Computational Materials Science 233 (2024) 112743

Contents lists available at ScienceDirect

Computational Materials Science

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

Full Length Article

Adaptive meshfree method for fourth-order phase-field model of fracture

using consistent integration schemes
Yulong Shao a, *, Qinglin Duan b, Rongfu Chen a
a
Key Laboratory of Mechanical Reliability for Heavy Equipments and Large Structures of Hebei Province, Yanshan University, Qinhuangdao 066004, China
b
State key Laboratory of Structural Analysis for Industrial Equipment, Department of Engineering Mechanics, Dalian University of Technology, Dalian, 116024, China

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

Keywords: An adaptive fourth-order phase-field model is proposed using consistent element-free Galerkin method (EFG).
Fourth-order phase-field model Convenient construction of high-order interpolation functions in EFG method is fully utilized. Requirement of C1
EFG method continuity of shape functions is satisfied in fourth-order phase-field model. Two different consistent integration
Consistent integration
schemes with high efficiency and high accuracy, namely quadratically consistent one-point integration scheme
Adaptivity
and three-point integration scheme, are employed to evaluate the fourth-order phase-field equation and
Fracture
displacement equation, respectively. To reduce computational cost, the adaptive strategy of local background
mesh refinement is established on basis of strain energy history and phase field. Background mesh is refined via
the insertion of nodes at the midpoints of each side of the integration element. Comparing with EFG method,
computational accuracy and efficiency are enhanced by the proposed method. Load-displacement response, the
smoothness of the resulting stress and crack paths are predicted well.

1. Introduction including crack initiation and branching, even in 3D [14]. Owing to

these advantages, the phase-field model is extensively investigated
As one of the major failure modes of engineering components, [15–21]. Among them, the fourth-order phase-field model [17] has
computational modeling and accurate prediction of fracture are of great gained wide attention due to more regular and faster convergent solu­
significance to understand the failure mechanisms of the engineering tions. Excellent performances are shown in hyperelastic materials
structures. The developed traditional numerical approaches including [22,23], strongly anisotropic materials [24–26], ductile fracture [27],
finite element method (FEM) [1], extended FEM (XFEM) [2,3], etc. hydraulic fracture propagation [28] and contact problem [29].
model the cracks as displacement discontinuities. These methods have However, C1 continuous shape functions are requested for numerical
been gained extensive popularity in fatigue fracture [4,5], ductile frac­ solution of fourth-order phase-field model. It is difficult to construct C1
ture [6–8], composites [9,10], and biomaterials [11,12]. However, these continuity using standard FEM. To solve this problem, Chiarelli et al.
numerical techniques are not able to predict crack initiation without [30] split the fourth-order phase-field equation into two second-order
additional criteria. Besides, some issues remain including the repre­ differential equations by the introduction of an additional variable.
sentation of crack surface, the determination of crack propagation This leads to the increase of the size of the resulting equations and the
criteria for complex fracture events in 3D, etc. computational cost. Recently, many researchers adopt isogeometric
These drawbacks of the classical numerical approaches could be analysis which satisfies C1 continuity to solve fourth-order phase-field
circumvented by the promising phase-field model of brittle fracture equation [25,31–40]. However, due to the tensor-product structure of
presented by Francfort and Marigo [13]. Discontinuous displacement at NURBS, only global refinement is allowed in standard NURBS dis­
cracks in phase-field model is smeared via a smooth fractured region cretization. Besides, small enough size of the elements is demanded to
(phase field) whose width is controlled by the length scale parameter. describe the crack accurately in phase-field model and plenty of ele­
The displacement field and phase field are evaluated by the principle of ments and nodes are involved [31–33]. The affordable computational
energy minimization at the same time. The most appealing superiority of scale poses a challenge to solving fourth-order phase-field model. To
this approach is that the complicated fracture events can be simulated reduce the computational expense, different techniques for local mesh

* Corresponding author.
E-mail address: [email protected] (Y. Shao).

https://doi.org/10.1016/j.commatsci.2023.112743
Received 12 October 2023; Received in revised form 13 December 2023; Accepted 14 December 2023
Available online 28 December 2023
0927-0256/© 2023 Published by Elsevier B.V.
Y. Shao et al. Computational Materials Science 233 (2024) 112743

(a) sharp crack (b) Diffuse crack

Fig. 1. One-dimensional crack representation at x = 0.

refinement in regions where a crack will propagate is adopted such as ∫

Hierarchical B-splines [29], Locally refined (LR) B-splines [40] and Eu (u, Γ) = Wu (ε)dΩ (2)
Virtual Uncommon-Knot-Inserted Master-Slave (VUKIMS) technique Ω\Γ

[34]. Some adaptive mesh refinement strategies are also proposed ∫

[25,36–39]. Es (Γ) = gc dΓ (3)
By contrast, the shape functions in meshfree methods [41–44] such Γ

as the element-free Galerkin (EFG) method [45,46], not only satisfy C1 ∫ ∫

continuity but also are independent of elements and nodes can be P(u) = b⋅udΩ + t⋅udΓ (4)
inserted freely. Thus, fourth-order phase-field equation is solved
Ω ∂Ωσ

straightway. Fundamental convenience is offered for local mesh where Eu (u, Γ) is the stored strain energy in fractured body Ω, Es (Γ) the
refinement and implementation of h-adaptivity without additional crack surface energy producing the crack Γ and P(u) the external po­
techniques, which is helpful to reduce the computational expense for tential energy. Wu is the strain energy density written in term of the
simulation of unpredicted crack paths in engineering structures. Apart strain ε. gc is the critical energy release rate and u is the displacement. It
from that, the construction of high-order interpolation is very easy for is found that the total potential energy in Eq. (1) is consistent with that
meshfree methods. Thus, high gradients of phase field at cracks are of the classical Griffith model. The difference is that the evolution of a
captured precisely. Some research for fourth-order phase-field model unknown crack in Eq. (1) can be evaluated only by the minimization of
has been done employing meshfree methods. Fourth-order phase-field Eq. (1). So a crack can initiate without defects and any additional
model is applied employing local maximum entropy approximants by criteria.
Amiri et al. [47,48]. Wu et al. [49] proposed a reproducing kernel However, the introduction of sharp discontinuities (discontinuous
gradient smoothing meshfree method for the fourth-order phase field displacement) leads to some numerical difficulties. To this end, a vari­
model of brittle fracture. However, only local mesh refinement in re­ able d(x) (phase field) is introduced by Bourdin et al. [53] and Miehe
gions where a crack will propagate is utilized and adaptivity is not et al. [19] to diffuse the sharp discontinuities. As Fig. 1 shows, the phase
implemented. To reduce the computational cost further, an adaptive field d(x) = 1 at cracks and d(x) = 0 in other areas. In this paper, fourth-
fourth-order phase-field model on basis of isogeometric meshfree order phase-field model of fracture [17] is adopted. Thus, the crack
collocation method is developed by Nguyen-Thanh et al. [25] to model surface energy in Eq. (3) can be computed by the domain integral as
crack propagation of polycrystalline materials. follows
The main goal of this paper is to develop a high-efficiency and high- ∫ ∫ ∫ ( )
precision meshfree method to evaluate fourth-order phase-field model. Es (d) = gc dΓ = gc γ(d, ∇d)dΩ =
gc l2 l4
d2 + |∇d|2 + |Δd|2 dΩ
Full use of C1 continuity and convenience to implement h-adaptivity in Γ Ω 2l Ω 2 16
EFG method is made. In addition, numerical integration schemes for (5)
fourth-order phase-field model is focused in this paper. Consistent
integration schemes presented by Wang et al. [50,51] and Duan et al. where γ(d, ∇d) and l represent the crack surface density function and the
[52] are employed to take place of the standard Gauss integration to length scale parameter, respectively.
compute the stiffness matrix, respectively. Consequently, an adaptive Moreover, the effect of regularized cracks needs to be taken into
consistent EFG (CEFG) method is presented to enhance computational account in the elastic strain energy Eu (u, Γ) defined by Eq. (2). The
efficiency of solving fourth-order phase-field model. material strength around the crack is degraded by the introduction of a
The rest of this paper is structured below. Fourth-order phase-field degradation function g(d) = (1 − d)2 into Eu (u, Γ). The tensile and
model of brittle fracture is briefly introduced in section 2. Meshfree compressive parts Ψ+ (ε) and Ψ− (ε) are obtained from the decomposition
discretization of the phase-field model and two different consistent of elastic energy density. The unphysical fracture under compression is
integration schemes are described in section 3. Adaptive strategy is prevented by the only degradation of Ψ+ (ε). Thus, the elastic strain
proposed in section 4. Some numerical examples are conducted in Sec­ energy density is written as
tion 5 and conclusions are given in Section 6.
Wu (ε(u), d) = [g(d) + k]Ψ+ (ε) + Ψ− (ε) (6)
2. Fourth-order phase-field model of brittle fracture
1 [ ]2 [ ]
Ψ+ (ε) = λ 〈tr(ε)〉+ + μtr (ε+ )2 (7)
2
Fourth-order phase-field model of brittle fracture is first briefly
presented by the variational principle [13]. The total potential energy is 1 [ ]2 [ ]
expressed by Ψ− (ε) = λ 〈tr(ε)〉− + μtr (ε− )2 (8)
2
Π(u, Γ) = Eu (u, Γ) + Es (Γ) − P(u) (1) where a numerical parameter k = 0 is employed. The total strain ε is split
into the following form
where

2
Y. Shao et al. Computational Materials Science 233 (2024) 112743

ε = ε+ + ε− (9) accurately, quadratic MLS approximations are used.

∑
3 ∑
3
ε+ = 〈εI 〉+ nI ⊗ nI , ε− = 〈εI 〉− nI ⊗ nI (10) 3.1. Discretization of phase-field equation and consistent one-point
I=1 I=1 integration scheme

where 〈 ∗ 〉± = (∗ ± | ∗ |)/2, εI and nI represent the eigenvalues and ei­ The phase-field equation in Eq. (16)1 is discretized below. The
genvectors of ε separately. following approximation of the phase field d and its variation δd are used
The total potential energy in Eq. (1) is given by combining Eqs. (5)
and (6) ∑
NP ∑
NP
d(x) = NdI (x)dI , δd(x) = NdI (x)δdI (17)
∫ ∫ ∫ ∫
I=1 I=1
Π(u, d) = Wu (ε(u), d)dΩ + gc γ(d, ∇d)dΩ − b⋅udΩ − t⋅udΓ
where NdI (x) and dI represent the MLS shape function and phase field of
Ω Ω Ω ∂Ωσ
(11)
node I separately. NP denotes the number of the approximation nodes.
Minimizing Eq. (11), the controlling equations of the fourth-order Using standard Galerkin discretization with Penalty Method imposing
phase-field model can be obtained by the boundary conditions, the phase-field equation is discretized by
⎧ ( )
(18)
2 4 p
⎨ 2(1 − d)Ψ+ (ε) − gc d − l Δd + l Δ(Δd) = 0
⎪ (βK pd + β1 K′d + K d )d = βFpd + Fd
l 2 16 (12)
⎪
⎩
divσ (u, d) + b = 0 where β and β1 are the penalty parameters.
∫ ∫
( )( )
(19)
p
along with the boundary condition K pd = NTd Nd dΓ, K′d = ∇NTd n nT ∇Nd dΓ
∂Ω ∂Ω
⎧
⎪
⎪ n⋅σ = t on ∂Ωσ ∫ {( }
⎪
⎪ gc ) T gc l gc l3 ′T ′
⎪
⎪
⎪
⎪ u = u on ∂Ωu Kd = 2Hn + Nd Nd + BTd Bd + B d B d dΩ (20)
⎪
⎪ Ω l 2 16
⎨
d(x) = 1 on Γ
(13) ∫ ∫
⎪
⎪
⎪
⎪ Δd = 0 on ∂Ω Fpd = NTd dΓ , Fd = 2NTd Hn dΩ (21)
⎪
⎪ (l2
⎪
⎪
) ∂Ω Ω
⎪
⎪
⎩∇ Δd − d ⋅n = 0 on ∂Ω [ ]
8 Nd = Nd1 (x) Nd2 (x) ⋯ NdNP (x) (22)

and [ ] [ ]T
Bd = B1d (x) B2d (x) ⋯ BNP
d (x) , BId = NdI,x NdI,y (23)
[ ]{ }
σ = (1 − d)2 + k λ〈tr(ε)〉+ G + 2με+ + λ〈tr(ε)〉− G + 2με− (14)
[ 1 ] [ ]T
(24)
2 NP I
B′d = B′d (x) B′d (x) ⋯ B′d (x) , B′d = NdI,xx NdI,xy NdI,yy
where ∂Ω is the boundary of Ω and tr(ε) is the trace of the total strain ε. G
is a second-order unit tensor. To reduce the complexity of solving Eq. where n denotes the vector of the unit normal to ∂Ω. In Eq. (20), Bd (x)
(12), Ψ+ (ε) in Eq. (12) is replaced by a strain energy history function and B′d (x) are matrices of first and second order derivatives of MLS nodal
+
H(x, t) = max Ψ (ε(x, τ)) (15) shape function for phase field, respectively.
τ∈[0,t) To save the computational time and enhance the accuracy of EFG
Finally, the controlling equations of fourth-order phase-field model method, the first and second order derivatives of standard MLS nodal
I,ij
can be given by shape function NdI,i in Eq. (23) and Nd in Eq. (24) are replaced by the
⎧ ( ) corrected ̃ I,i
N and ̃ I,ij
N to compute stiffness matrix K d in Eq. (20). N
̃ and I,i
2 4 d d d
⎨ 2(1 − d)H − gc d − l Δd + l Δ(Δd) = 0
⎪
l 2 16 (16) ̃ I,ij
N can be determined separately by
d
⎪
⎩ ∫ ∫
divσ (u, d) + b = 0
̃ I,i (x)dΩ =
N d NdI (x)ni dΓ (i, j = x, y) (25)
Two equations in Eq. (16) are decoupled. A convenient staggered ΩS ΓS

scheme is adopted to solve Eq. (16). ∫ ∫

̃ I,ij 1 [ I,i ]
N d q(x)dΩ = Nd q(x)nj + NdI,j q(x)ni dΓ
ΓS 2
3. Meshfree discretization for phase-field model ∫
ΩS
∫
1 I[ ]
− Nd q,j (x)ni + q,i (x)nj dΓ − NdI q,ij (x)dΩ (i,j = x, y)
In this section, the controlling equations of fourth-order phase-field ΓS 2 ΩS
model are discretized by EFG method. Owning to the limitation of the (26)
scope, only the final discretization of governing equations is given.
Because of the non-polynomial shape functions of EFG method, it has where ΩS is the background integration cell. ΓS denotes its boundary
low computational efficiency and accuracy using high-order Gauss with the unit normal vector ni .q(x) = p,xx (x) ∪ p,xy (x) ∪ p,yy (x) and p(x) is
integration. To solve this problem, several effective consistent numerical the basis vector for MLS shape functions. The following quadratic basis
integration schemes [50–52,54,55] have been proposed and recent ad­ is used in this paper
vances can be found in [54,55]. In this paper, two different consistent [ ]T
integration schemes by Wang et al. [50,51] and Duan et al. [52], namely p(x) = 1 x y x2 xy y2 , q(x) = [1]T (27)
quadratically consistent one-point integration scheme and three-point Eq. (26) degenerates to
integration scheme, are utilized to evaluate the domain integration for ∫ ∫
phase field equation and displacement field equation, respectively. 1 [ I,i ]
̃ I,ij
N d dΩ = Nd nj + NdI,j ni dΓ (28)
Their central ideas are to replace the standard derivatives of MLS nodal ΩS ΓS 2
shape function with corrected ones to compute stiffness matrix. In Different from Wang et al. [51], there is only one equation in Eqs.
addition, to capture the high gradients of phase filed at cracks (25) and (28), respectively. Thus, the first and second order derivatives

3
Y. Shao et al. Computational Materials Science 233 (2024) 112743

(a) One-point integration scheme (b) Three-point integration scheme

Fig. 2. Schematic diagram of quadratically consistent integration schemes (dark dots, blue clubs and magenta stars are approximation nodes, boundary integration
and domain integration points separately).

of nodal shape function at each integration point are evaluated only where
using the quadratically consistent one-point integration scheme shown ∫
in Fig. 2a, we have K u = BTu D(d)Bu dΩ (35)
Ω
3 ∑
∑ 2
∫
(29)
I,i
̃ d (xC ) =
AN NdI (xG )nLi wG [ ( ) ( )T ]
L=1 G=1 K ∂Ωu = NTu LTn D(d)Bu + LTn D(d)Bu Nu d∂Ωu (36)
∂Ωu

2 ( ) ∫
1∑ 3 ∑
(30)
I,ij
̃ d (xC ) =
AN NdI,i nLj + NdI,j nLi wG K pu = NTu Nu d∂Ωu (37)
2 L=1 G=1 ∂Ωu

∫ ∫
where A and xC are the area and the centroid coordinate of the cell
Fu = NTu bdΩ + NTu td∂Ωσ (38)
separately. xG and wG are the coordinate and weight at two Gauss points Ω ∂Ωσ
(G = 1 ∼ 2) on each edge (L = 1 ∼ 3), respectively. nLi denotes the unit ∫
normal vector of each edge of the cell. F∂Ωu =
( )T
LTn D(d)Bu ud∂Ωu (39)
For the corrected nodal derivatives in x direction, they can be ∂Ωu
determined by ∫
Fpu = NTu ud∂Ωu (40)
1∑ 3 ∑ 2
̃ I,x
N d (xC ) = N I (xG )nLx wG (31) ∂Ωu
A L=1 G=1 d
where
1∑ 3 ∑ 2 [ ]
̃ I,xx
N N I,x nL wG (32) Nu = N1u (x) N2u (x) ⋯ NNP
u (x) (41)
d (xC ) =
A L=1 G=1 d x
[ ]
Bu = B1u B2u ⋯ BNP (42)
Similarly, we can determine the corrected first and second order u

̃ I,y , N
derivatives N ̃ I,xy and N
̃ I,yy of MLS nodal shape function for phase [ ]T
d d d nx 0 ny
fields. The stiffness matrix K d defined by Eq. (20) can be computed by Ln = (43)
0 ny nx
these corrected derivatives.
where Bu is the strain matrix. Like the phase-field equation, consistent
3.2. Discretization of displacement field equation and consistent three- integration scheme is used to compute the corrected first order deriva­
point integration scheme tive Ñ I,i of nodal shape functions. N ̃ I,i is evaluated by the divergence
u u
theorem as follows
Next, the displacement field equation in Eq. (16)2 is discretized. The ∫ ∫ ∫
displacement u is approximated as Ñ I,i (x)q(x)dΩ = NuI (x)q(x)ni dΓ − NuI (x)q,i (x)dΩ (i, j = x, y)
u
ΩS ΓS ΩS
∑
NP ∑
NP
u(x) = NIu (x)uI , δu(x) = NIu (x)δuI (33) (44)
I I=1
where q(x)= p,x (x) ∪ p,y (x). In this paper, the following quadratic basis is
where NIu (x) denotes the matrix of MLS nodal shape functions for the adopted
displacement field. The essential boundary condition is enforced by [ ]T
(45)
T
Nitsche’s method [56] and the final discretization of the displacement p(x) = 1 x y x2 xy y2 , q(x) = [ 1 x y ]
field equation is obtained by Three equations are obtained in Eq. (44). Consistent three-point
{ }
K u − K ∂Ωu + βK pu u = Fu − F∂Ωu + βFpu (34) integration scheme shown in Fig. 2b is used. The corrected derivatives
of nodal shape functions at integration points in each cell are just

4
Y. Shao et al. Computational Materials Science 233 (2024) 112743

In this paper, quadratically consistent one-point and three-point

integration schemes are employed to compute the domain integration
for phase field equation and displacement field equation, respectively.
This method is called consistent element-free method (CEFG).

4. Adaption strategy

As described in Section 2, the sharp crack is smeared into the regu­

larized crack by the length scale parameter. The width of the regularized
crack is controlled by this parameter. When it tends to 0, the sharp crack
is recovered. This leads to the need for plenty of elements and unbear­
able computational cost. An adaptive strategy for second-order phase-
field model is developed by Shao et al. [57]. The amount of elements is
reduced and computational efficiency is enhanced. Besides, this method
is extended into 3D crack propagation [58]. In this section, this strategy
is extended into the fourth-order phase-field model. Corresponding
adaptive strategy is depicted below.
Fig. 3. Schematic diagram of the cracked area.

4.1. Refinement criterion

evaluated by
∑
3
I,i
3 ∑
∑ 2 ∑
3 Similar to the adaptive criterion for second-order phase-field model
̃ u (xH )q(xH ) =
WH N NuI (xG )q(xG )nLi wG − WH NuI (xH )q,i (xH ) [57], the phase field d is also regarded as a criterion which is employed
to determine the cracked area Ωd (green area in Fig. 3). However,
H=1 L=1 G=1 H=1

(46)
significantly, it is not enough large for crack initiation and propagation.
Namely, not only the cracked region Ωd but also the area ΩH (pink area
where WH and xH are the integration weight and coordinate of each
in Fig. 3) is pre-refined.
evaluation point (H = 1 ∼ 3) for domain integration.
The strain history function H(x, t) in Eq. (16)1 is the driving force of
For corrected nodal derivatives in x direction, they are given by
the evolution of phase field. When H exceeds certain value, the phase
Wdx = f x (47) field d = 1 and a new crack initiates or grows. The residual strain history
⎧ ⎫ (1 − d)2 H at crack is minimum (d = 1) and maximum at crack tip zone
⎪
⎪
⎪
̃ u (x1 ) ⎪
I,x ⎪ where the crack will propagate. Thus, the following criterion of back­
⎡ ⎤ ⎪ N ⎪
W1 W2 W3
⎪
⎪
⎨
⎪
⎪
⎬ ground mesh refinement is adopted.
(48)
I,x
W = ⎣ W1 x1 W2 x2 W3 x3 ⎦ , dx = Ñ u (x2 )
W1 y1 W2 y2 W3 y3
⎪
⎪
⎪
⎪
⎪
⎪ (1 − d)2 H > HC or d > dC (50)
⎪ I,x
⎪ ⎪
⎩Ñ u (x3 ) ⎪
⎪
⎭ 2
where HC and dC are parameters. Namely, when (1 − d) H or phase field
⎫ d at one integration point in one element reaches HC or dC , the back­
⎧
⎪ ∑3 ∑2 ⎪
⎪ ground mesh will be refined.
⎪
⎪ ⎪
⎪
⎪
⎪
⎪ L=1
N I (xG )nLx wG
G=1 u
⎪
⎪
⎪
⎪ ⎪
⎪
⎨ ∑3 ∑2 ∑3
⎪
⎬ 4.2. Refinement of background mesh
fx = NuI (xG )xG nLx wG − WH NuI (xH ) (49)
⎪
⎪ L=1 G=1 H=1 ⎪
⎪
⎪
⎪
⎪ ∑3 ∑2 ⎪
⎪
⎪ Refinement of background mesh via the addition of new computa­
⎪
⎪ ⎪
⎪
⎪
⎩ L=1 G=1
NuI (xG )yG nLx wG ⎪
⎪
⎭ tional nodes is shown in Fig. 4. If refinement criterion defined by Eq.
(50) is satisfied in a background integration cell, this cell will be divided
into four sub-sells by inserting nodes at mid-points of its each side. The
̃ I,x of nodal shape function at
The corrected first order derivative N u domain integration is conducted in new sub-cells. Arrangement of
I,y
integration points is obtained by Eq. (47). Similarly, N
̃ is also obtained.
u
integration points is shown in Fig. 4. As the approximation functions of
The stiffness matrices K u defined by Eq. (35) are computed by these EFG method only depend on nodes instead of elements, it is acceptable
corrected derivatives. that there are the hanging nodes in EFG method (nodes A, B, C in Fig. 4).
Some special treatments in standard FEM are not required.

Fig. 4. Schematic diagram of the local refinement of background mesh (red line denotes a crack).

5
Y. Shao et al. Computational Materials Science 233 (2024) 112743

Fig. 5. Schematic diagram of gradual transition of background mesh.

Fig. 6. Schematic diagram of 1D problem.

In addition, the gradual distribution of background mesh for EFG

method is also very easy. While the cell △edf (golden area in Fig. 5) is
refined, it has smallest nodal distance than that of its neighboring cells
(light yellow in Fig. 5). Refinement of these cells will be conducted
though refinement criterion Eq. (50) is not satisfied. As a result, a
smooth transition of nodes is produced between the different refined
regions.
The adaptive background mesh refinement will cease if the size of
the refined cell
h < hC (51)

where hC is a parameter.

5. Numerical examples

Some examples are implemented to study the validity and perfor­

mance of the proposed adaptive CEFG. For comparison, quadratic EFG
method with the same adaptive strategy is also carried out. Different
Fig. 7. Convergence for the derivative of phase field for 1D rod.
from adaptive CEFG, in each cell, 16 integration points are utilized. The
radius of the support for node I is
rI = chmax
I (52) d′e = −
4|x| −
e 2|x|/l
(54)
l2
where c is a constant and hmax
I is evaluated by The L2 norm for the error in the derivative of phase field is given by
hmax
I = max‖xI − xJ ‖ (53) √∫̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√
J⊂SI √ (d′h − d′e )T (d′h − d′e )dΩ
η=√ Ω ∫ ′T ′ (55)
where xI is the location of node I and SI is the set of neighboring points of d d dΩ
Ω e e
node I which construct a polygon surrounding node I.
where d′h and d′e are numerical and exact solutions of derivative of phase
field. Three different sizes of uniform background mesh h = 0.00625
5.1. 1D problem mm, 0.004 mm, 0.002 mm are adopted and the length scale parameter l
= 0.025 mm. The result in Fig. 7 for the convergence shows that the
Firstly, the ability of the developed approach to capture high proposed quadratic CEFG method has higher convergence than
gradient of phase field in fourth-order phase-field model is studied by a quadratic EFG method. The superiority of capturing high gradient of the
1D cracked rod in Fig. 6. There is a crack denoted by the phase field d = 1 proposed approach is validated.
in the middle of the rod. Numerical simulation is conducted in the half of
the rod by symmetry. Thus, the fourth-order phase-field equation is
merely evaluated along with the boundary d = 1. The derivative d′e of
exact solution is

6
Y. Shao et al. Computational Materials Science 233 (2024) 112743

(a) Schematic diagram (b) Initial background mesh

Fig. 8. Initial geometric setup of a square plate with a notch.

Fig. 9. Crack growth followed by adaptive background mesh refinement for a square plate with a notch given by the adaptive CEFG method.

mm2, μ=80.77 kN/mm2 andgc = 2.7 × 10− 3 kN/mm. The length scale
parameter l = 0.014 mm is used. A displacement increment Δu = 1 ×
10− 5 mm is imposed on the top boundary in x direction. The constant c =
2.4 in Eq. (52) for the radius of the support for each node is adopted.
To drive the adaptive mesh refinement, the parameters HC and dC
defined in Eq. (50) are used as follows
{ }
HC = 0.5max (1 − dk )2 Hk , k = 1, 2⋯, nqp , dC = 0.3 (56)

where nqp is the total amount of the integration points. Critical back­
ground mesh size hC = hmax /16 is adopted in the stopping criterion given
by Eq. (51) and hmax is the maximum cell size of the initial background
mesh.
The crack growth followed by adaptive background mesh refinement
and the final crack path obtained by the developed adaptive quadratic
CEFG are illustrated in Fig. 9 and Fig. 10 separately. These results are
consistent with those in [32]. The adaptive background mesh refinement

Fig. 10. Final crack path for a square plate with a notch given by the adaptive
CEFG method. Table 1
Computational time of different methods for a square plate with a notch.

5.2. Square plate with a notch Method Amount of Computational time

nodes (hours)

A square plate with a notch is then considered. The geometry and The adaptive EFG method 1781 (final 0.98
initial background mesh are illustrated in Fig. 8 and 138 initial step)
The proposed adaptive CEFG 1854 (final 0.63
approximation nodes are used. Material properties areλ = 121.15 kN/ method step)

7
Y. Shao et al. Computational Materials Science 233 (2024) 112743

(a) Adaptive CEFG method

(b) Adaptive EFG method

3
Fig. 11. σ xx fields of the square specimen with a notch with a displacement of u = 9 × 10− mm.

illustrated in Fig. 11. Compared with EFG method, the smoother stress
field is presented by the proposed CEFG method. Furthermore,
load–displacement response by the developed CEFG approach in Fig. 12
is consistent with the result in [59]. Due to the different degraded
functions, there is some difference after the peak of load–displacement
curve. Those results demonstrate the advantages of the developed
consistent integration schemes.

5.3. L-shaped panel

Finally, a L-shaped panel is considered for crack initiation and

propagation. The geometry and initial background mesh are shown in
Fig. 13 and 4626 initial approximation nodes are used. Material prop­
erties are: λ=6.16 kN/mm2, μ=10.95 kN/mm2 andgc = 95 × 10− 3N/
mm. The length scale parameter l = 3mm is used. A displacement
increment Δu = 2 × 10− 3 mm is imposed at the right in y direction. The
constant c = 2.6 in Eq. (52) for the radius of the support for each node is
adopted.
To drive the adaptive mesh refinement, the parameters HC and dC
Fig. 12. Load–displacement curve of the square specimen with a notch.
defined in Eq. (50) are used as follows
{ }
around the crack in Fig. 9 demonstrates the validity of the developed HC = 0.4max (1 − dk )2 Hk , k = 1, 2⋯, nqp , dC = 0.5 (57)
method. As a consequence, abundant computational nodes are reduced
in contrast to uniform mesh. Because of the employment of the consis­ The critical background mesh size hC = hmax /16 is adopted in Eq.
tent integration schemes, CEFG method consumes less time compared (51).
with EFG method, which is demonstrated in Table 1. The crack growth followed by adaptive background mesh refinement
Stress fields in x direction by the two different approaches are and the final crack path given by the developed method are illustrated in
Fig. 14 and Fig. 15a separately. Fig. 15 shows a good agreement between

8
Y. Shao et al. Computational Materials Science 233 (2024) 112743

( a) S chemati c di agram ( b) I ni ti al back ground mesh

Fig. 13. Initial geometric setup of the L-shaped panel.

Fig. 14. Crack growth followed by adaptive background mesh refinement for the L-shaped panel given by the adaptive CEFG method.

(a) Crack path given by the proposed CE FG method (b) Experimental result in[60]
Fig. 15. Comparison of the crack path of the L-shaped panel. (a) Crack path given by the proposed CEFG method. (b) Experimental result in [60].

the crack path and the experimental result [60]. Computational time consistent schemes.
and stress field of the proposed CEFG method and the EFG method are
compared in Fig. 16 and Table 2. The proposed CEFG method costs less 6. Conclusions
computational time and presents smoother stress fields. Those results
also conform the validity and superiority of using two different An adaptive fourth-order phase-field model of fracture is developed

9
Y. Shao et al. Computational Materials Science 233 (2024) 112743

(a) Adaptive CEFG method

(b) Adaptive EFG method

Fig. 16. σ xx fields of the L-shaped panel with a displacement of u = 0.33 mm.

Declaration of competing interest

Table 2
Computational time of different methods for the L-shaped panel. The authors declare that they have no known competing financial
Method Amount of Computational time interests or personal relationships that could have appeared to influence
nodes (hours) the work reported in this paper.
The adaptive EFG method 2001 (final 0.5
step) Data availability
The proposed adaptive CEFG 1916 (final 0.27
method step)
The authors do not have permission to share data.

using two different consistent integration schemes (quadratically Acknowledgement

consistent one-point integration scheme and three-point integration
scheme). The requirement of C1 continuity of the shape functions is This work is supported by Hebei Natural Science Foundation (Youth
satisfied for higher-order derivatives of the phase field. Several numer­ Science Foundation) through grant No. A2022203045 which is grate­
ical examples are carried out and the effectiveness and the advantages of fully acknowledged.
the developed adaptive CEFG are validated. Numerical results show the
high gradient of phase field is accurately captured by the proposed References
method and the refined regions are determined by the proposed mesh
[1] D. Ngo, A.C. Scordelis, Finite element analysis of reinforced concrete beams, J. ACI.
refinement criterion. The application of two different consistent inte­
64 (1967) 152–163.
gration schemes remarkably reduces the computational time of EFG [2] N. Moës, J. Dolbow, T. Belytschko, A finite element method for crack growth
method and improve its accuracy. without remeshing, Int. J. Numer. Methods Eng. 46 (1) (1999) 131–150.
[3] T. Belytschko, T. Black, Elastic crack growth in finite elements with minimal
remeshing, Int. J. Numer. Methods Eng. 45 (5) (1999) 601–620.
CRediT authorship contribution statement [4] H. Deng, B.o. Yan, T. Okabe, Fatigue crack propagation simulation method using
XFEM with variable-node element, Eng. Fract. Mech. 269 (2022) 108533.
Yulong Shao: Conceptualization, Investigation, Writing – original [5] S. Kumar, I.V. Singh, B.K. Mishra, A homogenized XFEM approach to simulate
fatigue crack growth problems, Comput. Struct. 150 (2015) 1–22.
draft, Writing – review & editing. Qinglin Duan: Methodology. Rongfu
Chen: Visualization.

10
Y. Shao et al. Computational Materials Science 233 (2024) 112743

[6] M. Ahmad, J.L. Curiel-Sosa, S. Arun, J.A. Rongong, An enhanced void-crack-based [34] K.D. Nguyen, C. Thanh, F. Vogel, H. Nguyen-Xuan, M. Abdel-Wahab, Crack
rousselier damage model for ductile fracture with the XFEM, Int. J. Damage Mech. propagation in quasi-brittle materials by fourth-order phase-field cohesive zone
28 (6) (2019) 943–969. model, Theor. Appl. Fract. Mech. 118 (2022) 103236.
[7] S. Kumar, G. Bhardwaj, A new enrichment scheme in XFEM to model crack growth [35] R. Makvandi, S. Duczek, D. Juhre, A phase-field fracture model based on strain
behavior in ductile materials, Theor. Appl. Fract. Mech. 96 (2018) 296–307. gradient elasticity, Eng. Fract. Mech. 220 (2019) 106648.
[8] S. Beese, S. Loehnert, P. Wriggers, 3D ductile crack propagation within a [36] K. Paul, C. Zimmermann, K.K. Mandadapu, T.J.R. Hughes, C.M. Landis, R.A. Sauer,
polycrystalline microstructure using XFEM, Comput. Mech. 61 (1) (2018) 71–88. An adaptive space-time phase field formulation for dynamic fracture of brittle
[9] L.B. Zhao, J. Zhi, J.Y. Zhang, Z.L. Liu, N. Hu, XFEM simulation of delamination in shells based on LR NURBS, Comput. Mech. 65 (4) (2020) 1039–1062.
composite laminates, Compos. Pt. A-Appl. Sci. Manuf. 80 (2016) 61–71. [37] S. Goswami, C. Anitescu, T. Rabczuk, Adaptive fourth-order phase field analysis for
[10] A. Nasirmanesh, S. Mohammadi, XFEM buckling analysis of cracked composite brittle fracture, Comput. Methods. Appl. Mech. Eng. 361 (2020) 112808.
plates, Compos. Struct. 131 (2015) 333–343. [38] F. Zhu, H. Tang, F. Liu, X. Zhang, Adaptive fourth-order phase field method for
[11] A.E. Vellwock, L. Vergani, F. Libonati, A multiscale XFEM approach to investigate rock fractures using novel refinement criteria and improved data transfer
the fracture behavior of bio-inspired composite materials, Compos. Pt. B-Eng. 141 operators, Comput. Geotech. 151 (2022) 104987.
(2018) 258–264. [39] W. Li, M. Ambati, N. Nguyen-Thanh, H. Du, K. Zhou, Adaptive fourth-order phase-
[12] A. Idkaidek, I. Jasiuk, Cortical bone fracture analysis using XFEM -case study, Int. field modeling of ductile fracture using an isogeometric-meshfree approach,
J. Numer. Meth. Biomed. 33 (4) (2017) e2809. Comput. Methods. Appl. Mech. Eng. 406 (2023) 115861.
[13] G.A. Francfort, J.J. Marigo, Revisiting brittle fracture as an energy minimization [40] K. Paul, C. Zimmermann, T.X. Duong, R.A. Sauer, Isogeometric continuity
problem, J. Mech. Phys. Solids. 46 (8) (1998) 1319–1342. constraints for multi-patch shells governed by fourth-order deformation and phase
[14] M. Ambati, T. Gerasimov, L. De Lorenzis, A review on phase-field models of brittle field models, Comput. Meth. Appl. Mech. Eng. 370 (2020) 113219.
fracture and a new fast hybrid formulation, Comput. Mech. 55 (2) (2015) 383–405. [41] G.R. Liu, An overview on meshfree methods: For computational solid mechanics,
[15] A. Kumar, B. Bourdin, G.A. Francfort, O. Lopez-Pamies, Revisiting nucleation in the Int. J. Comput. Methods. 13 (05) (2016) 1630001.
phase-field approach to brittle fracture, J. Mech. Phys. Solids. 142 (2020) 104027. [42] V.P. Nguyen, T. Rabczuk, S. Bordas, M. Duflot, Meshless methods: A review and
[16] J. Wu, A unified phase-field theory for the mechanics of damage and quasi-brittle computer implementation aspects, Math. Comput. Simul. 79 (2008) 763–813.
failure, J. Mech. Phys. Solids. 103 (2017) 72–99. [43] S. Li, W.K. Liu, Meshfree particle methods, springer, Berlin, 2004.
[17] M.J. Borden, T.J.R. Hughes, C.M. Landis, C.V. Verhoosel, A higher-order phase- [44] T. Belytschko, Y. Krongauz, D. Organ, M. Fleming, P. Krysl, Meshless methods: An
field model for brittle fracture: Formulation and analysis within the isogeometric overview and recent developments, Comput. Meth. Appl. Mech. Eng. 139 (1-4)
analysis framework, Comput. Meth. Appl. Mech. Eng. 273 (2014) 100–118. (1996) 3–47.
[18] C. Kuhn, R. Müller, A continuum phase field model for fracture, Eng. Fract. Mech. [45] P. Krysl, T. Belytschko, Element-free galerkin method: Convergence of the
77 (18) (2010) 3625–3634. continuous and discontinuous shape functions, Comput. Meth. Appl. Mech. Eng.
[19] C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent 148 (3–4) (1997) 257–277.
crack propagation: Robust algorithmic implementation based on operator splits, [46] T. Belytschko, Y.Y. Lu, L. Gu, Element-free Galerkin methods, Int. J. Numer.
Comput. Meth. Appl. Mech. Eng. 199 (45–48) (2010) 2765–2778. Methods Eng. 37 (2) (1994) 229–256.
[20] C. Miehe, F. Welschinger, M. Hofacker, Thermodynamically consistent phase-field [47] F. Amiri, D. Millán, M. Arroyo, M. Silani, T. Rabczuk, Fourth order phase-field
models of fracture: Variational principles and multi-field FE implementations, Int. model for local max-ent approximants applied to crack propagation, Comput.
J. Numer. Methods Eng. 83 (10) (2010) 1273–1311. Meth. Appl. Mech. Eng. 312 (2016) 254–275.
[21] H. Amor, J.J. Marigo, C. Maurini, Regularized formulation of the variational brittle [48] F. Amiri, High-order phase-field model with the local and second-order max-
fracture with unilateral contact: Numerical experiments, J. Mech. Phys. Solids. 57 entropy approximants, Front. Struct. Civ. Eng. 13 (2) (2019) 406–416.
(8) (2009) 1209–1229. [49] J. Wu, D. Wang, Z. Lin, D. Qi, An efficient gradient smoothing meshfree
[22] C. Hesch, S. Schuß, M. Dittmann, M. Franke, K. Weinberg, Isogeometric analysis formulation for the fourth-order phase field modeling of brittle fracture, Comput.
and hierarchical refinement for higher-order phase-field models, Comput. Meth. Part. Mech. 7 (2) (2020) 193–207.
Appl. Mech. Eng. 303 (2016) 185–207. [50] B.B. Wang, C. Lu, C.Y. Fan, M.H. Zhao, A stable and efficient meshfree Galerkin
[23] F. Peng, W. Huang, Y. Ma, Z.Q. Zhang, N. Fu, Fourth-order phase field model with method with consistent integration schemes for strain gradient thin beams and
spectral decomposition for simulating fracture in hyperelastic material, Fatigue plates, Thin-Walled Struct. 153 (2020) 106791.
Fract. Eng. Mater. Struct. 44 (9) (2021) 2372–2388. [51] B.B. Wang, C. Lu, C.Y. Fan, M.H. Zhao, Consistent integration schemes for meshfree
[24] S. Teichtmeister, D. Kienle, F. Aldakheel, M.A. Keip, Phase field modeling of analysis of strain gradient elasticity, Comput. Meth. Appl. Mech. Eng. 357 (Dec. 1)
fracture in anisotropic brittle solids, Int. J. Non-Linear Mech. 97 (2017) 1–21. (2019) 112601.
[25] N. Nguyen-Thanh, W. Li, J. Huang, K. Zhou, Adaptive higher-order phase-field [52] Q. Duan, X. Li, H. Zhang, T. Belytschko, Second-order accurate derivatives and
modeling of anisotropic brittle fracture in 3D polycrystalline materials, Comput. integration schemes for meshfree methods, Int. J. Numer. Methods Eng. 92 (4)
Meth. Appl. Mech. Eng. 372 (2020) 113434. (2012) 399–424.
[26] R. Ma, W. Sun, FFT-based solver for higher-order and multi-phase-field fracture [53] B. Bourdin, G.A. Francfort, J.J. Marigo, Numerical experiments in revisited brittle
models applied to strongly anisotropic brittle materials, Comput. Meth. Appl. fracture, J. Mech. Phys. Solids. 48 (4) (2000) 797–826.
Mech. Eng. 362 (2020) 112781. [54] D. Wang, J. Wu, An inherently consistent reproducing kernel gradient smoothing
[27] M. Dittmann, F. Aldakheel, J. Schulte, P. Wriggers, C. Hesch, Variational phase- framework toward efficient Galerkin meshfree formulation with explicit
field formulation of non-linear ductile fracture, Comput. Meth. Appl. Mech. Eng. quadrature, Comput. Meth. Appl. Mech. Eng. 349 (2019) 628–672.
342 (2018) 71–94. [55] H. Du, J. Wu, D. Wang, J. Chen, A unified reproducing kernel gradient smoothing
[28] Y. Sun, Z. Liu, X. Tang, A hybrid FEMM-Phase field method for fluid-driven fracture Galerkin meshfree approach to strain gradient elasticity, Comput. Mech. 70 (1)
propagation in three dimension, Eng. Anal. Bound. Elem. 113 (2020) 40–54. (2022) 73–100.
[29] C. Hesch, M. Franke, M. Dittmann, I. Temizer, Hierarchical NURBS and a higher- [56] S. Fernández-Méndez, A. Huerta, Imposing essential boundary conditions in mesh-
order phase-field approach to fracture for finite-deformation contact problems, free methods, Comput. Meth. Appl. Mech. Eng. 193 (12–14) (2004) 1257–1275.
Comput. Meth. Appl. Mech. Eng. 301 (2016) 242–258. [57] Y. Shao, Q. Duan, S. Qiu, Adaptive consistent element-free Galerkin method for
[30] L.R. Chiarelli, F.G. Fumes, E.A.B.D. Moraes, G.A. Haveroth, J.L. Boldrini, M. phase-field model of brittle fracture, Comput. Mech. 64 (3) (2019) 741–767.
L. Bittencourt, Comparison of high order finite element and discontinuous Galerkin [58] Y. Shao, Q. Duan, S. Qiu, Consistent element-free Galerkin method for three-
methods for phase field equations: application to structural damage, Comput. dimensional crack propagation based on a phase-field model, Comp. Mater. Sci.
Math. Appl. 74 (7) (2017) 1542–1564. 179 (2020) 109694.
[31] F. Zhu, H. Tang, X. Zhang, G. Papazafeiropoulos, Fourth-order hybrid phase field [59] C. Hesch, A.J. Gil, R. Ortigosa, M. Dittmann, C. Bilgen, P. Betsch, M. Franke,
analysis with non-equal order elements and dual meshes for simulating crack A. Janz, K. Weinberg, A framework for polyconvex large strain phase-field methods
propagation, Comput. Geotech. 142 (2022) 104587. to fracture, Comput. Methods Appl. Mech. Eng. 317 (2017) 649–683.
[32] T.K. Mandal, V.P. Nguyen, J. Wu, On a new high order phase field model for brittle [60] B. Winkler, Traglastuntersuchungen von unbewehrten undbewehrten
and cohesive fracture: Numerical efficiency, length scale convergence and crack Betonstrukturen auf der Grundlage eines objektiven Werkstoffgesetzes fürBeton,
kinking, Comput. Mater. Sci. 203 (2022) 111079. Dissertation University of Innsbruck, Austria (2001).
[33] K. Weinberg, C. Hesch, A high-order finite deformation phase-field approach to
fracture, Contin. Mech. Thermodyn. 29 (4) (2017) 935–945.

11