Available online at www.sciencedirect.

com
ScienceDirect

Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

www.elsevier.com/locate/cma

A coupling approach of state-based peridynamics with node-based

smoothed finite element method
Y.H. Bie a , X.Y. Cui a , b , ∗, Z.C. Li a
a State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Changsha, 410082, PR China
b Collaborative Innovation Center of Intelligent New Energy Vehicle, Shanghai, 200092, PR China

Received 8 July 2017; received in revised form 10 November 2017; accepted 13 November 2017
Available online 6 December 2017

Highlights

• Innovative strategy to couple OSPD with NS-FEM.

• No blending function, mapping parameters or interface elements are required.
• The coupling approach is free of ghost forces.
• It takes full advantages of both OSPD and NS-FEM.

Abstract

In this work, a novel approach to couple ordinary state-based peridynamics (OSPD) with node-based smoothed finite element
method (NS-FEM) is proposed. In present method, the solution domain is partitioned into two regions, one is discretized by OSPD,
the other by NS-FEM, and more importantly, no transition region is introduced. The physical information is transmitted mutually
from local to non-local regions, which is governed by the unified coupling equations of motion. The coupling takes full advantage
of the generality of OSPD and the efficiency of NS-FEM. The parts of regions where damage and fracture either exist or are
expected to propagate are described by OSPD, and the rest of regions are described by NS-FEM to reduce the computational cost
and surface effect. Additionally, the critical bond work in OSPD is assumed to depend on the bond length, which is derived by the
relation with the critical energy release rate in this study. Several numerical examples involving crack propagation are investigated
under either dynamic or quasi-static conditions and satisfactory results have been obtained demonstrating the validity and efficiency
of the proposed coupling approach.
⃝c 2017 Elsevier B.V. All rights reserved.

Keywords: Coupling approach; State-based peridynamics; Finite element method; Damage; Crack propagation

∗ Corresponding author at: State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Changsha,
410082, PR China.
E-mail address: [email protected] (X.Y. Cui).

https://doi.org/10.1016/j.cma.2017.11.022
0045-7825/⃝ c 2017 Elsevier B.V. All rights reserved.
676 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

1. Introduction
Accurate modeling of material failures and crack propagation is still a challenging issue within the framework of
classical continuum mechanics for a long time. Ever since the pioneering study by Irwin [1] in fracture mechanics,
various numerical methods have been proposed to investigate the criterion for the onset of crack growth and the
fracture processes. One of main available approach to deal with the crack propagation in the realm of classical
continuum theory is to re-mesh the body in such a way that the crack lies on the boundary [2], however, the
complexity of programming and the decrease of calculation precision arising from re-meshing approaches limit its
applications [3].
In an effort to eliminate re-meshing techniques, the partition of unity finite element method (PUFEM) [4,5] and
the extended finite element method (XFEM) [6–8] have been developed in modeling crack propagation. Owing to that
these methods allow the discontinuity to exist in the finite elements and the crack growth is described independently of
the mesh, consequently, the re-meshing techniques are not needed [9] in modeling the discontinuity, yet this method
requires additional criteria for the tracking of crack path and crack branching. Additionally, meshless methods have
also received much attention [10,11] for their high potential in numerical simulations of material failures. Although the
burdensome re-meshing is avoided, like classical FEM, these methods need fine meshes in the crack-tip to calculate the
accurate stresses used in the evaluation of the crack propagation criterion. In this sense, if the crack path is unknown
a priori, fine meshes will be needed for tracking crack path, which leads to greater computational cost. Moreover, the
coupling methods between the finite element methods and meshless methods have been also studied in [12,13].
Despite the much progress made in the above methods, the difficulties inherent in these methods are caused by
the fact that the spatial derivatives needed in the partial differential equations of continuum mechanics make no sense
on the crack tip or surface [14]. Therefore, any numerical methods based on these equations inherit this difficulty in
modeling problems involving crack propagation. And these methods are in the need of extra relations to govern the
initiation of cracks as well as their growth velocities and directions, which leads to the complexity of these methods,
especially when tackling the multiple cracks.
Recently, peridynamics was introduced by Silling [15] as a non-local theory of solid mechanics to deal with the
discontinuities. The theory is based on the spatial integral equations instead of the partial differential equations
used in the classical theory. Therefore, the peridynamic governing equations are valid at fracture surfaces. These
characteristics permit the crack initiation and propagation to be modeled without the additional treatment for crack
growth as the material failure is described through the breakage of the bonds between particles [16]. As a result, crack
nucleation, propagation, branching and multiple cracks interaction can be easily implemented by simulation [17].
Thanks to the generality of peridynamics, a series of applications of this method have been investigated. The first
numerical simulation of the fracture problems using bond-based peridynamics was studied by Silling et al. [18].
After that, this innovative method has been exploited onward for extensive applications including modeling extreme
loading of structures [19], fragmentation [20,21], brittle dynamic crack growth [22–26], heat transfer [27–29] and
composites [30,31]. It is worthwhile to mention that bond-based peridynamics suffers from significant restriction on
the constitutive behavior of isotropic materials with Poisson’s ratio of 1/3 in plane stress and 1/4 in plane strain or
three dimensions [32,33]. To overcome the restriction, a generalized formulation of peridynamics called state-based
was proposed in [34,35]. And subsequently, the linearized theory of peridynamic states was also studied in [36,37].
Additionally, state-based peridynamics has been applied to study the plasticity [38–40] and fractures [41,42].
However, the peridynamic model is computationally expensive compared with FEM, which limits its application
to the practical engineering problems. Moreover, it is a challenging task to cope with the issue of the surface
effect in peridynamics. To overcome these issues, the concept of coupling peridynamics with FEM has emerged.
Macek and Silling [43] proposed the first coupling scheme where the peridynamic model is implemented in a
conventional element analysis code using truss elements. Kilic and Madenci [44] introduced a coupling scheme
using an overlapping region in which both peridynamics and finite element equations are used simultaneously.
Agwai et al. [45] coupled the finite element method and peridynamics using the sub-modeling approach where the
global modeling is performed using FEM while the peridynamic theory is utilized for the sub-modeling and failure
prediction, and the same coupling concept was employed by [46]. Also, Liu and Hong [47] introduced interface
elements between FEM and peridynamic regions, in which the inverse isoparametric mapping technique is utilized.
Lubineau et al. [48] proposed the morphing functions to couple the local and non-local theories, and Azdoud [49]
also utilized this approach to study the static fracture. In addition to these coupling techniques, Seleson et al. [50]
proposed the force-based coupling scheme for peridynamics and classical elasticity using non-local weights composed
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 677

