You are on page 1of 21

Available online at www.sciencedirect.

com
ScienceDirect

Comput. Methods Appl. Mech. Engrg. 355 (2019) 492–512


www.elsevier.com/locate/cma

Localizing gradient damage model with micro inertia effect for


dynamic fracture
Zhao Wanga , Amit Subhash Shedbalea , Sachin Kumara,b , Leong Hien Poha ,∗
a Department of Civil and Environmental Engineering, National University of Singapore, 1 Engineering Drive 2,
E1A-07-03, Singapore 117576, Singapore
b Department of Mechanical Engineering, Indian Institute of Technology Ropar, Punjab, 140001, India

Received 22 November 2018; received in revised form 18 March 2019; accepted 19 June 2019
Available online xxxx

Abstract
A localizing gradient damage model with micro inertia effect is proposed for the dynamic fracture of quasi-brittle materials.
The objective is to achieve mesh independent solutions, and to avoid spurious effects associated with the conventional nonlocal
enhancement. The proposed localizing gradient damage model closely resembles the conventional gradient enhancement, albeit
with an interaction domain that decreases with damage, complemented by a micro inertia effect. We first consider a classical
crack branching problem, where the localizing gradient damage model is shown to resolve the mesh sensitivity issue, as well
as to correctly reproduce the crack profile. Moreover, the micro inertia effect is observed to retard the crack velocity. Next,
the tensile loading of a Polymethyl Methacrylate plate is considered. It is shown that the proposed model effectively captures
the experimentally observed transition of crack profiles as the loading rate increases, i.e. from a straight crack propagation,
to sub-branching, and finally to macro branching. Numerical results in terms of crack patterns, crack velocities, and fracture
energies are in good agreement with the experimental data. To furthermore demonstrate the superior performance of the
localizing gradient damage model, the macro branching problem is solved using the conventional gradient enhancement with
micro inertia. It is shown that a spurious damage growth and an erroneous interaction between closely spaced cracks suppress
the development of macro branching, even though reasonable values are obtained for the fracture energy and crack velocity.
The localizing gradient damage model is able to fully resolve these issues.
⃝c 2019 Elsevier B.V. All rights reserved.

Keywords: Gradient damage; Micro inertia; Dynamic fracture; Strain-rate effect; Micromorphic

1. Introduction
The propagation of a dynamic crack is governed by the evolution of dissipative mechanisms in the fracture
process zone [1], as well as the initiation, growth and coalescence of micro cracks [2]. These phenomena are
generally time and rate dependent, influenced by inertia effects at the crack tip and the viscous behavior of the bulk
material in between cracks. A continuum model for dynamic fracture aims to capture these characteristics well.
Standard continuum models are ill-suited for such analyses as they become mesh dependent during strain
softening [3]. To the limit of a vanishing element size, damage occurs without any energy dissipation. As a
∗ Corresponding author.
E-mail address: ceeplh@nus.edu.sg (L.H. Poh).

https://doi.org/10.1016/j.cma.2019.06.029
0045-7825/⃝ c 2019 Elsevier B.V. All rights reserved.
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 493

remedy, nonlocal/gradient enhancements are commonly adopted to address the issue of mesh dependency, physically
motivated by the interactions between micro cracks in the process zone [4–6]. With the integral approach, a spatial
averaging of the damage-driving variable(s) is done within an interaction domain, the size of which introduces a
length scale parameter to serve as a localization limiter [7–11]. The nonlocal integral formulation can also be re-
cast into a differential equation to give the so-called “implicit” gradient enhancement [12–16], where a length scale
parameter associated with the higher order term characterizes the size of the micro process zone. Alternatively, the
generalized micromorphic framework can be adopted for the development of regularized softening models [17–23].
Here, additional “morphic” fields capture the energetic contribution of rapidly fluctuating micro processes in the
thermodynamics framework, otherwise neglected in a standard continuum. The microforce balance equation in the
micromorphic framework governs the coupling between micro and macro, and very nicely recovers the differential
expression in a gradient enhancement, albeit with a different interpretation of the morphic (nonlocal) variable.
The conventional integral/gradient approach assumes a constant interaction domain. While the global responses
are now regularized, a wrong damage initiation and evolution process, as well as a spurious damage growth
phenomenon, have been reported for quasi-static [24–26] and dynamics problems [27–29]. To fully regularize the
softening response, several suggestions have been proposed in literature. One approach utilizes a weighted average
of both local and nonlocal variables to drive the damage process [20,30–35]. An alternative approach adopts an
evolving length scale parameter that increases with damage [26,36–38].
A recent homogenization framework in Sun and Poh [39] showed that the interactions between active microcracks
propagate onto the macro scale as a nonlocal effect, which decreases with damage. Motivated by this finding, a
localizing gradient damage model was developed in Poh and Sun [24] based on the generalized micromorphic
framework. Therewith, an interaction function that decreases with damage is incorporated into the thermodynamics
framework, to describe the vanishing nonlocal interaction within a decreasing bandwidth of active micro-cracks.
The resulting microforce balance equation bears a close resemblance to, and recovers the conventional gradient
expression as a special case. The localizing gradient enhancement produces sharp damage profiles, hence remedying
the limitations of a conventional approach. Moreover, the localizing gradient enhancement can also help to
address an incorrect crack initiation problem during mode I fracture, induced by the conventional nonlocal/gradient
enhancement as reported in Simone et al. [25]. With the localizing gradient enhancement utilizing a large length
scale parameter, the maximum damage in mode I fracture is still located slightly away from the crack tip at the
initial stage of strain softening. As deformation proceeds, however, the location of maximum damage moves rapidly
towards the crack tip. This incorrect crack initiation problem can be resolved when a reasonably smaller length scale
parameter is adopted [24]. Alternatively, a correct crack initiation can be achieved by extending the scalar interaction
function in the localizing gradient enhancement, to an anisotropic interaction function as shown in Vandoren and
Simone [40]. Note that in Poh and Sun [24], a strain-based isotropic gradient formulation was considered. This has
been extended to an anisotropic stress-based gradient formulation [40,41], drawing inspiration from the stress-based
integral approach of Giry et al. [27]. Note also that the stress-based integral approach has been utilized by Pereira
et al. [42,43,44] with good results.
Though not the focus of this paper, a brief mention on phase field damage is warranted, since it constitutes
another broad class of gradient enhancement. Phase field model for brittle fracture was initially formulated based
on an energy minimization concept [45,46], and later reformulated in a standard thermodynamics framework [47].
Typically, the length scale parameter in a phase field model for brittle fracture characterizes the bandwidth of a
regularized sharp crack, where Griffith’s theory is recovered in the limit of vanishing length scale. Since the length
scale parameter has an influence on the numerical response, an appropriate value for the length scale parameter
is often subjective. This issue was resolved recently in the approach of Wu [48], Wu and Nguyen [49], Wu
[50], where the phase field model is made energetically equivalent to a cohesive fracture, to give length scale
insensitive solutions. For dynamic fracture, phase field models have been reported to capture the crack branching
phenomena well [51–56]. Recently, Doan et al. [57] presented a new rate-dependent hybrid phase field model to
adequately capture the crack dynamics, a dissipated energy dependent crack velocity, through a kinetic coefficient
which controls the rate of energy dissipation. The phase field model is further extended by Doan et al. [58] to
study dynamic crack propagation in a class of composite materials such as functionally graded glass-filled epoxy
beams. To manage the computational resources required for the very fine mesh associated with a typical phase
field model (also the case for gradient enhanced damage models), some remedies in literature include adaptive
mesh refinement [59–61] or a parallelization scheme with time adaptivity [62]. Recently, Patil et al. [61] and Wu
494 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

