Available online at www.sciencedirect.

com
ScienceDirect

Comput. Methods Appl. Mech. Engrg. 383 (2021) 113893

www.elsevier.com/locate/cma

A simplified continuous–discontinuous approach to fracture based on

decoupled localizing gradient damage method
Subrato Sarkar, I.V. Singh ∗, B.K. Mishra
Computational Mechanics and Multiscale Modelling Lab, Department of Mechanical and Industrial Engineering, Indian Institute of Technology
Roorkee, Uttarakhand, 247667, India
Received 20 December 2020; received in revised form 15 April 2021; accepted 24 April 2021
Available online xxxx

Abstract
In this work, two fracture modelling methods i.e., extended FEM (XFEM) and localizing gradient damage method (LGDM),
are combined to get the advantages of both methods while eliminating their limitations. The LGDM is a micromechanics-based
method that avoids spurious damage growth and incorrect damage initiation observed in conventional gradient damage method
(CGDM). However, in the LGDM, a cracked/discontinuous domain is not fully realized due to its continuous nature, even at
high damage. This leads to numerous non-physical effects in LGDM like high strains, severe stress oscillations and incorrect
representation of secondary variables. These shortcomings are rectified in the combined continuous–discontinuous (XFEM +
LGDM) approach by introducing XFEM at the end of damage evolution to impose a discontinuous crack in the domain. To
simplify the implementation, a fully coupled LGDM is converted to a decoupled LGDM using an operator-split (staggered)
methodology. It is observed from numerical examples that the decoupling of LGDM leads to a reduction in the computational
effort without compromising accuracy. Besides, the XFEM+LGDM approach avoids use of cohesive zone modelling necessary
in existing XFEM+CGDM approaches. Moreover, the proposed method (combined XFEM + LGDM) is also investigated in the
multi-physical framework of thermoelasticity. The numerical results reveal that the shortcomings mentioned earlier are rectified
in a physically consistent manner under mode I and mixed-mode conditions.
⃝c 2021 Elsevier B.V. All rights reserved.

Keywords: FEM; XFEM; Gradient damage; Staggered algorithm; Operator split; Thermoelasticity

1. Introduction
Fracture modelling in materials through numerical methods has become an important research area in applied
mechanics. Many new approaches to model fracture have been developed recently. These approaches are broadly
classified into two types based on their representation of fracture i.e. continuous and discontinuous approaches.
Discontinuous approaches represent fracture through an abrupt change/jump in the displacement field. Some popular
discontinuous approaches are FEM [1], extended FEM [2–4], extended IGA [5], extended EFG methods [6,7] and
VEM [8]. In contrast to this, the continuous approaches represent the fracture through spreading or smearing the
discontinuity onto a finite width. In other words, the continuous fracture is depicted by a finite width region of high
∗ Corresponding author.
E-mail address: [email protected] (I.V. Singh).

https://doi.org/10.1016/j.cma.2021.113893
0045-7825/⃝ c 2021 Elsevier B.V. All rights reserved.
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

deformation and negligible stiffness. The most widely used continuous approaches are the phase field method [9]
and the gradient damage method [10].
In this work, the discussions are focused on the extended FEM (XFEM) as the discontinuous approach and
gradient damage method (GDM) as the continuous approach. Although both of these methods are self-sufficient
in describing fracture in materials, however some of their limitations restrict their applicability. The applicability
of XFEM is typically limited to cases having pre-existing crack and an ad-hoc criterion is required by XFEM for
crack propagation [11]. In other words, the XFEM is not capable of modelling crack initiation and propagation
inherently, despite its accurate representation of cracks or discontinuities. On the other hand, the GDM is capable
of modelling crack/damage initiation and propagation inherently without needing any additional criterion. However,
some serious drawbacks associated with the GDM are as follows,
• The spurious damage spreading leads to high deformation and stiffness reduction in a wider region than actual
fracture.
• The absence of explicit crack causes non-physical transmission of traction across the damaged region, which
leads to residual strength even after high damage.
• The abnormal oscillations in stresses and strains are observed at high loads.
• The incorrect representation of secondary variables (such as heat flux in thermo-mechanical applications) is
observed due to the absence of explicit cracks.
Several methods have been proposed to address the above-mentioned issues. These proposed methods are primarily
aimed to combine XFEM (discontinuous) and GDM (continuous) to alleviate the shortcomings. These methods can
be broadly classified in two categories, i.e., with and without cohesive zone transition. The cohesive zone transition
uses a traction separation law to make the transition (from GDM to XFEM) gradual and smooth. The cohesive
zone transition-based methods can be found in the works of Comi et al. [12], Cazes et al. [13], Cazes et al. [14],
Seabra et al. [15], Roth et al. [16], Wang and Waisman [17], Jin and Arson [18] and Tamayo-Mas et al. [19]. The
other classification is without the cohesive zone transition, which includes the works of Mazars and Pijaudier-Cabot
[20], Patzák and Jirásek [21] and Simone et al. [22]. A more recent work by Geelen et al. [23] uses the phase field
method as the continuous approach instead of GDM.
From the research mentioned above, it is evident that most of the work is focused on using the cohesive zone
model (CZM) to resolve the transition from continuous to discontinuous crack in GDM. CZM is essential for a
smooth and gradual transition because of several limitations associated with the GDM, which are shown later,
and also reported in Simone et al. [22]. However, the use of CZM inevitably makes the overall approach very
complicated by combining three conceptually different methods, viz. XFEM, GDM and CZM. Moreover, XFEM
being a coupled approach leads to more complexities when additional physical phenomena, like thermal or electrical
effects, need to be included. Hence, the focus of the present work is to simplify the implementation and increase
the applicability of the continuous–discontinuous approach.
In this work, XFEM is adopted as the discontinuous approach and localizing gradient damage method (LGDM)
as the continuous approach. The choice of LGDM is based on its ability to limit the spurious damage spreading
and correct representation of crack tip [24]; [25]. In this regard, the novelties of the present work are outlined as
follows,
• The fully coupled LGDM is converted to a decoupled LGDM through an operator-split methodology based
on a history strain parameter.
• A staggered solution scheme is proposed for decoupled LGDM, which led to a simple and computationally
efficient implementation.
• The integration of XFEM with LGDM is carried out to solve several issues mentioned earlier, like eliminating
stress/strain oscillations and preventing traction transmission.
• The use of XFEM + LGDM approach avoids the use of CZM necessary in existing XFEM + CGDM
approaches.
• The combined XFEM + LGDM is applied in the thermo-mechanical framework to resolve the incorrect
representation of the secondary variable (heat flux).
The present paper is organized into four sections. Section 1 introduces the background and motivation for the
research work carried out in this work. Section 2 presents the detailed formulation and implementation of various
2
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 1. A schematic representation of the fracture processes occurring within the micro-continuum at different loading stages.

concepts proposed in this work. The subsequent Section 3 demonstrates all the advantages of adopting the proposed
formulations through various numerical examples involving different types of loads. Finally, Section 4 concludes
the paper by summarizing the outcomes. Appendices A and B are provided at the end for a detailed description of
several parameters used in the formulation.

2. Formulation

In this section, the formulation for the localizing gradient damage method (LGDM) is presented. Starting from the
coupled LGDM [26], a decoupling methodology is outlined for the LGDM. Later, the decoupled/staggered LGDM
is integrated with the extended FEM (XFEM) to introduce discontinuous cracks at the end of damage evolution to
rectify several abnormalities associated with the LGDM. The XFEM with staggered LGDM is called the XLGDM.
Finally, the thermo-mechanical coupling is incorporated in the XLGDM to simulate thermo-mechanical fracture.

2.1. Localizing Gradient Damage Method (LGDM)

The initiation of a fracture in a quasi-brittle or ductile material is characterized by the formation of a fracture
process zone in the regions of high stresses caused by loading [27,28]. The fracture process zone in a quasi-brittle
material comprises a diffused microcrack network, which is marked by increased activity of micro level fracture
responses. When the loading is increased further, these microcracks coalesce into a narrow region. The concentration
of the microcracks to a finite bandwidth leads to localized fracture. Simultaneously, the microcracks outside of the
localized fracture bandwidth elastically unload. This phenomenon is schematically illustrated in Fig. 1.
In this work, the fracture processes mentioned above are modelled using the localizing gradient damage
method (LGDM). The LGDM is based on the generalized micromorphic framework [24,29]. The micromorphic
framework quantifies the fracture processes at the micro-continuum level (Fig. 1) and includes their effects in
the macro-continuum fracture. The micromorphic framework provides a sound physical background to derive
a thermodynamically consistent gradient damage models, which are otherwise phenomenological [29,30]. The
micromorphic framework is built using two variables, viz. a macroscopic internal variable (φ) and a micromorphic
variable (χ φ). The macroscopic internal (φ) variable is identified from a set of existing model variables such that the
targeted physical phenomena (fracture processes) are adequately considered. Subsequently, a micromorphic variable
(χ φ) is defined to capture the targeted effects at the microscopic level. This micromorphic variable has the same
tensor rank as that of the macroscopic internal variable.
Within the scope of strain-based damage models, the fracture processes are described through the equivalent
strain. Hence, the identified internal variable for the micromorphic framework is the macro-equivalent strain (φ ≡
e). Correspondingly, the micromorphic variable is the micro-equivalent strain (χ φ ≡ ē), which accounts for the
micro level fracture processes. In the following sub-sections, these two equivalent strains are used to develop the
formulation.
3
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

2.1.1. Balance laws and constitutive equations

Using the principle of virtual power, the internal power (Pint ) for a given strain rate (ε̇) can be defined in domain
Ω as,
∫ ( )
Pint = σ : ε̇ + σ ė + ξ · ∇ ė dΩ (1)
Ω
where σ is the Cauchy stress. The second and third terms in Eq. (1) account for the work contribution of micro
level fracture processes and their effects on the macro level through micro-equivalent strain rate (ė) and its gradient
(∇ ė). The corresponding stress conjugates are called the equivalent coupling stress (σ ) and moment stress (ξ ). The
virtual power due to external forces can be expressed as,
∫
t · u̇ + ζ ė d∂Ω
( )
Pext = (2)
∂Ω
where t is the surface traction, u is the displacement and ζ is the higher-order traction. By equating the internal
and external powers in Eqs. (1)–(2), and applying divergence theorem yields following balance laws,
∇ ·σ =0 (3)

