Localizing Gradient Damage Model With Decreasing Interactions

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
Int. J. Numer. Meth. Engng (2016)

Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/nme.5364
Localizing gradient damage model with decreasing interactions
Leong Hien Poh*,† and Gang Sun

Department of Civil and Environmental Engineering, National University of Singapore, 1 Engineering Drive 2,
E1A-07-03, S117576, Singapore
SUMMARY
Nonlocal integral and/or gradient enhancements are widely used to resolve the mesh dependency issue with
standard continuum damage models. However, it is reported that whereas the structural response is mesh
independent, a spurious damage growth is observed. Accordingly, a class of modified nonlocal enhancements
is developed in literature, where the interaction domain increases with damage. In this contribution, we
adopt a contrary view that the interaction domain decreases with damage. This is motivated by the fact that
the fracture of quasi-brittle materials typically starts as a diffuse network of microcracks, before localizing
into a macroscopic crack. To ensure thermodynamics consistency, the micromorphic theory is adopted in
the model development. The ensuing microforce balance resembles closely the Helmholtz expression in a
conventional gradient damage model. The superior performance of the localizing gradient damage model is
demonstrated through a one-dimensional problem, as well as mode I and II failure in plane deformation. For
all three cases, a localized deformation band at material failure is obtained. Copyright © 2016 John Wiley
& Sons, Ltd.
Received 23 March 2016; Revised 15 July 2016; Accepted 22 August 2016
KEY WORDS: damage; frature; finite element method; gradient damage; localization; micromorphic
continua
1. INTRODUCTION
It is well reported that standard continuum models predict mesh dependent solutions during strain
softening. In the limit of vanishing element size, material failure occurs without any dissipation.
A widely adopted regularization technique is to incorporate into the model the effect of interac-
tions between underlying micro-processes that drive the failure mechanism, either through nonlocal
integral or gradient enhancements. In the context of damage mechanics for quasi-brittle materials,
whereas the structural responses are now objective through these enriched formulations, the conven-
tional nonlocal enhancement suffers from spurious damage growth. Consequently, a large diffused
process zone is obtained at complete failure, instead of a localized macroscopic crack. In this paper,
a localizing gradient damage model with decreasing nonlocal interactions is formulated based on the
micromorphic framework, which fully regularizes the softening behavior, that is, mesh independent
structural response with a correct description of the failure process zone.
Nonlocal integral enhancements for quasi-brittle materials are motivated by the interactions
between microcracks within the fracture process zone [1]. To account for the influence of such
interactions, the weighted average of a variable driving the damage process is incorporated into the
constitutive relations. The interaction domain thus introduces a length scale parameter into the for-
mulation, which serves as a localization limiter during strain softening [2–5]. Another broad class of
regularization technique is the so-called ‘implicit’ gradient enhancement, where an enriched variable
*Correspondence to: L. H. Poh, Department of Civil and Environmental Engineering, National University of Singapore,
1 Engineering Drive 2, E1A-07-03, S117576, Singapore.
† E-mail: leonghien@nus.edu.sg
Copyright © 2016 John Wiley & Sons, Ltd.