et al. [63] proposed a strategy where additional degrees of freedom characterizing the crack behavior are associated
only with the nodes within a small sub-domain, where damage is expected to develop, while the region exterior to
this identified sub-domain remains elastic. This strategy was shown to be computationally efficient, provided that
a damage sub-domain is known a priori. In case of dynamic fracture at high loading rates, however, the damage
initiation and propagation processes are generally not known a priori.
In the following, we focus on the dynamic fracture of quasi-brittle materials, with an objective to adequately
capture the key characteristics of rate dependent fracture energy, crack velocity and crack pattern. Referring to the
1D homogenized model for dynamic fracture in Wang and Poh [29], the rapidly fluctuating velocity of underlying
micro-processes was shown to emerge energetically at the macro scale as a micro inertia effect. Motivated by the
results therewith, the localizing gradient damage model in Poh and Sun [24] is extended for dynamic fracture in this
paper by incorporating a micro inertia effect. It is shown that the proposed localizing gradient damage model is able
to regularize the material dynamic failure response, with the development of sharp damage profiles. Additionally,
the dynamic fracture characteristics of a Polymethyl Methacrylate (PMMA) plate at different loading rates are well
captured by the model. The paper is organized as follows. In Section 2, the micromorphic framework incorporating
micro inertia effect is presented. A comparison of the proposed model with the conventional gradient model is
provided in Section 3. In Section 4, a benchmark simplified single-edge notch test is considered to demonstrate the
regularizing effect of the proposed model. In Section 5, the good performance of the proposed model is illustrated
by considering a tension test of a PMMA plate at different loading rates. The numerical framework of the proposed
localizing gradient model with micro inertia effect is summarized in Appendix.

2. Model formulation
Considering a 1D micromechanical problem in the quasi-static setting, Sun and Poh [39] showed that the inter-
actions between micro cracks propagate onto the macro scale as a term characterized by a higher order kinematic
term, with an interaction domain that decreases with damage. This observation motivated the phenomenological
localizing gradient enhancement in Poh and Sun [24], based on the generalized micromorphic framework. Therewith,
the microforce balance resembled closely the gradient expression in a conventional gradient model, albeit with an
interaction length scale that decreases with damage. Extending the 1D micromechanical problem to the dynamic
fracture of quasi-brittle materials, it was further shown in Wang and Poh [29] that a micro inertia effect manifested
itself in the homogenized microforce balance, which captured naturally the interaction and competition between
the underlying potential and kinetic energies associated with the micro processes. Drawing reference from the
observations therewith, the generalized micromorphic framework in Poh and Sun [24] is extended below, to account
for the micro inertia effect during dynamic fracture.

2.1. Hamilton’s principle

In a generalized micromorphic framework, a micro process zone is embedded within a macro continuum at each
macro point, as depicted in Fig. 1(a). The standard kinematic field ε thus characterizes the deformation of the macro
continuum at each macro point, as per standard procedure. Since the micro process zone may deform differently
from the surrounding material, its average deformation is additionally captured via a morphic variable ẽ, while ∇ẽ
accounts for the rapidly fluctuating response within the process zone — see Fig. 1(b). Although a scalar morphic
variable is considered in this work, it can be extended to a tensorial description if necessary. For a given macro
strain ε, the corresponding potential energy density at each macro point is thus postulated as
φpe
den
= σ : ε + σ̃ẽ + ξ̃ · ∇ẽ , (1)
where σ , σ̃ , and ξ̃ are the stress and moment stress conjugates to the respective kinematic fields.
Extending the same concept to the generalized continuum depicted in Fig. 2(a), the transient movement of
the micro process zone in Fig. 2(b) may differ from that of the surrounding material. Analogously, the standard
kinematic field u̇ characterizes the average transient movement of the macro continuum. The non-uniformities within
the micro process zone are characterized via its average transient movement through ẽ, ˙ and its rapidly fluctuating
˙
response through ∇ ẽ. For a given macro velocity u̇, the kinetic energy density is thus
1[ ]
φke
den
= ρ(u̇ · u̇) + ρλ(ẽ) ˙2 ,
˙ 2 + ρλl 2 (∇ ẽ) (2)
2
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 495

Fig. 1. (a) A micro continuum embedded within a macro continuum. (b) Deformation of the macro continuum characterized via ε, while ẽ
and ∇ẽ characterize the deformation of the micro process zone.

Fig. 2. (a) A micro continuum embedded within a macro continuum. (b) Movement of the macro continuum characterized via u̇, while ẽ˙
and ∇ ẽ˙ characterize the movement of the micro process zone.

where ρ is the volumetric density of bulk material, λ is a scaling factor (with unit of length squared) mapping the
volumetric density onto the micro continuum, and l is a length scale parameter associated with the higher order term.
Note that a similar expression for kinetic energy emerges from the bottom up homogenization analysis in Wang
and Poh [29].
Consider an arbitrary region Ω with the external boundary ∂Ω , the external work done is given as

Wext = (t · u + ζẽ) d S , (3)
∂Ω
where t and ζ denote the standard and higher-order tractions respectively.
Hamilton’s principle states that an action functional Ih characterizes the system in Ω over a given time interval
(t1 ≤ t ≤ t2 ), given by
∫ t2 {[∫ ] }
Ih = den
(φke − φpe
den
)d V + Wext dt . (4)
t1 Ω
Among admissible motions, the actual motion of the system is such that the variation of action functional
vanishes. Substituting (1)–(3) into (4) and setting δ I h = 0, we obtain the standard equation of motion
∇ · σ = ρ ü , (5)
as well as a microforce balance given by
σ̃ − ∇ · ξ̃ = −ρλẽ¨ + ρλl 2 ∇ 2 ẽ¨ . (6)
The tractions at the outer boundary with outward normal n are extracted as
t =σ ·n, (7)

ζ = ξ̃ · n . (8)

2.2. Constitutive relations

At this juncture, whereas the governing equations are obtained, the constitutive relations have yet to be
determined. With a view towards the formulation of a scalar damage model, the extent of damage at a macro
496 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

point is characterized via an equivalent strain measure e. Drawing reference to Forest [18] and Poh and Sun [24],
the free energy is postulated as
1 1 1
ψ = (1 − ω)ε : C : ε + h̃(e − ẽ)2 + g h̃l 2 ∇ẽ · ∇ẽ , (9)
2 2 2
where 0 ≤ ω ≤ 1 is the damage variable, C is the fourth-order elasticity tensor, h̃ is a coupling modulus and g is
an interaction function.
The first term in (9) is standard for an elasticity-based scalar damage model, while the second term represents the
coupling between micro and macro. The last term accounts for the energetic contribution of the rapidly fluctuating
response within the active micro process zone, the size of which is characterized via the length scale parameter l
associated with the higher order term ∇ẽ, together with an interaction function g.
As discussed in Poh and Sun [24], the active micro process zone vanishes with the formation of a macro crack.
The interaction function g is thus made dependent on the extent of damage, i.e.
ω=0
{
1,
g = g(ω) = , (10)
R, ω→1
where R ≈ 0 describes a residual interaction between active micro cracks within a localized damage region.
Referring to the last term in (9), the incorporation of an interaction function (10) thus prevents a spurious transfer
of energy from a highly damaged region to its elastically unloading neighborhood. This is also the mechanism
that remedies the spurious damage growth phenomenon associated with the conventional gradient enhancement, as
elaborated in Poh and Sun [24].
In this work, we adopt the interaction function g proposed in Poh and Sun [24], given by
(1 − R)exp(−ηω) + R − exp(−η)
g= , (11)
1 − exp(−η)
where η is a material parameter characterizing the decrease rate of interaction size with damage. Throughout this
work, η = 4 and R = 0.005 are considered.
Following the standard Coleman–Noll procedure, the macro constitutive relations are extracted as
∂ψ ∂e
σ = = (1 − ω)C : ε + h̃(e − ẽ) , (12)
∂ε ∂ε
∂ψ
σ̃ = = h̃(ẽ − e) , (13)
∂ẽ
∂ψ
ξ̃ = = g h̃l 2 ∇ẽ , (14)
∂∇ẽ
where σ denotes the stress tensor, σ̃ an equivalent coupling stress arising from the micro–macro interactions, and
ξ̃ a moment stress resulting from the interactions between active micro processes in the damaged region.
Note that the coupling term in (12) may induce a residual component during material failure [13]. In order to
avoid a possible stress locking effect, a very small coupling modulus is adopted (h̃ = E×10−9 ) throughout the
paper, which also renders the constitutive relation in (12) to recover a standard damage model [24]. Since h̃ ≪ E,
we found the second term in (12) to have a negligible contribution. In our future work, this coupling term will be
omitted for ease of numerical implementation.
To ensure thermodynamic consistency, the dissipation inequality below has to be satisfied
∂g 2
( )
1
Ḋ = φ̇pe
den
− ψ̇ = ε:C :ε− h̃l ∇ẽ · ∇ẽ ω̇ = Y ω̇ ≥ 0 . (15)
2 ∂ω
∂g
Since ∂ω < 0, following (10), the thermodynamic force Y ≥ 0. Given the monotonicity of damage evolution
(ω̇ ≥ 0), it is easily observed that the dissipation inequality is always satisfied.

