Computers and Geotechnics 165 (2024) 105921

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

A finite-volume implementation of the phase-field model for brittle fracture

with adaptive mesh refinement
X.L. Yang a, N. Guo a, b, *, Z.X. Yang a, b
a
Computing Center for Geotechnical Engineering, Engineering Research Center of Urban Underground Space Development of Zhejiang Province, Department of Civil
Engineering, Zhejiang University, Hangzhou 310058, China
b
Zhejiang Provincial Engineering Research Center for Digital & Smart Maintenance of Highway, China

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

Keywords: The phase-field method (PFM) has advantages in modeling crack propagation in rock-like materials by treating
Phase field the crack surface as a continuous function, therefore avoiding dealing with the sharp discontinuous displacement
Finite volume method field. This study presents an implementation of PFM using the finite volume method (FVM) to discretize the
Adaptive mesh refinement
governing equations for the conservation of linear momentum and the evolution of phase field. The coupled
Brittle fracture
Iterative staggered scheme
equations are solved by an iterative staggered scheme. The adaptive mesh refinement (AMR) technique is further
employed to improve computational efficiency, which is relatively easy to achieve in FVM as the method can
naturally handle unstructured meshes with hanging nodes in both two- and three-dimensional problems, as
opposed to the finite element method (FEM). Several classical crack propagation problems, including the single-
edge notched tension and shear tests, the L-shaped panel test, and the test on a notched plate with a hole, are
simulated and compared with available experimental and FEM results, which demonstrate that the FVM-based
PFM can model crack propagation accurately and effectively and could be more efficient than the FEM-based
PFM.

1. Introduction (Xu and Needleman, 1994). Since crack propagation inside the elements
is not permitted, this method has strong mesh dependence. On the other
Modeling crack propagation in rock-like materials has attracted hand, the extended finite element method (XFEM) (Belytschko and
widespread interest but remains challenging due to the material het­ Black, 1999) has proven to be one of the most well-received discrete
erogeneity involving inherent defects at different scales and directions, methods so far, which introduces an enriched basis function to describe
as well as the complexity of the problem itself. Being an essentially the displacement discontinuity thus allowing the crack to propagate
discontinuous process, the crack propagation problem poses tremendous within an element and can track the crack path without remeshing.
difficulties for the traditional continuum-based numerical methods, Nevertheless, the tracking of crack surface is a formidable task, espe­
especially the mesh-based methods such as the finite element method cially in 3D problems.
(FEM) that is still dominant in computational solid mechanics. In contrast to the discrete methods that explicitly simulate the strong
Presently, two conceptually different methods are mainly used in discontinuity in displacements across cracks, the continuous methods
FEM to tackle crack propagation, corresponding to the discontinuous model the crack as a diffuse zone, so that the displacement fields remain
(discrete) and the continuous (smeared) categories, respectively. The continuous, and the tracking of crack surface is no longer required. An
former describes the crack through the discontinuous displacement example of such methods is the continuum damage mechanics (CDM)
field, for example, by simply deleting the elements that satisfy the crack model (Bažant and Cedolin, 1983), which expresses the progressive
condition (Belytschko and Lin, 1987). This element deletion method damage of a material through the evolution of some internal variables
(EDM) however cannot simulate the branch of crack correctly (Liu et al., such as the damage variable. Yet the evolution law of the damage var­
2014). Another option is to insert cohesive interface elements at the iable is often phenomenological and simply assumed to be a function of
element interfaces, so the crack can propagate through the interfaces the equivalent plastic strain. Therefore, it is difficult for CDM to

* Corresponding author at: Computing Center for Geotechnical Engineering, Engineering Research Center of Urban Underground Space Development of Zhejiang
Province, Department of Civil Engineering, Zhejiang University, Hangzhou 310058, China.
E-mail address: [email protected] (N. Guo).

https://doi.org/10.1016/j.compgeo.2023.105921
Received 1 October 2023; Received in revised form 4 November 2023; Accepted 7 November 2023
Available online 10 November 2023
0266-352X/© 2023 Elsevier Ltd. All rights reserved.
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