L. H. POH AND G. SUN
and its gradient characterizing the damage micro-process are incorporated into the formulation [6].
Mathematically, the implicit gradient enhancement can be considered as a reformulation of the inte-
gral approach, with its length scale parameter associated with the higher-order term characterizing
the interaction domain via a Helmholtz expression [7].
The conventional nonlocal integral and gradient enhancements assume a constant interaction
domain throughout the entire load history. While the structural softening response is made mesh
independent, the material exhibits spurious damage growth. This was first reported in [8] for gra-
dient enhanced models, where the damaged region of a bar in tension was found to expand with
deformation, instead of localizing into a macroscopic crack. The deficiency of gradient enhance-
ments is further exposed in [9], where the damage process zone initiates and propagates erroneously
in mode I fracture and shear band problems. Because the integral approach shares a similar regular-
ization mechanism, it suffers from the same issues. Such limitations arise because the assumption
of a constant interaction domain enables the transfer of energy from the damage process zone to
a neighbouring elastically unloading region. A localized deformation is thus precluded because of
this ‘damage diffusion’ process.
Several suggestions have been proposed in literature to avoid spurious damage growth at com-
plete material failure. A simple remedy is to adopt the weighted average of both local and nonlocal
variables to characterize the damage process. It is found that for many models, a localized deforma-
tion is recovered if the weight of the nonlocal variable is set larger than unity, hence, the so-called
‘over-nonlocal’ formulation with the integral [10, 11] or gradient enhancement [12–15]. Instead of
a constant weight factor, this idea has been extended by utilizing an evolving weight dependent on
the damage or deformation level [16, 17].
The concept of an evolving length scale parameter to prevent the smearing of failure process(es)
onto neighbouring undamaged region(s) was proposed for gradient damage models in [8]. There-
with, the gradient activity in the Helmholtz expression of gradient formulations is given as a function
of the local equivalent strain. Consequently, nonlocal interaction is activated only in region(s) of
large deformation (hence damage), and vanishes in the elastically unloading neighbourhood. The
original formulation of the strain-based transient gradient damage model requires an extra continuity
equation in its implementation, but which has been reformulated to avoid this additional require-
ment [18]. That the length scale parameter increases with damage is supported by micromechanical
arguments and acoustic emission analyses [19]. Moreover, for a different class of gradient damage
models, it was shown that the length scale parameter, if made dependent on damage, has to increase
with damage in order to satisfy the thermodynamics requirement [20]. It is also highlighted that
attempts to induce a localized deformation in gradient formulations through a decreasing length
scale parameter, such that the nonlocal variable approaches its local counterpart at high damage
level, cannot regularize the softening response effectively [21]. In the context of integral formula-
tion, an evolving interaction intensity based on the principal stress state is found to resolve well
the limitations of the conventional approach [22]. This has been reformulated as a stress-based
gradient damage model in [23], and extended to incorporate fourth-order terms in the Helmholtz
expression by [24]. Because the length scale parameter(s) vanishes as the principal stress(es) tends
towards zero, some forms of mesh dependency are observed during strain softening in [24]. Possible
remedies are suggested in [23].
In this paper, we focus on the development of a gradient damage model that addresses the
aforementioned limitations of conventional nonlocal enhancements. To this end, the micromorphic
framework in [25, 26] is adopted. The central idea behind the generalized continuum theory is to
embed a micro-continuum at each macroscopic point. The rapidly fluctuating response(s) within
each micro-continuum is next linearized and characterized through a ‘morphic’ variable and its
gradient. The net influence of the rapidly fluctuating micro-processes, otherwise not adequately
accounted for in a standard model, is incorporated directly into the thermodynamics framework
through the morphic variable. In addition to the standard equilibrium condition, a micro-force bal-
ance is obtained from the extended virtual work statement, which governs the interactions between
micro-processes and the coupling between micro and macro. Note that this micro-force balance
recovers the Helmholtz expression in a conventional implicit gradient formulation.
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
LOCALIZING GRADIENT DAMAGE MODEL WITH DECREASING INTERACTIONS
The conventional micromorphic model assumes a constant interaction domain for the underlying
micro-processes. Recognizing that the process zone bandwidth localizes towards a macroscopic
crack in quasi-brittle fracture, we adopt the view that interactions between micro-processes decrease
with deformation – a departure from most modified nonlocal models in literature where the length
scale parameter increases with deformation. The proposed phenomenological enhancement is also
motivated by the homogenization theory for intergranular fracture in [27], where it is shown that a
gradient damage model with decreasing nonlocal interactions is able to fully regularize the material
softening response.
In the following, we first discuss on the micromorphic thermodynamics framework in Section 2,
where interactions between micro-processes decrease with damage. The resulting model bears a
close resemblance to the conventional gradient model, except for an interaction function introduced
to effect a localizing process zone. The full regularizing capability of the proposed gradient damage
model is demonstrated through a one-dimensional problem in Section 3. Its superior performance
over the conventional enhancement is further illustrated in Sections 4 and 5, through the mode I
fracture and shear band failure problems discussed in [9]. The numerical framework of the pro-
posed localizing gradient damage model, summarized in Appendix A, has the same structure as a
conventional gradient enhancement and does not require any additional numerical treatment.
2. THERMODYNAMICS FRAMEWORK
During strain softening, the nucleation and propagation of microcracks result in rapidly fluctuating
material responses within the failure process zone. Standard continuum damage models assume a
uniform field within each unit volume of material. Consequently, they cannot adequately account
for the interactions between microcracks and/or the surrounding bulk material – the source of
nonlocality in an enriched continuum damage theory [1].
The evolution of the failure process zone is illustrated schematically in Figure 1. The initial stage
(a) is characterized by a diffuse network of microcracks. As strain softening proceeds to stage (b),
the propagation of microcracks is confined to a narrower band, beyond which the elastic unloading
of bulk material leads to the closure of microcracks therewith. As deformation proceeds, coales-
cence of microcracks occurs rapidly, with a corresponding decrease in the process zone bandwidth.
Towards the final stage of strain softening at point (c), the process zone is constrained to a very nar-
row band, where a localized macroscopic crack is formed. Such an evolution of the process zone –
for example, observed in concrete experiments [28, 29] and meso-scale simulations [30] – motivates
the localizing gradient damage formulation in this section.
Figure 1. Interaction between microcracks at different loading stages. Intensity of solid lines within each
micro-continuum represents the extent of crack openings. Dotted lines indicate crack closure.
DOI: 10.1002/nme
In the following, we adopt the generalized micromorphic framework in [26] to develop a thermo-
dynamically consistent gradient damage model. Apart from the standard kinematic fields associated
with the macroscopic deformation, the framework introduces additional morphic variable(s) to
characterize the rapidly fluctuating responses in the micro-continuum embedded at each macro-
scopic point. The micro-macro interactions and their overall effect at the macro-level, otherwise
neglected in a standard continuum model, are thus naturally accounted for through the micromorphic
formulation.
At each macroscopic point, an equivalent strain eQ is introduced as the morphic variable that
characterizes the microscopic deformation underlying the embedded micro-continuum. In [27],
the morphic variable is defined as the continuum description of the intergranular cracks smeared
over a unit cell. Apart from the standard stress power contribution, the internal power is appended
with additional terms to account for the work performed arising from interactions between micro-
processes, as well as their propagation onto the macro-scale, through ePQ and its gradient. For a
P the internal power expended by the material within an arbitrary region R is thus
given strain-rate ",
postulated as
Z
Pint D W "P C Q ePQ C e
r ePQ dV ; (1)
R
where , Q , and e
are the stress conjugates to the respective kinematic fields.
The external power, ignoring body forces, is given as
Z
Pext D t uP C ePQ dA ; (2)
@R
where u denotes displacement, and t, , the traction and higher-order traction, respectively.
Applying the divergence theorem and requiring that Pint D Pext for any arbitrary region, we
obtain the standard equilibrium condition
r D0; (3)
as well as a microforce balance governing the micro–macro interaction
Q D r e
: (4)
At the outer surface with outward normal n, the tractions are determined as
tD n ; D Q n : (5)
We consider only isothermal linear elasticity deformation, where a scalar damage variable (!)
describes the degradation process. The macroscopic deformation state is characterized with an
equivalent strain variable e, the definition of which will be provided later. As discussed ear-
lier, the morphic variable eQ characterizes the rapidly fluctuating micro-processes underlying each
macroscopic point, and through which their interactions and propagation onto the macro-scale are
captured. Against this backdrop, the free energy density is assumed as
Q 2C
D 12 .1 !/" W C W " C 12 h.e e/ 1
2
ghl 2 r eQ r eQ : (6)
The first term with C as the fourth-order elasticity tensor is standard in elasticity-based damage
mechanics. The remaining (non-standard) terms account for the additional contributions arising
from the rapidly fluctuating micro-processes. Specifically, the second term captures the micro–
macro interaction with h as the coupling modulus. Note that the difference in equivalent strains at
the macro-scales and micro-scales (e and e,
Q respectively) characterizes the extent of non-uniformity
in material responses because of the underlying failure mechanisms. In a uniform deformation,
e D eQ such that the coupling term vanishes. The last term in (6) accounts for the linearized inter-
actions between micro-cracks through r e,Q with the size of micro-continuum characterized through
the length scale parameter l. Recall from Figure 1 that the failure process zone decreases with the
DOI: 10.1002/nme
propagation and coalescence of microcracks. Departing from the standard micromorphic framework
in [26], we introduce an interaction function g to account for the decreasing nonlocal interactions
between active microcracks, such that
²
1 ; !D0
g D g.!/ D : (7)
R ; !!1
Note that the continuum description of a macroscopic crack is a very narrow band of localized
deformation. Hence, R 0 describes the residual interaction between micro-processes within this
band.
Following the Coleman-Noll procedure, the constitutive relations are obtained as
@
D D .1 !/C W " C h.e e/N
Q (8)
@"
@
Q D D h.eQ e/ (9)
@eQ
@
Q D D ghl 2 r eQ ; (10)
@r eQ
@e
where N D @" . The constitutive relation for in (8) consists of a classical term, augmented with
a coupling stress tensor. Similarly, an equivalent coupling stress Q is obtained in (9) because of the
interaction across the two scales, whereas the moment stress e defined in (10) arises because of the
interactions between microcracks in the underlying process zone.
The dissipation inequality is thus given as