2.3. Damage characterization and evolution

In a simple scalar damage model, an equivalent strain measure characterizes the extent of material failure
underlying a macro point. Here, we adopt the modified von Mises equivalent strain measure in de Vree et al.
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 497

Fig. 3. Variation of interaction function (g) with damage (ω) (a) different η values for R = 0.005; (b) different R values for η = 4.

[64], defined as
]0.5
1 (k − 1)2 2
[
k−1 2k
e= I1 + I + J2 , (16)
2k(1 − 2ν) 2k (1 − 2ν)2 1 (1 + ν)2
where ν is Poisson’s ratio, k the ratio between compressive and tensile strengths, I1 and J2 are strain invariants
given by
I1 = tr(ε) , (17)

J2 = 3tr(ε · ε) − tr2 (ε) . (18)


We adopt a commonly used exponential damage evolution law given below,
κ0
ω =1− {(1 − α) + α exp [−β(κ − κ0 )]} , (19)
κ
where α and β are material parameters characterizing the strain softening process, and κ0 is a damage initiation
parameter.
The history parameter κ at a given loading step t is defined as
κ(t) = max {ẽ(τ ), 0 ≤ τ ≤ t} . (20)
Note that the damage process is characterized via the morphic variable in (20), in order to regularize the softening
response.

2.4. Interaction function: discussions on parameter selection

Referring to (11), the parameters η and R control the degradation of interaction domain as damage progresses.
Specifically, η signifies the rate of decreasing interaction between microcracks within fracture process zone. For
better clarity, the variations in exponential decay for different η values are shown in Fig. 3(a). A larger η value
may induce solution instabilities due to an abrupt reduction of interaction domain, while a smaller η denotes
a gradual decrease of interaction domain that may lead to spurious damage growth. Therefore, a suitable value
should be chosen to sufficiently capture the material behavior observed in experiments. Similarly, the variations in
exponential decay for different R values are shown in Fig. 3(b). Here, R defines the residual interactions remaining
in the system at complete failure. Ideally, the residual interaction value should be zero when the material is fully
damaged. However for better numerical stability, a small residual value is considered.

3. Discussions on microforce balance


Substituting the constitutive relations in (13) and (14) into the microforce balance in (6), we obtain
1
¨ .
ẽ − e = ∇ · (gl 2 ∇ẽ) + (−ρλẽ¨ + ρλl 2 ∇ 2 ẽ) (21)

498 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

Fig. 4. Experimentally observed crack branching phenomenon in a specimen with a pre-existing crack [70].

Recall that the kinematic fields e, ẽ and ∇ẽ characterize the non-uniformity in deformation through the potential
energy in (9), whereas the acceleration terms capture the fluctuating velocity via the kinetic energy in (2). The
microforce balance (21) thus governs the interaction and competition between the potential and kinetic energies
underlying a dynamic fracture process. Introducing λ̃ = λh̃ in (21), we obtain
ẽ − e = ∇ · (gl 2 ∇ẽ) + λ̃(−ρ ẽ¨ + ρl 2 ∇ 2 ẽ)
¨ . (22)
In the quasi-static limit, the acceleration terms in (22) vanish. The ensuing microforce balance then closely resembles
the conventional gradient enhancement, albeit with an interaction domain that decreases with damage. For the special
case of a constant interaction function g = 1, the conventional gradient enhancement is recovered, i.e.
ẽ − e = l 2 ∇ 2 ẽ . (23)
In (23), the constant length scale parameter enforces the diffusion of damage from the process zone to the
neighboring undamaged regions, thus precluding the development of a localized deformation band. This anomaly
is resolved with the incorporation of an interaction function g in (22). As elaborated in Poh and Sun [24], the
interaction term vanishes upon complete failure (ω → 1) to effect a localized deformation ẽ → e, hence avoiding
the spurious spreading of damage beyond the active process zone.
For dynamic analyses, it will be shown later in Section 4 that a very small value for λ̃ can induce a non-negligible
micro inertia effect, to significantly influence the crack velocity. Since ρ λ̃ ≪ 1, upon complete failure (ω → 1), a
similar localizing mechanism is effected by having ẽ → e.

4. Crack branching
The localizing gradient damage model is implemented numerically, with details of the numerical framework
provided in Appendix. In this section, we investigate the performance of the model in capturing the crack branching
phenomenon, and also to demonstrate the influence of the micro inertia terms.
We consider a center cracked specimen subjected to a constant velocity tensile loading. Mesh convergence is
investigated in terms of the crack profile, and the total dissipation in the specimen. Similar tests have been performed
experimentally [65–70], though the results are often not reported in sufficient detail for direct comparisons with
numerical predictions [43]. Qualitatively, a crack branching phenomenon is observed experimentally – see Fig. 4
from Ramulu and Kobayashi [70] – and captured in many numerical examples [71–73].
In the following, a simplified dynamic tension test is considered. The set-up involves a rectangular Polymethyl
Methacrylate (PMMA) plate, with the geometrical dimensions provided in Fig. 5(a). A loading velocity of ∆v
= 3 m/s is applied as per Fig. 5(b), for a short duration of t0 = 0.1 µs, and held constant thereafter. Due to the
symmetric nature of the problem, only one-quarter of the specimen is considered. Linear quadrilateral elements
with four quadrature points are used for discretization. The discretized geometry along with boundary conditions
is shown in Fig. 5(c). The simulations are done in plane strain condition. Following Linder and Armero [71], Xu
and Needleman [72], Falk et al. [73], the material properties are adopted as ρ = 1190 kg/m3 , E = 3.24 GPa, ν =
0.35. Additionally, the softening parameters are adopted as α = 1 and β = 6, with an internal length scale parameter
l = 0.03 mm.

4.1. Mesh convergence study

For the mesh convergence study, we consider 3 different element sizes (0.015, 0.01 and 0.0086 mm), with
λ̃ = 2.5×10−21 m2 . All simulations are done for a total loading duration of 10 µs. Following Wolff et al. [28], Geelen
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 499

Fig. 5. (a) Geometrical details and boundary conditions for the center crack plate; (b) imposed velocity; (c) corresponding discretized
geometry with boundary conditions.

et al. [55], Liu et al. [74], convergence is investigated in terms of the total dissipation and computed as
∂g 2
∫ ∫ ( )
1
D= ε:C :ε− h̃l ∇ẽ · ∇ẽ dωd V . (24)
2 Ω ∂ω
∂g
Note that since h̃ ≪ E, the second term in the integrand ( ∂ω h̃l 2 ∇ẽ · ∇ẽ) has a negligible contribution.
The evolution of dissipation with time presented in Fig. 6 shows convergence upon mesh refinement. From the
crack profiles in Fig. 7, it is observed that the branching angle and damage bandwidth has converged for the two
smallest element sizes. This example thus demonstrates the regularization property of the localizing gradient damage
enhancement, and its capability to capture the crack branching phenomenon. Note that the converged element size
of 0.01 mm is one third the magnitude of the length scale parameter l. Despite a decreasing nonlocal effect induced
by the interaction function in (21), the results here suggest that mesh convergence can be achieved as per the
recommendation for conventional gradient enhancement, i.e. the smallest element size in the process zone should
be 2–3 times smaller than the length scale parameter.