of blending functions, they also generalized this coupling approach to couple peridynamics and higher-order gradient
models of any order. Moreover, Han [51] proposed an adaptive coupling between damage mechanics and peridynamics
to objectively simulate all the steps that lead to material failure. Other recent coupling works can be found [52–56].
It is worth noting that these mentioned coupling schemes are associated with bond-based peridynamics, to date, there
are few researches on the coupling between ordinary state-based peridynamics and classical continuum mechanics
except for one coupling model in [57], where the morphing-based coupling strategy is further developed.
Additionally, node-based smoothed finite element method (NS-FEM), as a kind of nodal integration scheme, is
proposed for solid mechanics problems [58–60]. In the NS-FEM, the value of shape function at points is utilized
and the integration of weak form is performed along the boundary of the smoothing cell associated with nodes,
which avoids the complexity involved in Gauss integration for Galerkin methods. Moreover, NS-FEM is usually
developed for triangular or tetrahedral elements which can be easily meshed for really complicated geometries, the
simple in theory and implementation makes it quite suitable for practical engineering problems. A series of studies
involving acoustic analysis [61,62], electromagnetic analysis [63–65], transient heat transfer analysis [66], stochastic
analysis [67] and fracture analysis [68,69] demonstrate the good accuracy and high efficiency of NS-FEM.
In this paper, we present a method for coupling ordinary state-based peridynamics (OSPD) with node-based
smoothed finite element method (NS-FEM) to take full advantage of the generality of OSPD and the computational
efficiency of NS-FEM. Different from the coupling approaches mentioned above, the OSPD subregion is directly
coupled with the NS-FE subregion in present method. As no overlapping region is introduced, the blending function,
mapping technique and interface element are not necessarily required. More importantly, the inverse isoparametric
mapping technique can be avoided due to the integral character of NS-FEM. And it is worthwhile to mention that the
solution domain is firstly discretized into linear triangular elements for 2D (or tetrahedral elements for 3D), and then
converted to polygonal (or polyhedral) integral domains associated with OSPD nodes and NS-FE nodes, which is an
important step forward for previous coupling concept where the quadrilateral and hexahedral elements are employed.
The remainder of this paper is outlined as follows. In Section 2, the theory of ordinary state-based peridynamics
and node-based smoothed finite element method are reviewed. In Section 3, the coupling method between OSPD and
NS-FEM is described in detail. In Section 4, a series of numerical examples are analyzed to validate the proposed
coupling approach. And some concluding remarks are made in the final section.

2. Theory

2.1. Ordinary state-based peridynamics

In ordinary state-based peridynamic theory, the force vector states having unequal magnitudes are parallel to the
direction of the deformation vector states as shown in Fig. 1, which are determined by the collective deformation of
the family of x [70]. And the nonlinear integro-differential equation of motion of a material point at x in the reference
configuration at time t is written as
∫
ρ(x)ü(x, t) = (T[x, t] x′ − x − T[x′ , t] x − x′ )dVx′ + b(x, t)
⟨ ⟩ ⟨ ⟩
(1)
Hx

where ρ is the mass density, u is the displacement vector, x′ is the another material ′
⟨ that′ ⟩is bonded to x, x − x is
point
the bond vector, b(x, t) is the body force density at point x, T[x, t] x − x − T[x , t] x − x is the force vector that
⟨ ′ ⟩ ′

the material point at x′ exerts on the material point at x; Hx is the neighborhood of x within the horizon size of a scalar
nonzero δ, which can be written as

Hx = x′ ∈ Ω , ⏐x′ − x⏐ ≤ δ
{ ⏐ ⏐ }
(2)
where Ω is the elastic body, |·| is the Euclidean norm, δ is the horizon length. It should be noted that each material
point is identified by its coordinates xi in which i denotes node number. xi − x j is a bond vector that is equivalent to
x − x′ if no otherwise specified in the following sections. The force vector state in ordinary state-based peridynamics
can be expressed as
⏐ ⏐
T[x, t] x′ − x = t ξ M(Y), M = Y/ ⏐Y⏐
⟨ ⟩ ⟨ ⟩
(3)
678 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 1. Deformation of PD material points x and x′ and developing unequal pairwise force densities [16].

where T is the force vector state field and assumed to be Riemann integrable, t is the scalar force vector state field, M
is the deformation direction vector state, Y is the deformation vector state field defined by

Y = (x′ + u′ ) − (x + u) = ξ + η (4)
where the relative position vector ξ in the reference configuration and the relative displacement vector η at the time t
can be expressed as

ξ = x′ − x, η = u′ (x′ , t) − u(x, t) (5)

For linearly elastic solids, the three-dimensional and two-dimensional ordinary state-based peridynamic models
have been developed by Silling [34] and Le [35] respectively. The force vector state is derived by calculating the
Fréchet Derivatives of the strain energy with respect to the deformation state in the setting of peridynamics. The
detailed derivation can be referred to [70]. Here, this paper lists the final expression of the force scalar-state as follows
θ 15µ d
t = 3k ωx + ωe for 3D (6)
m m
4kµ θ 8µ d
t= ωx + ωe for 2D plane stress (7)
3k + 4µ m m
4 (3k − µ) θ 8µ d
t= ωx + ωe for 2D plane strain (8)
9 m m
E E
k= ,µ= (9)
3 (1 − 2ν) 2 (1 + ν)
where k and µ are bulk and shear modulus, respectively; E and ν are Young’s modulus and Poisson’s ratio,
respectively. ω is a scalar weighting function called the influence function whose argument is the bond vector ξ
in the reference
⏐ ⏐ configuration, x is the reference position scalar state field whose value at a bond vector is the bond
length ⏐ξ⏐, the weighted volume m is defined as

m = (ωx) • x (10)
Note that the dot product of two states A and B is defined by
∫
⟨ ⟩ ⟨ ⟩
A•B= A ξ B ξ dVξ (11)
Hδ (x)

θ is a scalar-valued function called dilatation and defined mathematically as

3
θ= (ωx) · e (12)
m
The extension scalar state can be written as
e = y − x, y = ⏐Y⏐ , x = ⏐ξ⏐
⏐ ⏐ ⏐ ⏐
(13)
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 679

(a) 2D smoothing domain. (b) 3D smoothing domain.

Fig. 2. The schematic of a node-based smoothing domain for node k.