1 @g 2
D D Pint P D "WC W" hl r eQ r eQ !P D Y !P > 0 : (11)
2 @!
@g
Note that @! < 0, following (7). Because the damage energy release rate Y > 0, together with the
usual requirement that damage can only increase, that is, !P > 0, the dissipation inequality is always
satisfied.
For the special case where energetic interactions between micro-cracks is taken to occur over a
constant domain throughout the failure process .g D 1/, the micromorphic damage model in [26]
is recovered. The ensuing formulation then closely resembles the conventional implicit gradient
damage model in [6], except for an additional coupling stress contribution in the constitutive relation
for in (8). Note that the same damage model as [26] was developed in [31] by requiring that the
second law of thermodynamics be satisfied only in a global sense. Therewith, it was furthermore
shown that the conventional implicit gradient damage model is recovered by having a low coupling
modulus h. A larger value of h, however, results in a high residual stress as ! ! 1, because the
second term in (8) does not degrade rapidly with damage. Based on the discussions in [31], a low
coupling modulus h is adopted in this paper so that (8) resembles the conventional constitutive law
and also to avoid the stress locking phenomenon.
2.1. Damage characterisation and evolution

Following [32], the macroscopic deformation is characterized through a modified von Mises
equivalent strain in this paper, given by
0:5
k1 1 .k 1/2 2 2k
eD I1 C I C J2 ; (12)
2k.1 2/ 2k .1 2/2 1 .1 C /2
where D Poisson’s ratio and k denotes the ratio between the compressive and tensile strengths.
The invariants are defined as
I1 D t r."/ ; (13)
DOI: 10.1002/nme
J2 D 3t r." "/ t r 2 ."/ : (14)
To ensure numerical stability towards the end failure stage, a widely used exponential damage
evolution law is adopted,
0
! D1 ¹1 ˛ C ˛ expŒˇ. 0 /º ; (15)

where ˛ and ˇ are material parameters, 0 the damage initiation threshold, and a history parameter
given by
.t / D max¹e.
Q / j 0 6 6 tº :
The interaction function g is assumed to decrease exponentially with damage. To satisfy (7), it is
postulated as
.1 R/exp.!/ C R exp./
gD ; (16)
1 exp./
where is a material parameter. Throughout this paper, we adopt D 5 and R D 0:005.
2.2. Damage-based decreasing nonlocal interaction

Recall that a microforce balance in the current formulation governs the interactions between two
different stress quantities arising from the underlying micro-processes. Substituting the constitutive
relations in (9) and (10) into (4), we obtain

eQ e D r gl 2 r eQ ; (17)
which has a similar form to the Helmholtz expression in a conventional gradient damage model
eQ e D l 2 r 2 eQ : (18)
The microforce balance (17) departs from the conventional expression (18) in that its nonlocal
interaction evolves and vanishes with damage through (16).
The motivation for a damage-based localizing interaction behavior can be elaborated further
by considering the higher-order traction in (5b). In a continuum model, a macroscopic crack is
described as a narrow band of highly damaged material. Referring to the constitutive relation in
(10), 0 towards the end stage of loading in regions where ! ! 1, through the interaction func-
tion in (16). Hence, any interactions across a continuum description of a macroscopic crack, through
, is severely constrained – a desirable outcome without transiting to discontinuous formulations
where cracks are inserted explicitly, for example, with the discontinuous Galerkin framework [33],
remeshing techniques [34] or partition of unity based interpolation functions [35–37].
As discussed in Section 2, the conventional gradient damage model is recovered by setting g D 1
and h E (elastic modulus). Because it assumes a constant interaction domain, the higher-order
traction does not vanish in the highly damaged regions. Such non-physical energy exchanges
across the continuum description of macroscopic cracks result in spurious damage growth and
wrong failure mode as reported in [9], which will also be demonstrated in the following sections.
2.3. Comparison with phase-field damage model

The different mechanisms between the proposed localizing gradient enhancement and phase-field
models are briefly highlighted here. The basic principle of a phase-field model is to regularize a
discontinuity (crack) into a small, but finite bandwidth with sharp gradients. Accordingly, a very
small length scale parameter is usually adopted. The thermodynamics framework for phase-field
DOI: 10.1002/nme
models has been presented in [38]. In a phase field model, the regularized damage variable (dN ) has
to satisfy the following expression