4.2. Effect of micro inertia

The effect of micro inertia is investigated by comparing the models with and without micro inertia in terms of
dissipation, crack velocity and crack pattern. Crack velocity is calculated by tracking the propagation of the most
advanced point, elaborated later in Section 5. Consistent with the previous section, λ̃ = 2.5×10−21 m2 is adopted.
The same boundary and loading conditions in Fig. 5 are considered for the two models with and without micro
inertia, using a (converged) mesh size of 0.01 mm.
The numerical results in terms of dissipation and crack velocity are provided in Fig. 8. As observed from 8(a),
dissipations in the two models coincide at the initial stage, followed by a very small difference at the end stage.
500 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

Fig. 6. Dissipation in the entire specimen with time for the different element sizes.

Fig. 7. Damage (ω) profiles with element sizes of (a) 0.015 mm, (b) 0.01 mm, and (c) 0.0086 mm.

Fig. 8. (a) Dissipation and (b) crack velocity obtained from models with and without micro inertia. The average velocities with and without
micro inertia are 678 m/s and 722 m/s, respectively.

However, a considerable effect on the crack velocity can be observed in Fig. 8(b). This shows that the micro
inertia effect slows down the crack propagation process. The corresponding damage profiles of the two models are
depicted in Fig. 9. It is found that while the crack branching phenomenon is captured with or without micro inertia,
the location where crack branching initiates is slightly delayed when micro inertia is considered. This observation
correlates well with the dissipation difference in 8(a). At the initial stage where a single crack is observed, both
models result in coincident energy curves. Subsequently, when crack branching occurs, the delayed response induced
by micro inertia leads to a lower dissipation.
The main findings obtained from this benchmark problem are thus
• the proposed model with micro inertia can objectively reproduce the crack branching phenomenon, which is
similar to the typical experimental observations;
• a small micro inertia effect helps to slow down the speed of crack propagation.
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 501

Fig. 9. Damage (ω) profiles using the models (a) without and (b) with micro inertia terms. Initiation of crack branching is slightly delayed
when micro inertia is considered. The dashed and dotted lines indicate the location where crack branching initiates with and without micro
inertia, respectively.

Fig. 10. (a) Geometrical dimensions and boundary conditions of the PMMA plate with a pre-existing crack, subjected to dynamic tensile
loading; (b) corresponding discretized geometry with boundary conditions.

5. Rate dependent dynamic fracture


In this section, we investigate the capability of the proposed model in capturing the strain rate dependent fracture
characteristics, based on the dynamic fracture of PMMA plates conducted by Zhou [65].

5.1. Experimental setup and observations

We refer to the series of experiment with PMMA plates reported in Zhou [65], Zhou et al. [75], schematically
shown in Fig. 10(a). The rectangular specimen was clamped at the top and bottom boundaries, and loaded until
a desired level is achieved. Next, a razor is used to create a small crack at the mid point of the specimen edge.
During the crack propagation process, vertical electrical conductive lines were used to record the crack tip location,
which is next utilized to calculate the crack velocity. Based on the measurement method, the crack velocity actually
referred to its real value projected onto the horizontal direction. In the following, the crack velocity can thus be
∆L
computed as v = ∆tti p , where ∆L ti p is the horizontal incremental distance of crack tip for a given time interval
∆t.
The experimental results show different characteristic crack profiles (straight crack, sub-branching and macro
branching) at increasing loading rates. It is highlighted that crack branching is induced by an excessive flow of
energy to the crack tip, which cannot be dissipated by developing a single crack [53,70], hence resulting in dynamic
502 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

Table 1
Summary of the experimental measurements and observations [65].
Test Crack velocity Fracture energy Phenomenon
(m/s) (J/m2 )
Case 1 (No. 0320B) 335 1588 Straight crack
Case 2 (No. 0322B) 596 5063 Sub-branching
Case 3 (No. 0324B) 665 9420 Macro branching

instability at the crack tip [76]. In the following, the experimental observations in Sharon and Fineberg [69] for
a similar set-up are also provided to facilitate our understanding on the different crack patterns. In addition, the
experimentally obtained crack velocities and fracture energies from Sharon and Fineberg [69] and Zhou [65] are
plotted in Fig. 18 for reference.
Before proceeding to the numerical analysis, we highlight some discussions provided in Zhou [65]. Therewith,
the fracture energy G c is empirically related to the crack velocity v by fitting Fig. 18 to give
675
G c = 2000 ln , (25)
675 − v
where 675 m/s is the limiting velocity.
Based on the experimental observations with PMMA, Zhou [65] postulated that:
• A smooth straight crack can be observed at the macro scale when G c < 2483 J/m2 , or equivalently to
v < 480 m/s based on (25).
• Periodic patterns are observed when G c lies between 2483 and 4394 J/m2 , or equivalently when v is in the
range of 480–600 m/s. A single crack without branching is observed but with rougher surface than the first
case. This pattern is not investigated here since there is insufficient experimental data to support the numerical
analysis.
• Sub-branching occurs when G c lies between 4394 and 6592 J/m2 , or equivalently when v is in the range of
600–650 m/s.
• Macro branching occurs when G c > 6592 J/m2 (or v > 650 m/s).
As discussed earlier, energy can be considered as a criterion for the development of crack branching. Therefore, G c
= 6592 J/m2 is taken as a reference threshold for crack branching in the following problem with PMMA. Beyond
this threshold value, sub-branching(s) surrounding the main crack becomes unstable, leading to the formation of
macro branching.

5.2. Numerical implementation and results

We refer to the numerical models in Wolff et al. [28], Zhou et al. [75]. A miniature plate with dimensions 32 mm
long, 16 mm high and 3 mm thick is considered, as depicted in Fig. 10(a). A pre-existing crack of 4 mm long is
introduced to trigger the fracture process. Corresponding discretized geometry is shown in Fig. 10(b) where initial
crack is modeled as a notch. For each case, an initial quasi-static analysis is carried out by imposing the prescribed
displacement at top and bottom surfaces of the plate. The pre-computed displacement field is next utilized as the
initial condition for the dynamic analysis, without any change in the boundary conditions. Note that the left and right
surfaces of the plate are unconstrained. Following [28,75], three cases (1, 2 and 3) with corresponding prescribed
displacements of ∆u = 0.06, 0.10, and 0.14 mm, are considered in the following. We refer to Zhou [65] for the
relevant experimental series with similar crack velocities to the three cases considered, with details provided in
Table 1.
Following Wolff et al. [28], Zhou et al. [75], we assume a plane stress condition for our numerical examples.
As discussed in Section 5.1, the crack velocity is determined by tracking the most advanced crack tip location.
In literature, different damage threshold values have been adopted for the definition of a fully developed crack,
e.g. ω ≥ 0.3 in Bobaru and Zhang [77], ω ≥ 0.5 in Pereira et al. [43] and ω ≈ 1 in Wolff et al. [28], Zhou et al.
[75]. In the following, we consider ω ≥ 0.95 when computing the crack velocity.
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 503

