algorithm).

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 417

A very import issue is the determination of a crack propagation direction. Two groups of such
criteria can be distinguished: a local and global one (this classification comes originally, e.g. from
Dumstorff and Meschke [25]). The criteria are also called propagation and global tracking
algorithms [20]. The first approach to calculate this direction uses a local stress state around a crack
tip. The global methods [20] find this direction by solving the independently formulated problem
defined in the entire model. In the paper, we chose the first simpler way. The direction of the crack
extension was defined to be perpendicular to the direction of the maximum principal stress of the
stress vector σ tip at a crack tip. These stresses were calculated as the averaged values using the
so-called averaged stress criterion proposed originally by Wells and Sluys [61]:

σtip ¼ ∫V σwdV ; (32)

where V is the semi-circle domain at the front of the crack tip and w denotes the averaging function
!
1 r2
w ðr Þ ¼ exp  2 ; (33)
ð2π Þ3=2 l3av 2lav

where r is the distance from the crack tip and lav is the averaging length related to the size of finite
elements. Note that this averaging operation affects neighbouring points only, and it is not intended
to introduce an additional characteristic length of micro-structure into the constitutive model.

7. LABORATORY TESTS ON CONCRETE BY NOORU-MOHAMED

In the laboratory tests by Nooru-Mohamed [5], a square-shaped and double-edge notched concrete
specimen with different combinations of shear and tensile loads under the force/displacement control
was used. The length and height of the largest element were 200 mm, and the thickness was 50 mm
(Figure 3). Two notches with dimensions of 25 × 5 mm2 were located in the mid-points of vertical edges.
The loading was prescribed by rigid steel frames glued to concrete. Several deformation scenarios were
analyzed. In numerical simulations, the so-called experimental loading paths number ‘3’, ‘4’ or ‘6’ are
commonly reproduced. The most popular is the complex loading path number ‘4’, and it was also

Figure 3. Laboratory test by Nooru-Mohamed [5]: geometry and boundary conditions.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
418 J. BOBIŃSKI AND J. TEJCHMAN

simulated in the paper. During the analyzed load scenario, the horizontal shear force Ps was applied until it
reached a specified value, while the horizontal edges were free (vertical force P = 0). Then the shear force
remained constant, and the vertical tensile displacement was prescribed. Three different values of the shear
force were applied: Ps = 5 kN (path ‘4a’), Ps = 10 kN (path ‘4b’) and Ps = 27.5 kN (path ‘4c’) – the maximum
shear force carried by a concrete specimen. In the experiment, two curved cracks with an inclination
depending upon the shear force were obtained (for the small value of Ps, the cracks were almost
horizontal, and for the large value of Ps, the cracks were strongly curved) (Figure 4).
The force–displacement curves P–δ from experiments and numerical simulations were also
compared. The relative displacement δ was determined as an average value of two vertical
displacements measured between points L1 and L2 on the left side and between points R1 and R2 on
the right side (Figure 5a). To evaluate the quality of crack patterns, a measure called the crack height
hc was introduced. It was defined as a vertical distance between a horizontal line connecting notches
and the most far-distant point lying on crack lines (Figure 5b). In experiments, the height hc,
calculated as an average value for four cracks, was equal to 1.5 cm (range 1.1–2.0 cm), 3.5 cm (range
2.7–4.3 cm) and 5.3 cm (range 4.3–6.5 cm) for the shear forces of 5, 10 and 27.5 kN, respectively.

Figure 4. Laboratory test by Nooru-Mohamed [5]: crack pattern during different loading paths: (a) ‘4a’, (b)
path ‘4b’ and (c) path ‘4c’.

Figure 5. Laboratory test by Nooru-Mohamed [5]: (a) location of measure points on concrete specimen
(units in millimetre) and (b) definition of crack height hc.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 419

In all simulations under plane stress conditions, the following elastic constants were assumed: Young
modulus E = 32.8 GPa and Poisson ratio v = 0.2. The strength parameters, ft or κ0, were determined by a
trial-and-error method to obtain the best agreement with experimental peak forces (as it was reported by
some researchers, Cervera and Chiumenti [8] or Meschke and Dumstorff [26], the use of the original
value of the tensile strength resulted in overestimating the maximum capacities of specimens). The
same procedure was used to choose parameters describing softening. A characteristic length in
simulations with smeared crack models was equal to l = 2 mm, based mainly on experimental results
performed by Skarżynski and Tejchman [67] using the digital image correlation technique, in which
the width of a localized zone in notched concrete beams under three-point bending was determined to
be 5–6 mm. The calculated width of a localized zone with l = 2 mm was about 6 mm for the Rankine
plasticity and about 8 mm for the damage constitutive laws (thus, the smaller l should be assumed
with a damage approach). Two finite element meshes were assumed: with the 33′832 three-node
constant strain triangle elements and with the 16′825 four-node quad elements. The maximum size of
finite elements in a localized zone was not greater than 1.5 mm (for the effective regularization with
l = 2 mm). The same meshes were used also with XFEM simulations (although such fine meshes are
not usually required when applying XFEM [63]). The possibility of cracking at the top-right and
bottom-left corners was excluded by assuming a linear elastic relationship there (such undesired
outcomes were obtained in experiments and in numerical simulations [9, 13]).
In Section 8, the results of three different test loading paths (test ‘4a’, ‘4b’ and ‘4c’) were presented
for all constitutive models in Sections 5 and 6. The sensitivity of different numerical models to the
effect of the ratio between the shear and tensile loads was investigated. Note that in the literature,
only one test (mainly test ‘4b’) has been usually simulated as a validation benchmark (without
analyses of remaining two loading paths).

8. NUMERICAL CONTINUOUS FE RESULTS

8.1. Enhanced elasto-plasticity


The preliminary calculations showed that the Drucker–Prager criterion (Eqs 5–8) surprisingly had no
impact on the obtained results although shear deformation was initially prescribed in the laboratory
experiment. First, the FE simulations with the usual Rankine criterion (Eq. 3) enhanced by non-
locality with a characteristic length l were carried out. The tensile strength was equal to ft = 2.3 MPa,
parameter ku = 0.004 and non-locality parameter was m = 2 [53]. The obtained FE results are
presented in Figure 6 (the cracks are shown via the contours of the non-local softening parameter κR ).
In the case of the force–displacement diagrams, a very good agreement with the experimental
outcomes was obtained when the prescribed shear forces were 5 and 10 kN. A discrepancy was
obtained at the maximum shear force Ps = 23.5 kN (smaller than in the experiment Ps = 27.5 kN),
although a compressive tendency under tensile loading in the first part of the vertical loading was
properly simulated (Figure 6A). A crack propagation shape was satisfactorily reproduced in FE
analyses (Figure 6B). The distance hc, measured to the centre of a localized zone, was equal to 2.2,
3.3 and 5.2 cm for the small, medium and maximum shear forces, respectively. For the medium and
maximum shear forces, the results were very close to the mean experimental values: 3.5 and 5.3 cm,
respectively. A too high crack occurred for the shear force 5 kN only; the crack height hc was even
larger than the maximum experimental height of 2 cm. The only drawback of calculated cracks was
that they were too straight near the starting notches. The height of the cracks can be adjusted by
changing the parameter κu (which affects the amount of fracture energy). With κu = 0.002 (two times
smaller value than the initial one), the crack height hc was larger, that is, 2.3 and 3.6 cm for the shear
forces Ps = 5 and Ps = 10 kN. With κu = 0.008 (two times larger value than the initial one), the crack
height hc was smaller, that is, 1.9 and 2.9 cm for Ps = 5 and Ps = 10 kN. At the same time, the increase
of the parameter κu resulted in the larger peak values and more ductile force–displacement curves
(Figure 7).
Our FE results do not confirm some conclusions drawn by the other researchers. In the paper by
Pivonka et al. [14], only one horizontal crack for the loading paths ‘4a’ and ‘4b’ was obtained for

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
420 J. BOBIŃSKI AND J. TEJCHMAN

Figure 6. FE results within enhanced elasto-plastic model with Rankine criterion (Eq. 3): (A) calculated and
experimental force–displacement curves P–δ and (B) calculated smeared crack patterns for shear force Ps
equal to (a) 5 kN, (b) 10 kN and (c) 23.5 kN (dashed line – experimental cracks).

simulations performed with a similar model based on the coupling of the Rankine and Drucker–Prager
criterion. There are two possible explanations: the formulated model was not equipped by a
characteristic length (fracture energy concept was used) and a relatively coarse mesh was used
(element size was about 5 mm). One horizontal crack was also obtained for the path ‘4a’ by di
Prisco et al. [13] using a regularized gradient plasticity model. The probable reason for such
behaviour was the choice of a too large characteristic length l (the localized zone width was about
12 mm, twice as great as the obtained in our analyses). In their simulations of the loading path ‘4b’
(Ps = 10 kN), two straight cracks were properly reproduced (as in our simulations). By taking the
characteristic length of l = 4 mm in our model, solely one crack was also reproduced for Ps = 5 kN.
Figure 8 shows the crack trajectories for the shear force Ps = 5 kN and the characteristic length
of l = 3, l = 4 and l = 5 mm (to preserve the constant fracture energy, the parameter κu was
appropriately scaled). Only for the smallest characteristic length equal to l = 3 mm, two smeared

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 421

Figure 7. The force–displacement curves P–δ for different ultimate softening parameters κu and shear forces
Ps equal to (a) 5 kN and (b) 10 kN (enhanced elasto-plastic model with Rankine criterion by Eq. 3).

Figure 8. Calculated smeared crack patterns within enhanced elasto-plastic model with Rankine criterion
(Eq. 3) for shear force Ps = 5 kN and characteristic length of micro-structure equal to (a) 3 mm, (b) 4 mm
and (c) 5 mm (dotted line – experimental cracks).

cracks were obtained. Simulations with larger characteristic lengths indicated one localization
zone only.
Next, the simulations were performed with the modified Rankine criterion (Eqs 9 and 10) enhanced
by non-locality to describe better the experimental biaxial test envelope (Figures 9 and 10). It turned
out that the use of the positive coefficient kf provided improper results because usually one smeared
crack was formed and a negative vertical force was calculated at the end of the loading process even
for the smallest shear force equal to Ps = 5 kN. Figure 9 presents the crack patterns obtained with the
coefficients kf and kg for the shear force 10 kN. When assuming kf = 0.05 and kg = 0, two straight
cracks were generated with hc = 2.5 cm (Figure 9a). The crack height hc might be adjusted by
specifying kf = 0 and by changing the parameter kg. The smeared crack patterns for the shear force
10 kN and different values of coefficients kg are described in Figure 9b and 9c. The larger kg, the

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
422 J. BOBIŃSKI AND J. TEJCHMAN

Figure 9. The calculated smeared cracks for enhanced elasto-plastic model with modified Rankine criterion
(Eqs 9 and 10) and shear force Ps = 10 kN with different coefficients kf and kg: (a) kf = 0.05, kg = 0.0, (b)
kf = 0.0, kg = 0.05 and (c) kf = 0.0, kg = 0.05 (dashed line – experimental cracks).

higher was the height hc. The crack height was equal to 2.9 and 3.4 cm for the coefficient kg
defined as 0.05 and 0.05, respectively. Finally, the force–displacement curves for the different
kf and kg coefficients are presented in Figure 10. The change of the coefficient kf (with kg = 0)
had a large impact on results. The different values of kg (with kf = 0) almost did not alter the
calculated force–displacement curves. In conclusion, the proposed modification of the Rankine

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 423

Figure 10. The experimental force–displacement curves P–δ for enhanced elasto-plastic model with modi-
fied Rankine criterion (Eqs 9 and 10) and shear force Ps = 10 kN for different coefficients kf (a) and kg (b).

criterion (Eqs 9 and 10) in order to better capture the concrete failure surface in a tension–compression
regime did not improve the results.

8.2. Enhanced damage mechanics


Three series of the numerical calculations with the different equivalent strain were performed
(Figures 11–15).

8.2.1. Modified von Mises equivalent strain. The modified Mises definition by Eq. 14 was used
(Figure 11) with the following material parameters: κ0 = 7 × 105, α = 0.92, β = 250, k = 10 and
l = 2 mm to fit the experimental results. An almost perfect agreement was achieved for the shear
force Ps equal to 5 and 10 kN (Figure 11A). For the maximum shear force 29.0 kN (larger than in
the experiment), some differences occurred. The minimum vertical force was 4.66 kN, while in the
experiment the value 1.51 kN was obtained. Nevertheless, the compressive nature of the vertical
force was captured. The calculated smeared cracks (based on the contours of the non-local
parameter) were in general too curved; the crack height hc was 2.4, 3.8 and 6.1 cm (Figure 11B),
respectively (in experiments: 1.5, 3.5 and 5.3 cm). The crack height hc might be controlled in a very
limited range by changing the parameter k. For the shear force Ps = 10 kN and k = 5, the crack height
was hc = 3.5 cm, while for k = 25, the height was hc = 3.8 cm (the identical height as for k = 10). The
simulations with the shear force Ps = 5 kN and different characteristic lengths were performed to
detect crack patterns with only one horizontally oriented crack between notches. Even for large
characteristic lengths (e.g. l = 20 mm), two curved cracks were always obtained.

8.2.2. Rankine equivalent strain. The initial FE simulation results with Eq. 16 were not realistic
(κ0 = 6.5 × 105, α = 0.92, β = 50, l = 2 mm), because one horizontally oriented crack was calculated
only and a sudden drop of the force–displacement curve after the peak was observed (Figure 12A
and 12B). Two cracks solely formed for the maximum shear force (Figure 12B).

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
424 J. BOBIŃSKI AND J. TEJCHMAN

Figure 11. FE results within enhanced isotropic damage with modified von Mises equivalent strain (Eq. 14):
(A) calculated and experimental force–displacement curves P–δ and (B) calculated smeared crack patterns
for shear force Ps equal to (a) 5 kN, (b) 10 kN and (c) 29 kN (dashed line – experimental cracks).

Next, the FE calculations were repeated with l = 0.5 mm [68] (Figure 13). Each crack first goes almost
horizontally, and near the centre of the specimen curved part starts (Figure 13B). Very similar crack
trajectories were observed also by Grassl and Jirásek [6] with the use of a local Rankine criterion. The
application of a small characteristic length allows for obtaining two cracks in all simulations
(Figure 13B). The crack height is equal to 1.9 and 3.5 cm for the small and medium shear forces,
respectively. The agreement between numerical and experimental outcomes is only moderate; moreover,
some convergence problems occurred (Figure 13A). Because the characteristic length (l = 0.5 mm) was
too small with the respect to the finite element size, differences in results were noticeable; for
example, only one crack is created in simulations with the quad mesh and Ps = 5 kN. Using a very
fine mesh (element size not greater than 0.5 mm), l = 0.5 mm and Ps = 5 kN, one horizontally oriented
crack was again obtained.

8.2.3. Häuβler-Combe–Pröchtel equivalent strain. The formula proposed by Häuβler-Combe and


Pröchtel [45] (Eq. 18) was tested with the following parameters: α1 = 0.08, α2 = 1.16, α3 = 2.0, γ = 0.2,

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 425

Figure 12. FE results within enhanced isotropic damage with Rankine equivalent strain (Eq. 16): (A) calcu-
lated and experimental force–displacement curves P–δ and (B) calculated smeared crack patterns for shear
force Ps equal to (a) 5 kN, (b) 10 kN and (c) 20.5 kN (dashed line – experimental cracks).

κ0 = 6.0 × 105, α = 0.92 and β = 200 and l = 2 mm (Figure 14). In general, a similar behaviour as in
simulations with the Rankine equivalent strain happened. For Ps = 5 kN, only one crack developed,
and the simulation failed to converge shortly after the peak (Figure 14B). For larger values of Ps,
two relatively straight cracks were produced (Figure 14B). The crack height hc was equal to 2.2 and
3.6 cm for the medium and maximum shear forces, respectively (smaller than in the experiments).
The additional simulations with the smaller characteristic length l = 0.5 mm and β = 50 (the remaining
parameters remained unchanged) allowed for reproducing for triangular finite elements two cracks
with the crack height equal to 1.8 cm (Figure 15). However, only one crack was created in the same
problem with the quad mesh. This fact again indicates both the importance of a proper choice of a
characteristic length and a sufficiently fine mesh discretization.
The summary of all FE results with respect to the crack height hc and maximum shear forces Pmax
as compared with experiments is presented in Table I. In general, the FE results similar to the
experimental outcomes were obtained. The Rankine criterion in plasticity (Eq. 3) provided a too
small maximum shear force, but the computed crack pattern was realistic. The behaviour of

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
426 J. BOBIŃSKI AND J. TEJCHMAN

Figure 13. FE results within enhanced isotropic damage with Rankine equivalent strain (Eq. 16) with char-
acteristic length l = 0.5 mm: (A) calculated and experimental force–displacement curves P–δ and (B) calcu-
lated smeared crack patterns for shear force Ps – (a) 10 kN and coarse mesh (after [68]), (b) 5 kN and fine
mesh and (c) 10 kN and fine mesh (dashed line – experimental cracks).

isotropic damage models strongly depended upon its definition in a tension–compression regime. If
the material strength’s increase was not possible in this regime, a very brittle material behaviour
was obtained, and usually one crack developed for Ps = 5 kN only. If the unphysical strength’s
increase was allowed in this region (as in the von Mises equivalent strain formula (Eq. 14)), the
FE results were close to the experiments. An alternative option is the introduction of the
dependence of the isotropic stiffness reduction upon the tensile and compressive effective principal
stresses using the weight proposed by Lee and Fenves [41]. Unfortunately, this approach does not
always guarantee that the concrete tensile strength in the tension–compression regime is not
overestimated.

9. NUMERICAL RESULTS WITH XFEM

The following parameters were assumed: ft = 2.3 MPa (tensile strength), Gf = 75 N/m (fracture energy)
and lav = 5 mm (averaging length). The drop factor was taken as df = 104 and the stiffness in a tangential
direction as Ts = 0 (to avoid large tangential tractions and locking effects). In order to avoid a sudden
crack direction turn, the limit angle of the maximum direction change of 8o between segments was
imposed. A very good agreement for the force–displacement curves for the shear forces Ps = 5 and
Ps = 10 kN was achieved (Figure 16A). For the maximum shear force equal to Ps = 25.5 kN (smaller
than in the experiment), large discrepancies were obtained. The minimum vertical force was
0.92 kN (the experimental value was 1.51 kN), an unrealistic curve with two local minima, and a
re-hardening phase occurred. Despite it, a compressive character of the vertical force under tension

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 427

Figure 14. FE results within enhanced isotropic damage with Häuβler-Combe–Pröchtel equivalent strain
(Eq. 18): (A) calculated and experimental force–displacement curves P–δ and (B) calculated smeared crack
patterns for shear force Ps equal to (a) 5 kN, (b) 10 kN and (c) 19 kN (dashed line – experimental cracks).

was again properly reproduced. In all tests, two curved cracks were obtained with increasing curvature
with respect to the increasing shear force (Figure 16B). The crack height hc was equal to 3.1, 4.6 and
7.4 cm for the shear forces equal to 5, 10 and 25.5 kN (Figure 16B, Table II), respectively (in
experiments: 1.5, 3.5 and 5.3 cm). Thus, all cracks were too curved. The crack heights were even
larger than the maximum experimental values. Meschke and Dumstorff [26] used a global tracking
algorithm, but they obtained almost identical values of hc: 3.0, 4.5 and 7.6 cm for the loading paths
‘4a’, ‘4b’ and ‘4c’, respectively. Similar crack heights were also achieved by Gasser and Holzapfel
[24], 4.6 cm, and Cox [64], 4.3 cm, in the simulations of the loading path ‘4b’ at Ps = 10 kN. Only
the heights hc calculated by Unger et al. [27] with a minimum potential energy criterion were
smaller: 1.8 and 2.7 cm for the tests ‘4a’ and ‘4b’. The simulations with the modified relationship in
a tangential direction (Eq. 30) for large values of Ts (the order of 1012 N/m3) provided the same
crack patterns and maximum vertical force, but they produced at the same time the higher residual

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
428 J. BOBIŃSKI AND J. TEJCHMAN

Figure 15. The calculated and experimental force–displacement curves P–δ (a) and calculated smeared crack
pattern (b) within enhanced isotropic damage model with Häuβler-Combe–Pröchtel equivalent strain (Eq.
18) for shear force Ps = 5 kN and characteristic length l = 0.5 mm (dashed line – experimental cracks).

Table I. The calculated crack heights hc and maximum shear forces Psmax based on FE results and
experiment.
Crack height hc (cm)
Psmax
Numerical model Ps = 5 kN Ps = 10 kN Ps = Pmax (kN) (kN)
Plasticity 2.2 3.3 5.2 23.5
Damage (modified von Mises) 2.4 3.8 6.1 29.0
Damage (Rankine) 1.6 2.9 5.1 27.1
Damage (Häuβler-Combe–Pröchtel) One crack 2.2 3.6 19.0
Experiment 1.5 3.5 5.3 27.5

forces at the end of the loading test. For the smaller values of Ts, the results were similar to those
obtained without the tangential stiffness.
The influence of the fracture energy on force–displacement curves for shear forces equal to 5 and
10 kN are depicted in Figure 17. The larger the fracture energy, the larger are the maximum force
and ductility in a softening regime. The fracture energy influences slightly the crack pattern (the
smaller fracture energy, the larger is the crack height hc). With the shear force 5 kN, the crack height
hc was equal to 3.3, 3.1 and 3.1 cm for the fracture energy equal to 37.5, 75 and 150 N/m,
respectively. With the shear force 10 kN, the crack height hc was equal to 4.9, 4.6 and 4.5 cm for the
fracture energy of 37.5, 75 and 150 N/m, respectively.
In order to investigate the influence of the maximum direction change angle, some simulations for
the shear force Ps = 10 kN were repeated with the maximum direction change angle equal to 10°, 20°
and 30°. The crack height hc was identical for all cases and was equal to 4.6 cm (Figure 18). Some

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 429

Figure 16. Numerical results with extended FEM: (A) calculated and experimental force–displacement
curves P–δ and (B) calculated discrete crack patterns for shear force Ps equal to (a) 5 kN, (b) 10 kN and
(c) 25.5 kN (grey region – experimental cracks).

differences occurred in the final stage of a crack growth. For the larger allowable angles of a crack
direction change, some sudden turns occurred. These angles produced also (e.g. for 30°) a
re-hardening process in the final region of the force–displacement curve.
The mesh dependence of the crack trajectories was also examined. Two coarse finite element
meshes were defined: (1) with the 3′840 three-node constant strain triangle elements and (2) with
the 1′030 four-node quad elements. The maximum size of a finite element in the crack growth
region was 5 mm (the width of the notch). The simulations were performed for the shear force 10 kN
with the lack of the restriction on the maximum direction change angle and with the limit angle of
15°. The averaging length lav was 8 mm. For both approaches and both FE meshes, the crack height

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
430 J. BOBIŃSKI AND J. TEJCHMAN

Table II. The calculated crack heights hc and maximum shear forces Psmax based on XFEM and experiment.
Crack height hc (cm)
Psmax
XFEM Ps = 5 kN Ps = 10 kN Ps = Pmax (kN) (kN)

Our results 3.1 4.6 7.4 25.5


Cox [64] 4.6
Dumstorff and Meschke [25] 4.1
Gasser and Holzapfel [24] 4.6
Meschke and Dumstorff [26] 3.0 4.5 7.6 27.5
Unger et al. [27] 1.8 2.7
Experiment 1.5 3.5 5.3 27.5
XFEM, extended FEM.

Figure 17. The calculated force–displacement curves P–δ from extended FEM for different fracture energy
for shear force Ps: (a) 5 kN and (b) 10 kN.

Figure 18. Calculated discrete crack patterns from XFEM for shear force Ps = 10 kN and maximum direction
change angle equal to (a) 10°, (b) 20° and (c) 30° (grey region – experimental cracks).

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 431

hc was 4.5 cm (Figure 19). It was almost identical to the value of 4.6 cm obtained with the fine mesh.
The calculations without a limit angle suffered from sudden direction changes in the final stage of a
crack growth and a re-hardening phenomenon on the force–displacement diagrams (similar to the
results with the fine mesh).
Finally, Figure 20 shows the results with the different averaging length lav for the shear forces 5 and
10 kN, respectively. Note that in the simulations with lav = 0, Eq. 33 could not be used; the crack
growth direction was determined using the averaged stress vector in a finite element lying at the
front of the crack tip only (it was calculated as a weighted sum of stresses in all integration points in
this element). The crack patterns are similar except for the last part near the crack tip (Figure 20B
and 20C). The higher the averaging length, the less curved crack happens. These discrepancies have
almost no impact on force–displacement curves, except for the final part at the end of loading
(Figure 20A). With no stress averaging ahead the crack tip (lav = 0), the crack pattern is properly
reproduced, but obtained cracks are less smooth. These results confirm the proper choice of lav equal
to 5 mm (about four times the finite element size) following the suggestion by Wells and Sluys [61].
The summary of our and other XFEM results with respect to the crack height hc and maximum shear
forces Pmax as compared with experiments is presented in Table II.
In the next step, a coupled continuous and discontinuous model will be used [69, 70], because a
fracture process may be in general subdivided into two main stages: appearance of narrow regions
of intense deformation (including micro-cracks) and occurrence of macro-cracks. In order to
describe strain localization, which first appears, a continuum model with non-local softening will be
used (Rankine criterion), and next, discrete cracks will be captured by XFEM. A transition point

Figure 19. Calculated discrete crack patterns from extended FEM for coarse meshes of: (A) three-node
triangles and (B) four-node quads with (a) no limit on maximum direction change angle and (b) maximum
direction change angle equal to 15° (grey region – experimental cracks).

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
432 J. BOBIŃSKI AND J. TEJCHMAN

Figure 20. Numerical results with extended FEM: (A) calculated force–displacement curves P–δ for averag-
ing lengths lav for shear force Ps equal to (a) 5 kN and (b) 10 kN, (B) calculated discrete crack patterns for
shear force Ps = 5 kN and averaging length lav equal to (a) 0 mm, (b) 10 mm and (c) 15 mm and (C) calculated
discrete crack patterns for shear force Ps = 10 kN and averaging length lav equal to (a) 0 mm, (b) 10 mm and
(c) 15 mm (grey region – experimental cracks).

will be determined based on experiments on concrete beams using digital image correlation [71]. At
transition, energy dissipated during continuum mode will be transferred to a discrete crack [70].

10. CONCLUSIONS

The detailed comparison of numerical simulations of a concrete specimen under mixed-mode


conditions in the laboratory test by Nooru-Mohamed was presented. Two different continuum
constitutive laws defined within enhanced elasto-plasticity and enhanced continuum damage
mechanics were used to study the geometry of curved smeared cracks. A discrete crack propagation
was also simulated using XFEM.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 433

The enhanced elasto-plastic model with the Rankine criterion was found to be the most suitable for a
proper simulation of the experimental crack shape and curvature. The results produced with an
isotropic damage formulation strongly depended upon the equivalent strain definition. The
simulation crack results with the original Rankine and Häuβler-Combe–Pröchtel equivalent strain
were not able to properly reflect the experimental crack pattern. A very brittle behaviour was
calculated, and usually, only one crack was created for the shear force 5 kN. When using the von
Mises equivalent strain definition, the FE results were close to the experiments (this model,
however, includes the non-physical strength’s increase in a tensile–compressive regime).
The experimental force–displacement diagrams were properly reflected in the calculations by the
continuum elasto-plastic model with the Rankine criterion and isotropic damage constitutive laws
with the modified von Mises definition. The accordance with the experimental results was very good
for the small and medium shear forces and moderate for the maximum shear force. In the last case,
the calculated force was partially negative (compressive) in both approaches as in the experiments
although some remarkable quantitative differences occurred.
Numerical simulations showed that isotropic damage models were not able to produce realistic crack
patterns in the Nooru-Mohamed test and capture the outcomes from usual biaxial tests on concrete. The
definition of anisotropic constitutive laws seems to be the best method to simultaneously simulate these
two experiments
The results from numerical simulations with enhanced continuum constitutive laws emphasized the
importance of a proper choice of a characteristic length and finite element size. When assuming too
high values, even properly regularized constitutive models provided improper results. The use of
local constitutive models with coarse meshes might result in realistic outcomes (based on the FE
simulations by Grassl and Jirásek [6] with a local damage model and Rankine definition of the
equivalent strain), which were not confirmed with non-local approaches using a fine mesh.
The shape of discrete cracks from XFEM was in satisfactory agreement with experiments. However,
the calculated cracks were too curved (the crack heights were too high). A very good agreement for the
force–displacement curves for small and medium shear forces was achieved. For the maximum shear
force, a compressive character of the vertical force under tension was solely realistically reproduced.

ACKNOWLEDGEMENTS
Research work has been carried out as a part of the project ‘Innovative resources and effective methods of
safety improvement and durability of buildings and transport infrastructure in the sustainable development’
financed by the European Union (POIG.01.01.02-10-106/09-01) and the project ‘Experimental and numer-
ical analysis of coupled deterministic-statistical size effect in brittle materials’ financed by National
Research Centre NCN (UMO-2013/09/B/ST8/03598).
The calculations were carried out at the Academic Computer Centre in Gdańsk (TASK).

REFERENCES
1. Arrea M, Ingraffea AR. Mixed-mode crack propagation in mortar and concrete. Report No. 81-83, Department of
Structural Engineering, Cornel University, Ithaca, New York, 1982.
2. Schlangen E. Experimental and numerical analysis of fracture process in concrete, PhD Thesis, Delft University of
Technology, Delft, The Netherlands, 1993.
3. Carpinteri A, Valente S, Ferrara G, Melchiorri G. Is mode II fracture energy a real material property? Computers and
Structures 1993; 48(3):397–413.
4. Ferrara L. A contribution to the modelling of mixed mode fracture and shear transfer in plain and reinforced concrete.
PhD Thesis, Politechnico di Milano, Milano, Italy, 1998.
5. Nooru-Mohamed MB. Mixed-mode fracture of concrete: an experimental approach. PhD Thesis, Delft University of
Technology, Delft, The Netherlands, 1992.
6. Grassl P, Jirásek M. On mesh bias of local damage models for concrete, Proceedings of 5th International Conference
on Fracture Mechanics of Concrete and Concrete Structures, FraMCoS-5, Vail, Colorado, U.S.A., 2004.
7. Jirásek M, Grassl P. Evaluation of directional mesh bias in concrete fracture simulations using continuum damage
models. Engineering Fracture Mechanics 2008; 75(8):1921–1943.
8. Cervera M, Chiumenti M. Smeared crack approach: back to the original track. International Journal for Numerical
and Analytical Methods in Geomechanics 2006; 30(12):1173–1199.
9. Slobbe AT, Hendriks MAN, Rots JG. C1-continuous crack propagation for mixed-mode fracture problems, XII Inter-
national Conference on Computational Plasticity. Fundamentals and Applications COMPLAS XII, 2013.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
434 J. BOBIŃSKI AND J. TEJCHMAN

10. Ožbolt J, Reinhardt HW. Numerical study of mixed-mode fracture in concrete. International Journal of Fracture
2002; 118(2):145–161.
11. Patzak B, Jirásek M. Adaptive resolution of localized damage in quasi-brittle materials. Journal of Engineering
Mechanics ASCE 2004; 130(6):720–723.
12. Desmorat R, Gatuingt F, Ragueneau F. Nonlocal anisotropic damage model and related computational aspects for
quasi-brittle materials. Engineering Fracture Mechanics 2007; 74(10):1539–1560.
13. di Prisco M, Ferrara L, Meftah F, Pamin J, de Borst R, Mazars J, Reynouard JM. Mixed mode fracture in plain and
reinforced concrete: some results on benchmark tests. International Journal of Fracture 2000; 103(2):127–148.
14. Pivonka P, Ožbolt J, Lackner R, Mang HA. Comparative studies of 3D-constitutive models for concrete: application
to mixed-mode fracture. International Journal for Numerical Methods in Engineering 2004; 60(2):549–570.
15. Jefferson AD. Craft - a plastic-damage-contact model for concrete. II. Model implementation with implicit return-mapping
algorithm and consistent tangent matrix. International Journal of Solids and Structures 2003; 40(22):6001–6022.
16. Abu Al-Rub RK, Kim SM. Computational applications of a coupled plasticity-damage constitutive model for
simulating plain concrete fracture. Engineering Fracture Mechanics 2010; 77(10):1577–1603.
17. Ananiev S, Ožbolt J. Plastic-damage model for concrete in principal directions, Proceedings of 5th International
Conference on Fracture Mechanics in Concrete and Concrete Structures FraMCoS-5, 2004.
18. Wu JY, Xu SL. An augmented multicrack elastoplastic damage model for tensile cracking. International Journal of
Solids and Structures 2011; 48(18):2511–2528.
19. Lensa LN, Bittencourt E. Constitutive models for cohesive zones in mixed-mode fracture of plain concrete.
Engineering Fracture Mechanics 2009; 76(14):2281–2297.
20. Oliver J, Huespe AE, Samaniego E, Chaves EW. Continuum approach to the numerical simulation of material failure
in concrete. International Journal for Numerical and Analytical Methods in Geomechanics 2004; 28(7-8):609–632.
21. Dias-da-Costa D, Alfaiate J, Sluys LJ, Júlio E. A discrete strong discontinuity approach. Engineering Fracture
Mechanics 2009; 76(9):1176–1201.
22. Spencer BW, Shing PB. Rigid-plastic interface for an embedded crack. International Journal for Numerical Methods
in Engineering 2003; 56(14):2163–2182.
23. Feist C, Hofstetter G. An embedded strong discontinuity model for cracking of plain concrete. Computer Methods in
Applied Mechanics and Engineering 2006; 195(52):7115–7138.
24. Gasser TC, Holzapfel GA. 3D crack propagation in unreinforced concrete. A two-step algorithm for tracking 3D
crack paths. Computer Methods in Applied Mechanics and Engineering 2006; 195(37-40):5198–5219.
25. Dumstorff P, Meschke G. Crack propagation criteria in the framework of X-FEM-based structural analyses. Interna-
tional Journal for Numerical and Analytical Methods in Geomechanics 2007; 31(2):239–259.
26. Meschke G, Dumstorff P. Energy-based modeling of cohesive and cohesionless cracks via X-FEM. Computer
Methods in Applied Mechanics and Engineering 2007; 196(21-24):2338–2357.
27. Unger JF, Eckardt S, Könke C. Modelling of cohesive crack growth in concrete structures with the extended finite
element method. Computer Methods in Applied Mechanics and Engineering 2007; 196(41-44):4087–4100.
28. Dong Y, Wu S, Xu SS, Zhang Y, Fang S. Analysis of concrete fracture using a novel cohesive crack method. Applied
Mathematical Modelling 2010; 34(12):4219–4231.
29. Zhang H. Simulation of crack growth using cohesive crack method. Applied Mathematical Modelling 2010;
34(9):2508–2519.
30. Schlangen E, Garboczi EJ. New method for simulating fracture using an elastically uniform random geometry lattice.
International Journal of Engineering Science 1996; 34(10):1131–1144.
31. Schlangen E, Garboczi EJ. Fracture simulations of concrete using lattice models: computational aspects. Engineering
Fracture Mechanics 1997; 57(2-3):319–332.
32. Bolander JE Jr, Saito S. Fracture analyses using spring networks with random geometry. Engineering Fracture
Mechanics 1998; 61(5-6):569–591.
33. Chang CS, Wang TK, Sluys LJ, van Mier JGM. Fracture modeling using a microstructural mechanics approach - II.
Finite element analysis. Engineering Fracture Mechanics 2002; 69(17):1959–1976.
34. Delaplace A, Ibrahimbegovic A. Performance of time-stepping schemes for discrete models in fracture dynamic
analysis. International Journal for Numerical Methods in Engineering 2006; 65(9):1527–1544.
35. Cusatis G, Bazant ZP, Cedolin L. Confinement-shear lattice CSL model for fracture propagation in concrete.
Computer Methods in Applied Mechanics and Engineering 2006; 195(52):7154–7171.
36. Kozicki J, Tejchman J. Modelling of fracture process in concrete using a novel lattice model. Granular Matter 2008;
10(5):377–388.
37. Hordijk DA. Local approach to fatigue in concrete. PhD Thesis, Delft University of Technology, 1991.
38. Abaqus Theory Manual, Version 6.11, Dassault Systèmes, 1080 Main Street, Pawtucket, RI 02860–4847, U.S.A.,
2011.
39. Feenstra PH, de Borst R. A composite plasticity model for concrete. International Journal of Solids and Structures
1996; 33(5):707–730.
40. Meschke G. Consideration of aging of shotcrete in the context of a 3D viscoplastic material model. International
Journal for Numerical Methods in Engineering 1996; 39(19):3123–3143.
41. Lee J, Fenves GL. Plastic-damage model for cyclic loading of concrete structures. Journal of Engineering Mechanics
ASCE 1998; 124(8):892–900.
42. Menetrey P, Willam KJ. Triaxial failure criterion for concrete and its generalization. ACI Structural Journal 1995;
92(3):311–318.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag
CONTINUOUS AND DISCONTINUOUS CONSTITUTIVE MODELS 435

43. de Vree JHP, Brekelmans WAM, van Gils MAJ. Comparison of nonlocal approaches in continuum damage mechan-
ics. Computers and Structures 1995; 55(4):581–588.
44. Jirásek M. Non-local damage mechanics with application to concrete. Revue a française de génie civil 2004; 8
(5-6):683–707.
45. Häuβler-Combe U, Pröchtel P. Ein dreiaxiales Stoffgesetz für Betone mith normaler und hoher Festigkeit. Beton- und
Stahlbetonbau 2005; 100(1):52–62.
46. Pijaudier-Cabot G, Bazant ZP. Nonlocal damage theory. Journal of Engineering Mechanics ASCE 1987; 113
(10):1512–1533.
47. Mazars J. Application de la méanique de l’endommagement au comportement non linéaire et à la rupture du béton de
structure. PhD Thesis, University of Paris 6, 1984.
48. Rodriguez-Ferran A, Huerta A. Error estimation and adaptivity for nonlocal damage models. International Journal of
Solids and Structures 2000; 37(48-50):7501–7528.
49. Simone A, Wells GN, Sluys LJ. From continuous to discontinuous failure in a gradient-enhanced continuum damage
model. Computer Methods in Applied Mechanics and Engineering 2003; 192(41-42):4581–4607.
50. Jirásek M, Patzak B. Consistent tangent stiffness for nonlocal damage models. Computers and Structures 2002; 80
(14-15):1279–1293.
51. Hsieh SS, Ting EC, Chen WF. A plasticity-fracture model for concrete. International Journal of Solids and Struc-
tures 1982; 18(3):181–197.
52. Jirásek M. Modeling of Localized Inelastic Deformation. Lecture Notes: Prague, 2008.
53. Brinkgreve RB. Geomaterial models and numerical analysis of softening. PhD thesis, Delft University of Technol-
ogy, 1994.
54. Jirásek M, Rolshoven S. Comparison of integral-type nonlocal plasticity models for strain-softening materials. Inter-
national Journal of Engineering Science 2003; 41(13-14):1553–1602.
55. Bobiński J, Tejchman J. Numerical simulations of localization of deformation in quasi-brittle materials within non-
local softening plasticity. Computers and Concrete 2004; 1(4):433–455.
56. Strömberg L, Ristinmaa M. FE-formulation of nonlocal plasticity theory. Computer Methods in Applied Mechanics
and Engineering 1996; 136(1-2):127–144.
57. Rolshoven S. Nonlocal plasticity models for localized failure. PhD Thesis, École Polytechnique Fédérale de Lau-
sanne, 2003.
58. Borino G, Failla B, Parrinello F. A symmetric nonlocal damage theory. International Journal of Solids and Struc-
tures 2003; 40(13):3621–3645.
59. Rodriguez-Ferran A, Morata I, Huerta A. Efficient and reliable nonlocal damage models. Computer Methods in Ap-
plied Mechanics and Engineering 2004; 193(30-32):3431–3455.
60. Melenk JM, Babuška I. The partition of unity finite element method: basic theory and applications. Computer
Methods in Applied Mechanics and Engineering 1996; 139(1-4):289–314.
61. Wells GN, Sluys LJ. A new method for modelling cohesive cracks using finite elements. International Journal for
Numerical Methods in Engineering 2001; 50(12):2667–2682.
62. Zi G, Belytschko T. New crack-tip elements for XFEM and applications to cohesive cracks. International Journal for
Numerical Methods in Engineering 2003; 57(15):2221–2240.
63. Tejchman J, Bobiński J. Continuous and Discontinuous Modelling of Fracture in Concrete Using FEM. Springer:
Berlin-Heidelberg, 2013.
64. Cox JV. An extended finite element method with analytical enrichment for cohesive crack modelling. International
Journal for Numerical Methods in Engineering 2009; 78(1):48–83.
65. Asferg JL, Poulsen PN, Nielsen LO. A consistent partly cracked XFEM element for cohesive crack growth. Interna-
tional Journal for Numerical Methods in Engineering 2007; 72(4):464–485.
66. Dumstorff P. Modellierung und numerische Simulation von Rissfortschritt in spröden und quasi-spröden Materialien
auf Basis der Extended Finite Element Method. Dissertation, Ruhr-Universitat Bochum, 2006.
67. Skarżyński Ł, Tejchman J. Calculations of fracture process zones on meso-scale in notched concrete beams subjected
to three-point bending. European Journal of Mechanics A/Solids 2010; 29(4):746–760.
68. Bobiński J, Tejchman J. FE-modelling of strain localization in concrete under mixed mode conditions using different
approaches, Proceedings of the 1st International Symposium on Computational Mechanics (ComGeo I), Juan-les-
Pins, France, 2009.
69. Comi C, Mariani S, Perego U. An extended FE strategy for transition from continuum damage to mode I cohesive
crack propagation. International Journal for Numerical and Analytical Methods in Geomechanics 2007; 31
(2):213–238.
70. Bobiński J, Tejchman J. A constitutive model for concrete based on continuum theory with non-local softening
coupled and extended Finite Element Method. Computational Modelling of Concrete Structures, Proceedings of
the Euro-C 2014 Conference, St. Anton am Arlberg, Austria, 2014.
71. Skarżyński Ł, Tejchman J. Experimental investigations of fracture process using DIC in plain and reinforced concrete
beams under bending. Strain 2013; 49(6):521–543.

Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2016; 40:406–435
DOI: 10.1002/nag

Das könnte Ihnen auch gefallen