dN f dN ; Gc D l 2 r 2 dN ; (19)
where f .dN ; Gc / is a thermodynamic force driving the damage process defined in terms of dN and
fracture energy Gc . Note that in a conventional gradient enhancement, the source term e in (18)
increases with damage. In contrast, the driving force f .dN ; Gc / in (19) vanishes as dN ! 1. This
is the main mechanism that avoids spurious damage growth in a phase field model [39], which is
easily observed if (19) is written in the rate form to give
P
dPN f d ; Gc D l 2 r 2 dPN : (20)
Because the driving force rate approaches zero in highly damaged regions, the corresponding dam-
age rate vanishes therewith, hence preventing further spreading of the damage process zone [40]. On
the contrary, the proposed localizing gradient enhancement achieves a similar effect in highly dam-
aged regions by having a vanishing interaction domain therewith, characterized through the higher
order term in (17).
3. ONE-DIMENSIONAL LOCALIZATION BEHAVIOR
This section investigates the performance of the damage-based interaction model as compared with
the conventional model, by considering the localization behavior in a one-dimensional bar. Because
the proposed model resembles closely the conventional gradient approach, its numerical frame-
work has a similar structure as that presented in [6] for the conventional model. The numerical
implementation is thus not elaborated but briefly summarized in Appendix A.
The bar, as depicted in Figure 2, is of length L D 0:1 m with a uniform cross sectional area.
To trigger localization, a defect region of length 0.01 m is introduced. The material parameters are
Young’s modulus E D 100 MPa and coupling modulus h D E 109 . The length scale parameter
is taken as l D L=30.
The material is assumed to fail only in tension; hence, the equivalent strain measure becomes
e D j"j ; (21)
which is achieved in (12) by having k ! 1. The damage initiation threshold in (15) is taken as
0 D 0:0001, with a 10% reduction in defect region.
The problem is first solved with the conventional gradient damage model in [6], with ˛ D 0:99
and ˇ D 180 for the evolution law in (15). The same parameters are adopted for the localizing
gradient damage model, except for ˇ which is calibrated as ˇ D 15, so that its structural response
matches that of the conventional model as shown in Figure 3a. The mesh convergence study depicted
in Figure 3b demonstrates the objectivity of the localizing gradient enhancement.
Despite the similar structural responses depicted in Figure 3a, the failure mechanisms underlying
the two models differ significantly. For the conventional gradient model, an initial localized damage
profile over the defect region is observed in Figure 4b. This induces a corresponding localized strain
profile e, which is next smoothened out over an interaction domain characterized by l through the
implicit gradient expression in (18). Accordingly, the nonlocal strain eQ evolves beyond the defect
Figure 2. A one-dimensional bar with imperfection in the shaded region.
DOI: 10.1002/nme
Figure 3. (a) Structural response with conventional and localizing gradient damage models. (b) Mesh
convergence with the localizing gradient damage model.
Figure 4. Evolution of (a) eQ and (b) ! with the conventional gradient model.
Figure 5. Evolution of (a) eQ and (b) ! with the localizing gradient model.
region, as depicted in Figure 4a, which in turn provokes further damage growth beyond the initially
localized band. This local-nonlocal interaction over a constant domain is the cause of the non-
physical phenomenon in Figure 4b, where the damaged region expands with deformation, instead
of localizing into a macroscopic crack at complete material failure.
Note that the spurious damage growth with conventional gradient models was first illustrated
analytically in [41] via a Fourier transform of the Helmholtz expression in (18), where the nonlocal
DOI: 10.1002/nme
strain (hence damage !) is shown to spread over the entire bar as deformation proceeds. To resolve
this problem, the constant length scale parameter in (18) is replaced with a strain-based transient
gradient activity parameter, such that the nonlocal interaction increases in damaged regions [8].
As discussed in Section 1, many nonlocal models with an evolving length-scale parameter are also
motivated by an interaction domain that increases with damage. Conversely, adopting a length scale
parameter that decreases with damage in (18) results in mesh sensitive solutions [21].
The current work departs from existing models in that its nonlocal interaction is assumed to
decrease with damage. The full regularizing capability of the damage-based interaction model was
partially demonstrated through the mesh sensitivity study in Figure 3b, and which will be confirmed
below from its damage evolution. Similar to the conventional model, an initial localized damage
profile over the defect region is observed in Figure 5b, which results in a corresponding localized
strain profile e. As discussed in Figure 1, the coalescence and propagation of microcracks charac-
terized by ! occurs in a narrowing band of micro-processes, which is captured in the model through
a decreasing value of g in the microforce balance (17). This reduced interaction between micro
and macro strains (e,Q e) thus serves to circumvent the smoothening effect in (18), as revealed by
the localized evolution of eQ in Figure 5a. Note also that the values for eQ in Figure 5a are an order
larger than those in Figure 4a, hence further illustrating the localized nature of deformation. Con-
sequently, the damage evolution in Figure 5b is confined to a bandwidth that decreases with !,
consistent with the notion of a narrowing band of active micro-cracks. As deformation proceeds,
full damage (! ! 1) is achieved only within a very narrow region, hence describing the formation
of a macroscopic crack correctly.
It is highlighted that both conventional and localizing gradient enhancements are phenomeno-
logical in nature. Accordingly, their calibrated material parameters need not be identical. As an
illustration, we consider ˇ D 15 for the conventional model, that is the same value as adopted in the
localizing gradient model. As shown in Figure 3a, its structural response differs significantly from
the reference solution – hardly a surprise because there is no reason to adopt the same parameters
for two different phenomenological models.
4. FAILURE IN MODE I PROBLEMS
This section considers the failure of a compact tension specimen (CTS) with a pre-existing crack,
as shown in Figure 6, where only half the specimen is discretized because of symmetry. With the
conventional gradient damage model, it was shown in [9] that a wrong damage initiation and propa-
gation was predicted, even though the structural response is well-regularized during strain softening.
This problem is reproduced below in plane strain, where E D 1GPa, D 0:2, k D 10 and ˛ D 0:99.
For the localizing gradient damage model, the coupling modulus is adopted as h D E 109 .
In the following, several length scale parameters are considered to illustrate the limitations of the
Figure 6. Compact tension specimen (CTS).
DOI: 10.1002/nme
conventional approach. All other parameters remain the same, except for 0 and ˇ, which are
calibrated so that a similar structural response is obtained for each case considered, as shown in
Figure 7a.
4.1. Relatively large length scale parameter