The material properties of PMMA are: Young’s modulus E = 3.09 GPa and Poisson’s ratio ν = 0.35 as given
in the experiment [65], density ρ = 1180 kg/m3 , and a compression–tension sensitivity ratio of k = 1.48 in (16)
as adopted by Wolff et al. [28], Zhou et al. [75]. The material parameter α in (19) is set to 1.
In our numerical simulations, a rate-dependent damage initiation threshold is adopted. Referring to Zhou et al.
[75], the tensile strengths of the investigated PMMA specimens are in the range 50–72 MPa, which are also
consistent with the experimental measurements in Chen et al. [78]. The rate-dependent damage initiation threshold
is determined as follows:
• Since the tensile strength range is limited, we assign the two limiting reported values of 50 MPa and 72 MPa
to cases 1 and 3 respectively.
• The tensile strength σt can be related to the natural logarithm of strain rate as σt ∝ ln ε̇ [79], where ε̇ = L∆u ,
0 ∆t
L 0 is the original specimen length as shown in Fig. 10, ∆u is the applied displacement and ∆t is the loading
duration.
• The tensile strength for case 2 with a prescribed displacement of 0.10 mm is thus determined as 63 MPa.
The damage initiation threshold corresponding to cases 1, 2 and 3 can thus be determined from κ0 = σEt , to give
0.016, 0.020 and 0.023, respectively.
The internal length scale parameter l characterizes the influence domain size of micro processes. It has been
reported that the width of the main crack, before the occurrence of macro branching, is experimentally measured
as 0.2 mm [69]. Accordingly, we assume l to be half of the experimental measurement, i.e. l = 0.1 mm, which is
consistent with the value adopted in the numerical studies of Wolff et al. [28], Bleyer et al. [53].
In the numerical simulations, fracture energy G c , is computed from total dissipation (24), given as
D
Gc = , (26)
Ll
where L l denotes the ligament length shown in Fig. 10 (28 mm in this example).

5.2.1. Calibration of parameters


The material parameters β and λ̃ are calibrated based on the range of the experimentally measured fracture
energies and crack velocities summarized in Table 1.
Note that the β governs the softening behavior, which significantly influences the fracture energy. As demon-
strated in Section 4.2, a small value of λ̃ has a more significant influence on crack velocity. Our calibration strategy
is thus as follows:
• We first consider case 1, where β is calibrated against the reported fracture energy. Subsequently, we calibrate
for λ̃ based on the reported crack velocity. A very small λ̃ value is found to significantly influence the crack
velocity, similar to the observations in Section 4.2.
• To simplify the calibration process, the same λ̃ value is adopted for cases 2 and 3. For these two cases, β is
calibrated based on the respective fracture energy.
• As a final check for cases 2 and 3 (where λ̃ is not calibrated), the numerically obtained crack velocities and
damage profiles are compared against experimental results.
In the following, β = 4.1, 4.1 and 6.8 for cases 1, 2 and 3 respectively, with λ̃ = 1.3×10−23 m2 for all three cases.

5.3. Numerical results

First, a simple mesh convergence study is carried out for case 3, where the loading rate is the highest. As shown in
Fig. 11(a), the specimen dissipation converges upon mesh refinement, with a negligible difference in crack velocities
for the two smallest element sizes in Fig. 11(b). An element size of 0.045 mm is thus considered to be sufficient,
and is utilized for all three cases.
For case 1, the proposed model predicts a straight crack propagation as shown in Fig. 12, which is consistent
with available experimental observations. The fracture energy is computed numerically as 1549 J/m2 , a close match
with the corresponding experimental data of 1588 J/m2 in Table 1. Note also that this falls below the threshold
value (G c < 2483 J/m2 ) highlighted in Section 5.1, such that a straight crack is obtained. As depicted in Fig. 13,
the average crack velocity obtained numerically is also consistent with the experimental value.
504 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

Fig. 11. (a) Dissipation and (b) crack velocity determined with different element sizes for case 3. The numerical solutions converged upon
mesh refinement.

Fig. 12. A straight crack observed experimentally for case 1 in (a) Zhou [65] and (b) Sharon and Fineberg [69]. (c) Numerical results
obtained with the proposed model.

At a higher loading rate in case 2, sub-branching is observed experimentally in Fig. 14. This phenomenon is
well captured numerically, where the damage band is formed by the evolution and merging of several sub-branches
throughout the fracture process. Numerically, G c is computed as 4382 J/m2 , close to the corresponding experimental
data of 5063 J/m2 in Table 1. The numerically predicted crack velocity also matches well with the experimental
value, as depicted in Fig. 15.
Finally, at the highest loading rate in case 3, several macro branches developed to produce a complicated fracture
pattern in Fig. 16. The damage profile is able to describe this phenomenon well. Moreover, the numerically obtained
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 505

Fig. 13. Crack velocity history for case 1. The experimental average velocity from Zhou [65] is 335 m/s. The numerical average crack
velocity is 301 m/s.

Fig. 14. Sub-branching observed experimentally for case 2 in (a) Zhou [65] and (b) Sharon and Fineberg [69]. (c) Numerical results obtained
with the proposed model.

Fig. 15. Crack velocity history for case 2. The experimental average velocity from Zhou [65] is 596 m/s. The numerical average crack
velocity is 621 m/s.
506 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

Fig. 16. Macro branching observed experimentally for case 3 in (a) Zhou [65] and (b) Sharon and Fineberg [69]. (c) Numerical results
obtained with the proposed model.

fracture energy of resulting in G c = 8764 J/m2 also matches well with the experimental data of 9420 J/m2 in Table 1.
Note also that the fracture energy exceeds the empirically threshold value of 6592 J/m2 highlighted in Section 5.1.
Based on the energy criterion [53,70], it implies that a single crack is incapable of releasing the excessive energy,
hence leading to the development of crack branching. Furthermore, as seen from Fig. 17, the numerical crack
velocity fits closely with the experimental value.
Finally, the numerically obtained fracture energies G c are plotted together with experimental data in Fig. 18, to
illustrate clearly the rate dependent fracture energy. The three cases considered here thus demonstrate the capability
of the proposed model to adequately capture all dynamic fracture characteristics at different loading rates. Note
also that sharp damage profiles are obtained for all numerical examples.
Similar studies on dynamic fracture in PMMA have been reported using other approaches like cohesive
model [75] and nonlocal gradient damage [28]. A generalized cohesive law based on the physics of micro cracking
at the crack tip process zone has been proposed by Zhou et al. [75] which captured the crack propagation and
branching phenomena effectively. In a nonlocal damage model presented in Wolff et al. [28], the rate dependency
is incorporated through a damage-delay model which efficiently captured the crack dynamics under smaller loading
rate. However, for the case with highest loading rate, the damage-delay model needs to be further modified in order
to capture the main crack branching.

5.4. Conventional gradient enhancement with micro inertia

To furthermore demonstrate the need for a localizing gradient enhancement, in this section, we adopt the
conventional gradient model with micro inertia for solving case 3, where macro branching is observed. Without the
micro inertia effect in the conventional gradient model, when calibrated against the fracture energy, the crack velocity
cannot be reproduced correctly. For brevity, the results without micro inertia are not reported here. Instead, we
deliberately consider the conventional gradient model with micro inertia effect, to give a more stringent comparison.
Numerically, this is achieved by setting g ≈ 1 in (21).
To be consistent, the same value of λ̃ = 1.3×10−23 m2 is adopted here. The softening parameter β is calibrated
following the procedure in Section 5.2.1 to give a fracture energy that is larger than the empirical threshold of
6592 J/m2 , as discussed in Section 5.1. Recall that beyond this threshold value, crack branching is expected. With
the same length scale parameter (l = 0.1 mm), the conventional gradient enhancement with micro inertia exhibits
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 507

Fig. 17. Crack velocity history for case 3. The experimental average velocity from Zhou [65] is 665 m/s. The numerical average crack
velocity is 672 m/s.

Fig. 18. Relationship between fracture energy G c and crack velocity v. A comparison between the numerical results and the experimental
data. Blue circle denotes experimental data from Zhou [65], yellow square denotes experimental data from Sharon et al. [80], purple line
represents the empirical expression in (25).

a very diffusive damage profile without any macro branching, as depicted in Fig. 19. This is despite the fact that
reasonable values for the fracture energy and crack velocity are obtained, as shown in Fig. 18. With the conventional
gradient enhancement, a constant internal length scale parameter enforces a fixed interaction domain throughout the
fracture process. A spurious damage growth in this case suppresses the crack branching phenomenon.
As a further illustration on the limitations of conventional gradient enhancement, a smaller length scale value
(0.5l) is now adopted. Using the same parameters as above, the conventional gradient model with 0.5l predicts a
reasonable fracture energy and crack velocity, as depicted in Fig. 18. However, the damage profile shown in Fig. 20
is still unable to capture the branching phenomenon. At the initial stage in Fig. 20(a), a sharp localized crack can be
observed, comparable to that obtained with the localizing model in Fig. 16(b). This is a result of having a smaller
internal length scale. However, a spurious damage growth develops as the main crack propagates in Fig. 20(b)–
(d). In Fig. 20(d), an erroneous interaction between closely spaced sub-branchings can also be observed. These
two effects collectively prohibit the development of crack branching. This example thus clearly demonstrates the
superior performance of the localizing gradient enhancement.

6. Conclusion
In this contribution, a localizing gradient damage model with micro inertia effect is proposed to resolve the
limitations of conventional nonlocal approaches, for the dynamic fracture of quasi-brittle material. The proposed
gradient enhancement is developed based on the micromorphic framework. The ensuing microforce balance
resembles the screened Poisson equation in the conventional gradient model, albeit with an interaction domain
508 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

Fig. 19. Damage profile for case 3 with the conventional gradient enhancement incorporating micro inertia effect (l = 0.1 mm).

Fig. 20. Damage evolution for case 3 with the conventional gradient enhancement incorporating micro inertia effect, using a smaller length
scale parameter (0.5l), at (a) t = 5 µs, (b) t = 15 µs, (c) t = 25 µs, and (d) t = 50 µs.

that decreases with damage, complemented by a micro inertia effect. Through a series of examples, the localizing
gradient model with micro inertia is shown to capture well the dynamic characteristics in terms of rate dependent
fracture energy, crack velocity and crack pattern. For all cases, sharp localized damage profiles are obtained,
corresponding to the development of macro cracks.
Considering a two-dimensional PMMA benchmark problem, it is shown that crack branching can be reproduced
well, and that the solutions converge upon mesh refinement. The micro inertia effect is shown to delay the damage
process by retarding the crack propagation velocity. The excellent performance of the proposed model is further
demonstrated via another two-dimensional PMMA tension test at different loading rates. It is shown that the
proposed model effectively captures the experimentally observed transition of crack profiles as the loading rate
increases, i.e. from a straight crack propagation, to sub-branching, and finally to macro branching. The numerical
results in terms of crack patterns, crack velocities, and fracture energies are in good agreement with the experimental
data. The need for a localizing gradient enhancement is also clearly demonstrated by considering case 3 of the
PMMA test. With a conventional gradient enhancement, even if a small length scale parameter is adopted, a spurious
damage growth and erroneous interaction between closely spaced cracks may suppress the development of crack
branching. Accordingly, even if the material parameters are calibrated to give the correct fracture energy and crack
velocity, the conventional gradient enhancement cannot reproduce the correct crack pattern. This problem is fully
resolved with the localizing gradient enhancement.
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 509

Acknowledgment
This work is supported by the NUS, Singapore Academic Research Fund R302-000-146-112.

Appendix

A.1. Weak formulation

The governing equations (5) and (6) are satisfied in a weak sense with the respective weight functions δu and
δẽ, to give
∫ ∫
[ σ : ∇(δu) + ρ ü · δu ] d V = t · δu d S , (A.1)
Ω ∂Ω
∫ ∫ ∫
σ̃ + ρλẽ¨ δẽ d V + ξ̃ + ρλl 2 ∇ ẽ¨ · ∇(δẽ) d V = ζ + ρλl 2 ∇ ẽ¨ δẽ d S .
( ) ( ) ( )
(A.2)
Ω Ω ∂Ω
Since there is no impedance at the external boundaries due to the damage processes, the higher order traction and
inertia in (A.2) are assumed to vanish.
A consistent linearization of the stress quantities in (12)–(14) is carried out to give
∂w
dσ = (1 − ω)C : dε − C : ε dẽ, (A.3)
∂κ
∂e
dσ̃ = −h̃ dε + h̃dẽ, (A.4)
∂ε
∂g ∂ω
d ξ̃ = h̃l 2 ∇ẽ dẽ + g h̃l 2 d(∇ẽ). (A.5)
∂ω ∂κ
Note that since h̃ ≪ E, the coupling term h̃(e − ẽ) in (12) has been neglected in (A.3) for simplicity.

A.2. Discretization in space

The basic variables u and ẽ are discretized as


u = NuT au , ẽ = NeT ae . (A.6)
2×1 2×8 8×1 1×4 4×1

where Nu and Ne are linear shape functions, and au and ae are the nodal degrees of freedom. The transpose of a
matrix is indicated with superscript T .
The corresponding gradient terms can be obtained as
ε = BuT au , ∇ẽ = BeT ae . (A.7)
3×1 3×8 8×1 2×1 2×4 4×1

where BuT and BeT are the gradient operator matrices.


The accelerations of the basic variables are written as
ü = NuT äu , ẽ¨ = NeT äe . (A.8)
2×1 2×8 8×1 1×4 4×1

Substituting (A.3)–(A.8) to the governing equations (5) and (6), the numerical framework is obtained as
[ ][ ] [ ][ ] [ ]
M11 0 äu K 11 K 12 dau F1
+ = , (A.9)
0 M22 äe K 21 K 22 dae F2
              
M ä K da F
where

M11 = ρ NuT Nu dv,

510 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

M22 = (ρλNeT Ne + ρλl 2 BeT Be )dv,
∫Ω
K 11 = (1 − ω)BuT C Bu dv,
Ω∫
∂ω T
K 12 = − Bu C ε Ne dv,
Ω ∂κ
∂e T

K 21 = − h̃ NeT Bu dv,
Ω ∂ε
2 ∂g ∂ω T
∫ { }
T 2 T
K 22 = h̃ Ne Ne + g h̃l Be Be + h̃l B ∇ẽ Ne dv,
Ω ∂ω ∂κ e
∫ ∫
F1 = − BuT σ dv + NuT tds,

∫ [ ∂Ω
]
F2 = − NeT σ̃ + BeT ξ̃ dv.

A.3. Discretization in time

At a given time t, the variable a is discretized in time with the Newmark method as
(∆t)2 [
a (t) = a (t−∆t) + ∆t ȧ (t−∆t) + (1 − 2βn )ä (t−∆t) + 2βn ä (t) ,
]
(A.10)
2
where ∆t denotes the time increment, and βn = 0.25.
Rearranging (A.10), we obtain
ä (t) = c0 ∆a − c1 ȧ (t−∆t) − c2 ä (t−∆t) , (A.11)
1 1 1 (t) (t−∆t)
where c0 = c1 =
βn (∆t)2
, c2 =
βn ∆t
, − 1 and ∆a = a
2βn
−a .
Substituting (A.11) into (A.9), we have
M c0 ∆a − c1 ȧ (t−∆t) − c2 ä (t−∆t) + K da = F .
[ ]
(A.12)
Within an incremental-iterative procedure, the variable ∆a undergoes iterations until equilibrium is achieved at
time t. A consistent linearization of the incremental variable gives
∆a (i) = ∆a (i−1) + da , (A.13)
with superscript i denoting the iteration step.
Substituting (A.13) into (A.12), the numerical framework becomes
(K + c0 M)da = F − M c0 ∆a (i−1) − c1 ȧ (t−∆t) − c2 ä (t−∆t) .
( )
(A.14)