The deviatoric extension scalar-state is

θx
ed = e − for 3D (14)
3
(γ θ)x 4µ
ed = e − ,γ = for 2D plane stress (15)
3 3k + 4µ
(γ θ)x 2
ed = e − , γ = for 2D plane strain (16)
3 3

2.2. The node-based smoothed finite element method

In the NS-FEM, the domain discretization is firstly generated as traditional FEM (linear triangular elements for 2D
and tetrahedral elements for 3D). Based on the background mesh, the solution domain Ω is further divided into Nnode
∑ Nnode
smoothing cells Ωks (k = 1, 2, 3, . . . , Nnode ) associated with node k such that Ω = k=1 Ωks and Ωis ∩Ω sj = ∅, i ̸= j,
in which Nnode is the total number of nodes in the solution domain. The schematics of a typical node-based smoothing
cell for 2D and 3D problem domain are illustrated in Fig. 2, the smoothing cell Ωks associated with the node k is
generated by connecting sequentially the mid-edge-point to the central point of the surrounding triangular elements
of the node k [60]. Thus the supporting nodes nsk are formed by its adjacent nodes including itself, and the number of
supporting nodes associated with the node k is defined as Nks , the boundary of the smoothing cell Ωks is labeled as Γks .
The discrete equation of motion in a smoothing cell Ωks of NS-FEM is
int
m I ü I = fext
I − fI (17)
where m I is the mass of node I , and fext
I fint
I are the external and internal nodal forces of the node I that are calculated
as follows
∫ ∫
fext
I = φ I bdΩ + φ I tdΓ (18)
Ω Γ
∫ ∑
f Iint
i = φ I, j (x) · P ji (x)dΩ = φ I, j (xk )P ji (xk )Vks , i = x, y, z (19)
Ωks k

where b is the vector of external body forces, t is the prescribed traction vector on the natural boundary Γ , xk are
the coordinates of node k in the reference configuration, φ I is the shape function. Vks is the area or the volume of the
smoothing cell Ωks , which can be calculated specifically by
∫ k Ns
1∑
Vks = dΩ = V e for 2D triangular meshes (20)
Ωks 3 j=1 j
680 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

∫ Ns
k
1∑
Vks = dΩ = V e for 3D tetrahedral meshes (21)
Ωks 4 j=1 j

where V je is the area (or the volume) of the jth triangular (or tetrahedral) element around the node k.
The smoothed gradient of shape function φ I, j (xk ) can be expressed as
∫
1
φ I, j (xk ) = s φ I, j (xk )dΩ (22)
Vk Ωks
Using Green’s theorem, Eq. (22) can be rewritten as
∫
1
φ I, j (xk ) = s φ I (xk ) · n j dΓ (23)
Vk Γks
where n j is the outward normal vector on the boundaryΓks .
And the nominal stress tensor can be calculated by
−1
P ji = J F jk σ ki , J = det F (24)
Consequently, the internal nodal forces can be obtained once the smoothed deformation gradient F and the
smoothed Cauchy stress σ are calculated.
The deformation gradient F and the smoothed deformation gradient F are defined as
∂u i
Fi j = + δi j (25)
∂Xj

∑ ∂φ I ∑
Fi j = u I i + δi j = φ I, j (xk )u I i + δi j (26)
s ∂Xj
I ∈Nk s I ∈Nk

where u I is the displacement of the node I . δ is the Dirac delta function.

And the Cauchy stress tensor can be written as

σi j = λεkk δi j + 2µεi j (27)

1( ) 1
εi j = u i, j + u j,i = (F i j + F ji ) − δi j (28)
2 2
where λ and µ are the Lame constants, εi j is the strain tensor.

3. Coupling between OSPD and NS-FEM

To obtain the efficiency of NS-FEM and exploit the generality of peridynamics in the presence of discontinuities,
the solution domain is partitioned into a NS-FE subregion and a OSPD subregion as illustrated in Fig. 3. The subregion
where cracks either exist or are expected to propagate is modeled using peridynamics, and the remaining part of
domain is described by NS-FEM that requires less computational efforts that is due to the local nature of NS-FEM.
With this coupling method, the overall computational cost and surface effect can be greatly reduced.

3.1. The idea of coupling method

The solution domain is firstly discretized into linear triangular (or tetrahedral) elements, and then converted to
polygonal (or polyhedral) integral domains as previously explained in Section 2.2. To clarify this idea clearly, four
definitions are defined as

Definition 1. Define a union of nodes x j whose neighborhoods include node xk in the reference configuration, denoted
by
{ ⏐
Hx′ k = x j ⏐xk ∈ Hx j
}
(29)
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 681

Fig. 3. Partition of the domain into OSPD subregion and NS-FE subregion.

where j and k are the node numbers identified by their coordinates, x j and xk respectively; Hx j is the horizon (or the
family) of x j , Hx′ k is the dual-horizon of xk in OSPD, readers can consult references [71,72] to get more insight into
the details of dual-horizon concept.

Definition 2. Define a set of node numbers associated with node xk , denoted by

{ ⏐
Sk = j ⏐( j ∈ nsk ) ∩ (x j ∈ Ω 1)
}
(30)
nsk
where is a union of the supporting nodes for the node k, which are formed by its adjacent nodes including itself as
shown in Fig. 2, Ω 1 denotes NS-FE subregion called local region as illustrated in Fig. 3.

Definition 3. The local internal force flocal−int

k is a vector value, which denotes the net force on xk exerted by NS-FE
node in local region Ω 1 such that
∑∫
f kilocal−int = φ k,m (x j ) · P mi (x j )dΩ (31)
j∈Sk Ω sj

where Ω sj (also called the smoothing cell in the NS-FEM) is the integral region of integration point x j , and the other
variables have been defined in Section 2.2.

Definition 4. The non-local internal force fnonlocal−int

k of node xk is a vector state, which denotes the net force state
on xk exerted by OSPD node in the non-local region Ω 2 defined by
∫ ∫ ∫ ∫
nonlocal−int
T[xk , t] x j − xk dVx j dVxk − T[x j , t] xk − x j dVxk dVx j
⟨ ⟩ ⟨ ⟩
fk = (32)
Ωks Hxk Ω sj Hx′ k

where Vx j and Vxk are the incremental volumes⟨ associated ⟩ with node x j and xk respectively, which are calculated by
Eq. (20) for 2D (or Eq.⟨ (21) for⟩ 3D). T[xk , t] x j − xk is referred to as the direct force state on xk exerted by the
bond xk x j , and T[x j , t] xk − x j is viewed as the reaction force state on xk exerted by the dual-bond x j xk accordingly
to [71].
Note that the dual-bond x j xk is a bond from the dual-horizon Hx′ k , that is to say, the dual-bond x j xk from the
dual-horizon Hx′ k is the same as the bond x j xk from the horizon Hxk , if and only if x j ∈ Hx′ k . And the non-local
682 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

internal force can be simplified as follows if certain conditions are satisfied.