Relative to the pre-existing crack, a large length-scale parameter of l D 0:2L is adopted here to
better illustrate the improved performance of the localizing gradient enhancement. The softening
parameter is calibrated as ˇ D 205 and ˇ D 5 in the conventional and localizing gradient enhance-
ments, respectively. For both models, 0 D 0:002 is adopted. Mesh convergence with the localizing
gradient damage model is demonstrated in Figure 7b.
The damage evolution along the crack front from loading steps (i) to (ii) in Figure 7a is next
scrutinized. At the initial stage of softening, a large gradient in the strain profile develops, with a
localized deformation (discretized continuously) in front of the crack tip and an almost vanishing
strain field behind the crack tip. With the conventional gradient damage model, the constant interac-
tion domain extends to the region behind the crack tip, hence minimizing the impact of the localized
deformation in the failure process zone. Because of the large strain gradient, the smoothened nonlo-
cal field (hence damage) has a maximum value at a distance away from the crack tip. This damage
profile in turn induces a larger deformation away from the crack tip, which after the smoothen-
ing process, leads to a gradual shift in the location of maximum damage, as depicted in Figure 8a.
Because ! 1 describes the development of a macroscopic crack, this gives rise to the anomaly
where crack propagation occurs away from the crack tip in brittle fracture.
With the localizing gradient damage model, the nonlocal interaction is restricted to an evolv-
ing fracture process zone through the interaction function g. At the initial stage of softening, the
(reduced) interaction radius in front of the crack tip extends slightly over to the region behind.
Figure 7. (a) Structural responses with conventional and localizing gradient models. (b) Mesh convergence
with the localizing gradient model (l D 0:2L).
Figure 8. Damage .!/ evolution ahead of crack tip from loading points (i) to (ii) in Figure 7 with l D 0:2L,
for (a) conventional and (b) localizing gradient enhancements. Markers indicate the locations of maximum
damage at each instance.
DOI: 10.1002/nme
Consequently, there is a shift in the location of maximum damage away from the crack tip, as
depicted in Figure 8b. As the softening continues, however, the location of maximum damage
rapidly shifts towards the crack tip in Figure 8b. The propagation of a macroscopic crack is thus
correctly captured with ! 1 from the crack tip.
The improved performance of the localizing gradient damage model is furthermore illustrated
through the evolution of damage in the specimen. With the conventional approach, Figure 9a shows
a rapidly expanding process zone that extends to the region behind the crack tip – a non-physical
response. Moreover, such a diffused process zone at complete material failure is not representative of
brittle fracture. In contrast, the damage evolution in Figure 9b with decreasing nonlocal interactions
Figure 9. Damage (!) profiles at the three loading points indicated in Figure 7a. Rows (a) and (b) illustrate
damage evolution with conventional and localizing gradient models, respectively for l D 0:2L.
Figure 10. Damage (!) profiles at failure for two different mesh sizes with the localizing gradient
enhancement.
DOI: 10.1002/nme
is confined to a finite bandwidth preceding the crack tip – even when a relatively large length scale
parameter is adopted – which aligns well with experimental observations, for example, in concrete
CTS specimens with process zone bandwidths approximately the size of an average aggregate [29].
That the damage profiles are mesh independent is demonstrated in Figure 10 with two different
element sizes.
4.2. Relatively small length scale parameter