References
[1] E. Bouchbinder, T. Goldman, J. Fineberg, The dynamics of rapid fracture: instabilities, nonlinearities and length scales, Rep. Prog.
Phys. 77 (2014) 046501.
[2] K. Ravi-Chandar, B. Yang, On the role of microcracks in the dynamic fracture of brittle materials, J. Mech. Phys. Solids 45 (1997)
535–563.
[3] Z.P. Bažant, Instability, ductility and size effect in strain-softening concrete, J. Eng. Mech. Div. 102 (1975) 331–344.
[4] Z.P. Bažant, Why continuum damage is nonlocal: justification by quasiperiodic microcrack array, Mech. Res. Commun. 14 (1987)
407–419.
[5] Ožbolt J., Z.P. Bažant, Numerical smeared fracture analysis: Nonlocal microcrack interaction approach, Int. J. Numer. Methods Eng.
39 (1996) 635–661.
[6] Z.P. Bažant, Nonlocal damage theory based on micromechanics of crack interactions, J. Eng. Mech. 120 (1994) 593–617.
[7] T.B. Bažant, T.P. Chang, Continuum theory for strain-softening, J. Eng. Mech. 110 (1984) 1666–1692.
[8] Z.P. Bažant, M. Jirásek, Nonlocal integral formulations of plasticity and damage: survey of progress, J. Eng. Mech. 128 (2002)
1119–1149.
[9] Z.P. Bažant, G. Pijaudier-Cabot, Nonlocal continuum damage, localization instability and convergence, J. Appl. Mech. 55 (1988)
287–293.
Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512 511

[10] M. Jirásek, Nonlocal models for damage and fracture: comparison of approaches, Int. J. Solids Struct. 35 (1998) 4133–4145.
[11] G. Pijaudier-Cabot, Z.P. Bažant, Nonlocal damage theory, J. Eng. Mech. 113 (1987) 1512–1533.
[12] J. Pamin, Gradient plasticity and damage models: a short comparison, Comput. Mater. Sci. 32 (2005) 472–479.
[13] R.H.J. Peerlings, T.J. Massart, M.G.D. Geers, A thermodynamically motivated implicit gradient damage framework and its application
to brick masonry cracking, Comput. Method. Appl. M. 193 (2004) 3403–3417.
[14] R.H.J. Peerlings, R. de Borst, W.A.M. Brekelmans, J.H.P. De Vree, Gradient enhanced damage for quasi-brittle materials, Int. J. Numer.
Methods Eng. 39 (1996) 3391–3403.
[15] R.H.J. Peerlings, R. de Borst, W.A.M. Brekelmans, M.G.D. Geers, Gradient-enhanced damage modelling of concrete fracture, Mech.
Cohes.-frict. Mater. 3 (1998) 323–342.
[16] R.H.J. Peerlings, M.G.D. Geers, R. de Borst, W.A.M. Brekelmans, A critical comparison of nonlocal and gradient-enhanced softening
continua, Int. J. Solids Struct. 38 (2001) 7723–7746.
[17] T. Dillard, S. Forest, P. Ienny, Micromorphic continuum modelling of the deformation and fracture behaviour of nickel foams, Eur. J.
Mech. 25 (2006) 526–549.
[18] S. Forest, Micromorphic approach for gradient elasticity, viscoplasticity, and damage, J. Eng. Mech. 135 (2009) 117–131.
[19] S. Forest, Nonlinear regularization operators as derived from the micromorphic approach to gradient elasticity, viscoplasticity and
damage, Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 472 (2016) 201507.
[20] I. Zreid, M. Kaliske, A gradient enhanced plasticity–damage microplane model for concrete, Comput. Mech. 62 (2018) 1239–1257.
[21] G. Hütter, U. Mühlich, M. Kuna, Micromorphic homogenization of a porous medium: elastic behavior and quasi-brittle damage, Contin.
Mech. Thermodyn. 27 (2015) 1059–1072.
[22] T. Brepols, S. Wulfinghoff, S. Reese, Gradient-extended two-surface damage-plasticity: micromorphic formulation and numerical aspects,
Int. J. Plast. 97 (2017) 64–106.
[23] T. Brepols, S. Wulfinghoff, S. Reese, A micromorphic damage-plasticity model to counteract mesh dependence in finite element
simulations involving material softening, in: Multiscale Modeling of Heterogeneous Structures, Springer, 2018, pp. 235–255.
[24] L.H. Poh, G. Sun, Localizing gradient damage model with decreasing interactions, Int. J. Numer. Methods Eng. 110 (2017) 503–522.
[25] A. Simone, H. Askes, L.J. Sluys, Incorrect initiation and propagation of failure in non-local and gradient-enhanced media, Int. J. Solids
Struct. 41 (2004) 351–363.
[26] M.G.D. Geers, R. de Borst, W.A.M. Brekelmans, R.H.J. Peerlings, Strain-based transient-gradient damage model for failure analyses,
Comput. Methods Appl. Mech. Eng. 160 (1998) 133–153.
[27] C. Giry, F. Dufour, J. Mazars, Stress-based nonlocal damage model, Int. J. Solids Struct. 48 (2011) 3431–3443.
[28] C. Wolff, N. Richart, J.F. Molinari, A non-local continuum damage approach to model dynamic crack branching, Int. J. Numer. Methods
Eng. 101 (2015) 933–949.
[29] Z. Wang, L.H. Poh, A homogenized localizing gradient damage model with micro inertia effect, J. Mech. Phys. Solids 116 (2018)
370–390.
[30] G. Di Luzio, Z.P. Bažant, Spectral analysis of localization in nonlocal and over-nonlocal materials with softening plasticity or damage,
Int. J. Solids Struct. 42 (2005) 6071–6100.
[31] P.A. Vermeer, B.R.B. J, A new effective nonlocal strain mesure for softening plasticity, in: R. Chambon, J. Besrues, I. Vardoulakis
(Eds.), Localization and Bifurcation Theory for Solids and Rocks, 1994, pp. 89–100.
[32] L.H. Poh, S. Swaddiwudhipong, Gradient-enhanced softening material models, Int. J. Plast. 25 (2009) 2094–2121.
[33] H.S. Hosseini, M. Horák, P.K. Zysset, M. Jirásek, An over-nonlocal implicit gradient-enhanced damage-plastic model for trabecular
bone under large compressive strains, Int. J. Numer. Methods Biomed. Eng. 31 (2015).
[34] G.D. Nguyen, A damage model with evolving nonlocal interactions, Int. J. Solids Struct. 48 (2011) 1544–1559.
[35] M. Schreter, M. Neuner, G. Hofstetter, Evaluation of the implicit gradient-enhanced regularization of a damage-plasticity rock model,
Appl. Sci. 8 (1004) (2018).
[36] S. Saroukhani, R. Vafadari, A. Simone, A simplified implementation of a gradient-enhanced damage model with transient length scale
effects, Comput. Mech. 51 (2013) 899–909.
[37] G. Pijaudier-Cabot, K. Haidar, J.F. Dubé, Non-local damage model with evolving internal length, Int. J. Numer. Anal. Met. 28 (2004)
633–652.
[38] M.E. Mobasher, H. Waisman, L. Berger-Vergiat, Thermodynamic framework for non-local transport-damage modeling of fluid driven
fracture in porous media, Int. J. Rock Mech. Min. Sci. 111 (2018) 64–83.
[39] G. Sun, L.H. Poh, Homogenization of intergranular fracture towards a transient gradient damage model, J. Mech. Phys. Solids 95
(2016) 374–392.
[40] B. Vandoren, A. Simone, Modeling and simulation of quasi-brittle failure with continuous anisotropic stress-based gradient-enhanced
damage models, Comput. Methods Appl. Mech. Eng. 332 (2018) 644–685.
[41] T.H. Nguyen, T.Q. Bui, S. Hirose, Smoothing gradient damage model with evolving anisotropic nonlocal interactions tailored to
low-order finite elements, Comput. Methods Appl. Mech. Eng. 328 (2018) 498–541.
[42] L.F. Pereira, J. Weerheijm, L.J. Sluys, A new effective rate dependent damage model for dynamic tensile failure of concrete, Eng.
Fract. Mech. 176 (2017) 281–299.
[43] L.F. Pereira, J. Weerheijm, L.J. Sluys, A numerical study on crack branching in quasi-brittle materials with a new effective rate-dependent
nonlocal damage model, Eng. Fract. Mech. 182 (2017) 689–707.
[44] L. Pereira, J. Weerheijm, L. Sluys, A new rate-dependent stress-based nonlocal damage model to simulate dynamic tensile failure of
quasi-brittle materials, Int. J. Impact. Eng. 94 (2016) 83–95.
[45] G.A. Francfort, J.J. Marigo, Revisiting brittle fracture as an energy minimization problem, J. Mech. Phys. Solids 46 (1998) 1319–1342.
512 Z. Wang, A.S. Shedbale, S. Kumar et al. / Computer Methods in Applied Mechanics and Engineering 355 (2019) 492–512

[46] B. Bourdin, G.A. Francfort, J.J. Marigo, Numerical experiments in revisited brittle fracture, J. Mech. Phys. Solids 48 (2000) 797–826.
[47] C. Miehe, F. Welschinger, M. Hofacker, Thermodynamically consistent phase-field models of fracture: Variational principles and
multi-field FE implementations, Int. J. Numer. Methods Eng. 83 (2010) 1273–1311.
[48] J.Y. Wu, A geometrically regularized gradient-damage model with energetic equivalence, Comput. Methods Appl. Mech. Eng. 328
(2018) 612–637.
[49] J.Y. Wu, V.P. Nguyen, A length scale insensitive phase-field damage model for brittle fracture, J. Mech. Phys. Solids 119 (2018)
20–42.
[50] J.Y. Wu, A unified phase-field theory for the mechanics of damage and quasi-brittle failure, J. Mech. Phys. Solids 103 (2017) 72–99.
[51] M. Hofacker, C. Miehe, A phase field model of dynamic fracture: robust field updates for the analysis of complex crack patterns, Int.
J. Numer. Methods Eng. 93 (2013) 276–301.
[52] M.J. Borden, C.V. Verhoosel, M.A. Scott, T.J. Hughes, C.M. Landis, A phase-field description of dynamic brittle fracture, Comput.
Methods Appl. Mech. Eng. 217 (2012) 77–95.
[53] J. Bleyer, C. Roux-Langlois, J.F. Molinari, Dynamic crack propagation with a variational phase-field model: limiting speed, crack
branching and velocity-toughening mechanisms, Int. J. Fract. 204 (2017) 79–100.
[54] T. Li, J.J. Marigo, D. Guilbaud, S. Potapov, Gradient damage modeling of brittle fracture in an explicit dynamics context, Int. J.
Numer. Methods Eng. 108 (2016) 1381–1405.
[55] R.J.M. Geelen, Y. Liu, T. Hu, M.R. Tupek, J.E. Dolbow, A phase-field formulation for dynamic cohesive fracture, Comput. Methods
Appl. Mech. Eng. 348 (2019) 680–711.
[56] V.P. Nguyen, J.Y. Wu, Modeling dynamic fracture of solids with a phase-field regularized cohesive zone model, Comput. Methods
Appl. Mech. Eng. 340 (2018) 1000–1022.
[57] D.H. Doan, T.Q. Bui, T. Va. Do, N.D. Duc, A rate-dependent hybrid phase field model for dynamic crack propagation, J. Appl. Phys.
122 (2017) 115102.
[58] D.H. Doan, T.Q. Bui, N.D. Duc, K. Fushinobu, Hybrid phase field simulation of dynamic crack propagation in functionally graded
glass-filled epoxy, Compos. Part. B-Eng. 99 (2016) 266–276.
[59] R.U. Patil, B.K. Mishra, I.V. Singh, An adaptive multiscale phase field method for brittle fracture, Comput. Methods. Appl. Mech.
Eng. 329 (2018) 254–288.
[60] R.U. Patil, B.K. Mishra, I.V. Singh, T.Q. Bui, A new multiscale phase field method to simulate failure in composites, Adv. Eng. Softw.
126 (2018) 9–33.
[61] R.U. Patil, B.K. Mishra, I.V. Singh, A local moving extended phase field method (lmxpfm) for failure analysis of brittle materials,
Comput. Methods. Appl. Mech. Eng. 342 (2018) 674–709.
[62] V. Ziaei-Rad, Y. Shen, Massive parallelization of the phase field formulation for crack propagation with time adaptivity, Comput.
Methods Appl. Mech. Eng. 312 (2016) 224–253.
[63] J.Y. Wu, J.F. Qiu, V.P. Nguyen, T.K. Mandal, L.J. Zhuang, Computational modeling of localized failure in solids: Xfem vs pf-czm,
Comput. Methods Appl. Mech. Eng. 345 (2019) 618–643.
[64] J.H.P. de Vree, W.A.M. Brekelmans, M.A.J. Van Gils, Comparison of nonlocal approaches in continuum damage mechanics, Comput.
Struct. 55 (1995) 581–581.
[65] F. Zhou, Study on the macroscopic behavior and the microscopic process of dynamic crack propagation, (Ph.D. thesis), The University
of Tokyo, Tokyo, 1996.
[66] H. Satoh, On the crack propagation in brittle materials, (Ph.D. thesis), The University of Tokyo, Tokyo, 1996.
[67] N. Murphy, M. Ali, A. Ivankovic, Dynamic crack bifurcation in pmma, Eng. Fract. Mech. 73 (2006) 2569–2587.
[68] A. Kobayashi, B. Wade, W. Bradley, S. Chiu, Crack Branching in Homalite-100 Sheets, Technical Report, Washington Univ Seattle
Dept Of Mechanical Engineering, 1972.
[69] E. Sharon, J. Fineberg, Microbranching instability and the dynamic fracture of brittle materials, Phys. Rev. 54 (7128) (1996).
[70] M. Ramulu, A.S. Kobayashi, Mechanics of crack curving and branching - a dynamic fracture analysis, Int. J. Fract. 27 (1985) 187–201.
[71] C. Linder, F. Armero, Finite elements with embedded branching, Finite Elem. Anal. Des. 45 (2009) 280–293.
[72] X.P. Xu, A. Needleman, Numerical simulations of fast crack growth in brittle solids, J. Mech. Phys. Solids 42 (1994) 1397–1434.
[73] M.L. Falk, A. Needleman, J.R. Rice, A critical evaluation of cohesive zone models of dynamic fractur, J. Physique IV 11 (2001)
543–550.
[74] G. Liu, Q. Li, M.A. Msekh, Z. Zuo, Abaqus implementation of monolithic and staggered schemes for quasi-static and dynamic fracture
phase-field model, Comput. Mater. Sci. 121 (2016) 35–47.
[75] F. Zhou, J.F. Molinari, T. Shioya, A rate-dependent cohesive model for simulating dynamic crack propagation in brittle materials, Eng.
Fract. Mech. 72 (2005) 1383–1410.
[76] J. Fineberg, S.P. Gross, M. Marder, H.L. Swinney, Instability in dynamic fracture, Phys. Rev. Lett. 67 (457) (1991).
[77] F. Bobaru, G. Zhang, Why do cracks branch? a peridynamic investigation of dynamic brittle fracture, Int. J. Fract. 196 (2015) 59–98.
[78] W. Chen, F. Lu, M. Cheng, Tension and compression tests of two polymers under quasi-static and dynamic loading, Polym. Test. 21
(2002) 113–121.
[79] C. Bauwens-Crowet, The compression yield behaviour of polymethyl methacrylate over a wide range of temperatures and strain-rates,
J. Mater. Sci. 8 (1973) 968–979.
[80] E. Sharon, S.P. Gross, J. Fineberg, Energy dissipation in dynamic fracture, Phys. Rev. Lett. 76 (2117) (1996).

You might also like