accurately predict the crack initiation and propagation path. More

recently, the phase field method (PFM) (Aranson et al., 2000) has
emerged as a competitive alternative for modeling crack propagation
and can deal with single and multiple cracks with complex topologies in
both 2D and 3D conditions.
The PFM bears sound physical background rooted in the variational
approach proposed by Francfort and Marigo (1998) based on the Griffith
theory of fracture (Griffith, 1921). In the regularized variational
formulation (Bourdin et al., 2000; Tanné et al., 2018), the sharp crack
surface Г is regularized as a band with a certain length scale and
described by an order parameter, i.e., the phase-field variable. Previous
studies (e.g., Miehe et al., 2010a) have shown that when the length scale
tends to 0, the diffuse crack converges to a sharp crack. In the past
decade, many versions of PFM have been developed. For instance, Amor
et al. (2009) and Miehe et al. (2010a) successively proposed the
spherical-deviatoric decomposition of strain energy and the spectral
decomposition of strain energy to distinguish the effects of tensile and
compressive stresses. Borden et al. (2014) proposed a fourth-order
phase-field model that leads to higher regularity in the exact phase-
field solution and a faster convergence rate. Ambati et al. (2015) pro­
posed a hybrid isotropic-anisotropic phase-field formulation achieving a Fig. 1. Schematic diagram of diffuse cracks.
significant reduction of computational cost. Furthermore, many phase-
field models tailored for specific materials have been developed to refinement index. In this regard, the FVM is more flexible than the FEM
investigate crack propagation in various scenarios. Zhang et al. (2017) in terms of mesh generation and adaptive remeshing. Furthermore, the
proposed a modified phase-field model using different critical energy FVM is naturally conservative and particularly suitable for solving
release rates for different fracture modes and applied it to the study of continuity or transport equations, therefore popular in computational
mixed-mode crack in rock-like materials. In addition, Fei and Choo fluid dynamics. As a result, the proposed FVM-based phase-field model
(2019) creatively introduced a stress decomposition method based on could be potentially conveniently coupled with fluid flow to solve
the interface stress to model rock fracture propagation considering multiphysics problems such as hydraulic fracturing.
frictional contact. They have applied this model in geotechnical engi­ The rest of the paper is organized as follows. Section 2 provides an
neering problems such as the slope stability analysis (Fei and Choo, overview of the phase field formulation for brittle fracture with the two
2020). Li et al. (2023b) introduced an adaptive phase-field method strain energy decomposition methods proposed by Amor et al. (2009)
based on the isogeometric-meshfree approach to capture the dynamic and Miehe et al. (2010a), respectively. Section 3 presents the FVM dis­
crack propagation in composite materials under impact loading. Kiran cretization of the governing equations for the balance of linear mo­
et al. (2022, 2023) developed an adaptive phase-field model for the mentum and phase field evolution. Next, seven numerical examples
study of electromechanical fracture of piezoelectric ceramics and the including the single-edge notched tension and shear tests, the L-shaped
interfacial fracture in transversely isotropic piezoelectric materials panel test, and the test on a notched plate with a hole, are given in
under various electromechanical loading conditions. For the solution of Section 4 to demonstrate the correctness and efficiency of the proposed
the coupled displacement and phase field governing equations, the model. Section 5 concludes the study.
staggered scheme (Seleš et al., 2019) is usually adopted due to its proved
robustness. 2. Phase field formulation
Most of the existing phase-field models have been implemented in
FEM, whereas a recent study by Sargado et al. (2021) has demonstrated In this section, we briefly outline a typical phase field formulation for
a notable gain in computational efficiency by using a combined FEM- brittle fracture, which is mainly based on the variational formulation in
finite volume method (FVM), where the stress equilibrium equation is Miehe et al. (2010a) and derived from the classical fracture theory
discretized by standard FEM and the phase-field evolution equation by (Griffith, 1921) and the minimum energy variational method (Francfort
FVM. Their study has directly motivated the present study to implement and Marigo, 1998). The former assumes that the crack propagates only
the phase-field model entirely in FVM considering that FVM is equally when the energy release rate reaches a critical value, whereas the latter
applicable to solve linear and nonlinear solid mechanics problems assumes that the crack propagation process should minimize the free
(Demirdžić et al., 1998; Demirdžić and Martinović, 1993; Demirdžić and energy and follow irreversibility conditions. In the PFM, the crack is
Muzaferija, 1994; Demirdžić and Muzaferija, 1995; Demirdžić et al., described by a diffuse phase boundary, which is represented by a
1997). Cardiff et al. (2018) developed an open-source FVM toolbox continuous function of the order parameter d. The value of d is in the
called solids4foam for solid mechanics and fluid–structure interaction range [0, 1], with d = 0 for the intact region and d = 1 for the fully
problems based on OpenFOAM (Weller et al., 1998), which will be used damaged region.
in the present study. The unified FVM implementation of both stress Consider a brittle linear elastic solid Ω⊂Rn (n denotes the number of
equilibrium and phase field evolution equations will lead to several dimensions) with an embedded fracture Γ as shown in Fig. 1, whose
other benefits. For example, the cell-centered FVM used in the study outer boundary is ∂Ω⊂Rn− 1 and subjected to either the Dirichlet
finds solutions from conservation equations on local cells, thus allowing boundary condition ∂Ωu or the Neumann boundary condition ∂Ωt ,
for jumps in solutions at the cell interfaces. Consequently, it can satisfying ∂Ωu ∪ ∂Ωt = ∂Ω and ∂Ωu ∩ ∂Ωt = ∅ (applies to each compo­
conveniently handle general structured or unstructured polyhedral nent for the vector fields). The crack fracturing processes can be deter­
meshes with hanging nodes, which means that the method is friendly to mined using the following potential energy functional:
adaptive mesh refinement (AMR) (Meredith and Vukčević, 2018), a ∫ ∫
desired feature for phase field modeling, especially when the crack path Π(u, Γ) = Ψ(ε(u) )dΩ + Gc dΓ (1)
is not known a priori, thus predetermined local mesh refinement (Wu
Ω Γ

et al., 2021) is infeasible. In the literature, the phase field (Borden et al., where Ψ is the elastic strain energy density, ε(u)= ((∇u)T + ∇u)/2 is
2012) or its gradient (Li et al., 2023a) is commonly used as the

2
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

the small strain tensor, Gc is the critical energy release rate of the ma­ calculated as:
terial, and u is the displacement. The strain energy of an isotropic linear
[ ] ∂Ψ(ε)
elastic material can be defined by the following formula: σ = (1 − d)2 + k (11)
∂ε
λ
Ψ(ε) = tr(ε)2 + με : ε (2) Existing studies have shown that the nucleation and propagation of
2 cracks are often caused by tension, whereas compression will lead to the
where λ and μ are the Lamé constants. With the phase field variable d, closure of cracks and hinder their propagation. Hence, some ten­
the surface density of cracks per unit volume γ can be expressed as sion–compression decomposition schemes are developed. A comparison
(Miehe et al., 2010b): of different decomposition methods can be found in Ambati et al.
(2015). In general, they can be classified as isotropic and anisotropic
d2 l0
γ(d) = + |∇d|2 (3) decompositions. The isotropic ones (Bourdin et al., 2000; Yang, 2006)
2l0 2
are linear and therefore less computationally expensive, but they allow
where ∇d is the spatial gradient of the phase field, and l0 ∈ R+ is the cracks to grow in compression and may lead to nonphysical crack pat­
characteristic length that controls the width of the diffuse crack transi­ terns. On the contrary, the anisotropic formulations decompose the
tion region, as shown in Fig. 1. Therefore, the regularized crack surface strain energy density into positive and negative parts, and the stress
Γl0 (d) can be represented by a scalar phase field via the following equation is reformulated as:
integral:
[ ] ∂Ψ+ (ε) ∂Ψ− (ε)
∫ σ = (1 − d)2 + k + (12)
Γl 0 (d) = γ(d)dΩ (4)
∂ε ∂ε
Ω
The commonly used decomposition methods are the spherical-
The total crack surface energy of the elastic solid can be obtained deviatoric decomposition proposed by Amor et al. (2009) and the
from Eq. (4): spectral decomposition proposed by Miehe et al. (2010a). For the
∫ ∫ ( 2 ) spherical-deviatoric decomposition, the strain energy can be expressed
d l0 as:
Gc dΓ ≈ Gc + |∇d|2 dΩ (5)
Γ Ω 2l0 2 ( )
1 2μ
Therefore, the expansion of the total potential energy can be ob­ Ψ+ = λ+ 〈tr(ε)〉2+ + μ(εdev : εdev ) (13a)
2 n
tained as
( 2 ) ( )
∫ ∫ 1 2μ
Π(u, d, ∇d) = g(d)Ψ(ε(u) )dΩ + Gc
d l0
+ |∇d|2 dΩ (6) Ψ− = λ+ 〈tr(ε)〉2− (13b)
2l0 2 2 n
Ω Ω

where g(d) is the stress degradation function that reduces the stored where εdev is the deviatoric part of the strain tensor. The Macaulay
energy within the solid as the phase field d evolves. Here, it is specified brackets are defined as: 〈x〉± = (x ± |x|)/2. For the spectral decompo­
[ ] sition method, the strain tensor is decomposed into positive and nega­
as g(d) = (1 − d)2 +k , where k is a model parameter to avoid nu­ tive parts:
merical singularity when d approaches unity and taken as k = 10− 6
ε = ε+ + ε− (14)
throughout the study. Its influence is proved negligible according to the
existing studies, e.g., Miehe et al. (2010a). ∑n
The external force work Wext is expressed as (body force neglected in
ε+ = p=1
〈εp 〉+ np ⊗ np (15a)
the study): ∑n
∫ ε− = p=1
〈εp 〉− np ⊗ np (15b)
Wext = t⋅udA (7) { } { }
∂Ωt where εp p=1⋯n
and np p=1⋯n are the principal strains and their
where t is the surface traction. According to Eqs. (6) and Eq. (7), the directions, respectively. Therefore, the positive and negative parts of
governing equations of the phase field model can be obtained by mini­ elastic energy can be expressed as:
mizing the total energy functional, i.e., letting the variation of Eq. (8) be λ [ ]
zero. Ψ+ = 〈tr(ε)〉2+ + μtr ε2+ (16a)
2
∫ ∫ ( 2 ) ∫
d l0
Π − Wext = g(d)Ψ(ε(u) )dΩ + Gc + |∇d|2 dΩ − t⋅u dA λ [ ]
Ω Ω 2l0 2 ∂Ωt Ψ− = 〈tr(ε)〉2− + μtr ε2− (16b)
2
(8)
The elastic energy can be expressed by the following formula:
As a result, the following coupled partial differential equations are
reached: Ψ = g(d)Ψ+ + Ψ− (17)

∇⋅σ = 0 in Ω (9a) Substituting Eq. (17) into Eq. (8), and taking the variation with
respect to u and d, the governing equations can be obtained:
[ ]
Gc (18a)
− Gc l0 ∇2 d + + 2Ψ(ε) d = 2Ψ(ε) in Ω (9b) ∇⋅σ = 0 in Ω
l0
[ ]
with the following boundary conditions: − Gc l0 ∇2 d +
Gc
+ 2Ψ+ d = 2Ψ+ in Ω (18b)
l0
σ ⋅n = t on ∂Ωt (10a)
where the stress σ is given by Eq. (12).
u = u on ∂Ωu (10b) It is necessary to supplement the irreversibility condition to prevent
the crack healing of the elastic solid under compression or unloading.
∇d⋅n = 0 on ∂Ω (10c) The commonly used method is to introduce a history state variable H,
which is updated according to:
where t and u are the prescribed boundary traction and displace­
ment, respectively. n is the outward normal vector. The stress is

3
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

(1996).
By integrating Eq. (18a) over the volume of P, ΩP , and applying the
divergence theorem, it gives:
∫ ∮
∇ • σ dΩ = n⋅σ dA = 0 (21)
ΩP ∂ΩP

Given the stress term on the face σ f , the surface integral in Eq. (21)
can be approximated by:
∮ ∑
n⋅σ dA ≈ Γf ⋅σ f (22)
∂ΩP f

where the summation is over the faces of cell P.

Taking the spectral decomposition method for example, σ f can be
obtained from Eqs. (12) and (16) as:
{ ( ) } ( )
σ f = g(d) λ〈tr εf 〉+ I + 2μ(εf )+ + λ〈tr εf 〉− I + 2μ(εf )− (23)

where the subscript f indicates the quantity evaluated at the face

center, and I is the identity tensor. In the cell-centered FVM, the
displacement is assumed to vary linearly within a cell:
u(x) = uP + (x − xP )⋅(∇u)P (24)
where uP and (∇u)P are the center displacement and constant
Fig. 2. Polyhedral control volume in FVM (after Cardiff et al., 2014). displacement gradient of the cell, respectively. The constant displace­
( ) ment gradient (or strain) will result in a discontinuity at the face of
Hn+1 = max Hn , Ψ+
n+1 (19) neighboring cells. Hence, an interpolation weight method is used to
evaluate the quantity at the face shared by cells P and N:
It represents the maximum tensile elastic strain energy from the start
( )
of loading to the current time tn+1 . Replacing the elastic energy Ψ+ in Eq. (∇u)f ≈ γf (∇u)P + 1 − γf (∇u)N (25)
(18b) with H leads to a non-decreasing d. Then, Eq. (18b) is rewritten as:
[ ] where the weight γ f is in the range (0, 1) and usually obtained by the
− Gc l0 ∇2 d +
Gc
+ 2H d = 2H in Ω (20) following formula:
l0 ⃒ ⃒
⃒xN − xf ⃒
In summary, Eqs. (18a), (19) and (20) complete the phase field γf = (26)
|xN − xP |
model. It is worth noting that the two different decomposition methods
use either Eq. (13a) or (16a) to calculate Ψ+ . The spectral decomposition where xP, xf, and xN are the coordinates of P, f, and N, respectively, as
is mainly used in this study unless otherwise stated. shown in Fig. 2.
The displacement gradient at the cell center is usually obtained using
3. Finite volume implementation the least squares method due to its ease of implementation on unstruc­
tured grids:
We use the residual control staggered solution scheme (Seleš et al., [ ]− 1
∑ ∑[ ]
2019) to alternately solve the displacement and phase field coupling (∇u)P ≈ 2
wf df df w2f df (uN − uP ) (27)
equations by the implicit cell-centered FVM (Cardiff and Demirdžić, f f

2021). FVM converts a given partial differential equation representing

where the weight wf can be taken as either 1 (Demirdžić and
the conservation law into a system of discrete algebraic equations on a ⃒ ⃒− 1
finite control volume (CV). Like in other mesh-based methods, the Muzaferija, 1995) or ⃒df ⃒ (Jasak and Weller, 2000). To reach the im­
geometric domain is first discretized into non-overlapping cells or finite plicit formulation, the separation solution algorithm is adopted (Jasak
volumes. Next, the transformed algebraic equation system is solved to and Weller, 2000):
find the field variables of each cell. The discretization of the solution
Kf ∇2 u− Kf ∇2 u + ∇⋅σ = 0
domain and equations in a standard cell-centered FVM and the solution ⏟̅̅̅⏞⏞̅̅̅⏟⏟̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅⏟ (28)
procedure are introduced below.
implicit explicit

where the coefficient Kf controls the smoothing effect and is typically

3.1. Discretization taken as Kf = λ +2μ (Cardiff et al., 2017). The implicit term in Eq. (28)
will be used to assemble the matrix (see [Au] in Eq. (36a)) of the
The computational space is divided into a finite number of convex resulting algebraic linear system, and the explicit terms contribute to the
polyhedral CVs surrounded by convex polygon faces. The discretization source vector (see [bu] in Eq. (36a)) of the linear equation system using
of variables in space is generally linear, so the FVM discretization is known values from the last iterative step.
usually second-order accurate (Cardiff et al., 2018). A typical CV is The discretization however may suffer from the so-called checker-
shown in Fig. 2, where the point P is located at the center of the CV with boarding errors (Cardiff and Demirdžić, 2021), with displacement os­
the position vector r. N is the center of the neighboring ⃒CV⃒ that shares cillations close to the spurious singular modes that appear in the
the face f with P. Γf is the area normal vector of f, i.e., ⃒Γf ⃒ is the face reduced-integration FEM. A diffusion term based on the Rhie-Chow
area, and df is the vector connecting P and N. correction (Rhie and Chow, 1983) is usually added in the implicit cell-
centered scheme for stabilization, shown as the second term in the
The mesh is considered non-orthogonal if Γf and df are not parallel
left-hand side of Eq. (29). The updated discretized linear momentum
and skewed if df does not intersect f at its center. The non-orthogonal or
equation becomes:
skewed mesh will deteriorate the accuracy of FVM. Remedies are often
required to optimize the discretization, which are detailed in Jasak

4
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

Fig. 3. Adaptive mesh refinement for (a) triangle and (b) quadrilateral cells in 2D, and (c) polyhedral cells in 3D (after Meredith and Vukčević, 2018).

[ ]
∑ ∑ ⃒ ⃒ uN − uP • For the traction boundary, specify forces on the boundary surfaces:
Γf ⋅σ f + (λ + 2μ) ⃒Δf ⃒ ⃒⃒ ⃒⃒ − Δf ⋅(∇u)f = 0
f f df (29) gb = |Γb |t (35)
⏟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏟
Rhie-Chow correction term
where Γb is the outward-pointing area vector of the boundary face, t
where is the specified traction. gb will be added to Eq. (22) in calculating the
force balance of the control volume.
df ⃒⃒ ⃒⃒2
Δf = Γf (30) After the above discretization, the discrete forms of Eqs. (18a) and
df • Γf (20) on each control volume can be cast into four (for 3D) linear alge­
For the phase field equation, Eq. (20) is modified to: braic equations (one vector equation for the displacement and one scalar
( ) equation for the phase field). The equation system for the whole domain
− l20 ∇2 d +
2Hl0
+1 d =
2Hl0 can be obtained by assembling all the CVs, which presents in the
Gc Gc (31) following general form:
⏟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏟
implicit
[Au ][u] = [bu ] (36a)
By integrating Eq. (31) over ΩP and applying the divergence theo­
[ d] [ ]
rem, it gives: A [d] = bd (36b)
∫ ( ) ∫ ∫ []
( )
2Hl0
+ 1 d dΩ + − l20 ∇d • n dA =
2Hl0
dΩ (32) where [Au ] and Ad are both assembled sparse matrices with diag­
ΩP Gc ∂ΩP ΩP G c onal terms auP and adP and non-zero neighbor off-diagonal terms auN and
[ ]
The discrete equation can be obtained as: adN . [bu ] and bd are the source vectors with terms buP and bdP which cover
( )
2Hl0 ∑ 2Hl0 the contribution from the explicit terms and boundary conditions. The
+ 1 dP ΩP − l20 (∇d)f • Γf = ΩP (33) expressions of these terms are shown below:
Gc f
Gc
⃒ ⃒
⃒Δf ⃒
where (∇d)f • Γf can be approximated by: u
aN = − (2μ + λ) ⃒⃒ ⃒⃒ (37a)
df
⃒ ⃒ dN − dP
(∇d)f • Γf = ⃒Γf ⃒ ⃒⃒ ⃒⃒ (34) ∑
df auP = − auN (37b)
f

⃒ ⃒
3.2. Boundary conditions and system of equations ⃒Γf ⃒
adN = − l20 ⃒⃒ ⃒⃒ (37c)
df
The boundary conditions in FVM can be classified into three types
according to Jasak and Weller (2000): (
2Hl0
) ∑
adP = + 1 ΩP − adN (37d)
Gc
• For the displacement boundary, directly specify the displacement u on
f

the boundary face. The surface gradient can be calculated from the ∑ ∑
buP = Γf ⋅σ f − (2μ + λ)Δf ⋅(∇u)f (37d)
center values of adjacent cells, e.g., Eq. (25).
f f
• For the symmetric boundary, image the control volume on the other
side of the boundary as an adjacent control volume.

5
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

Fig. 4. Iterative staggered scheme of the FVM-based phase field model with adaptive mesh refinement.

2Hl0 fine meshes to capture the abrupt gradient changes, often leading to a
bdP = ΩP (37f)
Gc surge in computational cost especially when the crack path is unknown a
priori. Therefore, the adaptive mesh refinement technique (Gupta et al.,
Eq. (36) can then be assembled from each CV using the above
2022; Krishnan et al., 2022; Xu et al., 2022; Dinachandra and Alankar,
equations as:
2022; Patil et al., 2018; Assaf et al., 2022) is preferred. Compared with
∑
auP up + auN uN = buP FEM, FVM can naturally deal with polyhedral or polygonal cells with
(38a)
f hanging nodes, which means that it is more conducive to applying
∑ adaptive mesh refinement in FVM. The isotropic refinement algorithm
adP dP + adN dN = bdP (38b) proposed by Meredith and Vukčević (2018) for OpenFOAM has been
f used in this study. Generally, it first identifies the candidate cells that
It is worth noting that the model implementation is based on solid­ meet a certain refinement criterion (d ≥ 0.4 in the study) and then splits
s4foam (Cardiff et al., 2018), an OpenFOAM toolbox for solid me­ the cells into a set of sub-cells by inserting points at the centers of the
chanics. The preconditioned conjugate gradient solver is used to solve cells, the faces, and the edges. The process of point insertion will be
Eq. (36) in the study with default settings. repeated according to the maximum refinement level, and the sub-cells
will be marked so that they will not be further refined in later loading
steps. The refinement in 2D with initial triangle and quadrilateral cells
3.3. Adaptive mesh refinement and solution flow after local refinement are shown in Fig. 3(a) and (b), respectively.
Similarly, the refinement in 3D for different polyhedrons is illustrated in
When a discrete crack is modeled as a diffuse phase field, abrupt Fig. 3(c). Note that a posteriori error estimator (Walloth and Wollner,
gradient changes occur in the phase field. It is then necessary to use very

6
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

4. Numerical examples

To verify the correctness and effectiveness of the proposed algo­

rithm, we present seven classical crack tests and compare the results
with those available in the literature (mainly FEM simulations along
with some experimental data). Particularly, the improvement of effi­
ciency brought by the adaptive mesh refinement method is discussed. All
2D simulations presented in this section are performed in serial on an
AMD® Ryzen® CPU R7-5800X @ 3.80 GHz with 32 GB RAM for fair
comparisons with the Abaqus simulations. The parallel computing
capability readily available in OpenFOAM (Cardiff et al., 2018) is only
enforced in the 3D tests.

4.1. Single-edge notched tension test

The geometry and boundary conditions of the single-edge notched

plate are shown in Fig. 5, where the initial crack is located at the left-
center of the plate with a length of 0.5 mm and a width of 0.001 mm.
The Young’s modulus of the material is E = 210 GPa, the Poisson’s ratio
ν = 0.3, and the critical energy release rate Gc = 2.7 N/mm. The phase
field characteristic length is l0 = 0.015 mm. These parameters are
consistent with those used in Miehe et al. (2010a). The vertical
displacement increment applied to the top surface is Δu = 10− 5 mm for
the whole loading steps. Note that in this example, the AMR is not
Fig. 5. Geometry and boundary conditions for the single-edge notched tension activated. A fixed mesh with local refinement is used instead as shown in
and shear tests. Fig. 6 which contains 3864 quadrilateral cells with a minimum cell size
of h = 0.00625 mm. The initial crack is modeled as a gap in the mesh
domain (see Fig. 6) using the mesh-induced method (Krishnan et al.,
2022).
The phase field contour during loading is shown in Fig. 7, which
clearly shows a brittle fracture pattern: The crack starts to propagate
from the tip at around u = 5.54 × 10− 3 mm and quickly breaks the plate
at u = 5.92 × 10− 3 mm, forming a straight horizontal crack path. The
reaction force–displacement curves are further shown in Fig. 8 with
parametric studies on the influence of minimum mesh size h. In general,
the present results agree well with the FEM results of Miehe et al.
(2010a), both showing an initially elastic response up to a peak force of
about 700 N followed by a drastic collapse.
For the mesh sensitivity analysis, we consider two extra cases with h
= 0.003125 and 0.0015625 mm. It is seen that the peak force decreases
with decreasing h, consistent with the observation in Molnr and Grav­
ouil (2017), and exhibits a convergence trend. Additionally, it is worth
mentioning that although the result with h = 0.00625 mm matches that
of Miehe et al. (2010a) best, the latter employed a single-pass staggered
scheme without convergence control. The use of smaller mesh sizes (i.e.,
h = 0.003125 and 0.0015625 mm) can result in a more accurate post-
peak behavior, characterized by a straight drop in reaction force. Also
implied in Fig. 8 is that the advantage of residual-controlled iterative
staggered scheme is only effective when a certain minimum cell size is
used.
To further compare the accuracy and efficiency of the proposed
phase field model with existing FEM implementations, we use the code
developed by Navidtehrani et al. (2021) using UMAT and HETVAL
Fig. 6. Fixed mesh used in the single-edge notched tension test (l0/h = 2.4). subroutines in Abaqus and simulate the same single-edge notched ten­
sion test. The influence of the ratio l0/h on the peak reaction force is also
2022) may give more objective adaptive solution, which however is not studied. The following simulations adopt the same grid and residual-
considered in the present study but will be explored in the future. controlled staggered scheme for both FVM and FEM. Three cases of l0/
The calculation flow of the FVM-based phase filed model with h = 2.1, 4.0, and 8.2, respectively, are analyzed. Figs. 9 and 10 present
adaptive mesh refinement using the iterative staggered scheme is shown the fixed mesh and force–displacement curves, respectively, for the l0/h
in Fig. 4 (for one loading step from ti to ti+1), where the residual toler­ = 8.2 case. The peak reaction forces and the time spent in the simula­
ance is taken as Er = 10− 7, and the maximum number of iterations is set tions are summarized and compared in Table 1.
to nmax = 1000. Fig. 10 compares the force–displacement curves for the case l0/h =
8.2 simulated by Abaqus and the present FVM model, together with that
by Miehe et al. (2010a). It is observed that the present FVM result shows
much better agreement with that of Miehe et al. (2010a), especially in
the pre-peak stage and peak force. On the contrary, the Abaqus result

7
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

3 3
Fig. 7. Crack pattern of the single-edge notched tension test at displacements (l0/h = 2.4): (a) u = 5.54 × 10− mm, (b) u = 5.67 × 10− mm, and (c) u = 5.92 ×
10− 3 mm.

Fig. 8. Force-displacement curves of the single-edge notched tension test.

predicts a smaller elastic stiffness but a larger peak. The variations in

post-peak behaviors are primarily attributed to the coarse discretization
especially along the crack propagation path, as previously discussed. A
Fig. 9. Fixed mesh used in the single-edge notched tension test (l0/h = 8.2).
similar observation was made by Mandal et al. (2019).
In addition, it can also be seen from Table 1 that for the same grid
and staggered scheme, the FEM implementation is costlier than the parameters follow the previous test except that a different Poisson’s
present FVM model, especially when l0/h = 8.2 (i.e., the finest mesh), ratio ν = 0.2 is used in this shear test. To compare with the results in Wu
the time cost in FVM is only 37 % of that in FEM, which highlights the et al. (2020), the spherical-deviatoric decomposition of strain energy
advantage of using FVM for phase field modeling. Moreover, as the peak (Amor et al., 2009) is used in this test.
reaction force decreases with the decrease of minimum cell size or the The phase field evolution contour shown in Fig. 12 is consistent with
increase of l0/h, the FVM results show a better convergence performance that reported in Wu et al. (2020). In contrast to the straight crack path
than the FEM, which has also been confirmed by the study of Sargado under tension, the crack path under shear develops from the initial crack
et al. (2021). tip at u = 8 × 10− 3 mm but bends down to the lower right corner of the
plate. According to the shear force–displacement curve shown in Fig. 13
4.2. Single-edge notched shear test (the case l0 = 0.015 mm), the present model prediction highly agrees
with the result in Wu et al. (2020), both in the pre- and post-peak stages.
The geometry of this test is the same as the previous one, as shown in Compared with the curves under tension, the shear test results show less
Fig. 5. The top boundary is displaced only in the horizontal direction at brittle responses.
an increment of Δu = 10− 5 mm, and the vertical displacements of the We further investigate the effect of length scale on the simulation
lateral boundaries are prohibited as in Wu et al. (2020). Note that results. Based on the previous study (Bažant and Cedolin, 1983), another
traction-free lateral boundaries are also used in some other studies (e.g., two tests with l0 = 0.03 and 0.0075 mm are performed. To ensure
Ambati et al., 2015; Sargado et al., 2021) which will result in totally convergent results, the ratio l0/h of all the tests is controlled to be close
different post-peak responses. Likewise, a fixed mesh with initial local and greater than 2. The results of the force–displacement curves are
refinement as shown in Fig. 11 is used without AMR during loading. The collectively shown in Fig. 13, which confirms that the peak reaction

8
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

tension test is conducted with two fixed meshes and an adaptively

refined mesh. The two fixed meshes correspond to the initial uniform
mesh and the final adaptive mesh with local refinement, respectively.
The parameters are basically the same as in section 4.1, except that l0 =
0.0075 mm. The meshes are shown in Fig. 14. The uniform mesh has
16,965 quadrilateral cells and a uniform cell size of 0.0077 mm. In AMR,
the cell will be refined when its d ≥ 0.4, and the refinement level is set to
2, which means that one cell would be split into 16 sub-cells.
The force–displacement curves are comparatively shown in Fig. 15.
Clearly, the peak reaction force with AMR applied is smaller than that
using the fixed uniform mesh, and the result with AMR matches notably
better the result of Miehe et al. (2010a). A comparison between the
result of AMR and that of the fixed locally refined mesh indicates that
AMR maintains the computational accuracy. Compared with Fig. 10, the
peak reaction force here is larger due to the use of a smaller l0. The post-
peak behavior is not shown as a straight drop due to the use of insuffi­
ciently small cell sizes.
Finally, it is interesting to see from Table 2 that the final mesh in
AMR has 22,308 cells, 31.5% more than the fixed uniform mesh.
Fig. 10. Comparison of FEM and FVM solutions of the single-edge notched However, the simulation time with AMR is the least among the three
tension test (l0/h = 8.2). cases. The use of AMR not only improves the result accuracy signifi­
cantly but also saves computational time by reducing almost half of the
time cost compared with the use of a fixed uniform mesh and 1/3
Table 1 compared with the use of a fixed locally refined mesh.
Comparison of FEM and FVM for the single-edge notched tension test.
l0/h h (×10− 3
mm) Run time (s) Peak reaction force (N) 4.4. L-shaped panel test
FEM FVM FEM FVM
In this example, the crack propagation in a concrete L-shaped panel
2.1 7.200 858 530 759.15 714.12
4.0 3.740 1242 787 740.09 703.17
is simulated referring to the experimental test in (Winkler et al., 2001).
8.2 1.825 3067 1135 724.54 702.40 The model geometry and boundary conditions are shown in Fig. 16(a).
To approximate the point load in the experiment, a uniform displace­
ment is applied to a small width (across 2 cells) in the numerical test.
The Young’s modulus E = 25.85 GPa, Poisson’s ratio ν = 0.18, critical
energy release rate Gc = 0.089 N/mm, and length scale l0 = 1.1875 mm.
The vertical displacement increment Δu = 3.33 × 10− 4 mm is applied
until u = 1 mm. The model is initially uniformly discretized with a cell
size of 5 mm. The refinement level for AMR is 4. The final mesh is shown
in Fig. 16(b), where the minimum cell size is 0.3125 mm.
Fig. 17(a)-(c) show the crack path evolution at different loading
steps, roughly corresponding to the crack initiation, propagation, and
final stages, respectively. The crack profile from the experimental tests
(digitized from Ambati et al., 2015) is also superimposed in Fig. 17(c) for
comparison. It can be concluded that the simulated crack path nicely lies
in the range of the experimental results. Note that there is no predefined
crack in this test. The results also prove that the present FVM-based
phase field model can simulate arbitrary crack propagation problems.

4.5. Notched plate with a hole

This section simulates the crack initiation and propagation of a

notched plate with a hole, following the experimental test presented by
Ambati et al. (2015). The model geometry and boundary conditions are
shown in Fig. 18(a), where the plate is fixed at the lower left pin, and
loaded in the vertical direction through the upper left pin. The applied
displacement increment is Δu = 0.001 mm until complete failure. The
material of the plate is cement mortar, with the same parameters as
Fig. 11. Fixed mesh used in the single-edge notched shear test (l0 = 0.015 mm). those in the experiment (Ambati et al., 2015), i.e., the Poisson’s ratio ν
= 0.22, Young’s modulus E = 4.9 GPa, and critical energy release rate Gc
force decreases consistently as l0 increases. The present results are = 2.28 N/mm. The length scale in the phase field simulation is l0 = 0.1
generally consistent with the results given in Wu et al. (2020) and other mm.
similar studies. The initial mesh is generated using GMSH (Geuzaine and Remacle,
2009) with mixed triangle and quadrilateral cells, as shown in Fig. 18
(b). The AMR is activated during loading with a refinement level of 3.
4.3. Single-edge notched tension test with adaptive mesh refinement The final mesh is given in Fig. 18(c) with a minimum cell size of 0.042
mm.
To demonstrate the advantages of AMR, the single-edge notched Fig. 19(a)-(c) present the crack paths at different displacements,

9
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

3 3 3
Fig. 12. Crack pattern of the single-edge notched shear test at displacements: (a) u = 8 × 10− mm, (b) u = 10 × 10− mm, and (c) u = 11.8 × 10− mm.

Fig. 13. Force-displacement curves of the single-edge notched shear test with
different l0.

corresponding to the crack initiation, propagation, and complete failure

stages, respectively. The final crack paths are consistent with those ob­ Fig. 15. Force-displacement curves of the single-edge notched tension test with
tained in the experiment (Ambati et al., 2015), as shown in Fig. 19(d). or without AMR.

Fig. 14. Meshes used in the single-edge notched tension test: (a) fixed uniform mesh, and (b) final mesh after AMR (same as the fixed locally refined mesh).

10
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

Table 2 The model is initially uniformly discretized with the cell size h =
Summary of the single-edge notched tension test with or without AMR. 0.025 mm (the out-of-plane cell size is 0.02 mm), and the AMR is evoked
Mesh type Initial / final cell numbers Run time (s) Speedup during loading to accelerate the computation with the refinement level
of 2. Fig. 20 shows the meshes at different time steps. It is seen that the
AMR 16,965 / 22,308 3862 1.97
Fixed uniform mesh 16,965 / 16,965 7599 1 mesh refinement starts from around the crack tip at u = 6.016 × 10− 3
Fixed locally refined mesh 22,308 / 22,308 5790 1.50 mm and shortly develops to form a band around the full crack plane at u
= 6.205 × 10− 3 mm. The difference in mesh density afterward is
insignificant. The evolution of phase field iso-volume (d > 0.9) in Fig. 21
4.6. Three-dimensional single-edge notched tension and shear tests confirms the path of crack propagation, similar to the results in Xu et al.
(2022) and Dinachandra and Alankar (2022). The force–displacement
The last two examples demonstrate that the present FVM-based curve after normalization by thickness is found identical to that in the
phase field model can be easily extended to 3D simulations. The 2D test (i.e., Fig. 8), which is therefore not repeated here.
single-edge notched tension and shear tests are performed for this pur­ For the 3D shear test, the model has a thickness of 0.05 mm and the
pose. For the tension test, the 3D model has a thickness of 0.1 mm, and other parameters are the same as in the 2D test presented in section 4.2
the other parameters are the same as in the 2D test presented in section (l0 = 0.015 mm). The model is initially uniformly discretized with the
4.1 (l0 = 0.015 mm). cell size h = 0.01 mm (the out-of-plane cell size is 0.025 mm), and the

Fig. 16. L-shaped panel test: (a) geometry and boundary conditions, and (b) final mesh.

Fig. 17. Crack pattern of the L-shaped panel test at displacements: (a) u = 0.25 mm, (b) u = 0.55 mm, and (c) u = 1 mm. The experimental crack profile is
superimposed in (c).

11
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

Fig. 18. Notched plate with a hole: (a) geometry and boundary conditions, (b) initial mesh, and (c) final mesh.

Fig. 19. Crack pattern of the notched plate with a hole at displacements: (a) u = 0.4 mm, (b) u = 1.085 mm, (c) u = 2.145 mm, and (d) experimental crack profile
(adapted from Ambati et al., 2015).

12
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

3 3 3
Fig. 20. Meshes in the 3D single-edge notched tension test at displacements: (a) u = 6.016 × 10− mm, (b) u = 6.205 × 10− mm, and (c) u = 9.0 × 10− mm.

3 3
Fig. 21. Phase field iso-volume (d > 0.9) in the 3D single-edge notched tension test at displacements: (a) u = 6.016 × 10− mm, (b) u = 6.205 × 10− mm, and (c) u
= 9.0 × 10− 3 mm.

AMR is evoked during loading to accelerate the computation with the repeated here.
maximum refinement level set to 3. Fig. 22 shows the meshes at different
time steps. It is seen that the mesh refinement starts from around the 5. Conclusions
crack tip at u = 8.0 × 10− 3 mm and gradually evolves into a curved band
around the crack surface at u = 12.1 × 10− 3 mm. The evolution of phase A cell-centered FVM implementation of the phase-field model with
field iso-volume (d > 0.9) in Fig. 23 confirms the path of crack propa­ AMR for brittle fracture using the iterative staggered scheme is pre­
gation. By examining the morphology of the local meshes around the sented in the study. Two commonly used strain energy density decom­
crack tip, it can be observed that in the 3D AMR, a quadtree refinement position methods, namely, the spherical-deviatoric decomposition and
scheme is activated when a structural hexahedral mesh is used. The spectral decomposition, are considered. The proposed model has been
force–displacement curve after normalization by the thickness is found applied to simulating several classical crack propagation problems,
quite close to that in the 2D test (i.e., Fig. 13), which is therefore not including the single-edge notched tension and shear tests, L-shaped

13
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

3 3 3
Fig. 22. Meshes in the 3D single-edge notched shear test at displacements: (a) u = 8.0 × 10− mm, (b) u = 10.7 × 10− mm, and (c) u = 12.1 × 10− mm.

3 3
Fig. 23. Phase field iso-volume (d > 0.9) in the 3D single-edge notched shear test at displacements: (a) u = 8.0 × 10− mm, (b) u = 10.7 × 10− mm, and (c) u = 12.1
× 10− 3 mm.

panel test, and test on a notched plate with a hole. By comparing with CRediT authorship contribution statement
the results from available experiments and FEM studies, it is concluded
that the FVM-based phase-field model is a promising tool to solve these X.L. Yang: Conceptualization, Software, Data curation, Investiga­
crack propagation problems and could be advantageous to the alterna­ tion, Writing – original draft. N. Guo: Supervision, Conceptualization,
tive FEM-based model in terms of efficiency in some case studies. The Methodology, Funding acquisition, Investigation, Writing – review &
isotropic AMR algorithm adopted in the study can effectively improve editing. Z.X. Yang: Conceptualization, Investigation, Writing – review &
the computational accuracy meantime significantly reduce the compu­ editing.
tational cost.
Another feature of FVM is that it has been extensively applied in Declaration of Competing Interest
solving continuity or transport equations, which means that the FVM-
based phase-field model could be conveniently coupled with fluid flow The authors declare that they have no known competing financial
to solve multiphysics problems such as hydraulic fracturing. This will be interests or personal relationships that could have appeared to influence
pursued in a future study. the work reported in this paper.

14
X.L. Yang et al. Computers and Geotechnics 165 (2024) 105921

Data availability brittle, cohesive, and dynamic fracture. Comput. Methods Appl. Mech. Eng. 399,
115347.
Jasak, H., Weller, H.G., 2000. Application of the finite volume method and unstructured
Data will be made available on request. meshes to linear elasticity. Int. J. Numer. Meth. Eng. 48, 267–287.
Jasak, H. (1996). Error analysis and estimation for the finite volume method with
Acknowledgments applications to fluid flows. PhD thesis, Imperial College London, London.
Kiran, R., Nguyen-Thanh, N., Zhou, K., 2022. Adaptive isogeometric analysis-based
phase-field modeling of brittle electromechanical fracture in piezoceramics. Eng.
The study has been financially supported by the Key R&D Program of Fract. Mech. 274, 108738.
Zhejiang Province (No. 2022C03180), Fundamental Research Funds for Kiran, R., Nguyen-Thanh, N., Yu, H., Zhou, K., 2023. Adaptive isogeometric analysis-
based phase-field modeling of interfacial fracture in piezoelectric composites. Eng.
the Central Universities (No. 2021FZZX001-14), and ZJU-ZCCC Institute Fract. Mech. 288, 109181.
of Collaborative Innovation (No. ZDJG2021001). Krishnan, U.M., Gupta, A., Chowdhury, R., 2022. Adaptive phase-field modeling of
brittle fracture using a robust combination of error-estimator and markers. Eng.
Fract. Mech. 274, 108758.
References Li, W., Ambati, M., Nguyen-Thanh, N., Du, H., Zhou, K., 2023a. Adaptive fourth-order
phase-field modeling of ductile fracture using an isogeometric-meshfree approach.
Ambati, M., Gerasimov, T., Lorenzis, L.D., 2015. A review on phase-field models of brittle Comput. Methods Appl. Mech. Eng. 406, 115861.
fracture and a new fast hybrid formulation. Comput. Mech. 55, 383–405. Li, W., Nguyen-Thanh, N., Du, H., Zhou, K., 2023b. Adaptive phase-field modeling of
Amor, H., Marigo, J., Maurini, C., 2009. Regularized formulation of the variational dynamic brittle fracture in composite materials. Compos. Struct. 306, 116589.
brittle fracture with unilateral contact: Numerical experiments. J. Mech. Phys. Solids Liu, Y., Filonova, V., Hu, N., Yuan, Z., Fish, J., Yuan, Z., Belytschko, T., 2014.
57, 1209–1229. A regularized phenomenological multiscale damage model. Int. J. Numer. Meth.
Aranson, I.S., Kalatsky, V.A., Vinokur, V.M., 2000. Continuum field description of crack Eng. 99, 867–887.
propagation. Phys. Rev. Lett. 85, 118–121. Mandal, T.K., Nguyen, V.P., Wu, J.-Y., 2019. Length scale and mesh bias sensitivity of
Assaf, R., Birk, C., Natarajan, S., Gravenkamp, H., 2022. Three-dimensional phase-field phase-field models for brittle and cohesive fracture. Eng. Fract. Mech. 217, 106532.
modeling of brittle fracture using an adaptive octree-based scaled boundary finite Meredith K, Vukčević V. (2018). Resolving the near-field flow patterns of an idealized
element approach. Comput. Methods Appl. Mech. Eng. 399, 115364. fire sprinkler with VOF modeling and adaptive mesh refinement. In: 13th
Bažant, Z.P., Cedolin, L., 1983. Finite element modeling of crack band propagation. OpenFOAM workshop in Shanghai China.
J. Struct. Eng. 109, 69–92. Miehe, C., Hofacker, M., Welschinger, F., 2010a. A phase field model for rate-
Belytschko, T., Black, T., 1999. Elastic crack growth in finite elements with minimal independent crack propagation: Robust algorithmic implementation based on
remeshing. Int. J. Numer. Methods Eng. 45, 601–620. operator splits. Comput. Methods Appl. Mech. Eng. 199, 2765–2778.
Belytschko, T., Lin, J.I., 1987. A three-dimensional impact-penetration algorithm with Miehe, C., Welschinger, F., Hofacker, M., 2010b. Thermodynamically consistent phase-
erosion. Comput. Struct. 25, 95–104. field models of fracture: Variational principles and multi-field FE implementations.
Borden, M.J., Verhoosel, C.C., Scott, M.A., Hughes, T.J., Landis, C.M., 2012. A phase- Int. J. Numer. Meth. Eng. 83, 1273–1311.
field description of dynamic brittle fracture. Comput. Methods Appl. Mech. Eng. 217, Molnr, G., Gravouil, A., 2017. 2D and 3D Abaqus implementation of a robust staggered
77–95. phase-field solution for modeling brittle fracture. Finite Elem. Anal. Des. 130, 27–38.
Borden, M.J., Hughes, T.J., Landis, C.M., Verhoosel, C.C., 2014. A higher-order phase- Navidtehrani, Y., Betegón, C., Martínez-Pañeda, E., 2021. A simple and robust Abaqus
field model for brittle fracture: Formulation and analysis within the isogeometric implementation of the phase field fracture method. Applications in Engineering
analysis framework. Comput. Methods Appl. Mech. Eng. 273, 100–118. Science 6, 100050.
Bourdin, B., Francfort, G.A., Marigo, J., 2000. Numerical experiments in revisited brittle Patil, R.U., Mishra, B.K., Singh, I.V., 2018. An adaptive multiscale phase field method for
fracture. J. Mech. Phys. Solids 48, 797–826. brittle fracture. Comput. Methods Appl. Mech. Eng. 329, 254–288.
Cardiff, P., Demirdžić, I., 2021. Thirty years of the finite volume method for solid Rhie, C.M., Chow, W.L., 1983. Numerical study of the turbulent flow past an airfoil with
mechanics. Arch. Comput. Meth. Eng. 28, 3721–3780. trailing edge separation. AIAA J. 21, 1525–1532.
Cardiff, P., Karac, A., Ivanković, A., 2014. A large strain finite volume method for Sargado, J.M., Keilegavlen, E., Berre, I., Nordbotten, J.M., 2021. A combined finite
orthotropic bodies with general material orientations. Comput. Methods Appl. Mech. element–finite volume framework for phase-field fracture. Comput. Methods Appl.
Eng. 268, 318–335. Mech. Eng. 373, 113474.
Cardiff, P., Tuković, Ž., De Jaeger, P., Clancy, M., Ivanković, A., 2017. A Lagrangian cell- Seleš, K., Lesičar, T., Tonković, Z., Sorić, J., 2019. A residual control staggered solution
centred finite volume method for metal forming simulation. Int. J. Numer. Meth. scheme for the phase-field modeling of brittle fracture. Eng. Fract. Mech. 205,
Eng. 109, 1777–1803. 370–386.
Cardiff, P., Karač, A., De Jaeger, P., Jasak, H., Nagy, J., Ivanković, A., Tuković, Ž., 2018. Tanné, E., Li, T., Bourdin, B., Marigo, J., Maurini, C., 2018. Crack nucleation in
An open-source finite volume toolbox for solid mechanics and fluid-solid interaction variational phase-field models of brittle fracture. J. Mech. Phys. Solids 110, 80–99.
simulations. In: arXiv: Numerical Analysis, pp. 1–45. Walloth, M., Wollner, W., 2022. A posteriori estimator for the adaptive solution of a
Demirdžić, I., Martinović, D., Ivanković, A., 1988. Numerical simulation of thermal quasi-static fracture phase-field model with irreversibility constraints. SIAM J. Sci.
deformation in welded workpiece. Zavarivanje 31, 209–219 in Croatian. Comput. 44 (3), B479–B505.
Demirdžić, I., Martinović, D., 1993. Finite volume method for thermo-elasto-plastic stress Weller, H.G., Tabor, G.R., Jasak, H., Fureby, C., 1998. A tensorial approach to
analysis. Comput. Methods Appl. Mech. Eng. 109, 331–349. computational continuum mechanics using object-oriented techniques. Comput.
Demirdžić, I., Muzaferija, S., 1994. Finite volume method for stress analysis in complex Phys. 12, 620–631.
domains. Int. J. Numer. Meth. Eng. 37, 3751–3766. Winkler, B., Hofstetter, G., Niederwanger, G., 2001. Experimental verification of a
Demirdžić, I., Muzaferija, S., 1995. Numerical method for coupled fluid flow, heat constitutive model for concrete cracking. Proceedings of the Institution of
transfer and stress analysis using unstructured moving meshes with cells of arbitrary Mechanical Engineers, Part L: Journal of Materials: Design and Applications 215,
topology. Comput. Methods Appl. Mech. Eng. 125, 235–255. 75–86.
Demirdžić, I., Muzaferija, S., Perić, M., 1997. Benchmark solutions of some structural Wu, J., Huang, Y., Zhou, H., Nguyen, V.P., 2021. Three-dimensional phase-field
analysis problems using finite-volume method and multigrid acceleration. Int. J. modeling of mode I + II/III failure in solids. Comput. Methods Appl. Mech. Eng. 373,
Numer. Meth. Eng. 40, 1893–1908. 113537.
Dinachandra, M., Alankar, A., 2022. Adaptive finite element modeling of phase-field Wu, J.-Y., Nguyen, V.P., Nguyen, C.T., Sutula, D., Sinaie, S., Bordas, S.P.A., 2020. Phase-
fracture driven by hydrogen embrittlement. Comput. Methods Appl. Mech. Eng. 391, field modeling of fracture. Adv. Appl. Mech. 53, 1–183.
114509. Xu, W., Li, H., Li, Y., Wang, T., Lu, S., Qiang, S., Hua, X., 2022. A phase field method with
Fei, F., Choo, J., 2019. A phase-field method for modeling cracks with frictional contact. adaptive refinement strategy and virtual crack insertion technique. Eng. Fract. Mech.
Int. J. Numer. Meth. Eng. 121, 740–762. 271, 108669.
Fei, F., Choo, J., 2020. A phase-field model of frictional shear fracture in geologic Xu, X., Needleman, A., 1994. Numerical simulations of fast crack growth in brittle solids.
materials. Comput. Methods Appl. Mech. Eng. 369, 113265. J. Mech. Phys. Solids 42, 1397–1434.
Francfort, G.A., Marigo, J., 1998. Revisiting brittle fracture as an energy minimization Yang, Z., 2006. Fully automatic modelling of mixed-mode crack propagation using scaled
problem. J. Mech. Phys. Solids 46, 1319–1342. boundary finite element method. Eng. Fract. Mech. 73, 1711–1731.
Geuzaine, C., Remacle, J.-F., 2009. Gmsh: A 3-D finite element mesh generator with Zhang, X., Sloan, S.W., Vignes, C., Sheng, D., 2017. A modification of the phase-field
built-in pre- and post-processing facilities. Int. J. Numer. Meth. Eng. 79, 1309–1331. model for mixed mode crack propagation in rock-like materials. Comput. Methods
Griffith, A.A., 1921. The phenomena of rupture and flow in solids. Phil. Trans. R. Soc. A Appl. Mech. Eng. 322, 123–136.
221, 163–198.
Gupta, A., Krishnan, U.M., Mandal, T.K., Chowdhury, R., Nguyen, V.P., 2022. An
adaptive mesh refinement algorithm for phase-field fracture models: Application to

15