σ =∇ ·ξ (4)
Eqs. (3)–(4) represent the standard equilibrium equation and the microforce balance equation with boundary
conditions σ · n = t and ξ · n = ζ respectively on the domain boundary ∂Ω . Besides, all the contributions to
the free energy density (ψ) can be expressed as,
1 1 1
ψ= (1 − D) ε : C : ε + h (e − e)2 + ghc (∇e · ∇e) (5)
2 2 2
where, C is the elastic modulus. The model parameters g, h and c are called the coupling modulus, gradient
parameter and the interaction function respectively (see description in Appendix A). The effect of these parameters
on the LGDM has been analysed in other related works [24,31,32]. The material degradation is represented through
a scalar damage variable (D), which follows a negative exponential law (Appendix A). In Eq. (5), the first term is the
standard strain energy, while the second term (also called the coupling term) is responsible for interactions between
processes at the macro and micro levels characterized by the difference (e – ē). In other words, the coupling term in
Eq. (5) quantifies the coupling between the responses of macro and micro levels through the relative difference in
the values of e and ē. As pointed out in Peerlings et al. [30], the coupling term also indicates the amount of energy
that can be stored by creating the difference between e and ē. Finally, the third term in Eq. (5) incorporates the
interaction among the micro level processes through the gradient of micro-equivalent strain (∇ ē). The micro level
processes (representing the formation of microcracks) are typically most active at the start of damage. The activity
here means that the microcracks interact among themselves and affect each other’s responses. However, on further
evolution of damage, the activity of the micro level processes tends to concentrate in a narrow region, as shown in
Fig. 1. This concentration of activity leads to further localization and formation of a macrocrack. It is worth noting
that the decrease in the activity of micro level processes is accounted for by the interaction function (g) in Eq. (5).
The interaction function is adopted such that it decreases with damage evolution (see Appendix A for details), thus,
causing a decrease in the activity of micro level processes in the LGDM.
Following the Coleman–Noll procedure on free energy density in Eq. (5), the constitutive equations for the
LGDM are obtained as,
∂e
σ = (1 − D) Cε + h (e − e) (6)
∂ε
σ = h (e − e) (7)
ξ = ghc (∇e) (8)
From the constitutive relations in Eqs. (6)–(8) and balance laws in Eqs. (3)–(4), analogies for scale transfer
with concurrent multiscale modelling [33] can be drawn. Referring to Eqs. (A.1)–(A.2) (in Appendix A) for the
expressions of interaction function (g) and damage variable (D), it can be observed that the interaction function (g)
4
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

depends on damage (D), and this damage is further dependent on the micro-equivalent strain (ē). In addition, the
evolution of ē is affected by the interaction function (g) through the microforce balance equation (4) and constitutive
equations (7)-(8). These dependencies can be represented as,
g(D) ← D(ē) ← ē ← g(D) (9)
Hence, the scale transfer from micro to macro scale and vice-versa is realized through a set of variables (interaction
function, damage and micro-equivalent strain) used in the constitutive relations.

2.1.2. Decoupling Methodology (Staggered LGDM)

The weak forms of the governing equations (3) and (4) are expressed as,
∫ ∫
(∇wu ) : σ dΩ = wu · td∂Ω (10)
∫Ω ( ∂Ω
)
we σ + ∇we · ξ dΩ = 0 (11)
Ω
where wu and wē are the corresponding weight functions. The finite element (FE) discretization of u and ē is taken
as,
u = Nu û; ε = Bu û (12)

e = Ne ê; ∇e = Be ê (13)
N and B are the shape functions and their derivatives. The respective subscripts represent the associated primary
variables and the caps denote their nodal values. Applying the Galerkin approach and substituting FE discretization
in the weak forms in Eqs. (10)–(11) leads to the following,
∫ ∫
BuT σ dΩ = NuT td∂Ω (14)
Ω
∫ ( ∂Ω
)
NeT σ + BeT ξ dΩ = 0 (15)
Ω
The linearization of Eqs. (14)–(15) yields a coupled system which can be represented as,
Kuu Kue δ û
[ ]{ } [ ]
F
= u (16)
Keu Kee δ ê Fe
where K denotes the tangent stiffness matrix coefficients and F denotes the force vector components. The solution
is obtained as δ û and δ ê using the Newton–Raphson incremental-iterative procedure. It should be noted that the
coupling between primary variables is apparent from the presence of coupling terms Ku ē and Kēu . Moreover, the
tangent stiffness matrix in Eq. (16) is non-symmetric, for which a monolithic solution scheme is generally adopted
[10,32].
In the present work, it is proposed to decouple the system in Eq. (16) and implement a staggered solution scheme
for the LGDM. The decoupled system is obtained when the coupling terms (Ku ē and Kēu ) are absent in,
δ û
[ ]{ } [ ]
Kuu 0 F
= u (17)
0 Kee δ ê Fe
The decoupled system in Eq. (17) has the following advantages over the coupled system in Eq. (16),
• It leads to less computational effort due to symmetric diagonal stiffness terms (Kuu and Kēē ) and the absence
of coupling terms (Ku ē and Kēu ).
• The use of linear interpolations for both u and ē (unlike coupled system, which requires quadratic interpolation)
makes the decoupled system easily integrable in the already existing finite element codes without many
changes.
• The introduction of extended FEM (XFEM) is much easier for the decoupled system than the coupled system
[22].
5
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

The decoupling is achieved through an operator-split methodology based on the history parameter (κ). This
parameter κ is defined such that it stores the maximum value of the micro-equivalent strain field (ē) at any point
of time (t) during the loading history,
κ = max {e (t) |0 ≤ t ≤ T } (18)
The operator-split methodology shares similarities with the phase field method discussed in Miehe et al. (3). In this
methodology, the solution scheme is split into two steps, i.e., a displacement step and a damage step. These steps
are summarized in the following points,
• Displacement step: In this step, the displacement part of the system is solved, i.e. Kuu δu = Fu (Eq. (17)),
while keeping the damage (D) constant. The constant damage is the outcome of the constant history parameter
(κ) and constant micro-equivalent strain (ē).
• Damage step: In this step, the micro-equivalent strain part of the system is solved, i.e. Kēē δē = Fē
(Eq. (17)), while assuming constant displacement (u). The history parameter (κ) is updated in this step, which
consequently updates the damage variable (D), hence the name damage step.

2.1.3. Linearization
The non-linear governing equations (14) and (15) are linearized for the application of the incremental-iterative
procedure. Within a load step, the linearization is performed at iteration i with respect to known values at the
previous iteration i – 1. Hence, the stress terms in Eqs. (14)–(15) can be linearized as,
i i−1
σ i = σ i−1 + δσ ; σ i = σ i−1 + δσ ; ξ =ξ + δξ (19)
Substituting the linearized stress terms back in Eqs. (14)–(15) yields,
∫ ∫ ∫
BuT δσ dΩ = NuT td∂Ω − BuT σ i−1 dΩ (20)
∫ ( Ω ∂Ω
∫ ( Ω
) i−1
)
Ne δσ + Be δξ dΩ = −
T T
NeT σ i−1 + BeT ξ dΩ (21)
Ω Ω
It must be noted that Eqs. (20)–(21) arise from the standard equilibrium equation and the micro-force balance
equation described in Section 2.1.1. Hence, Eqs. (20)–(21) form the displacement and the micro-equivalent strain
parts of the system respectively. Consequently, as per the decoupling methodology in the previous sub-section,
Eq. (20) is solved in the displacement step and Eq. (21) is solved in the damage step. With this in consideration,
δσ in Eq. (20) can be expressed using the constitutive relation in Eq. (6) as,
∂e
( )
δσ = (1 − D) Cδε + h(δe − δe) (22)
∂ε
It is noted that in the above equation (22), δē = 0 because the micro-equivalent strain is assumed to be constant
in the displacement step. Hence, rewriting Eq. (22) and substituting appropriate FE representations of the variables
yield,
∂e ∂e
( )( )
δσ = (1 − D) CBu δ û + h Bu δ û (23)
∂ε ∂ε
Similarly, the terms δσ and δξ in Eq. (21) are expressed using the constitutive relations in Eqs. (7)–(8) as,
δσ = h (δe − δe) (24)
δξ = (δg) hc∇e + ghcδ (∇e) (25)
In Eq. (24), δe = 0, because the macro-equivalent strain (e) (which is evaluated from u) remains constant in the
damage step. Eq. (25) remains unchanged because it does not have any term containing e. Hence, rewriting Eq. (24)
and substituting FE representation of the variables in Eq. (24) and (25) yields,
δσ = −hNe δ ê (26)
∂g ∂κ
( )( )
δξ = hc Be ê Ne δ ê + ghc Be δ ê
( ) ( )
(27)
∂κ ∂e
6
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Subsequently, the substitution of δσ , δσ and δξ in the linearized forms (Eqs. (20)–(21)) leads to,
δ û
[ ]{ } [ ]
Kuu 0 F
= u (28)
0 Kee δ ê Fe
where the stiffness coefficients are expressed as,
⎧ [( ) ]T ( ) ⎫
∂e i−1
∂e i−1
∫ ⎨( ⎬
BuT 1 − D i−1 CBu − h
)
Kuu = Bu dΩ (28a)
Ω ⎩ ∂ε ∂ε ⎭

{( }
∂g ∂κ
∫ )i−1 ( )i−1
)i−1
BeT Ne + g hcBe − NeT hNe dΩ
i−1
(
Kee = hc Be ê (28b)
Ω ∂κ ∂e
and the force vector components are expressed as,
{ ( )i−1 }
∂e
∫ ∫
i−1
NuT td∂Ω − BuT (1 − D)i−1 CBu ûi−1 + h e − Ne ê
( )
Fu = dΩ (28c)
∂Ω Ω ∂ε
∫ { }
)i−1 i−1
NeT h e − Ne ê + BeT g i−1 hcBe ê
(
Fe = − dΩ (28d)
Ω

2.1.4. Staggered solution scheme

The decoupled system in Eq. (28) is solved using the incremental-iterative staggered solution scheme provided
in Fig. 2. Similar staggered solution schemes are used for the regularized fracture using screened Poisson equation
[34,35]. The scheme is such that the displacement and the damage steps are solved separately. The simulation starts
with the input of material properties and initial displacement load assuming zero micro-equivalent strain (ē). Within
the displacement step, the stiffness matrix (Kuu ) and force vector (Fu ) are evaluated and assembled. After obtaining
a solution of the linearized system (Kuu δu = Fu ), the variables u, ε and e are updated. These updated variables
are passed to the damage step. In the damage step, these passed variables are used to evaluate the stiffness matrix
(Kēē ) and force vector (Fē ). After solving the linearized system (Kēē δē = Fu ), the obtained solution δē is used to
update ē, κ and D.
( δe Then the
δu
) convergence of both displacement and damage steps is checked after each iteration. If the solution error
e
and u
is found less than tolerance, then the convergence of load step is reached and the updated variables are
passed on for the next load step with an increase in load. Otherwise, the updated variables are passed on for the
next iteration for convergence. Further, if the end of loading is reached, then the simulation is stopped and output
parameters are obtained.

2.2. Staggered LGDM with extended FEM (XFEM)

In this work, the proposed continuous–discontinuous transition is carried out by introducing a discontinuous
method (extended FEM/XFEM) at the end stages of damage in a continuous method (decoupled LGDM). It is noted
that several other continuous methods like peridynamics (PD) [36], dual-horizon peridynamics (DH-PD) ([37]) and
discontinuous methods such as extended element-free Galerkin (XEFG) method [38], cracking particle (CP) method
[7,39] could be possible alternatives of the proposed continuous–discontinuous approach. However, the proposed
continuous–discontinuous approach offers several key advantages over the existing methods like XEFGM and CPM,
• Unlike XEFGM and CPM, the additional criteria for crack initiation/propagation and determination of crack
length/direction are not needed in the proposed continuous–discontinuous approach.
• The introduction of discontinuous cracks (through XFEM) at the end of damage evolution avoids the
requirement of cohesive laws for accurate post-peak response in the proposed continuous–discontinuous
approach. In contrast, cohesive laws are needed in the XEFGM and CPM.
• The implementation complexities associated with the XEFGM and CPM are avoided by adopting XFEM,
which can be easily integrated into the existing finite element implementations of CGDM/LGDM.
7
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 2. A flowchart showing the staggered solution scheme used for solving the decoupled LGDM.

Therefore, in the present work, a discontinuous crack using XFEM is introduced at the end of damage evolution
i.e. when D ≈ 1. The introduction of discontinuous crack rectifies several abnormalities associated with the LGDM
at high damage values like high strains and severe stress oscillations. Moreover, as outlined in Section 1, the
discontinuous crack also solves the problem of incorrect representation of secondary variables, which is observed in
the multi-physics implementations. Similar to Simone et al. [22], the XFEM crack is introduced when the damage
value at all the integration points in an element reaches 0.9999. It is also emphasized that introducing XFEM
cracks at the end of damage evolution (i.e. at D ≈ 0.9999) eliminates the problem of brittle behaviour observed in
conventional GDM [22]. While introducing XFEM, the following points are noted,
• The crack length increments are performed such that the crack tip always propagates to the other adjacent
elements i.e. within element crack increments are not allowed.
• After the crack increment, a data transfer is performed from old integration points to new ones. Subsequently,
a null step is executed to redistribute the stresses and establish equilibrium.
• Only the split elements (completely cut by the crack) are modelled using the XFEM.
The XFEM crack introduction strategy is schematically shown in Fig. 3.
It is highlighted that the aforementioned continuous–discontinuous methodology avoids complexities associ-
ated with the standalone discontinuous methods (like XFEM, XEFGM and CPM) and leads to a simplified
implementation. The simplifications are outlined as,
• The introduction of cohesive forces after crack increment is not necessary because the material points (or
elements) near the crack increment have almost attained unity damage, D ≈ 1. Therefore, the introduction of
cohesive forces will have an insignificant effect, hence, not considered.
8
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 3. A schematic representation of the XFEM strategy used in the proposed continuous–discontinuous (LGDM + XFEM) approach.

• The enrichment of tip element can be safely avoided, as suggested in Simone et al. [22]. This is again due to
unity damage in the elements near the crack tip, which will render the effects of tip enrichment insignificant.
Hence, only step enrichment for split elements is used.
Moreover, the standard XFEM is used in the present implementation due to its relative simplicity in understanding
and straightforward implementation. Besides, it is important to note that the proposed methodology is not limited to
standard XFEM only, and more advanced XFEM methodologies (like the phantom node method in Rabczuk et al.
[40]) may be employed for improved performance and accuracy.
The XFEM approximation and the derivatives for the primary variables are written as [41,42],
u = Nu û + Nenr
u â; e = Ne ê + Neenr b̂ (29)

ε = Bu û + Benr
u â; ∇e = Be ê + Benr
e b̂ (30)
where â and b̂ are the extra nodal degrees of freedom arising out of the split enrichment. The superscript ‘enr’
on N and B represents their enriched forms. The shifted Heaviside step enrichment is used to represent the crack
in split elements. The details of enrichment and the resulting enriched shape functions/derivatives are outlined in
Appendix B. The substitution of XFEM approximation (Eqs. (29)–(30)) in the linearized system of staggered LGDM
(Eq. (28)) yields,

⎪δ û⎪
⎡ ⎤⎧ ⎫ ⎡ ⎤
Kuu Kua 0 0 ⎪ Fu
⎥ ⎨δ â ⎬ ⎢Fa ⎥
⎪
⎢Kau Kaa 0 0
=⎢ (31)
δ ê ⎪
⎢ ⎥ ⎥
⎣0 0 Kee Keb ⎦ ⎪ ⎪ ⎪
⎣Fe ⎦
0 0 Kbe Kbb δ b̂ Fb
⎩ ⎭

where the stiffness coefficients apart from Kuu and Kēē are expressed as,
⎧ [( ) ]T ( ) ⎫
∂e i−1
∂e i−1
∫ ⎨( ⎬
BuT 1 − D i−1 CBenr Benr
)
Kua = u −h dΩ (31a)
Ω ⎩ ∂ε ∂ε u
⎭
Kua = (Kau )T (31b)
⎧ [( ) ] T ( ) ⎫
∂e i−1
∂e i−1
∫
( enr )T ⎨ ⎬
Kaa = Bu (1 − D)i−1 CBenr − h Benr
dΩ (31c)
Ω ⎩ u
∂ε ∂ε u
⎭
{ }
∂g ∂κ
∫ ( )i−1 ( )i−1
T
( )i−1 enr
Keb = Be hc Be ê Ne + g i−1 hcBeenr − NeT hNeenr dΩ (31d)
Ω ∂κ ∂e
9
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Kbe = (Keb )T (31e)

{( ) ( ) }
∂g i−1 ∂κ i−1 ( )i−1 enr
∫
( enr )T i−1 enr
)T
− Nenr hNenr
(
Kbb = Be hc Be ê Ne + g hcBe e dΩ (31f)
Ω ∂κ ∂e e

and the force vector components apart from Fu and Fe as,

{ ( )i−1 }
enr i−1 ∂e
∫ ∫
( enr )T ( enr )T
(1 − D) CBu û + h e − Ne ê
i−1 enr i−1
( )
Fa = Nu td∂Ω − Bu dΩ (31g)
∂Ω Ω ∂ε
∫ { }
( enr )T ( )i−1 ( enr )T i−1 i−1
Fb = − Ne h e − Neenr ê + Be g hcBeenr ê dΩ (31h)
Ω

Eq. (31) is solved using the incremental-iterative procedure outlined in Fig. 2 to obtain solutions as δ û, δ â, δ ê and
δ b̂.

2.3. XFEM + LGDM (XLGDM) for thermo-mechanical fracture

In this sub-section, the XLGDM (XFEM + LGDM) proposed in the previous sub-section is extended to a
multi-physics framework of thermo-mechanical fracture [43]. This sub-section can be treated as the extension of
thermo-mechanical LGDM developed in Sarkar et al. [31] to a continuous–discontinuous (XLGDM) framework. In
this framework, the XLGDM constitutes the mechanical part, while the thermal equilibrium equation governs the
thermal part. The steady-state thermal equilibrium equation is expressed as,
∇ · Q − qgen = 0 in domain Ω (32)
with natural and essential boundary conditions Q·n = qs and θ = θ s on the boundary ∂Ω . The Fourier law gives the
conductive heat flow as Q = – ∇{(1 – D)Kθ }, where K is the isotropic thermal conductivity and θ is the temperature.
The θ s is the prescribed temperature, while qgen and qs denote the internal heat generation and heat flow across the
boundary. To model adiabatic cracks, thermal conductivity degradation is incorporated through the damage variable
as (1 – D)K. The weak form of the thermal equilibrium equation is written as [31],
∫ ∫ ∫
(∇wθ ) ∇ {(1 − D) Kθ } dΩ = wθ qgen dΩ + wθ qs dΩ (33)
Ω Ω ∂Ω
where wθ is the weight function. The thermo-mechanical coupling of temperature with the mechanical equilibrium
equation is achieved through a thermal force arising due to thermal strains. Due to thermal strains, the standard
constitutive equation (6) is modified as,
∂e
σ = (1 − D) Cε − εth + h (e − e)
( )
(34)
∂ε
where the thermal strain is expressed as εth = aβ∆θ with a = {1,1,1,0,0,0}T and β = Eα/(1 − 2ν). E and ν
are Young’s modulus and Poisson’s ratio respectively, whereas α is the coefficient of linear thermal expansion.
Following the XFEM formulation for thermal problems [44], the interpolation function and its derivatives (with
split enrichment) are written as,
θ = Nθ θ̂ + Nenr
θ ĉ; ∇θ = Bθ θ̂ + Benr
θ ĉ (35)
Consequently, substituting Eq. (35) in the weak form and applying the Galerkin approach yields the following linear
system of equations for the weakly coupled thermo-mechanical XLGDM,
δ û
⎡ ⎤⎧ ⎫ ⎡ ⎤
Kuu Kua 0 0 0 0 ⎪
⎪ ⎪ Fu
δ
⎪
⎢ Kau Kaa 0 0 0 0 ⎥⎪⎪ â ⎪ ⎢Fa ⎥
⎪
⎪ ⎢
⎢ ⎥⎪⎨δ ê ⎪
⎪ ⎬ ⎢ ⎥
⎢ 0 0 K ee K eb 0 0 ⎥ Fe ⎥
=⎢ (36)
δ
⎢ ⎥ ⎥
⎢ 0
⎢ 0 K be K bb 0 0 ⎥
⎥⎪⎪ b̂ ⎪
⎪
⎢
⎢ Fb⎥
⎥

⎩δ θ̂ ⎪
⎣ 0 0 0 0 Kθ θ Kθ c ⎦ ⎪ ⎪ ⎣Fθ ⎦
⎪ ⎪
⎪
⎪
⎪
0 0 0 0 Kcθ Kcc δĉ Fc
⎭

10
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

where ĉ contains the extra degree of freedoms due to enrichment of the temperature (θ). With consistent
linearization, the thermal coefficient terms of the stiffness matrix are expressed as,
∫
Kθθ = BθT (1 − D)i−1 KBθ dΩ (36a)
Ω
∫
Kθ c = BθT (1 − D)i−1 KBenr
θ dΩ (36b)
Ω

Kcθ = (Kθ c )T (36c)

∫
( enr )T
Kcc = Bθ (1 − D)i−1 KBenr
θ dΩ (36d)
Ω
In the thermo-mechanical framework, the force terms of the standard equilibrium, i.e. Fu and Fa are modified due
to a change in the constitutive relation outlined in Eq. (34). Hence, the modified force vector terms (including an
additional term for the thermal force) are written as,
∫ ∫
( )i−1
Fu = T
Nu td∂Ω + NuT (1 − D)i−1 εth dΩ
∂Ω Ω
∫ { ( ) }
)i−1 ∂e i−1 (36e)
Bu (1 − D) CBu û + h e − Ne ê
T i−1 i−1
(
− dΩ
Ω ∂ε
∫ ∫
)T )T ( )i−1
Nenr Nenr (1 − D)i−1 εth
( (
Fa = u td∂Ω + u dΩ
∂Ω Ω
{ )i−1 } (36f)
∂e
∫ (
)T )i−1
Benr (1 − D) i−1
CBenr i−1
Neenr ê
( (
− u û +h e− dΩ
Ω
u
∂ε
Also, the force vector terms of the thermal equilibrium equation Fθ and Fc are expressed as,
∫ ∫ ∫
i−1
Fθ = NθT qgen dΩ + NθT qs d∂Ω − BθT (1 − D)i−1 KBθ θ̂ dΩ (36g)
Ω ∂Ω Ω

∫ ∫ ∫
)T )T )T i−1
Nenr Nenr Benr θ θ̂
(1 − D)i−1 KBenr
( ( (
Fc = θ qgen dΩ + θ qs d∂Ω − θ dΩ (36h)
Ω ∂Ω Ω
Eq. (36) is solved in an incremental-iterative manner shown in Fig. 3. The flowchart in Fig. 4 is an extension of the
flowchart in Fig. 2. It should be noted that the thermal step (in Fig. 4) is appended to the existing displacement and
damage steps. The simulation starts with the definition of material properties (thermal and mechanical), boundary
conditions (thermal and mechanical) and thermal loads. Similar to the procedure in Section 2.2, the standard
equilibrium is solved first in the displacement step. Then, the updated variables from the displacement step are used
by the damage step to solve the micro-force balance equation. The solution of the micro-force balance equation
updates the damage variable (D). After this, the updated damage variable is used in the thermal step to determine
the temperature distribution (θ) from the thermal equilibrium. Subsequently, the updated temperature variable is
passed in the displacement step to evaluate thermal forces (Eqs. ((36)e–(36)f). This procedure is repeated till the
convergence is achieved for all three primary variables viz. u, ē and θ as shown in Fig. 4.

2.4. Summary

In summary, Section 2 presents the detailed formulation and implementation methodology for staggered LGDM,
XLGDM (XFEM + staggered LGDM) and the thermo-mechanical XLGDM. The details include,
• Conversion of LGDM from a fully coupled monolithic scheme to a decoupled staggered scheme.
• Use of XFEM to represent a discontinuous crack in the staggered LGDM at the end of damage evolution.
• Application of XLGDM to simulate the thermo-mechanical fracture.
Moreover, the numerical and material parameters used in the methods mentioned above are discussed in
Appendices A and B.
11
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 4. A flowchart showing the staggered solution scheme fused for solving thermo-mechanical (XFEM + LGDM) XLGDM.

3. Results and discussion

In this section, the formulation and implementation methodology presented in Section 2 is employed to ascertain
the benefits of the proposed methods. A total of five numerical examples are used such that they cover the Mode I
and mixed-mode fracture under mechanical and thermo-mechanical loading. The obtained results are validated with
the published literature and available experimental data. All the examples are solved using plane strain idealization.
Several common model parameters in the LGDM, XLGDM and XCGDM adopted from Sarkar et al. [45] are h = E
× 10−9 , a = 0.99, b = 5, R = 0.005 and n = 5.

3.1. Side edge notch specimen (mode I mechanical fracture)

In this sub-section, a side edge notch (SEN) specimen undergoing Mode I fracture is used to demonstrate various
aspects of the proposed method. The geometry, load, boundary conditions and parameters for this numerical example
are shown in Fig. 5 ([24]). In this example, the SEN specimens are subjected to a monotonic displacement loading of
u = 0.005 mm in each equal-sized load step. The modified von Mises equivalent strain is used for the simulations in
this sub-section. The biquadratic quadrilateral elements are used in the monolithic scheme and bilinear quadrilateral
elements are used in the staggered scheme.
Firstly, the results obtained from the proposed staggered LGDM are compared and validated with the already
existing monolithic LGDM (Section 3.1.1). After validation, the staggered LGDM is converted to XLGDM by
introducing a discontinuity/crack at the end of damage evolution using extended FEM (XFEM) in Section 3.1.2.
Finally, the advantages of XLGDM over the XCGDM (XFEM + Conventional gradient damage method [22] are
highlighted in Section 3.1.3.
12
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 5. A schematic representation of geometry, load, boundary conditions and parameters for the side edge notch specimen (dimensions
are in mm).

Fig. 6. A comparison of load–displacement and damage response obtained from the monolithic and staggered schemes for the LGDM.

3.1.1. Validation of staggered LGDM with the monolithic LGDM

In this sub-section, the staggered LGDM proposed in Section 2.1 is compared and validated with the existing
monolithic LGDM [24]; [45]. The load–displacement response and the damage profiles obtained from both
monolithic and staggered schemes are shown in Fig. 6. Three types of uniform mesh having 5000, 11 250 and
20 000 elements are used to check convergence. It is observed (in Fig. 6a) that both the schemes converge at the
mesh of 100×200 = 20 000 elements and their converged responses are similar. Moreover, the damage plots obtained
from both the schemes are almost identical, as shown in Fig. 6b. Hence, it is evident that the proposed formulation
for the staggered LGDM in Section 2.1 yields a consistent fracture response.
Besides, as mentioned in Section 2.1.2, the staggered scheme’s computational advantages are also ascertained in
this sub-section. The nonlinear incremental iterative Newton–Raphson solution procedure is used for both monolithic
and staggered schemes. The displacement load step size of 0.005 mm is used in both the schemes, as mentioned
earlier. The automatic load step size control is not adopted for the sake of simplicity and straightforward comparison.
13
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 7. A comparison of various convergence parameters for monolithic and staggered schemes in LGDM.

The load step size is kept constant and equal for the entire simulation. The iteration count and simulation time are
used for the comparison. The plot of iteration counts within each load step is shown in Fig. 7a. It is observed that the
number of iterations required for the convergence is more in the staggered scheme as compared to the monolithic
scheme. This observation is in line with Mediavilla et al. [46]. In Mediavilla et al. [46], it is suggested to use
small load step size to attain faster convergence in the staggered scheme of conventional GDM. In other words,
the large number of iterations observed in the staggered scheme (as compared to the monolithic scheme in Fig. 7a)
is a consequence of using the same load step size (0.005 mm) for both. However, the staggered scheme takes less
computation time to complete the total simulation despite more iterations. As a result, the staggered scheme’s time
per iteration is notably less than the monolithic scheme, as shown in Fig. 7b. The less computation time in the
staggered scheme is due to several factors such as,
• The symmetric tangent stiffness matrix.
• Absence of the coupling terms.
• Use of bilinear elements instead of biquadratic elements (required in fully coupled LGDM).
Hence, the staggered scheme is treated as computationally more efficient than the monolithic scheme.

3.1.2. Staggered LGDM + XFEM (XLGDM)

In this sub-section, the SEN specimen is solved using staggered LGDM coupled with the extended FEM (XFEM)
to introduce discontinuous cracks at the end of damage evolution. As mentioned in Section 2.2, the XFEM cracks
are inserted when the damage reaches a threshold value i.e. Dthr = 0.9999. The response obtained from XLGDM
(XFEM + staggered LGDM) is compared with previously validated staggered LGDM (without XFEM) to assess
the correctness. The staggered LGDM is referred to as LGDM from here on for simplicity in terminology. In the
XLGDM, the initial notch is modelled using an XFEM crack. The load–displacement response and the damage
plots obtained from both LGDM and XLGDM are shown in Fig. 8.
From Fig. 8, the following major observations are drawn,
• It is observed that the load–displacement responses (Fig. 8a) of LGDM and XLGDM are similar for most of
the loading. The observed drops in the load–displacement response of XLGDM are at the end of loading. The
drops at the end, eliminate the early drops and jumps reported in XCGDM (XFEM + CGDM) by Simone
et al. [22] even at high damage values.
• The residual strength at high damage values (D ≈ 1) is negligible in the XLGDM compared to the LGDM.
The negligible residual strength in XLGDM is due to the introduction of the XFEM crack, which leads to
complete unloading of the specimen.
14
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 8. A comparison of load–displacement and damage response obtained from LGDM and XLGDM (XFEM + LGDM).

Fig. 9. A comparison of deformed mesh configurations at initial and final stages of loading obtained from LGDM and XLGDM (XFEM +
LGDM).

• Unlike XLGDM, some residual strength remains in the LGDM (Fig. 9a) even at high damage values due to
the use of exponential damage law (Appendix A).
The damage plots in Fig. 8b and deformed configuration plots in Fig. 9 demonstrate the similarities in the LGDM
and XLGDM. These similarities prove the consistency of the results obtained from the XLGDM with the LGDM.
Apart from the similarities, it is reported in Vandoren and Simone [47] and Sarkar et al. [45] that the LGDM
suffers from the spurious stress oscillations which result from the oscillations in the micro-equivalent strains (ē).
Similar oscillations in the micro-equivalent strains near the damage region are clearly observed in LGDM (Fig. 10).
The oscillations are less at the start of the damage evolution and increase with the increasing loads, as observed
in Fig. 10b. Moreover, in LGDM, the micro-equivalent strain increases unboundedly with the applied load. This
unbounded increase is observed at the crack tip in Fig. 10b.
In contrast to this, the maximum value of the micro-equivalent strains is limited in the XLGDM due to the
damage threshold value (Dthr ). Due to Dthr , the value of micro-equivalent strain is limited to a fixed amount that
15
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 10. The equivalent strain plots at start and end of damage propagation obtained from LGDM.

Fig. 11. The equivalent strain plots at different stages of damage propagation obtained from XLGDM (XFEM + LGDM).

was attained before introducing the XFEM crack, as observed in Fig. 11. Additionally, the micro-equivalent strain
becomes zero behind the crack tip resulting in elastic unloading of the material. Consequently, due to the limited
values of the micro-equivalent strains, the oscillations are insignificant in the XLGDM, as observed in Fig. 11c.
It is emphasized that the spurious oscillations shown in Fig. 10 can be avoided by using quadratic shape functions
(with reduced integration) for the displacement (u) as demonstrated in Sarkar et al. [45]. However, the quadratic
shape functions are not used in the present work to avoid extra complexity and computational overhead involved
in the XFEM implementation.

3.1.3. Comparison of XLGDM and XCGDM

In this sub-section, the comparison of results obtained from the XLGDM and the XCGDM (XFEM + CGDM)
is presented. For this numerical example, the formulation of XCGDM was recovered by setting the interaction
function (Appendix A) as unity (g = 1) in the formulation of XLGDM as discussed in Sarkar et al. [45]. In order
to match the load–displacement response and limit the damage spreading in XCGDM, the values of c and β for
the XCGDM are taken as 4 mm 2 and 20 respectively. The rest of the parameters are the same as LGDM. The
16
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 12. A comparison of load–displacement and damage response obtained using XCGDM (XFEM + CGDM) and XLGDM (XFEM +
LGDM).

mesh of 20000 elements is considered for both methods based on the previous sub-section’s converged solution.
The load–displacement response and the damage plots obtained from the XCGDM and XLGDM are shown in
Fig. 12. The load–displacement response of both the methods agrees well for most of the loading. It is observed
from Fig. 12b that despite considering a low value of the gradient parameter c = 4 mm 2 , the spurious damage
spreading is still present in the XCGDM. Similar plots showing damage spread were reported in Simone et al. [22]
also.
Apart from this, Simone et al. [22] also reported that in the load–displacement response of XCGDM, abnor-
malities such as early drops and jumps were observed. The early drops in the load–displacement response can be
avoided by taking a higher value of damage threshold (Dthr ). However, the abnormal jumps observed in the load–
displacement response were due to the large-sized XFEM crack increments [22]. The root cause of the large-sized
XFEM crack increments in the XCGDM is the non-local averaging. This can be easily understood by the damage
evolution pattern observed in XCGDM and XLGDM shown in Fig. 13. Due to the non-local averaging, the point of
maximum damage (in XCGDM) occurs at some distance from the crack tip and with the increase in load, it further
moves away, as shown by a red arrow in Fig. 13a. Contrastingly, the maximum damage point in the LGDM occurs
near the crack tip, right from the beginning of damage evolution and moves towards the crack tip with further load
increments (Fig. 13b). The closeness of the maximum damage point near the crack tip in the XLGDM leads to
smaller XFEM crack increments, as shown in Fig. 14. On the contrary, the crack growth increments observed in
the XCGDM are larger and irregular.
Moreover, in Fig. 12a, a steep load drop is observed in the load–displacement response of XLGDM when the
XFEM is introduced. The steepness of the drop can be attributed to the damage evolution in XLGDM (Fig. 13b).
In XLGDM, the threshold value (Dthr ) is attained almost instantly along the entire crack length from the tip (at x =
50 mm) to the right end (at x = 100 mm). Unlike XLGDM, the load drop in XCGDM is more gradual (Fig. 12a) due
to the gradual evolution of damage. Hence, the XLGDM yields a more accurate fracture response of a catastrophic
failure in comparison to the XCGDM.
Additionally, in the XCGDM, a cohesive zone model (CZM) must be used for smooth transitioning from
continuous to discontinuous regime. The CZM enables introduction of XFEM at lower damage threshold value i.e.
Dthr [14,16]. The use of CZM eliminates some of the spurious effects in the XCGDM, such as (a) the elimination
of spurious damage spreading (Fig. 12b) during the later stages of damage evolution and (b) the abnormal drops
in the load–displacement response [22] become gradual by using a consistent cohesive law in the CZM. For
consistency, the cohesive law must be energetically equivalent to the CGDM. In other words, the energy dissipated
17
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 13. A comparison of damage evolution pattern along the horizontal midline of the SEN specimen (y = 100 mm) showing movement
of maximum damage in XCGDM and XLGDM.

Fig. 14. A comparison showing the position of XFEM crack tips (after crack increments) along the horizontal midline of the SEN specimen
(y = 100 mm) in XCGDM and XLGDM.

by the cohesive law (after introducing XFEM) must be equivalent to the energy dissipation by CGDM for similar
deformation. However, obtaining this energetic equivalence is not straightforward and simple, especially for the
mixed-mode fractures (e.g., [14]). In contrast, the XLGDM proposed in this work does not need such a transition
scheme due to following reasons,

(a) There is no damage spreading in the later stages of LGDM evolution. This allows a relatively higher damage
threshold value (Dthr ) for the introduction of XFEM.
(b) Due to the high damage threshold (Dthr = 0.9999), the observed drops in the load–displacement response (after
the introduction of XFEM) are small in XLGDM as compared to XCGDM [22].

Hence, the absence of CZM in the continuous–discontinuous scheme in the XLGDM leads to simpler and
straightforward implementation.

3.2. L shaped specimen (mixed-mode mechanical fracture)

In this sub-section, the capability of the XLGDM to accurately simulate mixed-mode fracture is demonstrated
through the example of L shaped specimen. The geometry, load, boundary conditions and parameters for the L
shaped specimen are shown in Fig. 15. The model and material parameters used in this numerical example are:
E = 21 GPa, ν = 0.18, κi = 0.00015, c = 150 mm 2 , β = 80. These parameters are adopted from Nguyen et al.
[48] and Meschke and Dumstorff [49] to match the experimental results in Winkler [50]. The Mazar’s definition of
equivalent strain is used in this example.
The simulations are performed using both LGDM and the XLGDM to signify the difference in their responses.
The L shaped specimen is subjected to a displacement-controlled load of u = 0.005 mm for 200 equal load steps.
18
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 15. A schematic representation of geometry, load, boundary conditions and parameters along with the experimental crack band for the
L shaped specimen (dimensions are in mm).

Fig. 16. A comparison of load–displacement response obtained using LGDM and XLGDM for the L shaped specimen.

Firstly, the load–displacement response obtained from LGDM is checked for convergence. Three types of non-
uniform mesh are considered with element size in the damage region equal to 3.125 mm (12800 elements), 2.7 mm
(15300 elements) and 2.5 mm (20000 elements). The convergence for the LGDM is observed at the mesh of 20000
elements, as shown in Fig. 16. The converged solution of LGDM lies well within the experimental band of Winkler
[50].
Moreover, the load–displacement response obtained from XLGDM is observed to be similar to LGDM until
the XFEM is introduced (Fig. 16). After the introduction of XFEM, the drop in the load is observed as expected.
However, the load drop is delayed due to the higher value of the damage threshold (Dthr ). A mesh of 20000 elements
is considered for the simulation using XLGDM.
A comparison of damage and deformed mesh plots obtained from LGDM and XLGDM is shown in Fig. 17. It
is observed that the LGDM exhibits some amount of spurious damage spreading as compared to the XLGDM. This
is a result of the unbounded increase in the micro-equivalent strain in the LGDM as discussed in Section 3.1.2. It
is also observed that the spread of damage in LGDM is over several elements in the lateral direction of the crack
band (Fig. 17a). Consequently, more elements in the lateral direction are excessively deformed as observed in the
19
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 17. A comparison of damage and deformed configuration plots obtained using LGDM and XLGDM for the L shape specimen.

Fig. 18. A schematic representation of geometry, load, boundary conditions and parameters for the square specimen with echelon cracks
(dimensions are in mm).

deformation plot of LGDM. Unlike this, the damage is concentrated on a narrow band along the XFEM crack in
the XLGDM (Fig. 17b). Moreover, the deformation is limited to only elements containing the XFEM crack, thus,
depicting the fracture in a physically consistent manner. Hence, the XLGDM is capable of modelling mixed-mode
fracture accurately as compared to LGDM.

3.3. Square specimen with echelon cracks (mixed-mode mechanical fracture)

In this sub-section, the capability of the XLGDM to simulate multiple crack initiations, propagations and
coalescence is demonstrated through the example of a square specimen with echelon cracks. This example is adopted
from the work of Patil et al. [51], in which, phase field method is used to simulate fracture. The geometry, loads,
boundary conditions and parameters are shown in Fig. 18. The model and material parameters used in this example
are: E = 1.0 GPa, ν = 0.2, κi = 0.002, c = 4 mm 2 , β = 5. The modified von Mises equivalent strain is used for the
simulations. The displacement-controlled loading is applied at the top edge of the specimen, as shown. A uniform
load step size of 0.001 mm is used in the simulations. The simulations are carried out using both LGDM and
XLGDM. The improvements offered by XLGDM over the LGDM are highlighted in the end.
The load–displacement plots for this numerical example are shown in Fig. 19. The mesh convergence obtained
for the LGDM is used as a reference for the XLGDM simulation. The mesh size is depicted in Fig. 19 through the
20
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 19. A comparison of load–displacement response obtained using LGDM and XLGDM for the square specimen with echelon cracks.

edge length (EL) of the square elements in the critical regions. The load drop after the introduction of XFEM at
the later stages of loading is observed in XLGDM.
Fig. 20 depicts the damage evolution and deformed configuration plots obtained using the LGDM. In the case of
LGDM, the echelon cracks are defined explicitly using a non-uniform conformal mesh and the mesh size near the
critical regions is kept fine. From the damage plots in Fig. 20a, it is observed that the damage initiates at the middle
section of the specimen i.e., at the inner tips of the echelon cracks. The damage initiation simultaneously starts at
both left and right inner tips. Subsequently, damage propagates and coalesces near the middle of the specimen.
Similarly, the simultaneous multiple damage initiations are observed in the outer tips of the echelon cracks, which
ultimately propagate to the outer boundaries on the left and right. The deformed configuration plots in Fig. 20b
confirm the consistency in the damage evolution. The elements in the critical regions of the specimen (in Fig. 20b)
have an edge length of 0.2 mm.
Further, the present problem is solved using XLGDM and the obtained damage and deformed configuration
plots are shown in Fig. 21. The initial echelon cracks in the XLGDM simulations are defined using enrichments,
which do not require conformal mesh, as shown. It is noted that an adaptively refined mesh could have been used
in the XLGDM simulations for computational efficiency. However, a uniform mesh is adopted for simplicity. The
damage plots in Fig. 21a depict the XFEM crack extensions in the specimen at different stages. Similar to LGDM,
the XFEM cracks first simultaneously initiate at the inner tips of the echelon cracks and coalesce in the middle.
Subsequently, the cracks initiate at the outer tips of the echelon cracks and propagate to the outer boundaries of the
specimen. As highlighted previously, the introduction of XFEM in XLGDM leads to a highly localized fracture as
compared to the LGDM, as shown in deformed configuration plots in Fig. 21b.
The present numerical example demonstrates that the proposed XLGDM can consistently model multiple crack
initiations, propagation, and coalescence.

3.4. Side edge notch specimen (mode I thermo-mechanical fracture)

In this sub-section, the capability of the proposed thermo-mechanically coupled XLGDM to simulate fracture
under combined mechanical and thermal loading is demonstrated through a side edge notch (SEN) specimen. The
XLGDM is compared with the LGDM and CGDM to highlight its capabilities. The geometry, load, boundary
conditions (mechanical and thermal) and parameters are shown in Fig. 22, where the thermal boundary conditions
are depicted in red. The simulations are performed for an adiabatic crack under quasi-static conditions. The thermal
loading is such that the entire SEN specimen’s temperature decreases uniformly by 1 K in each load step while
maintaining a temperature difference of 10 K along the vertical length, as shown in Fig. 22. This type of loading
21
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 20. The damage evolution and deformed configuration plots obtained using LGDM for square specimen with echelon cracks.

causes the SEN specimen to undergo thermal contraction. The mechanical boundary conditions at the top and
bottom edges constrain the thermal contraction leading to damage initiation at the notch and subsequent fracture.
The material and numerical parameters are kept similar to Section 3.1 as described in Fig. 21. The additional thermal
properties used for the simulation are α = 12 × 10−6 (coefficient of thermal expansion) and K = 2.0 N/s-K (thermal
conductivity). The von Mises equivalent strain is used in the simulation. Following the converged solutions for SEN
specimen in Section 3.1, a uniform mesh of 100×200 = 20000 elements is used for the simulations.
This numerical example is solved using three methods viz. CGDM, LGDM and XLGDM, to demonstrate the
differences in the results. The obtained load–CTOD (crack tip opening distance) responses are shown in Fig. 23. The
parameters c and β for the CGDM and LGDM are calibrated to yield a similar load–CTOD response for enabling
a proper comparison. In the XLGDM, a drop in the specimen’s load-carrying capacity occurs near the end of the
simulation due to the introduction of XFEM at a high damage threshold of Dthr = 0.9999.
A comparison of temperature and damage plots obtained from thermo-mechanical CGDM and LGDM is shown
in Fig. 24. The temperature plots are shown at the start and end of the simulations. From the plots, two major
observations are drawn,
(a) Damage plots: The damage band in LGDM is significantly narrower than the CGDM. This is despite LGDM
using a high value of the gradient parameter c (which defines the size of the interaction region) as compared
to CGDM (Fig. 24b).
(b) Temperature plots: At the end of the simulation, the transition from high temperature to low temperature
(appears in green colour along the horizontal midline) occurs through a narrower region in LGDM as
compared to the CGDM (Fig. 24a). The narrowness of the damage band leads to a thin temperature transition
region in LGDM.
From the observations mentioned above, it is evident that the LGDM is more suitable than CGDM for a continuous–
discontinuous approach because it has a narrower damage band that results in a thin temperature transition region.
22
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 21. The damage evolution and deformed configuration plots obtained using XLGDM for square specimen with echelon cracks.

Fig. 22. A schematic representation of geometry, loads, boundary conditions (mechanical & thermal) and parameters for the side edge notch
specimen (dimensions are in mm).

23
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 23. A comparison of load–CTOD (crack tip opening distance) response obtained using thermo-mechanical CGDM, LGDM and XLGDM
for the SEN specimen.

The requirement of a narrow damage band for a continuous–discontinuous approach is essential since a narrow
damage band avoids abrupt changes during the simulation, which may cause instabilities while switching from
continuous to discontinuous regime [22]. Moreover, it is observed from the temperature plots in Fig. 24a that the
LGDM depicts an adiabatic crack’s temperature transition more accurately than CGDM at the end of the simulation.
In other words, an adiabatic crack does not allow heat to flow across. Therefore, the temperature transition across
the crack must be abrupt, as correctly represented by LGDM.
The temperature and damage plots obtained from thermo-mechanical XLGDM are shown in Fig. 25. It is
observed from the temperature plots of XLGDM that the temperature transition region disappears completely after
the introduction of XFEM and an abrupt temperature change is observed across the crack. This type of temperature
distribution accurately characterizes an adiabatic crack. Hence, XLGDM exhibits a discontinuous temperature
distribution across the crack, which is even more accurate than LGDM or CGDM. The deformed configuration
plots in Fig. 26 further clarify this observation. The deformed configuration plots show the thermo-mechanical
fracture for the three methods (CGDM, LGDM and XLGDM) in a consistent manner. The consistency can be
observed from the fact that, due to mechanical constraints and thermal contraction, the width of the SEN specimen
decreases while its length remains constant. This causes the damage to initiate and propagate from the notch tip for
all three methods. However, the deformed configuration plots in Fig. 26 also highlight some important differences
in the response of three methods. In CGDM, the wider deformation region spans over several elements across the
crack (Fig. 26a). In contrast, the deformation in LGDM and XLGDM is localized to a narrow region (Fig. 26(b)
and (c)). Particularly, in XLGDM, the localized deformation is confined to only one element which is even better
than LGDM, where a localized deformation spans across two elements.
In another observation, the response of CGDM, LGDM and XLGDM is studied for the representation of the
secondary variable. The secondary variable in this numerical example is the heat flux (Q) depicted through the
contoured flux line plots in Fig. 27. The magnitude of the heat flux is shown by the colour bar and the arrows on
the flux lines depict the direction. In Fig. 27, the start and end written below each plot depict the start and end
of damage propagation. In CGDM, it is observed that the flux lines become straight when they enter into damage
band, as shown in Fig. 27a. The straightening of the flux lines is caused due to the wide damage band in the CGDM
as shown in Fig. 24b. On the contrary, the LGDM (Fig. 27b) does not manifest such spurious heat flux response
due to a narrower damage band.
However, both the continuous methods (CGDM and LGDM) fail to accurately represent the heat flux variable
(Q) at the end of damage propagation. This is evident from the response of CGDM and LGDM in Fig. 27a and 27b.
Despite the decrease in the magnitude of heat flux in these plots, the direction of heat flux lines remains almost
24
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 24. A comparison of temperature and damage plots obtained using thermo-mechanically coupled CGDM and LGDM for the SEN
specimen.

unchanged from the start to the end of damage propagation. This is an incorrect response because, for an adiabatic
crack, the flux lines’ direction must change according to the damage propagation. Hence, with the introduction
of XFEM in the XLGDM, the heat flux lines are accurately shown in Fig. 27c. In this plot, the direction of heat
flux lines changes according to the damage/crack propagation along with the decrease in the magnitude of the heat
flux. Unlike in CGDM/LGDM, the XLGDM has a discontinuous representation of crack, resulting in an accurate
description of the secondary variable i.e., heat flux (Q).
In the end, a comparison of convergence parameters for both monolithic and staggered schemes used in the
thermo-mechanical LGDM is carried out. This comparison highlights the computational advantages realized from
converting coupled monolithic scheme to a decoupled staggered scheme. The nonlinear thermo-mechanical problem
is solved using the incremental iterative Newton–Raphson scheme. As stated previously, the incremental temperature
load is applied at the rate of ∆t = –1 K per load step. The load step size is kept constant throughout the simulation.
The simulation takes 200 load steps to complete, as shown in Fig. 28a. It is noted that an adaptive load step size
control can be used for a faster and efficient simulation. However, the load step size is kept constant and equal
throughout the simulation for brevity and straightforward comparison.
25
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 25. The temperature and damage plots obtained using thermo-mechanically coupled XLGDM for the SEN specimen.

Fig. 28 shows the comparison plots of several convergence parameters. In Fig. 28a, the number of convergence
iterations required in each load step by both monolithic and staggered thermo-mechanical LGDM are shown. It is
observed that during the post-peak regime, the number of iterations required by the staggered scheme (green) is
more as compared to the monolithic scheme (red), as shown in Fig. 28a and 28b. However, the staggered scheme is
found computationally more efficient as compared to the monolithic scheme. This is evident from Fig. 28b which
shows less total time and short time per iterations for the staggered scheme as compared to the monolithic scheme.
These observations are similar to the results obtained with the mechanical LGDM in Section 3.1.1.

3.5. Electronic solder joint failure (mixed-mode thermo-mechanical fracture)

In this sub-section, the fracture of an electronic solder joint under a complex thermo-mechanical load is studied.
This numerical example also demonstrates the response of thermo-mechanical XLGDM under mixed-mode loading
conditions. The solder joints are primarily used to join electronic components to the printed circuit board (PCB).
They are subject to high temperatures during the operation due to the flow of electric current. When the electricity
is switched off, the decrease in temperature causes the joints to contract thermally. As a result, the mechanical
constraints on the solder joint restrict its contraction leading to a fracture along the weakest section. The actual
fracture in a solder joint is depicted in Fig. 29 [52].
The fracture of the solder joint is simulated using both LGDM and XLGDM. For this, an axisymmetric model
of the joint is considered, as shown in Fig. 30. In the solder joint model, the electronic component lead is Invar
alloy and the filler is lead–tin alloy. The material properties of Invar alloy and the lead–tin alloy, along with the
other parameters are tabulated in Fig. 30. Mazars’s definition of equivalent strain is used in this example.
In the simulation, non-uniform meshes of 2181, 6342 and 8817 elements are considered for convergence study.
The mesh size in the damage region is identified by the characteristic element size which is obtained by taking the
square root of the element areas. The characteristic mesh sizes for 2181, 6242 and 8817 elements mesh are about
0.1 mm, 0.04 mm and 0.05 mm. The obtained load–CMOD (crack mouth opening distance) plots shown in Fig. 31
clearly show convergence. For load–CMOD plots, the reaction force on the horizontal surface PQ (in axisymmetric
model of Fig. 30) is considered as the load and the crack mouth at point P is considered for the opening distance.
The thermal load on the solder joint is applied in such a way that its temperature decreases uniformly when it
cools. The cooling takes place from the operating temperature of 423 K to the room temperature of 300 K. The
load is applied quasi-statically in a temperature step of ∆t = – 0.1 K. The damage and deformed configuration plots
obtained from LGDM and XLGDM are shown in Fig. 32 and Fig. 33 respectively. From these plots, the following
observations are drawn,
26
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 26. A comparison of deformed configuration plots at the start and end of simulation obtained using thermo-mechanically coupled
CGDM, LGDM and XLGDM for the SEN specimen.

• The damage band obtained in LGDM is slightly wider than in XLGDM. The narrower band in XLGDM is
due to the presence of discontinuous/XFEM crack (Fig. 32b).
• On the rightmost corner of the solder joint, an abnormal damage growth is observed in the LGDM (Fig. 32a).
Contrastingly, the XLGDM yields a physically correct result with no abnormal damage in the unloaded regions
away from the crack (Fig. 32b).
• In XLGDM, a localized deformation is obtained in the damaged regions as depicted in the deformed
configuration plot (Fig. 33b). On the contrary, the deformation in and around the damaged region is more
diffused across multiple elements in LGDM.
The observations show that the XLGDM representation of the solder joint fracture is more realistic and accurate
based on the actual fracture observed in the electronic solder joints (Fig. 29).
27
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 27. A comparison of heat flux plots (W/mm2 ) obtained at the start and end of damage propagation using thermo-mechanically coupled
CGDM, LGDM and XLGDM for the SEN specimen.

4. Conclusion and future scope

This work shows that the individual limitations present in the continuous and discontinuous approaches can be
alleviated by combining them in a numerically consistent manner. For this, the localizing gradient damage method
(LGDM) is chosen as the continuous approach and extended FEM (XFEM) as the discontinuous approach in this
work. The choice of continuous approach as LGDM is based on its ability to avoid unphysical behaviour observed in
conventional gradient damage method (CGDM). Firstly, the fully coupled LGDM is converted to a decoupled LGDM
through the operator-split methodology based on history micro-equivalent strain (κ). The staggered solution scheme
is employed to solve the obtained decoupled LGDM. The decoupled LGDM is then integrated with the XFEM to
introduce discontinuity at the end of damage evolution i.e. when D ≈ 1. The integrated XFEM + LGDM approach
is called XLGDM. The XLGDM is finally applied in a multi-physics framework to simulate thermo-mechanical
fracture.
28
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 28. A comparison of various convergence parameters for monolithic and staggered schemes in the thermo-mechanical LGDM.

Fig. 29. A sectional view showing the failure of a solder joint ([52]).

From the present work, the following outcomes are highlighted,

• The conversion of fully coupled LGDM to decoupled LGDM led to a notable decrease in computation effort
and complexity without compromising accuracy.
• The use of XLGDM (instead of XCGDM) led to more accurate fracture representation and avoided several
abnormalities associated with the XCGDM.
• The XLGDM led to simplification of methodology by avoiding cohesive zone modelling necessary for
continuous to discontinuous transition in XCGDM.
• The fracture response obtained from XLGDM is more physically consistent than the LGDM (without XFEM)
in both Mode I and mixed-mode conditions.
• The application of XLGDM to simulate thermo-mechanical fracture rectified the representation of the
secondary variable (heat flux), which was incorrectly represented in LGDM.
The proposed continuous–discontinuous method (XLGDM) can be further extended to simulate three-dimensional
(3D) cases [7]. This extension is expected to be straightforward and simpler due to the presence of LGDM in the
proposed continuous–discontinuous method. The LGDM, being a continuous method, alleviates several difficulties
29
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 30. A schematic representation of geometry, loads, boundary conditions (mechanical and thermal) and parameters for the solder joint
(dimensions are in mm).

Fig. 31. A comparison of load–CTOD (Crack tip opening distance) response obtained using thermo-mechanical LGDM and XLGDM for
the solder joint.

associated with the 3D discontinuous approaches [6]. Besides this, recent work on the 3D extension of the phase field
method [53] confirms the extensibility of the LGDM, which has a similar implementational procedure. However,
the extension of XLGDM to simulate 3D problems will be taken up in the future work.
30
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Fig. 32. A comparison of damage plots obtained using thermo-mechanically coupled LGDM and XLGDM for the solder joint.

Fig. 33. A comparison of deformed configuration plots obtained using thermo-mechanically coupled LGDM and XLGDM for the solder
joint.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could
have appeared to influence the work reported in this paper.

Acknowledgements

This work is performed as part of the doctoral thesis work of Mr. Subrato Sarkar under the fellowship of the
Ministry of Human Resource Development (MHRD), Government of India. The authors would like to thank Dr.
Manish Kumar for his valuable inputs on implementing XFEM.
31
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

Appendix A. Model parameters, damage law and macro-equivalent strain definitions

• Coupling modulus (h): It is responsible for the coupling effect of micro-processes on) the macro fracture
∂e
(
processes. The effect is considered through the additional coupling term h(e − e) ∂ε in the constitutive
equations (6) and (34). The value adopted for the coupling modulus (h << E) is significantly smaller based
on discussions in Peerlings et al. (3).
• Gradient parameter (c): This parameter defines the size of the interaction region or the fracture process
bandwidth, as depicted in Fig. 1. It is within this region the fracture processes of microfracture take place. Its
value is generally adopted as the characteristic length squared (c = l 2 ). The values of c are generally decided
according to the material, structure and domain size of the problem.
• Interaction function (g): This function is defined as,
(1 − R) e(−n D) + R − e(−n)
g= (A.1)
1 − e(−n)
where R and n are constants whose values are taken as 0.005 and 5 respectively. The interaction function
multiplied with the gradient parameter (c) in Eq. (8) governs the size of the interaction region (Fig. 1).
• Damage law (D): In this work, an exponential damage law is adopted as,
if κ < κi
{
0
D (κ) = κi (A.2)
1 − {1 − a + a exp [−b (κ − κi )]} otherwise
κ
where a and b are constants whose values are decided as per the problem. The threshold value of the history
parameter κi is the value at which damage evolution begins.
• Micro-equivalent strain definitions (e): The von Mises definition of equivalent strain used in this work is
expressed as,
√
k−1 1 (k − 1)2 2 2k
e= I1 + I + J2 (A.3)
2k (1 − ν) 2k (1 − 2ν)2 1 (1 − ν)2
where I1 and J2 are the first and second invariants of the strain and deviatoric strain tensor respectively, ν is
the Poisson’s ratio and k is the ratio of compressive to tensile strength. For all the simulations in this work,
k = 10 is used. Another type of equivalent strain used in this work is the Mazars strain [54] expressed as,
√
( + )2 ( + )2 ( + )2
e= e1 + e2 + e3 (A.4)
where ‘+’ implies the positive part of the principal strains (e1,2,3 ).

Appendix B. XFEM shape functions and derivatives

The bilinear quadrilateral elements are used for the staggered LGDM and XLGDM simulations in this paper.
The shape functions and derivatives used in the formulation of XLGDM are expressed as [55],
Ni (H − Hi )
[ ] [ ]
Ni 0 enr 0
Nu = ; Nu = (B.1)
0 Ni 0 Ni (H − Hi )

∂ Ni ∂ Ni
⎡ ⎤ ⎡ ⎤
⎢ ∂x 0 ⎥ ⎢ ∂ x (H − Hi ) 0 ⎥
∂ Ni ⎥ ∂ Ni
⎢ ⎥ ⎢ ⎥
(H − Hi )⎥
⎢ ⎢ ⎥
⎢ 0
Bu = ⎢ ⎥; Benr =⎢ 0 (B.2)
∂y ⎥ u ⎢ ∂y ⎥
⎣ ∂ Ni ∂ Ni ⎦ ⎣ ∂ Ni ∂ Ni
⎢ ⎥ ⎢ ⎥
(H − Hi ) (H − Hi )
⎦
∂y ∂x ∂y ∂x

Ne/θ = [Ni ] ; e/θ = [Ni (H − Hi )]

Nenr (B.3)
32
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

∂ Ni ∂ Ni
⎡ ⎤ ⎡ ⎤
⎢ ∂x ⎥ ⎢ ∂ x (H − Hi )⎥
Be/θ =⎢
⎣ ∂ Ni ⎦ ;
⎥ Benr
e/θ =⎢
⎣ ∂ Ni
⎥ (B.4)
(H − Hi )
⎦
∂y ∂y
The shape function Ni represents the standard Lagrange bilinear quadrilateral element used in the discretization.
The shifted Heaviside enrichment is used in the formulation. The Heaviside function is expressed as,
+1 x > 0
{
H (x) = (B.5)
−1 x < 0
where x is the normal distance with respect to the discontinuity or crack. In equations (B.1)–(B.4), the value of the
Heaviside function is represented at integration and nodal points by H and Hi respectively.

References
[1] G.T. Camacho, M. Ortiz, Computational modelling of impact damage in brittle materials, Int. J. Solids Struct. 33 (20–22) (1996)
2899–2938.
[2] M. Bansal, I.V. Singh, B.K. Mishra, S.P.A. Bordas, A parallel and efficient multi-split XFEM for 3-D analysis of heterogeneous
materials, Comput. Methods Appl. Mech. Engrg. 347 (2019) 365–401.
[3] Z. Han, T. Yu, H. Phan, T.Q. Bui, Extended stochastic finite element method enhanced by local mesh refinement for random voids
analysis, Comput. Struct. 239 (2020) 106326.
[4] M. Kumar, I.V. Singh, Numerical investigation of creep crack growth in plastically graded materials using C(t) and XFEM, Eng. Fract.
Mech. 226 (2020) 106820.
[5] A. Yadav, R. Patil, S. Singh, R. Godara, G. Bhardwaj, A thermo-mechanical fracture analysis of linear elastic materials using XIGA,
Mech. Adv. Mater. Struct. (2020) https://doi.org/10.1080/15376494.2020.1838006.
[6] S. Bordas, T. Rabczuk, G. Zi, Three-dimensional crack initiation, propagation, branching and junction in non-linear materials by an
extended meshfree method without asymptotic enrichment, Eng. Fract. Mech. 75 (2008) 943–960.
[7] T. Rabczuk, G. Zi, S. Bordas, H. Nguyen-Xuan, A simple and robust three-dimensional cracking-particle method without enrichment,
Comput. Methods Appl. Mech. Engrg. 199 (2010) 2437–2455.
[8] V. Dhanush, S. Natarajan, Implementation of the virtual element method for coupled thermo-elasticity in Abaqus, Numer. Algorithms
80 (3) (2019) 1037–1058.
[9] C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent crack propagation: Robust algorithmic implementation
based on operator splits, Comput. Methods Appl. Mech. Engrg. 199 (45–48) (2010) 2765–2778.
[10] R.H.J. Peerlings, R. de Borst, W.A.M. Brekelmans, J.H.P. De Vree, Gradient enhanced damage for quasi-brittle materials, Internat. J.
Numer. Methods Engrg. 39 (19) (1996) 3391–3403.
[11] V.B. Pandey, S.S. Samant, I.V. Singh, B.K. Mishra, An improved methodology based on continuum damage mechanics and stress
triaxiality to capture the constraint effect during fatigue crack propagation, Int. J. Fatigue 140 (2020) 105823.
[12] C. Comi, S. Mariani, U. Perego, An extended FE strategy for transition from continuum damage to mode I cohesive crack propagation,
Int. J. Numer. Anal. Methods Geomech. 31 (2) (2007) 213–238.
[13] F. Cazes, M. Coret, A. Combescure, A. Gravouil, A thermodynamic method for the construction of a cohesive law from a nonlocal
damage model, Int. J. Solids Struct. 46 (6) (2009) 1476–1490.
[14] F. Cazes, A. Simatos, M. Coret, A. Combescure, A cohesive zone model which is energetically equivalent to a gradient-enhanced
coupled damage-plasticity model, Eur. J. Mech. A Solids 29 (6) (2010) 976–989.
[15] M.R.R. Seabra, P. Šuštarič, J.M.A.C. de Sa, T. Rodič, Damage driven crack initiation and propagation in ductile metals using XFEM,
Comput. Mech. 52 (1) (2013) 161–179.
[16] S.-N. Roth, P. Léger, A. Soulaïmani, A combined XFEM–damage mechanics approach for concrete crack propagation, Comput. Methods
Appl. Mech. Engrg. 283 (2015) 923–955.
[17] Y. Wang, H. Waisman, From diffuse damage to sharp cohesive cracks: A coupled XFEM framework for failure analysis of quasi-brittle
materials, Comput. Methods Appl. Mech. Engrg. 299 (2016) 57–89.
[18] W. Jin, C. Arson, XFEM to couple nonlocal micromechanics damage with discrete mode I cohesive fracture, Comput. Methods Appl.
Mech. Engrg. 357 (2019) 112617.
[19] E. Tamayo-Mas, J. Feliu-Fabà, M. Casado-Antolin, A. Rodríguez-Ferran, A continuous-discontinuous model for crack branching,
Internat. J. Numer. Methods Engrg. 120 (1) (2019) 86–104.
[20] J. Mazars, G. Pijaudier-Cabot, From damage to fracture mechanics and conversely: a combined approach, Int. J. Solids Struct. 33
(20–22) (1996) 3327–3342.
[21] B. Patzák, M. Jirásek, Process zone resolution by extended finite elements, Eng. Fract. Mech. 70 (7–8) (2003) 957–977.
[22] A. Simone, G.N. Wells, L.J. Sluys, From continuous to discontinuous failure in a gradient-enhanced continuum damage model, Comput.
Methods Appl. Mech. Engrg. 192 (41–42) (2003) 4581–4607.
[23] R.J.M. Geelen, Y. Liu, J.E. Dolbow, A. Rodríguez-Ferran, An optimization-based phase-field method for continuous-discontinuous crack
propagation, Internat. J. Numer. Methods Engrg. 116 (1) (2018) 1–20.
33
S. Sarkar, I.V. Singh and B.K. Mishra Computer Methods in Applied Mechanics and Engineering 383 (2021) 113893

[24] L.H. Poh, G. Sun, Localizing gradient damage model with decreasing interactions, Internat. J. Numer. Methods Engrg. 110 (6) (2017)
503–522.
[25] A. Negi, S. Kumar, L.H. Poh, A localizing gradient damage enhancement with micromorphic stress-based anisotropic nonlocal
interactions, Internat. J. Numer. Methods Engrg. 121 (18) (2020) 4003–4027.
[26] S. Sarkar, I.V. Singh, B.K. Mishra, A.S. Shedbale, L.H. Poh, Source codes and simulation data for the finite element implementation
of the conventional and localizing gradient damage methods in ABAQUS, Data Brief 26 (2019b) 104533.
[27] K. Otsuka, H. Date, Fracture process zone in concrete tension specimen, Eng. Fract. Mech. 65 (2–3) (2000) 111–131.
[28] Y. Shen, T.F. Morgeneyer, J. Garnier, L. Allais, L. Helfen, J. Crepin, Three-dimensional quantitative in situ study of crack initiation
and propagation in AA6061 aluminum alloy sheets via synchrotron laminography and finite-element simulations, Acta Mater. 61 (2013)
(2013) 2751–2782.
[29] S. Forest, Micromorphic approach for gradient elasticity, viscoplasticity, and damage, J. Eng. Mech. 135 (3) (2009) 117–131.
[30] 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. Methods Appl. Mech. Engrg. 193 (30–32) (2004) 3403–3417.
[31] S. Sarkar, I.V. Singh, B.K. Mishra, A thermo-mechanical gradient enhanced damage method for fracture, Comput. Mech. 66 (6) (2020b)
1399–1426.
[32] S. Sarkar, I.V. Singh, B.K. Mishra, Adaptive mesh refinement schemes for the localizing gradient damage method based on
biquadratic-bilinear coupled-field elements, Eng. Fract. Mech. 223 (2020a) 106790.
[33] H. Talebi, M. Silani, S.P.A. Bordas, P. Kerfriden, T. Rabczuk, A computational library for multiscale modeling of material failure,
Comput. Mech. 53 (2014) 1047–1071.
[34] P. Areias, M.A. Msekh, T. Rabczuk, Damage and fracture algorithm using the screened Poisson equation and local remeshing, Eng.
Fract. Mech. 158 (2016) 116–143.
[35] P. Areias, M.A. Msekh, T. Rabczuk, Effective 2D and 3D crack propagation with local mesh refinement and the screened Poisson
equation, Eng. Fract. Mech. 189 (2018) 339–360.
[36] A. Jenabidehkordi, R. Abadi, T. Rabczuk, Computational modeling of meso-scale fracture in polymer matrix composites employing
peridynamics, Compos. Struct. 253 (2020) 112740.
[37] T. Rabczuk, H. Ren, A peridynamics formulation for quasi-static fracture and contact in rock, Eng. Geol. 225 (2017) 42–48.
[38] T. Rabczuk, S. Bordas, G. Zi, A three-dimensional meshfree method for continuous multiple-crack initiation, propagation and junction
in statics and dynamics, Comput. Mech. 40 (2007) 473–495.
[39] T. Rabczuk, T. Belytschko, Cracking particles: a simplified meshfree method for arbitrary evolving cracks, Internat. J. Numer. Methods
Engrg. 61 (2004) 2316–2343.
[40] T. Rabczuk, G. Zi, A. Gerstenberger, W.A. Wall, A new crack tip element for the phantom-node method with arbitrary cohesive cracks,
Internat. J. Numer. Methods Engrg. 75 (2008) 577–599.
[41] T.P. Fries, T. Belytschko, The extended/generalized finite element method: an overview of the method and its applications, Internat. J.
Numer. Methods Engrg. 84 (3) (2010) 253–304.
[42] A.R. Khoei, Extended Finite Element Method: Theory and Applications, John Wiley & Sons, 2014.
[43] X. Zhuang, R. Huang, C. Liang, T. Rabczuk, A coupled thermo-hydro-mechanical model of jointed hard rock for compressed air
energy storage, Math. Probl. Eng. 2014 (2017) 179169.
[44] M. Duflot, The extended finite element method in thermoelastic fracture mechanics, Internat. J. Numer. Methods Engrg. 74 (5) (2008)
827–847.
[45] S. Sarkar, I.V. Singh, B.K. Mishra, A.S. Shedbale, L.H. Poh, A comparative study and ABAQUS implementation of conventional and
localizing gradient enhanced damage models, Finite Elem. Anal. Des. 160 (2019a) 1–31.
[46] J. Mediavilla, R.H.J. Peerlings, M.G.D. Geers, An integrated continuous–discontinuous approach towards damage engineering in sheet
metal forming processes, Eng. Fract. Mech. 73 (7) (2006) 895–916.
[47] B. Vandoren, A. Simone, Modeling and simulation of quasi-brittle failure with continuous anisotropic stress-based gradient-enhanced
damage models, Comput. Methods Appl. Mech. Engrg. 332 (2018) 644–685.
[48] T.H.A. 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. Engrg. 328 (2018) 498–541.
[49] G. Meschke, P. Dumstorff, Energy-based modeling of cohesive and cohesionless cracks via X-FEM, Comput. Methods Appl. Mech.
Engrg. 196 (21–24) (2007) 2338–2357.
[50] B.J. Winkler, Traglastuntersuchungen Von Unbewehrten Und Bewehrten Betonstukturen Auf Der Grundlage Eines Objektiven
Werkstoffgesetzes Fur Beton, Innsbruck University Press, 2001.
[51] R.U. Patil, B.K. Mishra, I.V. Singh, A local moving extended phase feld method (LMXPFM) for failure analysis of brittle materials,
Comput. Methods Appl. Mech. Engrg. 342 (2018) 674–709.
[52] C. Hillman, K. Rogers, A. Dasgupta, M. Pecht, R. Dusek, B. Lorence, Solder failure mechanisms in single-sided insertion-mount
printed wiring boards, Circuit World 25 (3) (1999) 28–38.
[53] J.Y. Wu, Y. Huang, H. Zhou, V.P. Nguyen, Three-dimensional phase-feld modeling of mode I + II/III failure in solids, Comput. Methods
Appl. Mech. Engrg. 373 (2021) 113537.
[54] J. Mazars, G. Pijaudier-Cabot, Continuum damage theory—application to concrete, J. Eng. Mech. 115 (2) (1989) 345–365.
[55] N. Duhan, R.U. Patil, B.K. Mishra, I.V. Singh, Y.E. Pak, Thermo-elastic analysis of edge dislocation using extended finite element
method, Int. J. Mech. Sci. 192 (2020) 106109.

34