∫ ∫
fnonlocal−int T[xk , t] x j − xk dVx j dVxk , if and only if (x j ∈ Ω 1) ∪ (x j ̸∈ Hx′ k )
⟨ ⟩
k = (33)
Ωks Hxk

To bridge ordinary state-based peridynamics and node-based smoothed FEM, Without loss of generality, the unified
coupling governing equation of motion of a material point at xk in the reference configuration at time t can be
expressed as
local−int
m k ük = fext
k − fk + fnonlocal−int
k (34)
where m k is the mass of node k, ük is the acceleration vector, fextk is the external force applied on the node k in Eq. (18),
flocal−int
k is the local internal force vector in Eq. (31), fnonlocal−int
k is the non-local internal force state in Eq. (32).

3.2. The discrete form of coupling governing equations

To more clearly describe the discrete form of the unified coupling governing equation for different nodes in different
subregion of solution domain, four representative nodes (as shown in Fig. 3) are considered as “a”, “b”, “c”and “d”so
that each typical node represents one of the four available types. It should be noted that “a”, “b”, “c”and “d” are the
four different unions of node numbers.
(1) For a node xa ∈ Ω 1, if and only if

∀xk ∈ Ω 2, xa ̸∈ Hxk (35)

Then this node strictly belongs to the local node (called type “a” node) and is only subjected to the local internal
force. Consequently, the discrete governing equation of typical node a1 in Fig. 3 can be expressed as
ext
m a1 üa1 = fa1 − flocal−int
a1 (36)
ext
where the external force fa1 is calculated by Eq. (18), and the local internal force can be written as
∑
local−int
f a1i = φ a1,m (x j ) · P mi (x j )V js (37)
j∈Sa1

(2) For a node xb ∈ Ω 1, if and only if

∃xk ∈ Ω 2, xb ∈ Hxk (38)

Then this node belongs to the local node (called type “b” node) but is subjected to both local and non-local internal
forces. Consequently, the discrete governing equation of typical node b1 can be expressed as
local−int
m b1 üb1 = fext
b1 − fb1 + fnonlocal−int
b1 (39)
where the external force fext
b1 is calculated by Eq. (18), the local internal force flocal−int
b1 and the non-local internal force
fnonlocal−int
b1 can be written as
∑
local−int
f b1i = φ b1,m (x j ) · P mi (x j )V js (40)
j∈Sb1

∑
fnonlocal−int
b1 =− T [xk , t] ⟨xb1 − xk ⟩ βkb1 · Vxb1 · Vxk (41)
Hx′ b1

where βkb1 is the volume correction factor in Appendix.

(3) For a node xc ∈ Ω 2, if and only if

∃x j ∈ Ω 1, c ∈ nsj (42)
Then this node belongs to the non-local node (called type “c” node), and is subjected to both local and non-local
internal forces. Consequently, the discrete governing equation of typical node c1 in Fig. 3 is the same as Eq. (39), the
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 683

Fig. 4. The relative location relationship between the four typical nodes (a1, b1, c1, d1).

only differences are the mathematical expressions of the local and non-local internal forces that are written as follows
∑
local−int
f c1i = φ c1,m (x j ) · P mi (x j )V js (43)
j∈Sc1

∑
fnonlocal−int T [xc1 , t] x j − xc1 βc1 j · Vx j · Vxc1 , for x j ∈ Ω 1
⟨ ⟩
c1 = (44)
Hxc1

∑
fnonlocal−int T [xc1 , t] x j − xc1 βc1 j · Vx j · Vxc1
⟨ ⟩
c1 =
Hxc1
∑ (45)
T [x j , t] xc1 − x j β jc1 · Vxc1 · Vx j , for x j ∈ Ω 2
⟨ ⟩
−
Hx′ c1

Note that the volume correction factors,βc1 j and β jc1 have different values for general cases in the proposed
coupling model.
(4) For a node xd ∈ Ω 2, if and only if

∀x j ∈ Ω 1, d ̸∈ nsj (46)
Then this node strictly belongs to the non-local node (called type “d” node), and is only subjected to the non-local
internal force. Consequently, the discrete governing equation of typical node d1 in Fig. 3 can be written as
nonlocal−int
m d1 üd1 = fext
d1 + fd1 (47)
where fnonlocal−int
d1 can be calculated by Eqs. (44) and (45) on one condition that the subscript c1 is changed to d1.
To explain the four cases mentioned above more clearly, the relative location relationship between the four typical
nodes (a1, b1, c1, d1) is plotted in Fig. 4, all circles are horizons, all ellipses are sets of the supporting nodes denoted
by ns . a1 and b1 are local node numbers, c1 and d1 are nonlocal node numbers, such that {xa1 , xb1 } ∈ Ω 1, {xc1 , xd1 } ∈
Ω 2. Suppose there are two types of nodes whose node numbers are denoted by α (xα ∈ Ω 1) and β(xβ ∈ Ω 2) : (1)
If xα is not neighborhood of xβ and xβ is not the supporting node of xα , that is xα ̸∈ Hxβ , β ̸∈ nsα , then α and β
are equivalent to a1 and d1 respectively in Fig. 4(a). (2) If xα is neighborhood of xβ and xβ is not the supporting
node of xα , that is xα ∈ Hxβ , β ̸∈ nsα , then α and β are equivalent to b1 and d1 respectively in Fig. 4(b). (3) If xα is
neighborhood of xβ and xβ is the supporting node of xα , that is xα ∈ Hxβ , β ∈ nsα , then α and β are equivalent to b1
and c1 respectively in Fig. 4(c).

3.3. Failure criterion of the coupling method

In the numerical simulation of fracture many methods have been utilized to characterize when a material might
fail. Among these available methods, either the stress or the strain histories have been used to describe fracture. But in
peridynamics, the simplest approach to model fracture is through bond breakage, which is related to the critical bond
stretch [18] for bond-based peridynamics. However, in ordinary state-based peridynamics, considering that each bond
responds to the collective deformation of the entire family, there is no one-to-one relation between the bond strain
684 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 5. A schematic of fracture surface.

and work done on the bond [70], therefore, the bond length dependent energy based failure criterion is proposed to
characterize damage in the present coupling model (also called OSPD-NS-FEM model in the following sections), the
concept of which is the same with [39] that damage dependencies can be introduced into the critical bond work ωc .
It is worth noting that the failure criterion is based on the assumption that the critical bond work is not a constant but
a linear function of reciprocal of bond length, 1/ξ in the reference configuration. Here, let us attempt to create the
relations between the critical bond work ωc and the critical energy release rate G c (the critical strain energy required
to create a new fracture surface of unit area). Refer to [16] as shown in Fig. 5, G c can be written as follows
∫ δ ∫ δ ∫ cos−1 z/ξ
G c = 2h ωc ξ dφdξ dz, for 2D (48)
0 z 0

∫ δ ∫ 2π ∫ δ ∫ cos−1 z/ξ
Gc = ωc ξ 2 sin φdφdξ dθ dz, for 3D (49)
0 0 z 0

where ωc = ω0 /ξ , ω0 is constant. When evaluated and solved for ωc , the Eqs. (48) and (49) can be reduced to
Gc 3G c
ωc = , for 2D; ωc = , for 3D (50)
hδ 2 ξ π δ3ξ

To numerically implement the proposed failure criterion for the OSPD-NS-FEM model, the easiest way is to set
the influence function to zero. Consequently, the failure criterion can be expressed mathematically as follows
0 ωξ > ωc /2
{
ω(⏐ξ⏐) =
⏐ ⏐
(51)
1/ |ξ | otherwise
where ω(⏐ξ⏐) is the influence function. ωξ is the work done on the bond ξ computed at each time in the simulation,
⏐ ⏐

denoted by
∫ t
ωξ = ω(ξ, t) =
⟨ ⟩ ⟨ ⟩
T[x, t] ξ · Ẏ ξ dt (52)
0

Note that the factor of 1/2 appears in Eq. (51) because in the OSPD-NS-FEM model, the bond energy is determined
by both horizon and dual-horizon, and the bond and dual-bond break separately in the simulation.
To specify mathematically whether a bond breaks or not, the scalar valued function µξ is introduced as follows
1 ω(⏐⏐ξ⏐⏐) = 1/ |ξ |
{ ⏐ ⏐
µξ (ξ, t) = (53)
0 ω(⏐ξ⏐) = 0
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 685

Fig. 6. The geometry of the rectangular plate.

And the local damage index of a node is defined as

H µξ dVξ
∫
ϕ(x, t) = 1 − ∫x (54)
Hx dVξ

3.4. Time integration scheme of the coupling method

In the present work, the explicit Velocity-Verlet scheme [17] is adopted to carry out the Eq. (34).
∆t n n+1 ∆t n+1
u̇n+1/2 = u̇n + ü , u = un + ∆t u̇n+1/2 , u̇n+1 = u̇n+1/2 + ü (55)
2 2
In which ∆t is the numerical time step that must be smaller than the critical time step defined as:
∆min
∆t < ∆tc = (56)
ck
where ∆min is the minimum nodal distance in the discretized domain, and ck is the speed of sound in the material.

4. Numerical applications
4.1. Ghost force analysis

4.1.1. Two-dimensional wave propagation in a rectangular plate

Considering a rectangular plate with length of L = 0.1 m, and width of W = 0.04 m as shown in Fig. 6. The
material properties are Young’s modulus E = 1 Pa, density ρ = 1 kg/m3 , and Poisson’s ratio ν = 0. And the plane
stress is assumed in this example. It should be noted that this is actually a 1D problem solved in 2D [71]. The initial
displacement field can be expressed by the following equation:
u 0 (x, y) = 0.0002 exp(−(100x)2 ), v0 (x, y) = 0, x ∈ [0, 0.1], y ∈ [−0.02, 0.02] (57)

√ u 0 and v0 are the displacements in x and y directions, respectively. And the wave speed can be calculated
In which
as v = E/ρ = 1 m/s.
The objective of this example is to testify that the present coupling method is free of ghost force and is capable of
solving dynamic problems. Firstly, the rectangular plate is discretized with 8062 triangular elements and 4172 nodes
as shown in Fig. 7, and then the finite element mesh is converted into the material points and correlative polygonal
integral domains as discussed in Section 2.2. And after that the integral domains are partitioned into OSPD subregions
and NS-FEM subregions in Fig. 6. The horizon radius associated with each particle is set to be four times the size
of smoothing cell in all numerical simulations if no otherwise specified. The time step can be chosen according to
Eq. (56), ∆t = 0.5 ms, and the total solution time is set to be 1s for the sake of revealing the efficiency of the proposed
coupling method.
Figs. 8 and 9 show the displacement of all the particles in x direction. As can be seen, the displacement wave
profiles obtained by the coupling method are in almost agreement with that of NS-FEM and are much smoother than
686 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 7. The background mesh of the rectangular plate.

(a) 60 time steps. (b) 260 time steps.

Fig. 8. Displacement wave u x solved by OSPD-NS-FEM versus x coordinate.

(a) 60 time steps. (b) 260 time steps.

Fig. 9. Displacement wave u x solved by NS-FEM versus x coordinate.

 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 687

Fig. 10. The CPU time of the computation for three different methods.

that of [72] where dual-horizon peridynamics is used. Moreover, there is no significant spurious wave reflection in
the process of wave propagation solved by the proposed coupling method from Fig. 6. Additionally, the differences
between the displacement waves obtained by OSPD-NS-FEM and NS-FEM are also plotted in Fig. 9. The difference
is defined as u xo − u xc , where u xo and u xc are the displacements in x direction solved by the two methods above,
respectively. From Fig. 9(a), although the difference of incident waves between the two methods can be seen when the
peak of incident wave passes the left coupling boundary (x = 0.025 m), this difference is very small. When the peak
of reflected wave passes the right coupling boundary (x = 0.075 m), the difference of reflected waves between the
two methods in Fig. 9(b) is relatively larger than that in Fig. 9(a), which results from two main factors: the coupling
effect and the coarser mesh used in this example. But fortunately, the difference of reflected waves remains small, and
with the mesh refining, the coupling effect and the difference will be reduced. Therefore, it can come to a conclusion
that the proposed coupling method is free from significant influence of ghost forces. Additionally, the CPU time of
the computation for three different methods (including OSPD, NS-FEM and OSPD-NS-FEM) are shown in Fig. 10.
The hardware and software features of the computer are:

• Intel(R) Xeon(R) CPU E3-1231 v3 @ 3.40 GHz

• RAM: 8GB
• OS: Windows 7 Pro 64 bit

The result indicates that the OSPD-NS-FEM inherits the efficiency of NS-FEM and can largely reduce the
computational cost relative to OSPD. It should be pointed that the same original mesh in Fig. 7 is utilized for different
methods. And the following examples are simulated on the same computer above.

4.1.2. Uniform deformation of a plate

To present an accurate depiction of the effects of the proposed coupling method, a uniform homogeneous plate
is simulated under uniform extensions: pure traction deformation and pure shear deformation. The plane stress is
assumed and the material parameters of the OSPD-NS-FEM are Young’s modulus E = 189 GPa and Poisson’s ratio
ν = 0.4, respectively. The dimensions of this plate shown in Fig. 11(a) are same as [57]. The OSPD model is found
in the circular zone Ω 2 colored dark gray in the center of the plate and the rest of the plate Ω 1 is described by the
NS-FEM model. The plate is discretized with triangular elements and minimal size of grids is 0.03 µm. The traction
conditions are shown in Fig. 11(b), where the imposed displacement is u y = 0.1 µm. In Fig. 11(c), the plate is under
pure shear deformation and the boundary conditions are set as follows
{
u x = 0.1y
(58)
u y = 0.1x
Fig. 12(a) shows the relative error of the displacement component, u y , obtained for the pure traction test. The
relative error is defined by (u y − u y )/u y , where u y is the analytical displacement in y direction. It can be seen that
688 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 11. Dimensions and boundary conditions of a plate under uniform deformations.

Fig. 12. The relative error on the displacement for a plate under either traction or shear conditions. (a) The relative error on the displacement
component, u y , under traction conditions. (b) The relative error on the displacement component, u x , under shear conditions. (c) The relative error
on the displacement component, u y , under shear conditions.

the spurious effects occur close to the boundary of local and nonlocal regions, but, overall, the majority of the plate
shows homogeneous displacement fields where the range of relative errors for the majority of Fig. 12(a) is only from
−0.2% to 0.2%.
The (results of
) the pure( shear deformation are shown in Fig. 12(b) and (c), where the relative displacement
errors, u x − u x /u x and u y − u y /u y , are shown, where u x and u y are the analytical solution of the displacement,
)
respectively. One can see that the relative displacement errors share similar characteristics with that of the traction test
in Fig. 12(a), moreover, these errors are relatively smaller.
Overall, these two benchmarks demonstrate that the error of the proposed method is localized to the boundary of
local and nonlocal regions and the range of relative error remains small. More importantly, the coupling method can
provide homogeneous displacement fields and the coupling effects are in general negligible. It can be concluded that
the coupling method is free from ghost force to some degree.

4.2. Quasi-static analysis of a cantilever beam

To verify the ability of the coupling method to tackle the problems posed by the surface effect in peridynamics,
a cantilever beam subjected to a concentrated force is examined. The dimensions and geometry of the beam are
illustrated in Fig. 13. The plane stress is assumed. Young’s modulus and Poisson’s ratio used for the beam are
E = 200 GPa and ν = 0.3, respectively. The analytic solution to this problem can be calculated by [73].
H2
[ ( )]
Py
u x (x, y) = (6L − 3x) x + (2 + ν) y −2
(59)
6E I 4
H 2x
[ ]
−P
u y (x, y) = 3νy 2 (L − x) + (4 + 5ν) + (3L − x)x 2 (60)
6E I 4
where I = B H 3 /12, (B = 1) is the second moment of area of the cantilever beam.
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 689

Fig. 13. The geometry of the cantilever beam subjected to the concentrated force.

(a) u x (x = 0.50, y). (b) u y (x, y = 0).

Fig. 14. Displacement variations along the center lines under the concentrated force.

For the sake of comparison, two coupling models for this numerical simulation, namely coupling 1 and coupling 2,
are devised (as shown in Fig. 13 ). In these models, the solution domain is firstly discretized with triangular elements
and then converted into 1111 particles and polygonal integral domains associated them.
Given that the equations of motion of the coupling method and OSPD are in dynamic form, the Adaptive Dynamic
Relaxation (ADR) [74] is utilized to obtain the static solution of this problem. Fig. 14 shows the displacement
variations solved by different models along the center lines. It can be seen that the solutions calculated by OSPD
present relatively large errors especially near the free boundaries, which is called the skin effect [35] rooted in
peridynamics. When the coupling 1 model in Fig. 13 is used, the errors are reduced a little. Moreover, the solution
solved by the coupling 2 model can match up to the analytic solution very well, which means that the coupling
method can be viewed as a useful way to tackle the problems caused by the skin effect and the imposition of boundary
conditions in peridynamics.

4.3. Traction of pre-cracked plate

The objective of this example is to illustrate the quality of the proposed coupling method in the solution of problems
involving crack propagation. We now present a crack branching in a plate with a initial crack and an applied step load
as shown in Fig. 15. It is worth noting that this load (σ = 12 MPa) is constant for the whole duration of numerical
simulation. The same case was simulated by bond-based peridynamics [22,24] and XFEM [7].
690 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 15. Three coupling models for pre-cracked plate (the shadow areas denote the NS-FE subregion and the remaining parts are OSPD subregion).

Table 1
The proportion of OSPD nodes for the studied models.
Model OSPD nodes NS-FE nodes Proportion of OSPD nodes
Coupling 1 14 688 2040 87.80%
Coupling 2 11 760 4968 70.30%
Coupling 3 8832 7896 52.80%
OSPD 16 728 0 100.00%

The plate is made of soda-lime glass whose material properties are E = 72 GPa, ρ = 2440 kg/m3 , ν = 0.22
and G 0 = 135 J/m2 [22]. In the present study, three different coupling models are applied: Coupling 1, Coupling 2,
Coupling 3 as shown in Fig. 15 and the plane stress condition is assumed. For all simulations in this section, the plate
is discretized with 16 728 particles and polygonal integral domains associated them as Section 4.2, and the proportion
of OSPD nodes for different numerical models are shown in Table 1. A uniform time step ∆t = 50 ns calculated
by Eq. (56) is used and the total simulation time is set to t = 46 µs. Fig. 16 compares the temporal evolution of
damage distributions solved by OSPD model and the coupling 3 model, respectively. It can be seen that the crack
path obtained by the present coupling method is in perfect agreement with that obtained by OSPD. It should be
noted that the damage index bigger than 0.5 occurs in Fig. 16, which means that the fragment appears, that is to say
that multiple micro-cracks have emerged around a small body or a material point. This phenomenon usually occurs in
crack branching and multiple cracks interaction, which is very common in researches [24,47,51]. Additionally, another
important information that this example can provide is the propagation speed of the crack, which can be estimated by
computing the location of the crack tip (after branching the upper branch is followed) at the time when the data-dumps
are performed. The data-dumps are done every 40 time-steps starting from the initial time step. And the crack tip is
determined at each time step with the right-most node whose damage index is above 0.35 [24]. Consequently, the
crack propagation speed can be expressed as follows
∥xl − xl−1 ∥
Vl = (61)
tl − tl−1
where xl and xl−1 are the crack-tip positions at times tl and tl−1 respectively.
Fig. 17 shows the comparison of the crack propagation speed obtained by different numerical models and
experiment [75]. It can be noticed that the trends of the crack propagation speed results for all the numerical models
are very similar to what has been experimentally observed in [76]. One can observe in Fig. 17 that the crack is firstly
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 691

(a) Step 200. (b) Step 200.

(c) Step 500. (d) Step 500.

(e) Step 800. (f) Step 800.

Fig. 16. The temporal evolution of the material failure (the color bar denotes the local damage index).

Fig. 17. Comparison of the crack tip propagation speed estimated by different models, 1580 m/s is the maximum speed measured experimentally
by [75].

speeding, slowing down and accelerating to the maximum speed immediately before the crack branching about 22 µs
that is the same as [24], and then slowing down in the presence of branching, which demonstrates that the coupling
method possesses the ability to deal with the problems involving crack propagation. It must be pointed out that the
coupling models can largely reduce the computational cost (as shown in Fig. 18) in the wake of that more OSPD
692 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 18. The computational cost for different models.

Fig. 19. The schematics of double cantilever test specimen.

nodes is converted to NS-FE nodes, that is to say, for a given solution domain and total number of nodes, the greater
the area of NS-FE subregion is, the lower the computational cost will be. Overall, we can arrive at the conclusion that
the coupling method can provide an accurate result at a reduced computational cost.

4.4. Double cantilever test on a pre-cracked plate

This section concerns the double cantilever test on a pre-cracked plate. The dimensions and boundary conditions are
illustrated in Fig. 19 where local and nonlocal regions are partitioned. The material properties are Young’s modulus:
E = 28.3 GPa, Poisson’s ratio: ν = 0.2 and the fracture energy: G 0 = 490 J/m2 . Triangular elements are utilized to
discretize the solution domain in which the minimal particle size is 0.032 m, and then converted to polygonal integral
domains as shown in Fig. 19. To simulate quasi-static examples, the Adaptive Dynamic Relaxation (ADR) [74] is
employed, and the traction condition is imposed through prescribed displacement at each time step (∆t = 1 s), the
increment of displacement is set to be u y = 0.01 µm, the final displacement of loading point is u y = 2 mm.
The double cantilever test is simulated by both the coupling model and OSPD model with the same original meshes.
The curves of the applied external force versus the crack mouth opening displacement (CMOD) are plotted in Fig. 20.
It can be seen that the damage localization occurs at very similar time step with different models, and steep decline
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 693

Fig. 20. Load–CMOD curves obtained by different models.

Fig. 21. The deformed structures with the crack opening obtained by different models at u y = 1 mm (a scale factor of 100 has been applied to all
the displacement on these images).

of force occurs immediately after the peaks of both curves. It should be noted that the very small difference between
force curves lies in the fact that ordinary state-based peridynamics suffers from surface effect. However, the coupling
approach can deal with this issue, which is demonstrated in the absence of damages in Section 4.2. The deformed
structures with the crack opening are shown in Fig. 21, one can see that the deformation of double cantilever solved
by the coupling model is in good agreement with that of OSPD. These comparisons reveal that the coupling model is
able to deal with examples involving the crack propagation in the replacement of OSPD under quasi-static conditions,
more than anything, the computational cost is greatly reduced, which is obvious in Fig. 22.

4.5. Edge-cracked plate under impulsive loading

This section concerns an experiment studied by Kalthoff and Winkler [77] in which a plate with two initial
edge notches is impacted by a projectile as depicted in Fig. 23(a). In the experiment, a brittle fracture mode with
a propagation angle of about 70◦ is observed at lower strain rates, which is focused on this section.
694 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 22. Computational cost for different models.

(a) Geometry of the specimen. (b) Subregions of the coupling model.

Fig. 23. Experimental set-up for crack growth in an edge-cracked plate.

Considering the twofold symmetry of the configuration, only the upper half of the plate is modeled as shown in
Fig. 23(b). The initial impact loading is applied on the left edge on the segment 0 ≤ y ≤ 0.025 m, and a symmetry
condition is applied on the bottom edge of the finite element model (u y = 0); the other edges are traction-free.
The plane stress condition is assumed, the material properties are representative of steel 18Ni1900 used in [71], the
material properties are E = 190 GPa, ρ = 7800 kg/m3 , ν = 0.25 and energy release rate G 0 = 69 000 J/m2 ,
the impact loading velocity is v0 = 22 m/s. The solution domain is partitioned into OSPD subregion and NS-FE
subregion as shown in Fig. 23(b), which is discretized by triangular elements and then converted into 40 804 material
points and correlative integral domains as Section 4.4, where the minimal particle size is 5 × 10−4 m. In the numerical
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 695

(a) Step 240. (b) Step 480. (c) Step 1200.

(d) Step 240. (e) Step 480. (f) Step 1200.

Fig. 24. The temporal evolution of damage distributions calculated by different methods at certain steps.

calculation, the time step is ∆t = 8 × 10−8 s, which is less than the critical time step for the explicit time integration
according to Eq. (56), and the total time steps are 1200. The crack propagation speed is calculated by Eq. (61), and
the Rayleigh wave speed c R [78] is expressed as
cR 0.87 + 1.12ν
= (62)
cs 1+ν
√
In which ν is Poisson’s ratio, cs = µ/ρ is the shear wave speed and µ is shear modulus.
Fig. 24 compares the temporal evolution of damage distributions solved by OSPD and the coupling model, the color
range bar denotes the local damage index calculated by Eq. (54). The damage distribution obtained by the coupling
method at a certain time step is very similar to that of OSPD, which indicates that the performance of the coupling
method is good. Moreover, the crack patterns solved by different methods and experiment are compared as shown in
Fig. 25, the crack propagation angle of the coupling method is about 66. 5◦ , this satisfactory result is better than that
of other methods (63.8◦ in [7] and 65.1◦ in [79]), and the small difference between solutions obtained by the coupling
model and OSPD model is due to the surface effect existed in the OSPD model. It is worth mentioning that the crack
propagation angle profiles of the coupling model have a very similar pattern to that of experiment, which demonstrates
the validity and accuracy of the proposed coupling approach.
The crack propagation speed in the simulations of Kalthoff–Winkler experiment is shown in Fig. 26, the trend of
the crack speed profiles is in a perfect agreement with [72], and more importantly, the curve of the crack propagation
speed obtained by the coupling method is much smoother than the one obtained by OSPD model. The crack begins
to propagate at 20 µs and the maximum crack propagation speeds are 1390 m/s and 1616 m/sfor the coupling model
and OSPD model, respectively. All of them are within the limit of 75% of Rayleigh speed calculated by Eq. (62).
The CPU time of the computation for both models are reported in Fig. 27, this result indicates that the computational
times for the coupling model (OSPD-NS-FEM) are the almost half of that of OSPD model, which demonstrates the
efficiency of the proposed coupling method. It should be noted that the same original mesh is used for both models.
696 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

Fig. 25. The crack paths of Kalthoff–Winkler problem predicted by different numerical algorithms and experimental result.

Fig. 26. The crack propagation speed in the simulations of Kalthoff–Winkler experiment.

Fig. 27. Computational costs for the coupling model and OSPD model.
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 697

Fig. A.1. Circle to circle intersection. (For interpretation of the references to color in this figure legend, the reader is referred to the web version
of this article.)

5. Conclusions
In this paper, a novel coupling approach to couple ordinary state-based peridynamics (OSPD) with node-based
smoothed finite element method (NS-FEM) is developed. The solution domain is partitioned into OSPD subregions
and NS-FE subregions, and then the coupling is achieved by the unified coupling governing equations of motion.
The validity and efficiency of the proposed coupling method is examined through some numerical examples under
dynamic or quasi-static conditions, and some remarks can be made as follows:
(I) The solution domain is discretized with triangular elements and polygon integral regions are constructed through
the techniques used in NS-FEM, the concept of which is firstly incorporated into coupling of local with non-local
theories. Moreover, no transition regions are introduced, no morphing parameters or blending function is required in
the present coupling method.
(II) The proposed coupling strategy takes full advantage of the generality of OSPD and the computational efficiency
of NS-FEM while avoiding their disadvantages, and enables the description of cracks without any limitation of
Poisson’s ratio.
(III) The coupling method is governed by the unified coupling equations of motion, which is convenient for further
mathematical manipulation and implementation in programming. Moreover, this coupling is free of ghost forces.
(IV) The coupling method not only provides the accuracy of the solutions in simulating dynamic and quasi-static
examples either involving damages or not, but also greatly reduces the computational cost and surface effect inherent
in ordinary state-based peridynamics.

Acknowledgments
The support of National Key R&D Program of China (2017YFB1002704) and National Science Foundation of
China (11472101) are gratefully acknowledged.

Appendix. Volume correction

In order to account for the volume reduction of a node that has an intersection with the horizon boundary,
the volume correction scheme [18] is proposed in constant-horizon peridynamics. But for the OSPD-NS-FEM
model, because the horizon varies with coordinates, so the volume correction method in the variable-horizon
peridynamics [71] is introduced in this paper. For instance, one general case as illustrated in Fig. A.1, the material
point x j falls outside of the neighborhood of xi (x j ̸∈ Hxi ), but the volume associated with x j is partially inside the
horizon of xi . To consider this partial volume (red domain in Fig. A.1) in calculating the non-local internal force, the
698 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

intersection volume (red domain in Fig. A.1) between two circles (or spheres) can be approximately calculated as
follows
di2j +r 2j −Ri2 di2j +Ri2 −r 2j
Vint = r 2j cos−1 ( 2di j r j
) + Ri2 cos−1 ( 2di j Ri
) − 21 [(−di j + r j + Ri ) (A.1)
(di j + r j − Ri )(di j − r j + Ri )(di j + r j + Ri )]1/2 , for 2D
π
Vint = (Ri + r j − di j )2 (di2j + 2di j r j − 3r 2j + 2di j Ri + 6r j Ri − 3Ri2 ), for 3D (A.2)
12di j
where Ri (Ri = n · ri , n is constant) is the radius of Hxi , di j is the bond length ⏐ξi j ⏐ in the reference configuration, r j
⏐ ⏐

is the equivalent radius that can be expressed as

√ √
V js 3 3V j
s
rj = , for 2D; r j = , for 3D (A.3)
π 4π
Therefore, the volume correction factor can be written as
⎨Vint /V jS if (Ri − r j ) < di j < (Ri + r j )
⎧

βi j = 1 if di j < (Ri − r j ) (A.4)

0 otherwise
⎩

References
[1] G.R. Irwin, Linear fracture mechanics, fracture transition, and fracture control, Eng. Fract. Mech. 1 (1968) 241–257.
[2] P.A. Wawrzynek, A.R. Ingraffea, An interactive approach to local remeshing around a propagating crack, Finite Elem. Anal. Des. 5 (1989)
87–96.
[3] L.G. Olson, G.C. Georgiou, W.W. Schultz, An efficient finite element method for treating singularities in Laplace’s equation, J. Comput. Phys.
96 (1991) 391–410.
[4] J.M. Melenk, I. Babuška, The partition of unity finite element method: Basic theory and applications, Comput. Methods Appl. Mech. Engrg.
139 (1996) 289–314.
[5] T.C. Gasser, G.A. Holzapfel, Modeling 3D crack propagation in unreinforced concrete using PUFEM, Comput. Methods Appl. Mech. Eng.
194 (2005) 2859–2896.
[6] G. Zi, T. Belytschko, New crack-tip elements for XFEM and applications to cohesive cracks, Internat. J. Numer. Methods Engrg. 57 (2003)
2221–2240.
[7] T. Belytschko, H. Chen, J. Xu, G. Zi, Dynamic crack propagation based on loss of hyperbolicity and a new discontinuous enrichment, Internat.
J. Numer. Methods Engrg. 58 (2003) 1873–1905.
[8] P.M.A. Areias, T. Belytschko, Analysis of three-dimensional crack initiation and propagation using the extended finite element method,
Internat. J. Numer. Methods Engrg. 63 (2005) 760–788.
[9] J.V. Cox, An extended finite element method with analytical enrichment for cohesive crack modeling, Internat. J. Numer. Methods Engrg. 78
(2009) 48–83.
[10] G.R. Liu, Meshfree Methods: Moving beyond the Finite Element Method, second ed., Taylor & Francis/CRC Press, Boca Raton, USA, 2009.
[11] B.N. Rao, S. Rahman, An efficient meshless method for fracture analysis of cracks, Comput. Mech. 26 (2000) 398–408.
[12] G.R. Liu, Y.T. Gu, Coupling of element free Galerkin and hybrid boundary element methods using modified variational formulation, Comput.
Mech. 26 (2000) 166–173.
[13] J.W. Hong, K.J. Bathe, Coupling and enrichment schemes for finite element and finite sphere discretizations, Comput. Struct. 83 (2005)
1386–1395.
[14] F. Mossaiby, M. Bazrpach, A. Shojaei, Extending the method of exponential basis functions to problems with singularities, