Computers and Mathematics with Applications 125 (2022) 193–212

Contents lists available at ScienceDirect

Computers and Mathematics with Applications

www.elsevier.com/locate/camwa

Numerical simulations of dynamic fracture and fragmentation problems

by a novel diffusive damage model
Tinh Quoc Bui a,b,∗ , Hung Thanh Tran c
a
Duy Tan Research Institute for Computational Engineering (DTRICE), Duy Tan University, 6 Tran Nhat Duat, Dist. 1, Ho Chi Minh City 700000, Viet Nam
b
Faculty of Civil Engineering, Duy Tan University, Da Nang 550000, Viet Nam
c
Department of Civil and Environmental Engineering, Tokyo Institute of Technology, 2-12-1-W8-22, Ookayama, Meguro-ku, Tokyo 152-8552, Japan

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

Keywords: In this paper, we report numerical simulations for dynamic fracture and fragmentation problems in brittle
Dynamic fracture materials using a recently developed split strain energy diffusive mass-field damage model in terms of standard
Localized mass loss finite element method (FEM). The enhanced constitutive laws for mass source and mass flux by means of strain
Damage
energy decomposition are adopted to overcome certain drawbacks of the original non-split model. We present
Split strain energy
theoretical formulations in a more general way that is suitable and valid for both small and finite deformation
FEM
regimes. The coupled system of equations is derived in which the momentum is governed to describe the time-
dependent deformation of brittle solids whereas the mass-diffusion balance equations are for the evolution of
mass density. We also develop simple techniques for determining crack velocity and estimating amount of mass
loss. Numerical examples are considered for both small and finite deformations. The accuracy and performance
of the developed theory for dynamic brittle fracture are illustrated through comparison and verification, which
are to compare the computed results with respect to experimental data and other numerical methods in terms of
crack path trajectories, dynamic response, energy dissipation, and other relevant numerical aspects.

1. Introduction branching, arrest) have been introduced in the literature, either by dis-
crete approaches (discontinuity) or diffuse damage (continuity), see
Dynamic fracture problems have been attracted much attention from e.g., Refs. [3–12] and references therein. In terms of diffusive ap-
scientists and engineers for the past few decades. Rapid, impact loads proaches, the nonlocal gradient damage model [3,13] and the phase-
in particular and time-dependent loads in general are common and of- field model [4–6] are popular. The phase-field models are relatively
ten encountered in many real applications, and in some cases such a new compared to conventional continuum damage models but have
dynamic loading may be applied deliberately, for instance, fragmenta-
shown to be applicable to topologically complex cracks in both two-
tion and blasting, or may arise from accidental conditions, e.g., traffic
dimensional (2D) and three-dimensional (3D) bodies in an elegant way.
collisions. Regardless of the source of time-dependent applied loads, un-
In line with existing smeared models, Volokh [14,15] recently
derstanding the fracture behavior and failure mechanisms under time-
present an alternative damage model without the need for using any
dependent loading conditions is necessary and important to the design
of engineering applications. For experimental and theoretical aspects, additional internal damage variables (e.g., phase-field). They describe
curious readers can refer to an excellent review paper [1] for compre- the fracture phenomenon in solids as material sinks, which is governed
hensive information and knowledge on modern topics and challenges by local mass-diffusive balance equations. The crucial concept of the
in dynamic fracture, and some other recent developments and related material sinks should be tightened with the macroscopic description
perspectives in the work by Fineberg et al. [2]. of material stiffness degradation, which is the consequence of massive
Computational modeling for prediction of dynamic brittle fracture breakage of atomic bonds. As stated in Volokh [14], such bond breakage
has been extensively studied by many researchers, and different mod- must be gone with the mass loss, diffusing in a volume of characteristic
els for modeling dynamic fracture mechanism (e.g., crack initiation, size, leading to a highly localized mass loss area. It means that the local-

* Corresponding author.
E-mail addresses: [email protected], [email protected] (T.Q. Bui).

https://doi.org/10.1016/j.camwa.2022.08.036
Received 6 October 2021; Received in revised form 17 August 2022; Accepted 23 August 2022

0898-1221/© 2022 Elsevier Ltd. All rights reserved.

T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

ization can be thought as the material sink in the damaged region, the amples of quasi-static/dynamic brittle fracture problems for small and
mass loss follows out of the system, and finally the cracks form. The de- finite deformations with available reference data. In our numerical ex-
graded domain contains particles of materials but without stiffness, nor periments, we restrict our simulations by only considering the density
interaction with the bulk media [16]. The local mass-diffusion balance effects on the stiffness. This simplification is for convenience in the nu-
equations are defined to describe the mass flow in the degraded area merical investigation/representation and that does not alter the main
where, in general, the traditional law of conservation of mass is locally purpose or the original goal of our analysis. To be more precise, the
violated. It is important to underline here that the mass loss associated density is defined as a problem parameter that affects the stiffness, and
with the evolution of damage localization is assumed to be as small as not the inertia. The assumption of the density effect on the stiffness
compared to the mass of entire body. Therefore, the densities, during may have some similar characteristics to the artificial damage variable
the analysis, are assumed to be constant before and after the damage. defined in the conventional damage models, e.g., local and nonlocal
The objective of this contribution is to present a dynamic time- damage models as well as phase-field methods. We assume to consider a
dependent description of the recently developed diffusive localized compressible neo-Hookean free energy [21] in all numerical examples.
mass-field (MF) damage model enhanced by the energy decomposi- In a similar context of dynamic fracture at finite deformation regime,
tion technique. The model was applied to quasi-static brittle fracture this type of neo-Hookean material model was used in other previous
in the work [17] and here is intended for modeling of dynamic fracture studies, e.g., see Refs. [19] and [22].
and fragmentation simulations. This means that we focus our attention An outline of the article is as follows. Section 2 presents basic equa-
on considering dynamic crack propagation in brittle materials under tions for dynamic localized MF models, in which the momenta, material
time-dependent loading conditions, and its computational aspects in the constitutive laws and local mass-balance equations are addressed. Finite
context of the standard finite element method (FEM). In other words, element discretization with the weak forms, time-integration, and lin-
we develop the new dynamic diffusive damage scheme with the signifi- earization of the governing equations are then given in Section 3. Model
cant advantage in that no geometric description of cracks is required for validation for energy decomposition in terms of quasi-static fracture
assessing the susceptibility to dynamic brittle fracture and fragmenta- for plane stress is described in Section 5. We conduct five numeri-
tion, and distinguishing the tension and compression fracture behavior cal examples of dynamic fracture with complex geometries including a
by means of split strain-energy technique. The theoretical formulation double-edge notched (styrene butadiene rubber) specimen, and discuss
of this new computational damage framework is however derived in a the computed results in Section 6, in which fracture patterns, structural
more general way, which is suitable and valid for both small and fi- load-displacement response, dissipated fracture energy and other rele-
nite deformation regimes. It means that, in overall, the theory can be vant numerical aspects are analyzed. Comparison between the present
possibly applied to simulate the dynamic brittle fracture problems at results and reference solutions (e.g., experiment, and other numerical
any level of deformation without significant modification and effort. methods) is analyzed. Our conclusions and some discussions for further
Two key objectives are the development of a mathematically sound so- developments of this study are then given in Section 7.
lution for dynamic brittle fracture problems where the cracks can be
treated without additional effort, and demonstration of the accuracy 2. Theoretical description of dynamic localized mass-field model
and effectiveness of this numerical treatment by means of numerical
In this section, we present the formulations of momentum equation,
experiments in which validation of the numerical results is conducted
material constitutive laws of compressible neo-Hooken materials, in-
with the aid of the experimental data and other approaches. Our numer-
cluding general 3D case and plane strain and plain stress states, and the
ical experiments concern modeling quasi-static/dynamic fracture and
brief local mass-diffusion balance equation.
fragmentation from small strain to large deformation in brittle materi-
als.
2.1. Momentum equations
As a specific description, it is referred to the recent study by Tang et
al. [18] where an energy decomposition at finite deformation for neo-
In terms of the present diffusive damage theory, the basic assump-
Hookean models using phase-field model is developed. Here, we thus
tion is that the failure and, consequently, mass flow are very localized
adopt the same split strain-energy technique accounting for the con-
and hence momentum and energy balance equations can be written
stitutive laws of mass source and mass flux, but in the context of our
in standard forms without adding momentum and energy due to the
developed MF damage model instead, as presented in [17]. It is no- change in mass [15]. To this end, the momentum equation for the time,
ticed that the split operator technique [18] is valid for the compressible 𝑡 ≥ 0, in the spatial description and associated boundary conditions of
neo-Hookean materials only. For other hyperelastic material models the initial-boundary value problems in a domain Ω can be expressed as
such as Mooney–Rivlin, Odgen, and Yeoh, other appropriate techniques
𝑑 (𝜌𝒗)
for instance Hesch and Weinberg [19] and Ye et al. [20] should be = ∇ ⋅ 𝝈 + 𝒃𝑔 in Ω, (1a)
adopted instead. We will numerically show that the dynamic diffusive 𝑑𝑡
damage formulation presented here is highly suitable for simulation of 𝒖 = 𝒖̄ on Γ𝑢 , (1b)
dynamic brittle fracture problems including fragmentation simulation. 𝝈𝒏 = 𝒕̄ on Γ𝑡 , (1c)
Our numerical experiments also indicate the significant advantage of
the present theory over the phase-field model [6] in terms of dissipated 𝒙(𝑡 = 0) = 𝑿, 𝒗(𝑡 = 0) = 𝑽 0 in Ω, (1d)
fracture energy. In other words, although the phase-field model nicely where 𝒗 is the spatial velocity of a material point; 𝜌 is the current mass
offers reasonable crack patterns in most considered problems, it fails density; 𝝈 is the Cauchy stress tensor; 𝒃𝑔 is the spatial body force per
to yield the dissipated fracture energy in terms of dynamic brittle frac- unit volume; 𝒖̄ and 𝒕̄ represent the prescribed displacement and traction
ture context, which has been numerically revealed in the study [6]. on the essential boundary (Γ𝑢 ) and natural boundary (Γ𝑡 ), respectively;
This is particularly critical as the dissipated fracture energy predicted 𝒖 is the displacement field; 𝒏 is the unit outward normal vector on the
by the phase-field technique is approximately twice higher than the ref- boundary (𝜕Ω); 𝒙 is the current coordinate; 𝑿 is the reference coordi-
erence theoretical value. The current diffusive damage approach, in the nate, and 𝑽 𝟎 is the initial condition of velocity at time 𝑡 = 0. It is noticed
contrary, offers acceptable numerical solutions for not only the crack for 𝒖̄ and 𝒕̄ that, in this work, 𝒖̄ is for the constraint (fixed) condition of
patterns but also the dissipated fracture energy, which is analyzed in the displacement field which is not dependent on time, while 𝒕̄ stands
the numerical experiments. for the dynamic loading which is a time-dependent quantity.
The remarkable capabilities in distinction of different fracture be- The material point 𝑿 in Ω0 can be mapped onto the point 𝒙 in the
havior under tension and compression modes of this split strain-energy spatial configuration Ω through 𝒙 = 𝝌(𝑿, 𝑡). The deformation gradient
localized mass-field model are illustrated in a series of numerical ex- 𝑭 is thus expressed as