In this section, we investigate the effect of adopting a smaller length scale parameter as a remedy
for the conventional gradient enhancement, with l D 0:04L, 0 D 0:0031, and ˇ D 39. As a com-
parison, we furthermore illustrate the improved performance of the localizing gradient enhancement
with l D 0:08L, 0 D 0:0028, and ˇ D 5.
The damage evolution with the conventional approach is shown in Figure 11a. With a length
scale parameter that is 5 smaller than that in Section 4.1, the damage process is now confined to a
smaller bandwidth as compared with Figure 9a. However, the spurious damage growth phenomenon
persists, which can also be observed in the evolution of eQ in Figure 11b. In addition, the location of
maximum damage shifts away from the crack tip during strain softening, as depicted in Figure 12a,
such that a macroscopic crack (! 1) develops ahead of the tip. These results thus indicate that
the conventional approach is problematic even with a smaller length scale parameter.
For the localizing gradient enhancement with a length scale parameter 2.5 smaller than that in
Section 4.1, the damage process zone in Figure 13a exhibits a smaller bandwidth, as expected. More
interestingly, Figure 12b reveals that the maximum damage is located at the crack tip throughout the
softening process. A similar evolution process for eQ is depicted in Figure 13b. This indicates that the
localizing gradient enhancement is able to correctly capture the damage initiation and propagation
process with a reasonably small length scale parameter. In case a relatively large length scale param-
eter is adopted, the localizing gradient enhancement is still able to avoid spurious damage growth,
with a correct description of macroscopic crack propagation at ! 1, as discussed in Section 4.1.
Figure 11. (a) Damage (!) and (b) eQ profiles at the three loading points in Figure 7a with the conventional
gradient enhancement (l D 0:04L).
DOI: 10.1002/nme
Figure 12. Damage .!/ evolution ahead of crack tip from loading points (i) to (ii) in Figure 7 for (a) con-
ventional and (b) localizing gradient enhancements. The length scale parameter l is adopted as 0:04L and
0:08L, respectively. Markers indicate the locations of maximum damage at each instance.
Figure 13. (a) Damage (!) and (b) eQ profiles at the three loading points in Figure 7a with the localizing
5. FAILURE IN SHEAR BAND PROBLEMS
Many geo-materials fail under compressive loading where deformation localizes into narrow
inclined band(s). The formation of such shear bands leads to the strain softening of material, which
then triggers possible collapse mechanisms in a structure. Because deformation localizes and propa-
gates along a shear band, it has been modelled as a line with discontinuous displacement fields [42].
Adopting the conventional gradient enhancement to regularize the softening response, however,
results in an erroneous shear band formation [9]. This is illustrated in the succeeding discussions by
considering a specimen in plane compression, as depicted in Figure 14, with a small defect region
introduced to induce localization.
Unless otherwise stated, the material parameters used in both conventional and localizing gradient
damage models are E D 20GPa, h D E 109 , D 0:2, k D 1, 0 D 0:0001 (50% reduc-
tion in defect region), and ˛ D 0:99. Additionally, ˇ D 100 is adopted in the localizing gradient
DOI: 10.1002/nme
damage model, and calibrated as ˇ D 720 in the conventional model to give a similar structural
response, as shown in Figure 15a. Relative to the size of imperfection, a large length-scale parame-
ter of l D 0:05L is chosen to better demonstrate the deficiency of the conventional approach. The
mesh sensitivity study performed for the localizing gradient damage model in Figure 15b shows
convergence with respect to the element size.
The evolution of damage at the three loading points indicated in Figure 15a with the conven-
tional approach is illustrated in Figure 16a. At the initial stage of softening, a wide shear band,
consequence of the relatively large value for l, is observed. The corresponding eQ profile depicted in
Figure 15b shows a large strain gradient emanating from the defect region. As discussed in Section 4,
the smoothening mechanism over a constant interaction domain serves to reduce the impact of local-
ized deformation within the shear band, as well as to smear out the localization effect to regions
beyond the shear band. Consequently, the bandwidth grows with deformation. Moreover, it is eas-
ily observed from Figures 16a and b that the failure process zone propagates horizontally across the
‘shear band’, to the extent that the profile at final failure resembles a mode I problem – clearly a
non-physical behavior. Note that a similar response for the shear band formation in metallic materi-
als is also observed in plasticity models enriched with the conventional gradient enhancement [43].
Furthermore, it was mentioned in the introduction of [22] that the strain-based transient gradient
damage model in [8] does not resolve the limitation on shear band evolution. For the nonlocal inte-
gral formulation with evolving interactions in [17], an inclined shear band is formed only after a
(high) threshold of softening response. These examples thus highlight the difficulty of capturing the
shear band formation with nonlocal models.
The effect of adopting a smaller length scale parameter in the conventional approach is next
investigated with l D 0:01L, ˇ D 205, and 0 D 0:000104 (50% reduction in defect region). As
shown in Figure 17a, a smaller shear band at the initial stage of softening is obtained, compared with
Figure 16a. Because the smoothening mechanism is constrained to a smaller interaction domain,
the smearing out of localized deformation is now reduced, though not fully eliminated. This can be
Figure 14. Specimen under compression, with a defect region (shaded) introduced to induce localization.
Figure 15. (a) Structural responses with conventional and localizing gradient models for l D 0:05L. (b)
Mesh convergence with the localizing gradient model.
DOI: 10.1002/nme
gradient enhancement .l D 0:01L/.
DOI: 10.1002/nme
observed in Figure 17a – while damage develops at an inclined angle, it is coupled with a spurious
growth in the direction perpendicular to the shear band. Moreover, the failure process zone drifts
horizontally away from the defect region, before propagating at an inclined angle after significant
strain softening has developed. This is easily observed from the evolution of eQ in Figure 17b. The
results here thus reinforce those in Section 4.2 that a smaller length scale parameter cannot resolve
the limitations of a conventional gradient enhancement.
The performance of the localizing gradient damage model for such problems is shown in
Figure 18, with l D 0:05L. At the initial stage of softening, a wide shear band is observed,
within which deformation localizes. Note that damage ! and eQ initiate from the defect, and propa-
Figure 18. (a) Damage (!) and (b) eQ profiles at the three loading points in Figure 15a with the localizing
Figure 19. Damage (!) profiles at failure for two different mesh sizes with the localizing gradient
enhancement.
DOI: 10.1002/nme
gates along the shear band during strain softening. At complete material failure, a highly localized
deformation is obtained that lies within a narrow region along the shear band, that is a continuous
description of shear fracture. The failure process and profile agree well with experimental observa-
tions, for example in sand [44]. Mesh sensitivity is also shown to be fully resolved from the damage
profiles in Figure 19 with two different discretization sizes. This problem thus demonstrates the
capability of the localizing gradient damage model to correctly capture shear band failures.
6. CONCLUSION
In the context of damage mechanics for quasi-brittle materials, whereas the structure behavior is
regularized with the conventional nonlocal enhancement, a spurious damage growth is predicted in
the failure process zone. Moreover, a wrong damage initiation and propagation is observed in mode I
and II problems. In this contribution, a localizing gradient damage model is developed to remedy the
limitations of a conventional nonlocal enhancement, by having an interaction domain that decreases
with damage. This is motivated by the fact that the fracture of quasi-brittle materials typically starts
as a diffuse network of microcracks, before localizing into a macroscopic crack. Because nonlocal
interactions originate from the underlying micro-cracks, the interaction domain is assumed here to
be dependent on the intensity of active micro-cracks.
To this end, we adopt the micromorphic theory to develop a new class of gradient enhancement
that is consistent with the stated physical motivation, and which fully regularizes the softening
behavior. Departing from the thermodynamics framework in [26], energetic contributions arising
from the interactions between micro-processes, is now appended with an interaction function that
decreases with damage. This nonlocal contribution in the free energy potential thus vanishes in
regions of high damage levels. The ensuing microforce balance of the micromorphic continuum
recovers a Helmholtz expression that resembles closely the conventional gradient enhancement.
Accordingly, the numerical framework of the developed model has a similar structure as the
conventional gradient damage model, with only minor modifications required.
Through a one-dimensional problem in Section 3, it was shown that the localizing gradient
damage model is able to recover a localized failure, hence avoiding the spurious damage growth
experienced in the conventional enhancement. For the mode I and II problems considered in
Sections 4 and 5, it is shown that the conventional approach is problematic. A smaller length scale
parameter merely serves to alleviate the erroneous damage initiation and propagation phenomena,
but cannot fully resolve these limitations.
The localizing gradient damage model is able to circumvent spurious damage growth in both
mode I and II problems considered, even when a relatively large length scale parameter is adopted.
For the mode I problem, a large length scale parameter induces a shift of maximum damage away
from the crack tip at the initial softening stages. However, the location of maximum damage rapidly
moves towards the crack tip during the softening process, such that a macroscopic crack (! 1)
correctly propagates from the tip. This issue is fully resolved when a slightly smaller length scale
parameter is adopted, that is, damage initiates and propagates correctly from the crack tip throughout
the softening process. For the mode II problem, the localizing gradient damage model is able to
correctly capture the initiation of the shear band, as well as the subsequent localization process along
the shear band, even with a relatively large length scale parameter. These examples suggest that a
gradient enhancement with decreasing nonlocal interaction is able to fully regularize the material
failure behavior.
ACKNOWLEDGEMENT
This work is supported by the NUS Academic Research Fund R-302-000-146-112.
APPENDIX A: NUMERICAL FRAMEWORK IN PLANE DEFORMATION
The implementation of the gradient damage model with localizing interaction follows that of a
conventional gradient enhancement. Because only minor modifications are required to account for
DOI: 10.1002/nme
the different constitutive laws, we refer to [6] for the implementation details. In the following, the
numerical framework for the developed model in plane deformation is summarized. To facilitate its
implementation, the expressions are written in Voight notations.
The basic variables u and ee are discretized as
u D Nu au ; e
e D Ne ae ; (A.1)
21 216 161 14 41
where Nu and Ne are quadratic and linear shape functions, respectively, with au and ae the nodal
degrees of freedom. The gradient terms are similarly discretized as
" D Bu au ; re
e D Be ae (A.2)
31 316 161 21 24 41
where Bu and Be are the gradient operator matrices.

The governing equations in (3) and (4) are satisfied in the weak form, with the respective kine-
matic variables as the weight functions. Performing a consistent linearization of the stress quantities,
the system of equations at the elemental level reduce to
Z Z Z
T T
Bu d dv D Nu t da BuT dv ; (A.3)
163 31 162 21 163 31
Z Z
C BeT de dv D
NeT de BeT e dv :
NeT e (A.4)
41 42 21 41 42 21
Note that the surface integral involving higher-order traction in (A.4) is neglected, since there is no
impedance to the damage process at external surfaces.
Based on the equivalent strain adopted in (12), the constitutive relation for in (8) is given by
ij D .1 !/Cij kl "kl C e

h.e e e/Hij kl "kl ;
e/ıij C .e e (A.5)
where

e h.k 1/ k 1 0:5
hD 1C A I1 ; (A.6)
2k.1 2/ 1 2v
h
Hij kl D 2
A0:5 3ıi k ıj l ıkl ıij ; (A.7)
.1 C /
2
k1 2k
AD I1 C J2 : (A.8)
1 2 .1 C /2
Substituting the constitutive relations in (A.5), (9) and (10) into (A.3) and (A.4), the numerical
framework in plane deformation is obtained as

Kuu Kue dau Fu
D (A.9)
Keu Kee dae Fe
where
Z Z
Fu D NuT t da BuT dv ; (A.10)
161 162 21 163 31
Z !
Fe D NeT e
BeT e dv ; (A.11)
41 41 42 21
DOI: 10.1002/nme
Z ´
T
@e T @e
Kuu D BuT .1 !/ C C e
hı @"
C .e e
e/ ı h
@"
1616 163 33 31 31
13 13
" #μ (A.12)
h 0:5 e @e T 1 @A T
C .1C/2A H .e e
e/ C " @"
2A
.e e
e/ " @"
Bu dv ;
33 31 31 316
13 13
Z
Kue D BuT e
hı h
.1C/2
e
A0:5 H " @!
@
C " Ne dv (A.13)
164 163 31 33 31 33 31 14
Z !
@e T
Keu D NeT h @"
Bu dv (A.14)
416 41 13 316
Z " ! #
Kee D NeT h C BeT hl 2 @g
@
e Ne C BeT ghl 2 Be
re dv ; (A.15)
44 41 42 21 14 42 24
with ı D Œ1 1 0T .
31
REFERENCES
1. Bazant ZP. Why continuum damage is nonlocal: Micromechanics arguments. Journal of Engineering Mechanics
1991; 117:1070–1087.
2. Bazant ZP, Jirasek M. Nonlocal integral formulations of plasticity and damage: survey of progress. Journal of
Engineering Mechanics 2002; 11:1119–1149.
3. Bazant ZP, Pijaudier-Cabot G. Nonlocal continuum damage, localization instability and convergence. Journal of
Applied Mechanics 1988; 55:287–293.
4. Toti J, Marfia S, Sacco E. Coupled body-interface nonlocal damage model for FRP detachment. Computer Methods
in Applied Mechanics and Engineering 2013; 260:1–23.
5. Nguyen GD, Houlsby GT. A coupled damage-plasticity model for concrete based on thermodynamic principles:
Part II: non-local regularization and numerical implementation. International Journal for Numerical and Analytical
Methods in Geomechanics 2008; 32:391–413.
6. Peerlings RHJ, de Borst R, Brekelmans WAM, de Vree JHP. Gradient-enhanced damage for quasi-brittle materials.
International Journal for Numerical Methods in Engineering 1996; 39:3391–3403.
7. Peerlings RHJ, Geers MGD, de Borst R, Brekelmans WAM. A critical comparison of nonlocal and gradient-enhanced
softening continua. International Journal of Solids and Structures 2001; 38:7723–7746.
8. Geers MGD, de Borst R, Brekelmans WAM, Peerlings RHJ. Strain-based transient-gradient damage model for failure
analyses. Computer Methods in Applied Mechanics and Engineering 1998; 160:133–153.
9. Simone A, Askes H, Sluys LJ. Incorrect initiation and propagation of failure in non-local and gradient-enhanced
media. International Journal of Solids and Structures 2004; 41:351–363.
10. Di Luzio G, Bazant ZP. Spectral analysis of localization in nonlocal and over-nonlocal materials with softening
plasticity or damage. International Journal of Solids and Structures 2005; 42:6071–6100.
11. Vermeer PA, Brinkgreve RBJ. A New Effective Non-Local Strain Measure for Softening Plasticity. In Localisation
and Bifurcation Theory for Soils and Rocks, Chambon R, Besrues J, Vardoulakis I (eds). Balkema: Rotterdam, 1994;
89-100.
12. Poh LH, Swaddiwudhipong S. Gradient enhanced softening material models. International Journal of Plasticity
2009; 25:2094–2121.
13. Poh LH, Swaddiwudhipong S. Over-nonlocal gradient enhanced plastic-damage model for concrete. International
Journal of Solids and Structures 2009; 46:4369–4378.
14. Challamel N. A variationally based nonlocal damage model to predict diffuse microcracking evolution. International
Journal of Mechanical Sciences 2010; 52:1783–1800.
15. Hosseini HS, Hoŕak M, Zysset PK, Jirásek M. An over-nonlocal implicit gradient-enhanced damage-plastic model
for trabecular bone under large compressive strains. International Journal for Numerical Methods in Biomedical
Engineering 2015; 31:e02728.
16. Bui QV. Initiation of damage with implicit gradient-enhanced damage models. International Journal of Solids and
Structures 2010; 47:2425–2435.
17. Nguyen GD. A damage model with evolving nonlocal interactions. International Journal of Solids and Structures
2011; 48:1544–1559.
DOI: 10.1002/nme
18. Saroukhani S, Vafadari R, Simone A. A simplified implementation of a gradient-enhanced damage model with
transient length scale effects. Computational Mechanics 2013; 51:899–909.
19. Pijaudier-Cabot G, Haidar K, Dubé JF. Non-local damage model with evolving internal length. International Journal
Numerical Analysis Geomechanics 2004; 28:633–652.
20. Triantafyllou A, Perdikaris PC, Giannakopoulos AE. Gradient elastodamage model for quasi-brittle materials with
an evolving internal length. Journal of Engineering Mechanics 2015; 141:04014139.
21. Geers MGD, Peerlings RHJ, Brekelmans WAM, de Borst R. Phenomenological nonlocal approaches based on
implicit gradient-enhanced damage. Acta Mechanica 2000; 144:1–15.
22. Giry C, Dufour F, Mazars J. Stress-based nonlocal damage model. International Journal of Solids and Structures
2011; 48:3431–3443.
23. Bongers G. A Stress-Based Gradient-Enhanced Damage Model. Master Thesis, Supervisor: Simone A, Delft
University of Technology 2011.
24. Thai TQ, Rabczuk T, Bazilevs Y, Meschke G. A higher-order stress-based gradient-enhanced damage model based
on isogeometric analysis. Computer Methods in Applied Mechanics and Engineering 2016; 304:584–604.
25. Dillard T, Forest S, Ienny P. Micromorphic continuum modelling of the deformation and fracture behavior of nickel
foams. European Journal of Mechanics 2006; 25:526–549.
26. Forest S. Micromorphic approach for gradient elasticity, viscoplasticity and damage. Journal of Engineering
Mechanics 2009; 135:117–131.
27. Sun G, Poh LH. Homogenization of intergranular fracture towards a transient gradient damage model. Journal of the
Mechanics and Physics of Solids 2016; 95:374–392.
28. Carrasquillo RL, Slate FO, Nilson AH. Microcracking and behavior of high strength concrete subject to short-term
loading. ACI Journal Proceedings 1981; 78:179–186.
29. Otsuka K, Date H. Fracture process zone in concrete tension specimen. Engineering Fracture Mechanics 2000;
65:111–131.
30. Grassl P, Jirásek M. Meso-scale approach to modelling the fracture process zone of concrete subjected to uniaxial
tension. International Journal of Solids and Structures 2010; 47:957–968.
31. Peerlings RHJ, Massart TJ, Geers MGD. A thermodynamically motivated implicit gradient damage framework and
its application to brick masonry cracking. Computer Methods in Applied Mechanics and Engineering 2004; 193:
3403–3417.
32. de Vree JHP, Brekelmans WAM, van Gils MAJ. Comparison of nonlocal approaches in continuum damage
mechanics. Computers and Structures 1995; 55:581–588.
33. Wu L, Noels L. Elastic damage to crack transition in a coupled non-local implicit discontinuous Galerkin/extrinsic
cohesive law framework. Computer Methods in Applied Mechanics and Engineering 2014; 279:379–409.
34. Mediavilla J, Peerlings RHJ, Geers MGD. Discrete crack modelling of ductile fracture driven by non-local softening
plasticity. International Journal for Numerical Methods in Engineering 2006; 66:661–688.
35. Seabra MRR, de Sa J. MAC, Andrade FXC, Pires F. FMA. Continuous-discontinuous formulation for ductile fracture.
International Journal of Material Forming 2010; 4:271–281.
36. 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; 192:4581–4607.
37. Wang Y, Waisman H. From diffuse damage to sharp cohesive cracks: a coupled XFEM framework for failure analysis
of quasi-brittle materials. Computer Methods in Applied Mechanics and Engineering 2016; 299:57–89.
38. Miehe C, Welschinger F, Hofacker M. Thermodynamically consistent phase-field models of fracture: Varia-
tional principles and multi-field FE implementations. International Journal for Numerical Engineering 2010; 83:
1273–1311.
39. de Borst R, Verhoosel CV. Gradient damage vs phase-field approaches for fracture: similarities and differences.
Computer Methods in Applied Mechanics and Engineering. In Press; doi:10.1016/j.cma.2016.05.015.
40. Duda FP, Ciarbonetti A, Sánchex PJ, Huespe AE. A phase-field/gradient damage model for brittle fracture in elastic-
plastic solids. International Journal of Plasticity 2015; 65:269–296.
41. Geers MGD. Experimental analysis and computational modelling of damage and fracture. Ph.D. Thesis, Eindhoven
University of Technology, 1997.
42. Borja RI. A finite element model for strain localization analysis of strongly discontinuous fields based on standard
Galerkin approximation. Computer Methods in Applied Mechanics and Engineering 2000; 190:1529–1549.
43. Engelen RAB, Geers MGD, Baaijens FPT. Nonlocal implicit gradient-enhanced elasto-plasticity for the modelling
of softening behavior. International Journal of Plasticity 2003; 19:403–433.
44. Alshibli KA, Sture S. Shear band formation in plane strain experiments of sand. Journal of Geotechnical and
Geoenviromental Engineering 2000; 126:495–503.
DOI: 10.1002/nme

Localizing Gradient Damage Model With Decreasing Interactions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Localizing Gradient Damage Model With Decreasing Interactions

Uploaded by

Copyright:

Available Formats

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng (2016)

Localizing gradient damage model with decreasing interactions

Leong Hien Poh*,† and Gang Sun

Received 23 March 2016; Revised 15 July 2016; Accepted 22 August 2016

Copyright © 2016 John Wiley & Sons, Ltd.

as well as a microforce balance governing the micro–macro interaction

2.1. Damage characterisation and evolution

J2 D 3t r." "/ t r 2 ."/ : (14)

2.2. Damage-based decreasing nonlocal interaction

2.3. Comparison with phase-field damage model

3. ONE-DIMENSIONAL LOCALIZATION BEHAVIOR

Figure 2. A one-dimensional bar with imperfection in the shaded region.

4. FAILURE IN MODE I PROBLEMS

Figure 6. Compact tension specimen (CTS).

4.1. Relatively large length scale parameter

4.2. Relatively small length scale parameter

5. FAILURE IN SHEAR BAND PROBLEMS

APPENDIX A: NUMERICAL FRAMEWORK IN PLANE DEFORMATION

where Bu and Be are the gradient operator matrices.

ij D .1 !/Cij kl "kl C e

You might also like