194
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

𝑭 = 𝑰 + ∇0 𝒖, (2) 2.2.2. Plane stress states

𝜕(⋅) In general, the derivation of stress tensors as well as elasticity tensors
where 𝑰 is the 2nd -order unit tensor; ∇0 (⋅) = denotes the spatial
𝜕𝑿 in cases of the plane stress state for hyperelastic models is not straight-
derivatives related to the reference configuration. The right, 𝑪, and the forward as that of the plane strain case. In literature, there are many
left, 𝒃, Cauchy-Green deformation tensors are given by existing algorithms that have been introduced to calculate plane stress
states for hyperelastic models, see e.g., Refs. [21,23,24]. Here, the tech-
𝑪 = 𝑭𝑇𝑭; 𝒃 = 𝑭𝑭𝑇. (3)
nique introduced by Pascon [25,26] is used to approximate the plane
The material constitutive laws for the second Piola-Kirchhoff, 𝑺, and stress states in terms of the compressible neo-Hookean model. Some
Cauchy stress tensors are given by [17] necessary formulae pertaining to the derivation of constitutive equa-
( ) tions for the plane stress state are quoted here. The deformation tensor,
𝜌 𝜕𝑊 𝑭 , and its in-plane components, 𝑭̃ are:
𝑺 =2 , (4)
𝜌0 𝜕𝑪
( ) ⎡ 𝐹11 𝐹12 0 ⎤ [ ]
2 𝜌 𝜕𝑊 𝑇 𝐹 𝐹12
𝝈= 𝑭 𝑭 , (5) 𝑭 = ⎢ 𝐹21 𝐹22 0 ⎥ ; 𝑭̃ = 11 ; 𝐽̃ = det(𝑭̃ ). (8)
𝐽 𝜌0 𝜕𝑪 ⎢ ⎥ 𝐹21 𝐹22
⎣ 0 0 𝐹33 ⎦
where 𝜌0 is the constant initial mass density, 𝐽 is the determinant of
The right Cauchy-Green deformation tensor, 𝑪 and its in-plane com-
𝑭 , and 𝑊 = 𝜌𝑜 𝑤 defines the strain energy density (SED) function of
ponents, 𝑪̃ are:
virgin materials as energy per unit volume with 𝑤 being the SED per
unit mass. ⎡ 𝐶11 𝐶12 0 ⎤ [ ]
𝐶11 𝐶12
It is important to point out again here that the ratio (𝜌∕𝜌0 ) has 𝑪 = ⎢ 𝐶21 𝐶22 0 ⎥ ; 𝑪̃ = ̃
; 𝐽̃2 = det(𝑪). (9)
⎢ ⎥ 𝐶21 𝐶22
now been incorporated into the constitutive relations, which makes the ⎣ 0 0 𝐶33 ⎦
present work different with the conventional counterpart. The current And the left Cauchy-Green deformation tensor, 𝒃, and its in-plane
density, 𝜌, plays a key role in the modeling of the developed damage components, 𝒃̃ are:
approach. More specifically, the current density alters the degradation
of material stiffness, and two obvious states should thus be defined, i.e., ⎡ 𝑏11 𝑏12 0 ⎤ [ ]
𝑏 𝑏12
𝜌 = 𝜌0 for the initial state of loading condition and for the state at which 𝒃 = ⎢ 𝑏21 𝑏22 0 ⎥ ; 𝒃̃ = 11 ̃
; 𝐽̃2 = det(𝒃). (10)
⎢ ⎥ 𝑏21 𝑏22
the material has not reached the damage yet; and 𝜌 = 0 for the state at ⎣ 0 0 𝑏33 ⎦
which, usually during the deformation and up to a certain point, the Note that in this analysis, the symbol tilde on the hat (̃∙) is used to
material gets completely damaged, or the mass is lost in other words. indicate the 2 × 2 components of a tensor (∙) and its related terms. The
In the present investigation, it is additionally assumed that the super relation between the initial thickness, 𝐻0 , and the current thickness, ℎ,
fast process of the massive breakage of atomic bonds as well as the rate at a material point of the 2D-model is determined through
in time of the degraded transition between virgin material to failure are
ℎ ℎ2
neglected. 𝐹33 = or 𝐶33 = 𝑏33 = . (11)
𝐻0 𝐻02
2.2. Material constitutive laws In the plane-stress state, the determinant of the deformation tensor
𝑭 can be calculated as
The compressible neo-Hookean material model is only adopted in √ √
all the numerical applications, which is given by [21] 𝐽 = det(𝑭 ) = 𝐽̃𝐹33 = 𝐽̃ 𝐶33 = 𝐽̃ 𝑏33 . (12)
𝜇 𝜆 To obtain stress and elasticity tensors of the plane-stress condition
𝑊 = (𝐼 − 3) − 𝜇ln𝐽 + (ln𝐽 )2 , (6)
2 1 2 for the compressible neo-Hookean model, we first derive 𝑺 for the gen-
where 𝐼1 is the first invariant of 𝑪; 𝜇 and 𝜆 are the Lamé constants of eral case as already shown in Eq. (7a). Next, we impose the plane-stress
considered materials which can be determined from Young modulus, 𝐸, condition (the out of plane stress field is equal to zero) to find the re-
and Poisson’s ratio, 𝜈. lations between the stretch in the third direction and the others, and
In this study, 2D problems are considered either in plane-strain or obtaining the in-plane components of 𝑺. Then, the in-plane components
plane-stress conditions. We refer to the spatial description, stress and of the Cauchy stress tensor are calculated from the in-plane components
elasticity tensors are then all written in the Eulerian setting. of 𝑺. For the in-plane components of the material elasticity tensor, they
are calculated from the in-plane components of 𝑺. Finally, the in-plane
components of the spatial elasticity tensor are obtained from the in-
2.2.1. General cases & plane strain states
plane components of the material elasticity tensor.
From Eqs. (4), (5) and the Ref. [21], the second Piola-Kirchhoff
Let us consider Eqs. (7a), and (9), and using the plane-stress condi-
stress, the Cauchy stress, the material elasticity tensor (ℂ) and the spa-
tion (𝑆33 = 0) to obtain
tial elasticity tensor in component form (ℂ𝑖𝑗𝑘𝑙 ) are given, respectively,
[ ( ) ]
by 𝜌 1 1
𝑆33 = 𝜇 1− + 𝜆 ln 𝐽 = 0. (13)
𝜌 𝜕𝑊 𝜌 [ ] 𝜌0 𝐶33 𝐶33
𝑺 =2 = 𝜇(𝑰 − 𝑪 −1 ) + 𝜆(ln𝐽 )𝑪 −1 , (7a)
𝜌0 𝜕𝑪 𝜌0 From Eq. (13), one gets
1 𝜌 [𝜇 𝜆
]
𝝈 = 𝑭 𝑺𝑭 𝑇 = (𝒃 − 𝑰) + (ln𝐽 )𝑰 , (7b) ( )
𝐽 𝜌0 𝐽 𝐽 𝜆 ln 𝐽 = 𝜇 1 − 𝐶33 . (14)
𝜕𝑺 𝜌 [ ]
ℂ=2 = 𝜆(𝑪 ⊗ 𝑪 ) + 2(𝜇 − 𝜆ln𝐽 )(𝑪 −1 ⊙ 𝑪 −1 ) ,
−1 −1
(7c) Substituting Eq. (14) into Eq. (7a) to yield the in-plane components
𝜕𝑪 𝜌0
𝑺̃ of 𝑺 as
1 𝜌 [ ′ ]
ℂ𝑖𝑗𝑘𝑙 = 𝐹 𝐹 𝐹 𝐹 ℂ
𝐽 𝑖𝐼 𝑗𝐽 𝑘𝐾 𝑙𝐿 𝐼𝐽 𝐾𝐿 𝜌0
= 𝜆 (𝛿𝑖𝑗 𝛿𝑘𝑙 ) + 𝜇′ (𝛿𝑖𝑘 𝛿𝑗𝑙 + 𝛿𝑖𝑙 𝛿𝑗𝑘 ) . (7d) 𝜌 [ ( ̃ ̃ −1 ) ( ) −1 ] 𝜌 [ ( )]
𝑺̃ = + 𝜇 1 − 𝐶33 𝑪̃ 𝜇 𝑰̃ − 𝐶33 𝑪̃
−1
𝜇 𝑰 −𝑪 = . (15)
𝜌0 𝜌0
In the above expressions, 𝜆′ = 𝐽𝜆 and 𝜇′ = 𝜇 − 𝐽𝜆ln𝐽 . From Eq. (15), the in-plane components 𝝈̃ of 𝝈 are calculated as
The constitutive laws described above are valid for both plane strain [ ]
states and general 3D problems. They are later used in the next subsec- 1 ̃ ̃ ̃𝑇 𝜌 𝜇 (̃ ̃
)
𝝈̃ = 𝑭 𝑺 𝑭 = 𝒃 − 𝐶33 𝑰 . (16)
tion in deriving the constitutive laws for plane stress states. 𝐽 𝜌0 𝐽

195
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

( )
𝜌
The in-plane components of the Lagrangian elasticity tensor are de- 𝜁̇ = −𝐻 𝜖 − , 𝜁 (𝑡 = 0) = 0, (23)
rived as 𝜌0
̃ [ ( ) ( )] where a non-dimensional precision constant, 0 < 𝜖 ≪ 1, is defined as
ℂ̃ = 2 𝜕 𝑺 = 𝜌 𝑚 𝑪̃ −1 ⊙ 𝑪̃ −1 + 𝑚 𝑪̃ −1 ⊗ 𝑪̃ −1 , (17)
𝜕 𝑪̃ 𝜌0 1 2 the criteria for deletion.
The constitutive law of the spatial mass flux with 𝜅 > 0 as the mass
with
conductivity for isotropic material is expressed as [15]
[ ( )𝑚 ]
𝑚1 = 2𝜇𝐶33 , (18a) 𝑤
𝒔 = 𝜅𝐻(𝜁 ) exp − ∇(𝜌). (24)
𝐽̃2 𝜙
𝑚2 = −2𝜇 , (18b)
𝑚3
{ [ ( Substituting Eqs. (22) and (24) into Eq. (21a), yields
)] }2
exp 𝜇𝜆 1 − 𝐶33 ( ) { [ ( )𝑚 ] ( )}
1 𝜇 𝑤 𝜌
𝑚3 = − +2 . (18c) ∇ ⋅ 𝑙 𝐻(𝜁 ) exp −
2
∇
𝐶33 𝐶33 𝜆 𝜙 𝜌0
[ ( )𝑚 ] ( ) (25)
It is noticed in Eq. (18) that the terms 𝑚1 , 𝑚2 , and 𝑚3 are just 𝜌
𝑤
sub-variables used to avoid long expressions for Eq. (17), and other + 𝐻(𝜁 ) exp − − = 0,
𝜙 𝜌0
following equations.
The in-plane components of the Eulerian elasticity tensor in the com- with 𝑙 being
√ the characteristic length and determined through the 𝜅 and
ponent form are given by 𝛽 as 𝑙 = 𝜅∕𝛽 to control the band width of damaged zones.
In addition, the energy limiter per unit referential volume is deter-
1 ̃ ̃ ̃ ̃ ̃
̃ 𝑖𝑗𝑘𝑙
ℂ = 𝐹 𝐹 𝐹 𝐹 ℂ mined as [14]
𝐽 𝑖𝐼 𝑗𝐽 𝑘𝐾 𝑙𝐿 𝐼𝐽 𝐾𝐿
{ }
𝜌 1 [ ( )] ( ) (19)
1 ̃ ̃ Φ = 𝜌0 𝜙. (26)
= 𝑚1 𝛿𝑖𝑘 𝛿𝑗𝑙 + 𝛿̃𝑖𝑙 𝛿̃𝑗𝑘 + 𝑚2 𝛿̃𝑖𝑗 𝛿̃𝑘𝑙 .
𝜌0 𝐽 2
With 𝑊 = 𝜌𝑜 𝑤 associated with Eq. (26), Eq. (25) can then be rewrit-
To complete the expressions of the stress and elasticity tensors in ten as
the cases of the plane stress condition for the neo-Hookean model, the { [ ( )𝑚 ] ( )}
𝑊 𝜌
component of the right Cauchy-Green deformation tensor in the third ∇ ⋅ 𝑙2 𝐻(𝜁 ) exp − ∇
direction, 𝐶33 , needs to be determined. As, there is no explicit expres- Φ 𝜌0
[ ( )𝑚 ] ( ) (27)
sion for 𝐶33 from the other components. A nonlinear equation needs to 𝑊 𝜌
+ 𝐻(𝜁 ) exp − − = 0.
be solved to obtain 𝐶33 . From Eq. (12), and after some mathematical Φ 𝜌0
manipulations, the expression for 𝐶33 is obtained from Eq. (14) as
In this paper, the element deletion [15] is used in numerical simu-
{ [𝜇 ( }
)] 2 lations, however this technique is not the same as the element deletion
𝐽̃2 𝐶33 = exp 1 − 𝐶33 . (20) method in [7] or element erosion in [11]. The element deletion adopted
𝜆
here is not only used for postprocessing purpose, i.e., to visualize the
In numerical applications, Eq. (20) is solved at each Gauss quadra-
damaged zone, but also employed for fulfilling the global irreversibil-
ture point by using Newton-Raphson method. Input data for solving
ity constraint of crack evolution. The step function 𝐻(𝜁 ) in Eq. (27) is
Eq. (20) is just the determinant (𝐽̃) of the in-plane deformation matrix
thus replaced by an indicator . Basically, this indicator describes the
(𝐹̃ ) and Lamé constants. The tolerance of 10−8 for the Newton-Raphson
completely damaged (or deleted) elements, we thus rewrite Eq. (27) as
loop when solving Eq. (20) is used in this work. Our numerical appli-
cation reveals that the loop just needs a few iterations (around 4) to { [ ( )𝑚 ] ( )} [ ( )𝑚 ] ( )
𝑊 𝜌 𝑊 𝜌
satisfy the given tolerance. ∇ ⋅ 𝑙2  exp − ∇ +  exp − − = 0.
Φ 𝜌0 Φ 𝜌0
2.3. Local mass-diffusion balance equations (28)
Eq. (28) is denoted as the original model (i.e., non-decomposed set-
The detailed derivation of the local mass-diffusion balance equa-
ting) [14,15] that will subsequently be used in the numerical experi-
tions enhanced by energy decomposition was presented in [17]. For
ments.
self-contained purpose, we briefly present some basic equations in this
The enhanced model using the energy decomposition, which was
subsection.
derived in [17], is denoted as the decomposed model in our numerical
By defining 𝜉 and 𝒔 as the spatial mass source (sink) and spatial mass
simulations:
flux, respectively, the spatial form of the local mass-diffusion balance
{ [ ( + )𝑚 ] ( )} [ ( + )𝑚 ] ( )
equation is then given by [15,17] 𝑊 𝜌 𝑊 𝜌
∇ ⋅ 𝑙2  exp − ∇ +  exp − − = 0,
𝑑𝜌 Φ 𝜌0 Φ 𝜌0
=∇⋅𝒔+𝜉 =0 in Ω, (21a)
𝑑𝑡 (29)
𝒔⋅𝒏=0 on 𝜕Ω. (21b)
where 𝑊+ is the positive part of SED as described in Ref. [17] which is
The mass source in [15] is adopted for this analysis
given as follows
{ [ ( )𝑚 ] }
𝑤
1 ∑[ + 2
3
𝜉(𝜌, 𝜌0 , 𝑤, 𝜙) = 𝛽 𝜌0 𝐻(𝜁 ) exp − −𝜌 , (22) ] 𝜆
𝜙 𝑊+= 𝜇 𝑖 + 2 (ln𝐽 ) ,
(𝜆 ) − 1 − 2ln𝜆+ + 2
(30)
2 𝑖=1 𝑖
in which 𝜙 indicates the specific energy limiter per unit mass; 𝛽 > 0
where
defines a material constant; while 𝑚 is a dimensionless parameter con-
{ {
trolling the sharpness of failure transition between virgin material and 𝜆𝑖 for 𝜆𝑖 > 1 𝐽 if 𝐽 > 1
damaged state via the stress-strain relation. 𝜆𝑖 =
+
𝐽 =
+
, (31)
1 for 𝜆𝑖 ≤ 1 1 if 𝐽 ≤ 1
In Eq. (22), 𝐻(𝜁 ) is defined as a step function, and 𝐻(𝜁 ) = 0 if 𝜁 < 0
and 𝐻(𝜁 ) = 1 otherwise. To prevent the non-physical heading of mate- with 𝜆𝑖 representing the 𝑖𝑡ℎ principal stretch. They can be determined by
rials after damage, the switch parameter 𝜁 ∈ (−∞, 0] and its evolution solving the eigenvalue problem for the right Cauchy-Green deformation
characteristic are additionally defined as [15] tensor, 𝑪.

196
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

The associated boundary condition for Eqs. (28) and (29) is 𝒂 = 𝑵 𝒖 𝒂, ̂ 𝒖 = 𝑵 𝒖 𝒖,

̂ 𝒗 = 𝑵 𝒖 𝒗, ̂

̂ ∇𝒖 = 𝑩 𝒖 𝒖,
𝛿𝒖 = 𝑵 𝒖 𝛿 𝒖, ̂
̂ ∇𝛿𝒖 = 𝑩 𝒖 𝛿 𝒖, (36)
∇𝜌 ⋅ 𝒏 = 0 on 𝜕Ω. (32)
𝜌 = 𝑵 𝜌 𝝆, ̂ ∇𝜌 = 𝑩 𝜌 𝝆,
̂ 𝛿𝜌 = 𝑵 𝜌 𝛿 𝝆, ̂
̂ ∇𝛿𝜌 = 𝑩 𝜌 𝛿 𝝆,
Remark #1: As already addressed in [17], it is important to stress
with
out here again that the local mass-diffusion balance equations consist of
five material-dependent quantities, i.e., 𝜌0 , 𝑚, 𝑙, Φ and 𝑊 , which can [ ] ⎡ 𝑁1,1 0 ... 𝑁4,1 0 ⎤
𝑁1 0 ... 𝑁4 0
be calibrated from the experiments. The above SED, 𝑊 , is a function of 𝑵𝒖 = , 𝑩𝒖 = ⎢ 0 𝑁1,2 ... 0 𝑁4,2 ⎥,
0 𝑁1 ... 0 𝑁4 ⎢ ⎥
material constants and the gradient deformation tensor. ⎣ 𝑁1,2 𝑁1,1 ... 𝑁4,2 𝑁4,1 ⎦
[ ]
[ ] 𝑁1,1 ... 𝑁4,1
3. Dynamic discretized equations and their linearization 𝑵 𝜌 = 𝑁1 ... 𝑁4 , 𝑩 𝜌 = ,
𝑁1,2 ... 𝑁4,2
(37)
The weak forms of momentum and mass-diffusion equations and
detailed implementation using FEM are presented. Also, a brief on stag- where 𝒂, ̂ and 𝒖̂ are element nodal acceleration, velocity and dis-
̂ 𝒗,
gered algorithm in [17] for solving governing equations of the coupled placement vectors, respectively; and 𝝆̂ is the nodal density vector; 𝑁𝐼
localized MF damage model is given, additionally discussing some other (𝐼 = 1 ∶ 4) denote the interpolation shape functions of the Q4 element;
𝜕𝑁
numerical issues. 𝑁𝐼,𝑖 = 𝜕𝑥 𝐼 (𝑖 = 1, 2) are the spatial derivatives of the shape functions
𝑖
with respect to directions 𝑥𝑖 .
3.1. Weak form By substituting Eq. (36) into Eqs. (33) and (35), they yield1
( )
The momentum equation (1a) and the mass-diffusion balance equa- 𝛿 𝒖̂ 𝑇 ⋅ ̂ + 𝛿 𝒖̂ 𝑇 ⋅
𝜌 𝑵 𝑇𝒖 𝑵 𝒖 𝒂d𝑣 𝑩 𝑇𝒖 {𝝈}d𝑣
∫ ∫
tions (28) and (29) are transformed to their weak forms by applying Ω𝑒 Ω𝑒
(38a)
the standard Bubnov-Galerkin procedure with the test functions 𝛿𝒖 for 𝑇
= 𝛿 𝒖̂ ⋅ 𝜌𝑵 𝑇𝒖 𝒃𝑔 d𝑣 + 𝛿 𝒖̂ 𝑇 ⋅ 𝑵 𝑇𝒖 𝒕̄d𝑠,
displacement field and 𝛿𝜌 for density field. ∫ ∫
Ω𝑒 Γ𝑒𝑡
It is important to point out here that, since the mass balance equa-
tions of the original model and the decomposed one are similar to each ⎛ ⎞
⎜ 𝑙2 𝑇 𝑵 𝑇𝜌 𝑵 𝜌 ⎟
other, only the weak form of the decomposed model is discussed. 𝛿 𝝆̂ 𝑇 ⋅ 𝑩 𝑩 ̂ = 𝛿 𝝆̂ 𝑇 ⋅ 𝑵 𝑇𝜌 d𝑣,
∫ ⎜⎜ 𝜌0 𝜌 𝜌
+ [ ( )𝑚 ] ⎟ 𝝆d𝑣 ∫
𝑊 +,𝑒
The weak form of the momentum equation (1a) following the stan-
Ω𝑒 ⎝ 𝜌0  exp − Φ +𝜀 ⎟ Ω𝑒
dard Bubnov-Galerkin procedure is written as follows: ⎠
(38b)
𝜌𝒂 ⋅ 𝛿𝒖d𝑣 + 𝝈 ∶ ∇𝛿𝒖d𝑣 = 𝒃𝑔 ⋅ 𝛿𝒖d𝑣 + 𝒕̄ ⋅ 𝛿𝒖d𝑠, (33) where
∫ ∫ ∫ ∫
Ω Ω Ω Γ𝑡 [ ]𝑇
{𝝈} = 𝜎11 𝜎22 𝜎12 , (39)
where 𝒂 is the spatial acceleration of a material point. As noticed from
is the Voigt notation form of the Cauchy stress. is the SED per unit 𝑊 +,𝑒
Eq. (21a) that 𝑑𝜌
𝑑𝑡
= 0, therefore in Eq. (1a), 𝑑(𝜌𝒗)
𝑑𝑡
= 𝜌 𝑑(𝒗)
𝑑𝑡
= 𝜌𝒂. The first
referential volume for virgin materials of the element [27]:
term in the above expression is accordingly derived.
By applying the same procedure to Eq. (29), the weak form can be 1
𝑊 +,𝑒 = 𝑊 + d𝑉 , (40)
obtained as 𝑉0𝑒 ∫
Ω𝑒0
( 2 [ ( + )𝑚 ]
𝑙 𝑊
 exp − ∇2 (𝜌)𝛿𝜌 with 𝑉0𝑒 being the initial volume of the element.
∫ 𝜌0 Φ
Ω The final system of discrete equations of the governing equations is
[ ( + )𝑚 ] ) (34)
obtained
𝑊 1
+ exp − 𝛿𝜌 − 𝜌𝛿𝜌 d𝑣 = 0.
Φ 𝜌0
𝑴 𝒂̂ + 𝑭 int = 𝑭 ext , (41a)
The staggered algorithm developed in [17] is used to solve the cou- 𝜌 𝜌
𝑲 𝝆̂ = 𝑭 , (41b)
pled momentum - local mass-diffusion balance equations. As a result,
int ext
at the time when the mass balance equation is solved, two variables where 𝑴 , 𝑭 , 𝑭 are the consistent mass matrix, the internal force
𝑊 + and  of an element are known. With these notices, by applying vector and the external force vector, respectively; and 𝑲 𝜌 , 𝑭 𝜌 are, re-
the integration by parts to Eq. (34) and using the associated boundary spectively, the stiffness matrix and the force vector of the mass balance
condition Eq. (32), one can obtain equation, and are determined as

⎛ ⎞ 𝑴= 𝜌𝑵 𝑇𝒖 𝑵 𝒖 d𝑣, (42a)
⎜ 𝑙2 𝜌𝛿𝜌 ⎟ ∫
⎜ ∇𝜌 ⋅ ∇𝛿𝜌 + [ ( + )𝑚 ] ⎟ d𝑣 = ∫ 𝛿𝜌d𝑣, (35) Ω𝑒
∫ ⎜ 𝜌0 𝑊
𝜌0  exp − Φ +𝜀⎟
Ω ⎝ ⎠ Ω 𝑭 int = 𝑩 𝑇𝒖 {𝝈}d𝑣, (42b)
∫
where 0 < 𝜀 ≪ 1 is a small positive number that is used for avoiding Ω𝑒

numerical instabilities i.e., the case of damaged elements  = 0, and in 𝑭 ext = 𝑵 𝑇𝒖 𝒃𝑔 d𝑣 + 𝑵 𝑇𝒖 𝒕̄d𝑠, (42c)
the numerical simulations 𝜀 = 10−10 is typically used. ∫ ∫
Ω𝑒 Γ𝑒𝑡

⎛ ⎞
3.2. FEM discretization
⎜ 𝑙2 𝑇 𝑵 𝑇𝜌 𝑵 𝜌 ⎟
𝜌
𝑲 = ⎜ 𝑩𝜌 𝑩𝜌 + [ ( +,𝑒 )𝑚 ] ⎟ d𝑣, (42d)
∫ ⎜ 𝜌0 𝑊
In the numerical experiments, both the four-node quadrilateral (Q4) Ω𝑒 ⎝ 𝜌0  exp − Φ +𝜀⎟
⎠
and isoparametric eight-node brick (B8) elements are used for 2D and
3D problems. Here, only the formulae related to Q4 element are de-
scribed. Extension to 3D is however straightforward, and not given. The 1 Without loss of generality, the discrete forms of the weak forms are given

discretization formulae are at element level.

197
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

𝑭𝜌 = 𝑵 𝑇𝜌 d𝑣. (42e) In Eq. (49), the spatial constitutive matrix 𝑫 is constructed from
∫
Ω𝑒 the components of the fourth-order tensor ℂ𝑖𝑗𝑘𝑙 . The details of 𝑫, 𝑩 𝐺 ,
Σ𝝈 in cases of 2D problems can be found in [17]. Curious readers can
One should notice that, to derive the weak form of the local mass-
also refer to the appendix A for a brief derivation of 𝐾𝑇 and its related
diffusion balance equation in the case of the original model, the term
terms.
𝑊 +,𝑒 in Eq. (42d) is simply replaced by 𝑊 𝑒 . In a similar manner, the
term 𝑊 is used in Eq. (40) instead of 𝑊 + . 3.4. A staggered algorithm
Since, Eq. (41a) is the semi-discrete equation of the weak form of the
momentum equation, Eq. (41a) needs further discretization in time in The staggered scheme developed in [17] is adopted here. Consid-
order to be solved. In this study, the Newmark time-integration method ering a finite time increment [𝑡𝑛 , 𝑡𝑛+1 ], with time step Δ𝑡 = 𝑡𝑛+1 − 𝑡𝑛 . At
is adopted for our time discretization purpose. Eq. (41a) for a specific the loading time 𝑡𝑛 , the staggered solution assumes all variables includ-
time step 𝑡𝑛+1 is given by ing 𝒂𝑛 , 𝒗𝑛 , 𝒖𝑛 , 𝜌𝑛 , 𝑛 are known a priori. Then, at time increment step
𝑡𝑛+1 , the variables/fields are updated accordingly as described in the
𝑴 𝑛+1 𝒂̂ 𝑛+1 + 𝒁 𝑛+1 𝒗̂ 𝑛+1 + 𝑭 int
𝑛+1
= 𝑭 ext
𝑛+1
. (43) following general procedure: (a) In the first step, the acceleration, 𝒂𝑛+1 ,
Note that in Eq. (43) we have additionally introduced the damping velocity, 𝒗𝑛+1 and displacement 𝒖𝑛+1 fields are updated by solving the
effect, which is represented by 𝒁. This introduction is only applied for momentum equation and using the frozen density field 𝜌𝑛 and flag vari-
the numerical example by Grégoire’s experiment, which is analyzed in able 𝑛 ; (b) The term, 𝑊𝑛+1 /𝑊𝑛+1 +
, is then updated, as the next step,
subsection 6.4. using the displacement field 𝒖𝑛+1 and the flag variable 𝑛 ; (c) We next
With the aid of Newmark method [28], the approximations of the account for the density field 𝜌𝑛+1 , which is calculated from the displace-
accelerations and velocities at the time step 𝑡𝑛+1 can be expressed as ment field 𝒖𝑛+1 , the flag variable 𝑛 and the SED 𝑊𝑛+1 /𝑊𝑛+1 +
; In the last
step, step (d), the flag variable 𝑛+1 is estimated using the density field
( )
𝒂̂ 𝑛+1 = 𝑎0 𝒖̂ 𝑛+1 − 𝒖̂ 𝑛 − 𝑎2 𝒗̂ 𝑛 − 𝑎3 𝒂̂ 𝑛 , (44a) 𝜌𝑛+1 .
( )
𝒗̂ 𝑛+1 = 𝑎1 𝒖̂ 𝑛+1 − 𝒖̂ 𝑛 − 𝑎4 𝒗̂ 𝑛 − 𝑎5 𝒂̂ 𝑛 , (44b) 4. MF damage model’s parameter calibration & estimation of
1 𝛼𝑁 1 1 𝛼𝑁 crack tip velocity, dissipated energy, and amount of mass loss
where 𝑎0 = 𝑎1 =
𝛽𝑁 (Δ𝑡)2
,𝑎2 = 𝑎3 =
𝛽𝑁 Δ𝑡
, − 1, 𝑎4 =
𝛽𝑁 Δ𝑡
− 1, 𝑎5 =
, 𝛽𝑁
( ) 2𝛽𝑁
𝛼𝑁
Δ𝑡 2𝛽 − 1 , with 𝛼𝑁 and 𝛽𝑁 being the parameters of the Newmark- 4.1. MF damage model’s parameter calibration
𝑁
1 1
beta method, and in this study 𝛼𝑁 = and 𝛽𝑁 = are used.
2 4 Basically, the developed MF damage models require five material
Substituting Eqs. (44) into Eq. (43) to obtain
constants such as 𝜌0 , 𝑚, 𝑙, Φ, 𝑊 ∕𝑊 + . The determination of these
[ ( ) ]
𝑴 𝑛+1 𝑎0 𝒖̂ 𝑛+1 − 𝒖̂ 𝑛 − 𝑎2 𝒗̂ 𝑛 − 𝑎3 𝒂̂ 𝑛 parameters is the same as reported in [17] for finite deformations, in
[ ( ) ] (45) which the characteristic length 𝑙 can be determined as [30]
+ 𝒁 𝑛+1 𝑎1 𝒖̂ 𝑛+1 − 𝒖̂ 𝑛 − 𝑎4 𝒗̂ 𝑛 − 𝑎5 𝒂̂ 𝑛 + 𝑭 int
𝑛+1
= 𝑭 ext
𝑛+1
.
𝛾 𝛾
𝑙= = ( ), (50)
In the descriptions above, (∙)𝑛 means the field/term (∙) at the time 𝜓𝑓 Φ
Γ 1
,0
step 𝑡𝑛 and similarly, (∙)𝑛+1 implies the field/term (∙) at the time step 𝑚 𝑚

𝑡𝑛+1 . in which 𝛾 denotes the surface failure energy and can be obtained from
experiment2 ; 𝜓𝑓 represents the volumetric energy dissipated during the
3.3. Linearization damage or it can be understood as the constant bulk failure energy
defined in [33]; while Γ(𝑠, 𝑥) = ∫𝑥 𝑡𝑠−1 𝑒−𝑡 d𝑡 is the upper incomplete
∞

The staggered algorithm [17] is adopted to solve the coupled gamma function.
momenta-localized mass balance equations with, first solving the dis- In terms of small deformations, and the case where the compressible
placement field, then determining the mass density field. The staggered neo-Hookean model Eq. (6) is considered, the process to achieve the re-
scheme leads the discretized equations of the local mass balance laws quired parameters of MF damage model can be simply done by firstly
to linear algebraic equations, which can be solved by directly inverting determining Young modulus 𝐸, Poisson ratio 𝜈, initial density 𝜌0 , sur-
the stiffness mass matrix. However, the discrete form of the momentum face fracture energy 𝛾 and critical stress 𝜎𝑐 of the material, and then
equation (45) is naturally a nonlinear algebraic equation which could conducting the uniaxial tensile test numerically and fitting 𝑚 and Φ to
be solved by using Newton-Raphson iterative process. For more details the stress-stretch curve, finally, determining the characteristic length
of the Newton-Raphson algorithm used to solve the nonlinear behavior from Φ, 𝛾, 𝑚 through Eq. (50).
in hyperelasticity, one can refer to Refs. [21,29] for example. As an example, we particularly consider a unit cube under uniaxial
The weak form of the momentum equation needs to be linearized tensions, see Fig. 1. The cube shown in Fig. 1 with dimension of 1 m
by means of Newton-Raphson algorithm. The linearization combining on each edge is modeled as one B8 element. Face 4 (i.e., nodes 2-3-
with the FEM discretization leads Eq. (45) to the form of finite element 7-6) is extended in the positive 𝑥1 direction. The following boundary
matrix as conditions are given: 𝑢1 = 0 at Face 6 (i.e., nodes 1-4-8-5), 𝑢2 = 0 at
Face 3 (i.e., nodes 1-2-6-5), and 𝑢3 = 0 at Face 1 (i.e., nodes 1-2-3-4).
𝑲 𝒖𝒖 Δ𝒖̂ 𝑛+1 = −𝑹𝒖𝒖 , (46) This test will be used later in the numerical applications of Section 6 to
determine material constants of MF damage models when the critical
where 𝑲 𝒖𝒖 , 𝑹𝒖𝒖 are the effective tangent matrix and the residual vector,
stress of the material is given.
accordingly, which are calculated through
4.2. Estimation of crack tip velocity, dissipated energy and amount of mass
𝑲 𝒖𝒖 = 𝑴 𝑛+1 𝑎0 + 𝒁 𝑛+1 𝑎1 + 𝑲 𝑇 𝑛+1 , (47)
[ ( ) ] loss
𝑹𝒖𝒖 = 𝑴 𝑛+1 𝑎0 𝒖̂ 𝑛+1 − 𝒖̂ 𝑛 − 𝑎2 𝒗̂ 𝑛 − 𝑎3 𝒂̂ 𝑛
[ ( ) ] (48) In the present formulation, the crack tip velocity 𝑣𝐶
𝑛+1
at time step
+𝒁 𝑛+1 𝑎1 𝒖̂ 𝑛+1 − 𝒖̂ 𝑛 − 𝑎4 𝒗̂ 𝑛 − 𝑎5 𝒂̂ 𝑛 + 𝑭 int − 𝑭 ext ,
𝑛+1 𝑛+1 𝑡𝑛+1 is simply calculated as
in which 𝑲 𝑇 is the spatial tangent stiffness matrix, and defined as
( 𝑇 )
𝑩 𝒖 𝑫𝑩 𝒖 + 𝑩 𝑇𝐺 Σ𝝈 𝑩 𝐺 d𝑣.
2 𝛾 should be understood as the fracture energy in terms of phase-field model
𝑲𝑇 = (49)
∫ 𝐺𝑐 [31,32].
Ω𝑒

198
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 1. A unit cube under tensions. (a) Geometry, (b) Reference and current states.

Fig. 2. Schematic of tracking the crack-tip, with (a) the crack tip at time 𝑡𝑛 and (b) the crack tip at time 𝑡𝑛+1 . The crack tip is easily located from detecting deleted
elements.

Δ𝑙𝑛+1 Δ𝑙𝑛+1
𝑉 total sink 𝑉 total sink Φ 𝑈𝑑
total
𝑚total sink 𝑉
total sink 𝜌
𝑣𝐶
𝑛+1
= = , (51) 𝑚loss = = 0
= = = , (53)
Δ𝑡𝑛+1 𝑡𝑛+1 − 𝑡𝑛 𝑚0 𝑉0 𝜌0 𝑉0 𝑉0 Φ 𝑉0 Φ
where Δ𝑙𝑛+1 is the travel distance of the crack √ tip within a time step where 𝑚total sink , 𝑚0 are the total mass loss and the initial mass of the
[𝑡𝑛 , 𝑡𝑛+1 ] which is determined by Δ𝑙𝑛+1 = (𝑥𝑛+1 − 𝑥𝑛 )2 + (𝑦𝑛+1 − 𝑦𝑛 )2 , sample, respectively; 𝑉 total sink , 𝑉0 are the total volume generated by the
where [𝑥𝑛 , 𝑦𝑛 ] and [𝑥𝑛+1 , 𝑦𝑛+1 ] are the current coordinates of the crack deleted elements and the initial volume of the specimen, accordingly;
tips at time 𝑡𝑛 and 𝑡𝑛+1 in plane problems, respectively. Basically, when and 𝑈𝑑total is the total amount of energy dissipation. It is noticed that
tracking the crack-tip, we first have the information of deleted elements in the case when the thickness is not considered, the term 𝑉0 will be
(from the step function, ), then the crack-tip is traced by on the co- replaced by 𝐴0 with 𝐴0 being the corresponding reference area of 2D
ordinates of nodes belonging to the totally damaged area. The crack-tip specimen.
is simply assumed to be coincident to the tip of the crack path depicted
by deleted elements whose are schematically depicted in Fig. 2. 5. Validation: plane stress vs plane strain under quasi-static
The dissipated energy during the numerical simulation is also calcu- shear loading
lated. The energy dissipation (𝑈𝑑 ) when cracks occur could be simply
calculated as [15] Although the validation for the decomposed model has been ana-
lyzed in [17] for the plane strain, yet the plane stress assumption has
𝐸𝑁
∑ not been examined. To this end, a single notched edge under shear
𝑈𝑑 = Φ 𝑉0𝑖 , (52)
𝑖=1 loading is examined [6,31,34,35] and both 2D state assumptions (plane
stress and plane strain) are studied.
where 𝐸𝑁 is the number of deleted elements (the elements with  = The geometry and boundary conditions of the single notched edge
0); 𝑉0𝑖 is the referential volume of the 𝑖𝑡ℎ element. In the cases of 2D are sketched in Fig. 3 (left). The material parameters of the SED
analysis and if the thickness of the specimen is not considered when function and the surface failure energy are 𝜆 = 121.15 kN/mm2 , 𝜇 =
calculating 𝑈𝑑 , the referential volumes of deleted element, 𝑉0𝑖 , will be 80.77 kN/mm2 , 𝛾 = 2.7 × 10−3 kN/mm [34]. Other parameters required
replaced by the corresponding referential areas, 𝐴𝑖0 . Under this setting, for the MF damage models are chosen as 𝑚 = 5, 𝑙 = 0.015 mm, 𝜌0 =
∑𝐸𝑁 𝑖
𝑈𝑑 would be the energy per unit depth, i.e., 𝑈𝑑 = Φ 𝑖=1 𝐴0 which has 2400 kg/m3 , 𝜖 = 0.001. The energy limiter, Φ, is deduced from 𝛾, 𝑚 and
the unit of J/m. 𝑙 through Eq. (50). A mesh of 48261 Q4 elements with an effective size
The percentage of mass loss compared to the whole body can be ℎ ≈ 0.0025 mm shown in Fig. 3 (right) is adopted. The displacement con-
quantitatively estimated as follows: trol with a fixed increment Δ𝑢 = 1 × 10−4 mm is used. The simulation

199
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 3. Single notched edge quasi-static shear test. Geometry (unit of length: mm) and boundary conditions (left); FEM mesh (right).

Fig. 4. Single notched edge quasi-static shear test. Comparison of the crack paths (normalized density fields) at the final stage of the simulations between the original
and decomposed models corresponding to plane strain and plane stress assumptions.

is set to be automatically terminated once either cracks reach the ex- that both structural load-displacement curves and final crack patterns
ternal boundary or the distance between the crack-tip and the external are not much different from the plane strain and plane stress states.
boundary is around 0.004 mm.
The crack paths, which are represented by the normalized density 6. Numerical experiments and discussions
fields, for both plane-strain and plane-stress assumptions are calculated
and given in Fig. 4. In this test, it is expected that cracks should grow To show the accuracy and performance of the developed decom-
in the region where are mainly subjected to tension, i.e., bottom part posed approach for dynamic description of the localized mass-field
damage model in modeling brittle dynamic fracture using standard
of the specimen. Any cracks that occur in the areas subjected to com-
FEM, five numerical examples with complex configurations including
pression, i.e., upper part, should be unrealistic. Simulation results for
fragmentation simulation are considered in this section. Validation of
both plane stress and strain are shown in Fig. 4, which clearly exhibit
the computed dynamic damage results (e.g., crack patterns, crack veloc-
the advantages of the developed decomposed model over the original
ity, dissipated energy) is based on comparisons with reference solutions
counterpart. When the original model is used, generally, non-physical obtained by either experiments or other numerical techniques.
cracks develop in the compression domains.
We further show in Fig. 5 a comparison of the structural load- 6.1. Kalthoff and Winkler’s experimental test
displacement curves between the present theory and the phase-field
[35]. The figure confirms the accuracy of the decomposed model as The first numerical example of dynamic brittle fracture analysis
compared with the reference result. The original model however yields deals with a well-established experiment reported by Kalthoff and Win-
less accurate results. It is interesting to observe in the discussed results kler [36] and Kalthoff [37]. In the experiment, a double pre-notched

200
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 6. Kalthoff’s experiment. Geometry (unit of length: mm) and experimental setup (left); Upper half of the specimen modeled due to twofold symmetry (right).
The experimental result of the crack path by Kalthoff [37] is represented by a thick blue line.

ing the structure to be stable. If the same boundary condition in Fig. 6

is used for quasi-static analysis or in the dynamic simulation with a
low impact velocity (nearly quasi-static condition), it will not work as
the specimen will have rigid body movement following the horizon-
tal direction. However, under the dynamic loading (with high impact
loading velocities), it can be figured out that the inertia would create
the resistance which makes the dynamic fracture occurs. The boundary
conditions shown in Fig. 6 are the nature of the impact problem and in
consistent with the experimental testing setups by Kalthoff and Winkler
[36] and Kalthoff [37]. The constraints used in Fig. 6 (right) are just for
the symmetrical analysis as only half of the specimen is employed.
The material parameters of maraging steel are 𝐸 = 190 GPa and
𝜈 = 0.3, 𝜌0 = 8000 kg/m3 [6]. This leads to dilatational, shear, and
Rayleigh wave speeds of 𝑣𝑑 = 5654 m/s, 𝑣𝑠 = 3022 m/s, 𝑣𝑅 = 2803
m/s, respectively. To determine the other required parameters includ-
ing 𝑙, 𝑚, Φ, the surface failure energy 𝛾 = 2.213 × 104 N/m, and critical
stress 𝜎𝑐 = 1.07 GPa [6] are also considered. With the given 𝐸, 𝜈 and
𝜎𝑐 , the uniaxial tensile test is conducted numerically by using a unit
cube as mentioned on subsection 4.1. Fig. 7 (right) shows the Cauchy
stress-stretch curve and necessary information about the other required
Fig. 5. Single notched edge quasi-static shear test. Comparison of the global
material parameters for the MF damage models obtained from numeri-
load–displacement curves between the two MF damage approaches and the
phase-field model [35]. cally uniaxial tensile test with 100 steps of calculation. In the figure, a
thick dashed line indicates the Cauchy stress-stretch curve of the intact
material which can be obtained from elastic theory.
steel plate was subjected to impact loading with different impact veloc-
In addition to the quasi-static loading conditions study in Section 5,
ities as shown in Fig. 6 (left). The authors pointed out that the failure
here in terms of dynamic damage analysis, the developed decomposed
transits from brittle to ductile fracture when increasing the velocity of
model is again verified to exhibit its major advantage over the origi-
the striker. In literature, different numerical schemes have been applied
nal counterpart in eliminating the developing of crack in compression
to investigate this experiment, see for example Refs. [6–8,35,38–40]. In
domains. To this end, a finite element mesh consisting of 80110 Q4 ele-
this paper, only the brittle mode is modeled with an impact velocity
ments with refinement of ℎ ≈ 0.3 mm is used. The adopted mesh used for
𝑣∗ = 33 m/s. Brittle fracture was experimentally observed with the an-
the analysis is given in Fig. 7 (left). The simulation time is 33 μs with
gles of crack path about 70◦ from the original crack plane.
a fixed step size of time chosen as Δ𝑡 = 0.2 μs. As observed from the
The plane strain state is assumed. The input geometry and load-
Kalthoff’s experiment sketched in Fig. 6, one should notice that under
ing conditions for numerical simulations are shown in Fig. 6, where
such boundary conditions, the part below the pre-crack mainly suffers
symmetry condition is employed to reduce the computational cost. The
from compression while the upper part is dominantly subjected to ten-
impactor is modeled by applying the kinematic velocity instead, see [6]
sion. This observation is important for the performance analysis of the
as
decomposed model. The simulation results (e.g., stress and normalized
{𝑡
𝑣0 if 𝑡 ≤ 𝑡0 , density fields) are finally represented in Fig. 8. Not surprisingly, the
𝑣 = 𝑡0 (54) original model, similar to quasi-static loading test, is not able to repro-
𝑣0 if 𝑡 > 𝑡0 ,
duce the crack patterns correctly, i.e., cracks grow in the compression
where 𝑣0 = 16.5 m/s and 𝑡0 = 1 μs. There are no other constraints domains, whereas the decomposed model is capable of well capturing
applied to all unspecified surfaces. It is worth mentioning here that for the evolution of crack. Therefore the decomposed model is used for the
the dynamic impact loading, the inertia plays an important role as help- rest of analysis.

201
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 7. Kalthoff’s experiment. A mesh using for verifying models (left), and Cauchy stress versus stretch curve and necessary information about the other required
parameters for MF damage models (numerical tensile test) (right).

Fig. 8. Kalthoff’s experiment. Comparison of the computed stress and normalized density fields of the two models at time 𝑡 = 33 μs. Note that the stress fields are
plotted in the current configuration while the normalized density fields are shown in the reference configuration. Here, the deleted elements are removed from the
mesh to plot the crack paths in the stress fields.

Next, three different refined meshes are particularly considered, and Table 1
they are detailed in Table 1 and Fig. 9. The total simulation time is Kalthoff’s experiment. Details of the meshes.
100 μs with a fixed step size of time chosen as Δ𝑡 = 0.2 μs. The com- Mesh Numbers of Q4 element Effective size, ℎ [mm]
puted final crack patterns of the Kalthoff’s experiment are then plotted #1 31809 0.2
in Fig. 9. The average angles of the simulated cracks are around 68◦ #2 55180 0.15
which are in excellent agreements to the experimental result of 70◦ . For #3 122967 0.1

three considered meshes, the differences in the obtained final crack pat-
terns are indistinguishable, shedding light on good performance of the
low 60% of the Rayleigh wave speed, up to the time 𝑡 ≈ 80 μs. Then it
developed model. In addition, the evolution of the normalized density decreases when cracks reach to the top surface. The simulated crack tip
fields and stress fields in horizontal (𝑥1 ) direction at different time steps velocities also agree well with those derived from phase-field models,
are given in Fig. 10. To represent the crack path in the stress fields, the see [6,35] for example.
deleted elements are removed from the mesh. Here, the stress fields are The dissipated energy for three given meshes in comparison with
plotted in current configurations while the others are shown in the ref- the results from phase field model [6] and cracking node method [8]
erence state. is depicted in Fig. 12. The results show the mesh convergence in terms
The crack tip velocities for three considered meshes are also evalu- of dissipated energy of the proposed model. In the figure, 𝑈𝑑theoretical is
ated and represented in Fig. 11. We obverse from our numerical experi- calculated for the ideal case of a straight crack propagating at 70◦ from
ment, also seen in the figure, that the cracks start propagating at slightly the initial crack. In that case the total length of the crack is about 79.81
different time. However, all results in general agree well to each other. mm, with the given surface fracture energy of 𝛾 = 2.213 × 104 N/m,
The cracks accelerate at the time 𝑡 ≈ 18 μs to a velocity of 1250 m/s be- the total dissipated energy is 1766.27 J/m. Interestingly, the compared

202
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 9. Kalthoff’s experiment. Three considered meshes (top) and final crack patterns corresponding to each mesh are computed using the decomposed model.

Fig. 10. Kalthoff’s experiment. Stress fields and normalized density fields for Mesh 3 at times (a) 𝑡 = 16 μs, (b) 𝑡 = 43.6 μs, (c) 𝑡 = 71.6 μs, (d) at the end of the
simulation, 𝑡 = 100 μs. For the stress fields, two color bars are used: the left color bar is for the time steps a, b, c while the other is for the last time step.

results of the dissipated energy clearly indicate that the present work
and also the cracking node method by Song and Belytschko [8] are able
to predict the total dissipated energy whereas the phase field model by
Borden et al. [6] fails to do so. It yields a large dissipated energy (by
about > 40%) as compared to the estimated theoretical value. In fact,
other phase field models, see [35,38,40] also yield the same overesti-
mation, and that wrong estimation of the dissipated energy by the phase
field models is critical, nor concrete explanations about this unrealistic
issue have been provided in the literature. For estimation of the mass
loss for this dynamic shear problem, it is determined from Eq. (53) and
𝑈𝑑total 1691.7
Fig. 12 as: 𝑚loss = 𝐴0 Φ
= (0.12 )×(16.3540×106 )
= 1.03%. In other words, as
compared with the half of the specimen, the mass loss quantified in this
example is about 1.03%.

6.2. Rittel’s experiment

Next example is devoted to comparison study of an experiment

on a compact compression specimen made out of PMMA (Polymethyl
methacrylate) as shown in Fig. 13 (left). The information of this Rit-
Fig. 11. Kalthoff’s experiment. Crack tip velocity. tel’s experiment is based on Ref. [41]. Curious readers can further

203
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Table 2
Fragmentation simulation. Comparison on the numbers of large fragments
among this study and other reference results reported in literature.
Result of Numbers of major fragments

Cracking particles method [46] 12, 15, and 16

Hybrid discontinuity tracking-fitting method [8] 18, 19, and 20
Adaptive cohesive zone [43] 20, 23, and 24
Adaptive splitting of polygonal finite elements [48] 18 and 23
This study: Case 1 12
This study: Case 2 15

tively. The evolution of the stress fields and normalized density fields
for this example are also provided and depicted in Fig. 16.
In addition, in this example, the calculated results using the original
non-split SED model are shown to reconfirm the advance of the de-
composed model against its non-split counterpart for dynamic fracture
analysis. From Fig. 17, one can easily see that, the original non-split
strain energy model fails to simulate this test. In other words, the
obtained crack pattern from the original model is different from the
Fig. 12. Kalthoff’s experiment. Comparison of the dissipated energy for dif-
experimental one.
ferent meshes among the present model, cracking node method by Song and
Belytschko [8], and phase field model by Borden et al. [6]. Note that in Song
and Belytschko [8], a slightly difference in the value of surface failure energy 6.3. Fragmentation simulation
was used as 𝛾 = 2 × 104 N/m. On the legend, Borden et al. (Mesh 2) and Bor-
den et al. (Mesh 5) mean the results from meshes 2 and 5, respectively used in Fragmentation simulation of a thick cylinder subjected to an im-
the work of Borden et al. [6]. pact load as depicted in Fig. 18 is investigated. The main objective of
this fragmentation simulation is to show the capability of the decom-
refer to a similar experiment, which was also conducted by Rittel and posed model in modeling catastrophic multiple cracks problem. Again,
Maigre [42]. Numerical solutions predicted for this experiment have the plane stress condition is assumed. The geometry and boundary con-
previously been considered by several researchers using different ap- ditions of the problem are shown in Fig. 18 (left). The internal impact
proaches, for instance, a discontinuous Galerkin FE setting with Nitsche pressure applied on the internal surface decreasing with time can be
flux Ref. [41], dynamic cohesive models using nodal perturbation and seen in Fig. 18 (right). The material parameters are 𝐸 = 210 GPa, 𝜈 = 0.3,
edge-swap operators [43], and XFEM [44,45]. However, only the exper- and 𝜌0 = 7850 kg/m3 [43]. For this example, we adopted the surface fail-
imental result and numerical solution reported by Hirmand et al. [41] ure energy 𝛾 = 2000 N/m and critical stress 𝜎𝑐 = 850 MPa from [43] to
are used for our comparison purpose. determine the other required parameters, i.e., 𝑙, 𝑚, Φ, via a uniaxial
The input geometry and loading conditions for our numerical sim- tensile test described in subsection 4.1.
ulations are detailed in Fig. 13 (right). The Hopkinson bar strike is This fragmentation problem has been previously simulated by dif-
modeled by applying a time-dependent uniform pressure on the con- ferent methods, for instance, a meshfree cracking particle method [46],
tact surface between the Hopkinson bar (diameter 12.7 mm) and the cohesive elements [47], cracking node method [8], dynamic cohesive
specimen. The resultant impact load used in the numerical calculation, models using nodal perturbation and edge-swap operator [43], and
shown in Fig. 14 (left), was obtained from Rittel’s experiment [41]. No adaptive polygonal finite elements [48]. These reference results are thus
traction boundary conditions are applied to all unspecified surfaces. used for comparison purpose.
The plane stress condition is assumed. The material parameters are The Cauchy stress-stretch curve and values determined for other re-
𝐸 = 5.76 GPa, 𝜈 = 0.42, and 𝜌0 = 1180 kg/m3 [41]. To determine the quired material parameters with 100 steps of calculation are plotted
other required necessary parameters including 𝑙, 𝑚, Φ, we adopted the in Fig. 19 (right). Similarly, a thick dashed line indicates the Cauchy
surface failure energy 𝛾 = 352.3 N/m and critical stress 𝜎𝑐 = 105 MPa stress-stretch curve of the intact material which can be obtained from
from [41]. With the given data of 𝐸, 𝜈 and 𝜎𝑐 , the uniaxial tensile elastic theory.
test is conducted numerically by using a unit cube as mentioned in An unstructured mesh consisting of 231226 Q4 elements with the
subsection 4.1 with 100 steps of calculation. Fig. 14 (right) illustrates size of the mesh is around 0.5 mm is used for the simulation. We thus
the Cauchy stress-stretch curve and determined values for the other re- consider two cases of the implementation: (Case 1) has no special treat-
quired material parameters. In the figure, a thick dashed line indicates ment for the model’s symmetry, and (Case 2), similar to [8], uses a ±5%
the Cauchy stress-stretch curve of the intact material which can be ob- perturbation in the elastic modulus to break the symmetry of the model.
tained from elastic theory. An example of the random field of 𝐸 using in numerical simulations is
The spatial discretization of the model is conducted using a set of depicted in Fig. 19 (left). The total simulation time is 110 μs and a step
9528 Q4 irregular elements. To improve the computational efficiency, size of time chosen as Δ𝑡 = 0.5 μs.
as usual, a local refined mesh is used at zones where the crack is ex- The normalized density fields for two mentioned cases are computed
pected to propagate with an effective size of the mesh ℎ ≈ 0.1 mm. by the decomposed MF model and the obtained results of crack pat-
The total simulation time is 120 μs with a step size of time chosen as terns are then given in Table 2 and Fig. 20. Generally, the numbers
Δ𝑡 = 1 μs. of fragments for two specified cases are not the same. The numbers of
The final deformed shape of the specimen at the end of the simu- major fragments for Cases 1 and 2 are, respectively, 12 and 15. We ob-
lation in comparison to the numerical and experimental results [41] is serve that the present method yields reasonable numbers of fragments
given in Fig. 15. The present approach offers acceptable deformed shape as other reference models do.
and curved crack patterns as compared with the reference solutions. In The variation of stress fields, strain energy fields as well as the prop-
this study, the crack started propagating at time 𝑡0 = 63 μs which also agation of cracks for Case 2 are described in Fig. 21. For Case 2, during
agrees well with the reference results derived from the experiment and the loading process up to the time 𝑡 = 58.5 μs, cracks start propagating
the numerical simulation in [41], 𝑡0 = 66 μs and 𝑡0 = 56.4 μs, respec- from the inner surface almost uniformly. Some cracks arrested, and sev-

204
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 13. Rittel’s experiment. Experimental setup of the compact compression test (left); Geometry (unit of length: mm) and boundary conditions for numerical study
(right).

Fig. 14. Rittel’s experiment. Time-dependent history of total impact load 𝐹 (𝑡), see Ref. [41] (left), and Cauchy stress versus stretch curve and the other required
parameters for MF damage models (numerical tensile test) (right).

Fig. 15. Rittel’s experiment. Comparison of the deformed shape and crack pattern of the specimen at the final stage among the present study, experiment and a
discontinuous Galerkin FE [41]: (a) numerical prediction by this study (deleted elements are removed from the mesh), (b) the discontinuous Galerkin FE numerical
result [41], and (c) the experiment [41].

eral other cracks continued propagating until reaching the outer surface words, as compared with the whole specimen, the mass loss quantified
of the thick ring as shown in Fig. 21. in this example is about 14.3%.
The energy dissipation during the simulation of this fragmentation
example for the two considered cases is plotted in Fig. 22. Case 2 yields 6.4. Grégoire et al.’s experiment
the dissipated energy higher than that of Case 1. This phenomenon
comes out from the fact that more cracks occur in Case 2 than they Next example deals with the mixed-mode dynamic propagation of
a plate specimen made of PMMA with one crack and one hole. This
do in Case 1, see Table 2 and Fig. 20. For estimation of the mass loss for
problem was experimentally conducted by Grégoire et al. [49] and sim-
this cylinder specimen, it is determined from Eq. (53) and Fig. 22 for
ulated by XFEM [9,49,50]. The plane stress condition is assumed. The
𝑈𝑑total 1688
Case 2 as: 𝑚loss = 𝑉0 Φ
= [ ]
𝜋(0.152 −0.082 )0.07 ×(3.3278×106 )
= 14.3%. In other geometry and the experimental setup are shown in Fig. 23. The material

205
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 16. Rittel’s experiment. Stress fields and normalized density fields at times (a) 𝑡 = 65 μs, (b) 𝑡 = 83 μs, (c) 𝑡 = 102 μs, (d) at the end of the simulation, 𝑡 = 120 μs.
Note that in the case of the results of stress fields, deleted elements are removed from the mesh to show the cracks. The stress fields are plotted in current
configurations while the others are shown in the reference state.

Fig. 17. Rittel’s experiment. Comparison of the simulated results between the original non-split model and the experimental data: (a) the simulated deformed shape,
(b) the obtained crack pattern, and (c) the experimental reference [41]. It is obvious that the original non-split energy model seriously fails to predict the crack path
trajectory, which is one source of motivations inspiring us to develop the decomposed approach.

Fig. 18. Fragmentation simulation. Geometry (unit of length: mm) and boundary conditions (left); Applied impact pressure with respect to time (right).

206
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 19. Fragmentation simulation. Random distribution of Young modulus 𝐸 (left), and Cauchy stress versus stretch curve and the other required parameters for
MF damage models (numerical tensile test) (right).
√
nylon 𝐸 nylon
output bar which is calculated by 𝑐𝐿 = ; 𝝈 f and 𝒗f are the
𝜌nylon
stress (pressure) and velocity fields on the contact surface between the
specimen and the output bar, respectively.
Integrating Eq. (56a) into the output interface surface results in the
damping matrix 𝒁 to the momentum equation. Note that, since the
plane stress condition is used in this study, it is assumed that the cross-
section of the output bar is a rectangle shape with Height × Width =
40 × 10𝜋 (mm2 ). Here, the number 10𝜋 deduces from the area of the
cross section of the output bar per the height of the output interface.
Therefore, when integrating Eq. (56a) the thickness of the assumed out-
Fig. 20. Fragmentation simulation. Comparison of the normalized density fields put bar is used instead of the thickness of the specimen.
(crack paths) at the end of the simulations between Case 1 and Case 2. The crack patterns of the present work are estimated at two dif-
ferent time steps, 𝑡 = 500 μs and 𝑡 = 600 μs and then compared with
parameters of the specimen are 𝐸 = 2.4 GPa,√𝜈 = 0.42, 𝜌0 = 1180 kg/m3 , reference solutions derived from the XFEM [50] (at time 𝑡 = 500 μs) and
fracture toughness energy, 𝐾𝐼𝐶 = 1.33 MPa m [49]. The surface fail- experimental data [49] (at time 𝑡 = 600 μs) in Fig. 25. Compared results
ure energy 𝛾 is calculated from 𝐾𝐼𝐶 by using the Irwin’s formula for the indicate good agreements among the mentioned solutions, reflecting the
plane-stress state as accuracy and capability of the developed method in capturing dynamic
crack patterns of PMMA materials.
𝐾𝐼𝐶
2
We further compare the crack tip abscissa (horizontal position of
𝛾= ≈ 737 𝑁∕𝑚. (55)
𝐸 crack’s tip in time) between the present numerical simulation and the
The material parameters of the output bar are also based on [49] experimental data in Fig. 26 (left). The initiation, arrest and restart of
as 𝐸 nylon = 3.6 GPa and 𝜈 nylon = 0.41, 𝜌nylon = 1145 kg/m3 , mainly used the crack reported from the experiment [49] are well reproduced by the
for modeling the impedance boundary condition on the interface be- proposed model. In addition, the simulated output velocity of this study
tween the specimen and the output bar. Since, there is no data for the is also compared with respect to the experimental result as illustrated
maximal tensile stress of PMMA from the experiment, the other param- in Fig. 26 (right). It is obviously that from the beginning to the point
eters required for the MF damage model are chosen as 𝑚 = 5, 𝑙 = 1 mm, when crack starts propagating, 𝑡 ≈ 208 μs, the numerical output velocity
𝜖 = 0.00001. The energy limiter, Φ, is deduced from 𝛾, 𝑚 and 𝑙 by the agrees well with the experimental data. However, during the evolu-
aid of Eq. (50). tion of the crack, numerical results of the output velocity are shown
A finite element mesh consisting of 7065 Q4 irregular elements with to be fluctuated. This phenomenon may come out from the impedance
an effective size of the mesh ℎ ≈ 0.5 mm is used. The simulation requires boundary condition which is used to model the output bar, the problem
600 time steps with a step size of time chosen as Δ𝑡 = 1 μs. with oscillation of the numerical output velocity was also reported in
The boundary conditions of the example are shown in Fig. 24 (right). [50]. In general, the present numerical scheme is capable of reproduc-
In numerical study, the input velocity depicted in Fig. 24 (left) [49] is ing the output velocity as compared to what had been observed during
applied on the interface between the input bar (a diameter of 40 mm) the experiment.
and the specimen. On the opposite side, the impedance condition, which
was used in Grégoire et al. [49], is used to model the output bar. No
6.5. Finite deformation simulation: a double edge-notched specimen under
other constraints are required.
tension
Based on [49], at each interface node on the contact surface between
the specimen and the output bar, the stress, 𝝈 f , and velocity, 𝒗f , have
the following relation The last numerical example aims to illustrate the ability of the
present theory in modeling localized failure at finite strain. We thus
𝝈 f ⋅ 𝒏 = 𝑧(𝒗f ⋅ 𝒏)𝒏 on the contact face, (56a) consider a double-edge notched (styrene butadiene rubber) specimen
nylon under tensile load as shown in Fig. 27 (left). This problem was ex-
𝑧 = 𝜌nylon 𝑐𝐿 , (56b)
perimentally and numerically investigated by Hocine et al. [51]. The
where 𝑧 is the impedance which is related to the material properties geometry and the associated boundary condition of this problem are
nylon
of the output bar; 𝑐𝐿 is one-dimensional elastic wave speed of the also described in Fig. 27 (left).

207
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 21. Fragmentation simulation. Hydrostatic stress, ( 12 trace(𝝈)), normalized strain energy and normalized density fields at times (a) 𝑡 = 60 μs, (b) 𝑡 = 76 μs, (c)
𝑡 = 92 μs, (d) at the end of the simulation, 𝑡 = 110 μs for Case 2. Note that in the case of the results of stress fields and normalized density fields, deleted elements are
removed from the mesh to show the cracks. The stress fields are plotted in current configurations while the others are shown in the reference state.

solver is employed [53], i.e., the displacement control with a fixed in-
crement Δ𝑢 = 1 × 10−4 m is used up to the point when the crack initiates,
then, the top edge of the specimen is kept and the dynamic solver takes
place. The fixed time step of Δ𝑡 = 0.5 μs is utilized for the dynamic cal-
culation.
A comparison of the structural load-displacement responses between
this work and the experimental data [51] is then given in Fig. 28. Al-
though the peak load is slightly different between each solution, it is not
significant and acceptable. Overall, the force-displacement curve com-
puted by the developed model is in good agreement with the reference
experimental data.
Furthermore, the evolution of crack in the deformed configuration at
different stages of the loading is depicted in Fig. 29. From this example,
it reveals the capacity of the developed MF model in terms of modeling
fracture behavior of materials at large deformation.

7. Conclusions

In this study, we have conducted numerical simulations of quasi-

Fig. 22. Fragmentation simulation. The energy dissipation for the two consid- static/dynamic brittle fracture at small and finite strain in brittle mate-
ered cases. rials (e.g., PMMA, styrene butadiene rubber) by using a diffusive local-
ized MF damage model with the standard FEM. The developed method
Similar to the previous work by Miehe and Schänzel [52], the plane integrates three ingredients: (i) the theoretical formulation is described
strain state is thus assumed and the material parameters are specified in a general way that is valid and suitable for any level of deformation
as 𝜇 = 0.203 N/mm2 , 𝜈 = 0.45 and 𝛾 = 2.67 N/mm. Other additional pa- under dynamic time-dependent loading conditions; (ii) using the strain
rameters that are required for the current MF damage model are chosen energy density decomposition technique for mass source and mass flux
as 𝑚 = 1, 𝑙 = 5 mm, 𝜌0 = 1550 kg/m3 , and 𝜖 = 0.001. The energy limiter, to compressible neo-Hookean material models [17]; and (iii) FEM based
Φ, can be estimated from 𝛾, 𝑚 and 𝑙 via Eq. (50). In numerical imple- discretization technique associated with a staggered algorithm. Such in-
mentation, a locally refined mesh of 6138 elements with an effective tegrated desirable features make the developed approach more general
size ℎ ≈ 0.8 mm as given in Fig. 27 (right) is adopted. To capture the and effective than the one reported in [15]. It provides a rigorous tech-
fast propagation of crack at large deformation the quasi-static/dynamic nique to a diffusive fracture modeling based on the introduction of a

208
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 23. Grégoire et al.’s experiment. Geometry (unit of length: mm) and experimental setup [49].

Fig. 24. Grégoire et al.’s experiment. Experimental input velocity [49] (left), and boundary conditions (right).

Fig. 25. Grégoire et al.’s experiment. Comparison of the dynamic crack path among the present MF damage model, XFEM [50] at time 𝑡 = 500 μs and the experimental
result [49] at time 𝑡 = 600 μs. The present results obtained at times (a) 𝑡 = 500 μs and (c) 𝑡 = 600 μs.

localized mass-diffusive balance equation. The developed decomposed The performance and accuracy of the proposed model are validated
method overcomes difficulties of the original non-decomposed model through several examples including four experiments and fragmenta-
in distinguishing the fracture behavior under tension and compression, tion simulation. The computed dynamic fracture results confirm the
and, similar to phase-field models, it eliminates the drawbacks of the realistic crack patterns, and crack propagation velocities and dissipated
computational discrete cracks particularly when it comes to problems energies in close agreement with the experimental data and other nu-
with catastrophic multiple cracks (e.g., fragmentation simulation in sub- merical results. One significant finding, as already investigated in the
section 6.3). In other words, in the present approach, neither artificial first numerical example (subsection 6.1), is that our method dominates
internal damage variables, nor ad-hoc fracture criteria are required as over the phase field models in estimation of the dissipated energy (see
the damage localization is driven by the physically sound law of mass- Fig. 12). All the mentioned phase field models in terms of dynamic anal-
diffusive balance. ysis yield quite large energy dissipation, by about > 40% for Kalthoff

209
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 26. Grégoire et al.’s experiment. Comparison of the crack tip abscissa vs. time (left), and output velocities vs. time (right) between the present MF damage
model and Grégoire’s experiment [49].

Fig. 28. Double edge-notched specimen under tension. Comparison on the

global load–displacement curves between the present work and the experimen-
tal data [51].

Fig. 27. Double edge-notched specimen under tension. Geometry (unit of Data availability
length: mm) and experimental setup of the Hocine et al.’s experiment [51] (left)
and the considered mesh (right).
The authors are unable or have chosen not to specify which data has
been used.

Acknowledgements

and Winkler problem, as compared to the theoretical value, and such

Hung Thanh Tran is grateful to the Japanese Government MEXT
overestimation is once again critical.
scholarship for his Integrated Doctoral Education Program (Scholarship
This work is our first attempt towards the modeling of dynamic ID No. 193229).
brittle fracture based on a diffusive damage theory enhanced by en-
ergy decomposition using the standard FEM, opening an alternative Appendix A. Tangent matrix derivation
research direction in simulating dynamic fracture problems. Limita-
tions of the model are certainly out there, while we should consider In the appendix, the derivation of the spatial tangent matrix, 𝐾𝑇 in
other techniques to be able to handle other nonlinear material models, Eq. (49), which is required for the Newton-Raphson algorithm to update
e.g., Mooney–Rivlin, Odgen, and Yeoh, the effect of the density around the displacement field, is briefly discussed. To find the spatial tangent
crack’s tip as [15] should be deeply studied as well. In the long run, stiffness matrix, one can follow the linearization of the principle of vir-
we also plan to combine, e.g., material point methods, with the present tual work in spatial description, which is well documented in Chapter
localized MF damage model to form a new computational approach 8 of Ref. [27]. Following the reference, we need to derive the lineariza-
which may own several desirable features and significant advantages tion of the internal work in the material description first. Then, the
for dynamic fracture at large deformation in rubber-like materials. linearized terms are pushed forward to the spatial description.

210
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

Fig. 29. Double edge-notched specimen under tension. Snapshots of the normalized density fields shown in deformed shapes at different stages of the loading
(corresponding to the point (a) to (f) in Fig. 28). Deleted elements are removed to show the crack.

For a background of the linearizarion technique to solve a nonlinear where Grad(∙) ≡ ∇0 (∙) is the gradient operator in the material descrip-
problem, let’s consider a nonlinear function 𝑓 = 𝑓 (𝒖) with a single argu- tion.
ment of the displacement field, 𝒖. Then, the fundamental relationship In the index notation, Eq. (A.4) can be rewritten as
for the linearization of the function 𝑓 based on the standard Newton- ( )
Raphson method is defined as 𝜕𝛿𝑢𝑖 𝜕Δ𝑢𝑖 𝜕𝛿𝑢𝑖 𝜕𝛿𝑢𝑗
𝐷Δ𝒖 𝛿𝑊int (𝒖, 𝛿𝒖) = 𝑆 + 𝐹𝑖𝐼 ℂ 𝐹 d𝑉 .
∫ 𝜕𝑋𝐽 𝜕𝑋𝐿 𝐽 𝐿 𝜕𝑋𝐽 𝐼𝐽 𝐾𝐿 𝑗𝐾 𝜕𝑋𝐿
𝑓 (𝒖, Δ𝒖) = 𝑓 (𝒖) + Δ𝑓 (𝒖, Δ𝒖), (A.1) Ω0

(A.5)
where Δ𝒖 stands for the increment of the displacement field; and Δ(∙)
is the linearization operator which is expressed from the directional Then, the linearized terms in the material configuration are pushed
derivative given for 𝑓 as forward to the spatial description as [27] to obtain
( )
d 𝜕𝛿𝑢𝑖 𝜕Δ𝑢𝑖 𝜕𝛿𝑢𝑖 𝜕𝛿𝑢𝑘
Δ𝑓 (𝒖, Δ𝒖) = 𝐷Δ𝒖 𝑓 (𝒖) = 𝑓 (𝒖 + 𝜀Δ𝒖)|𝜀=0 , (A.2) 𝐷Δ𝒖 𝛿𝑊int (𝒖, 𝛿𝒖) = 𝜎𝑗𝑙 + ℂ𝑖𝑗𝑘𝑙 d𝑣
d𝜀 ∫ 𝜕𝑥𝑗 𝜕𝑥𝑙 𝜕𝑥𝑗 𝜕𝑥𝑙
Ω
with 𝐷(∙) is the Gateaux operator with respect to the incremental dis- ( ) (A.6)
placement field, Δ𝒖. 𝜕𝛿𝑢𝑖 𝜕𝛿𝑢𝑘 𝜕𝛿𝑢𝑖 𝜕Δ𝑢𝑖
= ℂ𝑖𝑗𝑘𝑙 + 𝜎𝑗𝑙 d𝑣.
Next, we consider the internal virtual work, 𝛿𝑊int . The expressions ∫ 𝜕𝑥𝑗 𝜕𝑥𝑙 𝜕𝑥𝑗 𝜕𝑥𝑙
Ω
of 𝛿𝑊int in the spatial and material configurations are given by [27]
Up to here, one can employ the discretization formulae given in
𝛿𝑊int (𝒖, 𝛿𝒖) = 𝝈(𝒖) ∶ 𝛿𝒆(𝒖)d𝑣 = 𝑺 (𝑬(𝒖)) ∶ 𝛿𝑬(𝒖)d𝑉 , (A.3) Eq. (36) together with the use of Voigt notation and some arrangements,
∫ ∫
Ω Ω0 the FEM matrix form of the spatial tangent stiffness matrix, 𝐾𝑇 , is finally
derived corresponding to Eq. (A.6) as
where 𝑬 and 𝒆 are the Green-Lagrange and Euler-Almansi finite strain
tensors, respectively; 𝛿𝑬 and 𝛿𝒆 are the variations of 𝑬 and 𝒆; 𝑑𝑉 and ( 𝑇 )
𝑲𝑇 = 𝑩 𝒖 𝑫𝑩 𝒖 + 𝑩 𝑇𝐺 Σ𝝈 𝑩 𝐺 d𝑣. (A.7)
𝑑𝑣 are the volume elements of material and spatial configurations, ac- ∫
Ω
cordingly, and they have the relation as 𝑑𝑣 = 𝐽 𝑑𝑉 . It is noticed that ( )
the internal virtual work given above is related to the second term on In the literature, the first part 𝑩 𝑇𝒖 𝑫𝑩 𝒖 and the second part
( 𝑇 )
the left-hand-side of Eq. (33) where the tangent stiffness matrix could 𝑩 𝐺 Σ𝝈 𝑩 𝐺 of the tangent stiffness matrix, 𝐾𝑇 , are normally called
also be derived. However, for simplicity to find directional derivatives as the material stiffness matrix and the geometrical (or initial stress)
of the stress related to the displacement field, the notations as expressed stiffness matrix, accordingly. In Eq. (A.7), the spatial constitutive ma-
in Eq. (A.3) are referred here. trix, 𝑫, is constructed from the components of the fourth-order tensor,
According to [27] (Chapter 8), the linearization of the internal work ℂ𝑖𝑗𝑘𝑙 ; 𝑩 𝐺 is defined from the spatial derivatives of the FE shape func-
in the undeformed state is expressed as tion, 𝑁𝐼,𝑖 ; while Σ𝝈 is obtained from components of the Cauchy stress
tensor, 𝝈. Details for the derivations of 𝑫, 𝑩 𝐺 , Σ𝝈 can be found in
𝐷Δ𝒖 𝛿𝑊int (𝒖, 𝛿𝒖) Refs. [21,29,54,55]. For 2D plane strain problems, they are defined as
d
= 𝛿𝑊int (𝒖 + 𝜀Δ𝒖)|𝜀=0
d𝜀 ⎡ ℂ1111 ℂ1122 ℂ1121 ⎤
d 𝑫 = ⎢ ℂ2211 ℂ2222 ℂ2221 ⎥, (A.8)
= [ 𝑺 (𝑬(𝒖 + 𝜀Δ𝒖)) ∶ 𝛿𝑬 (𝒖 + 𝜀Δ𝒖) d𝑉 ]|𝜀=0 ⎢ ⎥
d𝜀 ∫ ⎣ ℂ1211 ℂ1222 ℂ1221 ⎦
Ω0
(A.4)
and
= [𝑺 (𝑬(𝒖)) ∶ 𝐷Δ𝒖 𝛿𝑬(𝒖) + 𝛿𝑬(𝒖) ∶ 𝐷Δ𝒖 𝑺 (𝑬(𝒖))]d𝑉
∫ ⎡ 𝑁1,1 0 ... 𝑁4,1 0 ⎤
Ω0 ⎢ ⎥
𝑁 0 ... 𝑁4,2 0
( ) 𝑩 𝐺 = ⎢ 1,2 ⎥, (A.9)
= Grad𝛿𝒖 ∶ GradΔ𝒖𝑺 + 𝑭 𝑇 Grad𝛿𝒖 ∶ ℂ ∶ 𝑭 𝑇 GradΔ𝒖 d𝑉 , ⎢ 0 𝑁1,1 ... 0 𝑁4,1 ⎥
∫ ⎢ 0 ⎥
Ω0 ⎣ 𝑁1,2 ... 0 𝑁4,2 ⎦

211
T.Q. Bui and H.T. Tran Computers and Mathematics with Applications 125 (2022) 193–212

⎡ 𝜎11 𝜎12 0 0 ⎤ [28] N.M. Newmark, A method of computation for structural dynamics, in: Proceedings
⎢ ⎥
𝜎 𝜎22 0 0 of ASCE, J. Eng. Mech. 85 (1959) 67–94.
Σ𝝈 = ⎢ 12 ⎥. (A.10)
⎢ 0 0 𝜎11 𝜎12 ⎥ [29] E.A. de Souza Neto, D. Peric, D.R.J. Owen, Computational Methods for Plasticity:
⎢ 0 0 𝜎12 𝜎22 ⎥ Theory and Applications, John Wiley & Sons, Ltd, 2008.
⎣ ⎦ [30] K.Y. Volokh, Characteristic length of damage localization in rubber, Int. J. Fract.
168 (2011) 113–116.
References [31] M. Ambati, T. Gerasimov, L.D. Lorenzis, A review on phase-field models of brittle
fracture and a new fast hybrid formulation, Comput. Mech. 55 (2015) 383–405.
[1] B.N. Cox, H. Gao, D. Gross, D. Rittel, Modern topics and challenges in dynamic [32] H. Amor, J.-J. Marigo, C. Maurini, Regularized formulation of the variational brittle
fracture, J. Mech. Phys. Solids 53 (2005) 565–596. fracture with unilateral contact: numerical experiments, J. Mech. Phys. Solids 57
[2] J. Fineberg, E. Bouchbinder, Recent developments in dynamic fracture: some per- (2009) 1209–1229.
spectives, Int. J. Fract. 196 (2015) 33–57. [33] K.Y. Volokh, On modeling failure of rubber-like materials, Mech. Res. Commun.
[3] C. Wolff, N. Richart, J.-F. Molinari, A non-local continuum damage approach to 37 (8) (2010) 684–689.
model dynamic crack branching, Int. J. Numer. Methods Eng. 101 (2014) 933–949. [34] C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent
[4] M. Hofacker, C. Miehe, Continuum phase field modeling of dynamic fracture: crack propagation: robust algorithmic implementation based on operator splits,
variational principles and staggered FE implementation, Int. J. Fract. 178 (2012) Comput. Methods Appl. Mech. Eng. 199 (45–48) (2010) 2765–2778.
113–129. [35] G. Liu, Q. Li, M.A. Msekh, Z. Zuo, Abaqus implementation of monolithic and stag-
[5] D.H. Doan, T.Q. Bui, T.V. Do, D.D. Nguyen, A rate-dependent hybrid phase field gered schemes for quasi-static and dynamic fracture phase-field model, Comput.
model for dynamic crack propagation, J. Appl. Phys. 122 (2017) 115102. Mater. Sci. 121 (2016) 35–47.
[6] M.J. Borden, C.V. Verhoosel, M.A. Scott, T.J. Hughes, C.M. Landis, A phase-field [36] J. Kalthoff, S. Winkler, Failure mode transition at high rates of shear loading, in:
description of dynamic brittle fracture, Comput. Methods Appl. Mech. Eng. 217–220 International Conference on Impact Loading and Dynamic Behavior of Materials,
(2012) 77–95. vol. 1, 1987, pp. 185–195.
[7] J.-H. Song, H. Wang, T. Belytschko, A comparative study on finite element methods [37] J. Kalthoff, Modes of dynamic shear failure in solids, Int. J. Fract. 101 (2000) 1–31.
for dynamic fracture, Comput. Mech. 42 (2008) 239–250. [38] S. Zhou, T. Rabczuk, X. Zhuang, Phase field modeling of quasi-static and dynamic
[8] J. Song, T. Belytschko, Cracking node method for dynamic fracture with finite ele- crack propagation: COMSOL implementation and case studies, Adv. Eng. Softw. 122
ments, Int. J. Numer. Methods Eng. 77 (2009) 360–385. (2018) 31–49.
[9] A. Combescure, A. Gravouil, D. Grégoire, J. Réthoréb, X-FEM a good candidate for [39] M.R. Hirmand, K.D. Papoulia, Block coordinate descent energy minimization for
energy conservation in simulation of brittle dynamic crack propagation, Comput. dynamic cohesive fracture, Comput. Methods Appl. Mech. Eng. 354 (2019) 663–688.
Methods Appl. Mech. Eng. 197 (5) (2008) 309–318. [40] R.J. Geelen, Y. Liu, T. Hu, M.R. Tupek, J.E. Dolbow, A phase-field formulation for
[10] T. Li, J.-J. Marigo, D. Guilbaud, S. Potapov, Gradient damage modeling of brittle dynamic cohesive fracture, Comput. Methods Appl. Mech. Eng. 348 (2019) 680–711.
fracture in an explicit dynamics context, Int. J. Numer. Methods Eng. 108 (2016) [41] M.R. Hirmand, K.D. Papoulia, A continuation method for rigid-cohesive fracture in
1381–1405. a discontinuous Galerkin finite element setting, Int. J. Numer. Methods Eng. 115 (5)
[11] P. Areias, M.A. Msekh, T. Rabczuk, Damage and fracture algorithm using the (2018) 627–650.
screened Poisson equation and local remeshing, Eng. Fract. Mech. 158 (2016) [42] D. Rittel, H. Maigre, An investigation of dynamic crack initiation in PMMA, Mech.
116–143. Mater. 23 (3) (1996) 229–239.
[12] P. Areias, J. Reinoso, P.P. Camanho, J.C. de Sá, T. Rabczuk, Effective 2D and 3D [43] G.H. Paulino, K. Park, W. Celes, R. Espinha, Adaptive dynamic cohesive fracture
crack propagation with local mesh refinement and the screened Poisson equation, simulation using nodal perturbation and edge-swap operators, Int. J. Numer. Meth-
Eng. Fract. Mech. 189 (2018) 339–360. ods Eng. 84 (11) (2010) 1303–1343.
[13] V. Lyakhovsky, Y. Hamiel, Y. Ben-Zion, A non-local visco-elastic damage model and [44] I. Asareh, J. Song, R.L. Mullen, Y. Qian, A general mass lumping scheme for the vari-
dynamic fracturing, J. Mech. Phys. Solids 59 (2011) 1752–1776. ants of the extended finite element method, Int. J. Numer. Methods Eng. 121 (10)
[14] K.Y. Volokh, Fracture as a material sink, Mater. Theory 1 (3) (2017) 9. (2019) 2262–2284.
[15] A. Faye, Y. Lev, K.Y. Volokh, The effect of local inertia around the crack-tip in [45] T. Menouillard, J. Rethore, N. Moes, A. Combescure, H. Bung, Mass lumping strate-
dynamic fracture of soft materials, Mech. Soft Mater. 1 (1) (2019) 4. gies for X-FEM explicit dynamics: application to crack propagation, Int. J. Numer.
[16] V. Agrawal, K. Dayal, Dependence of equilibrium Griffith surface energy on crack Methods Eng. 74 (3) (2008) 447–474.
speed in phase-field models for fracture coupled to elastodynamics, Int. J. Fract. 207 [46] T. Rabczuk, T. Belytschko, Cracking particles: a simplified meshfree method for
(2017) 243–249. arbitrary evolving cracks, Int. J. Numer. Methods Eng. 61 (13) (2004) 2316–2343.
[17] T.Q. Bui, H.T. Tran, A localized mass-field damage model with energy decomposi- [47] F. Zhou, J.-F. Molinari, Dynamic crack propagation with cohesive elements: a
tion: formulation and FE implementation, Comput. Methods Appl. Mech. Eng. 387 methodology to address mesh dependency, Int. J. Numer. Methods Eng. 59 (1)
(2021) 114134. (2004) 1–24.
[18] S. Tang, G. Zhang, T.F. Guo, X. Gou, W.K. Liu, Phase field modeling of fracture in [48] S. Leon, D. Spring, G. Paulino, Reduction in mesh bias for dynamic fracture using
nonlinearly elastic solids via energy decomposition, Comput. Methods Appl. Mech. adaptive splitting of polygonal finite elements, Int. J. Numer. Methods Eng. 100 (8)
Eng. 347 (2019) 477–494. (2014) 555–576.
[19] C. Hesch, K. Weinberg, Thermodynamically consistent algorithms for a finite- [49] D. Grégoire, H. Maigre, J. Rethore, A. Combescure, Dynamic crack propagation un-
deformation phase-field approach to fracture, Int. J. Numer. Methods Eng. 99 (2014) der mixed-mode loading–comparison between experiments and X-FEM simulations,
906–924. Int. J. Solids Struct. 44 (20) (2007) 6517–6534.
[20] J.Y. Ye, L.W. Zhang, J.N. Reddy, Large strained fracture of nearly incompressible hy- [50] A. Gravouil, T. Elguedj, H. Maigre, An explicit dynamics extended finite element
perelastic materials: enhanced assumed strain methods and energy decomposition, method. Part 2: element-by-element stable-explicit/explicit dynamic scheme, Com-
J. Mech. Phys. Solids 139 (2020) 103939. put. Methods Appl. Mech. Eng. 198 (30–32) (2009) 2318–2328.
[21] J. Bonet, R.D. Wood, Nonlinear Continuum Mechanics for Finite Element Analysis, [51] N.A. Hocine, M. Abdelaziz, A. Imad, Fracture problems of rubbers: 𝐽 -integral es-
Cambridge University Press, 2008. timation based upon 𝜂 factors and an investigation on the strain energy density
[22] M. Fagerström, R. Larrson, Approaches to dynamic fracture modelling at finite de- distribution as a local criterion, Int. J. Fract. 117 (2002) 1–23.
formations, J. Mech. Phys. Solids 56 (2008) 613–639. [52] C. Miehe, L.-M. Schänzel, Phase field modeling of fracture in rubbery polymers.
[23] E. Kirchner, S. Reese, P. Wriggers, A finite element method for plane stress problems Part I: finite elasticity coupled with brittle failure, J. Mech. Phys. Solids 65 (2014)
with large elastic and plastic deformations, Commun. Numer. Methods Eng. 13 (12) 93–113.
(1998) 963–976. [53] T.K. Mandal, A. Gupta, V.P. Nguyen, R. Chowdhury, A. de Vaucorbeil, A length scale
[24] S. Klinkel, S. Govindjee, Using finite strain 3D-material models in beam and shell insensitive phase field model for brittle fracture of hyperelastic solids, Eng. Fract.
elements, Eng. Comput. 19 (3) (2002) 254–271. Mech. 236 (2020) 107196.
[25] J.P. Pascon, Finite element analysis of functionally graded hyperelastic beams under [54] N.-H. Kim, Introduction to Nonlinear Finite Element Analysis, Springer, 2014.
plane stress, Eng. Comput. 36 (2020) 1265–1288. [55] T. Belytschko, W.K. Liu, B. Moran, K. Elkhodary, Nonlinear Finite Elements for Con-
[26] J.P. Pascon, Large deformation analysis of plane-stress hyperelastic problems via tinua and Structures, John Wiley & Sons, Ltd, 2000.
triangular membrane finite elements, Int. J. Adv. Struct. Eng. 11 (2019) 331–350.
[27] G.A. Holzapfel, Nonlinear Solid Mechanics: A Continuum Approach for Engineering,
John Wiley & Sons, Ltd, 2000.

212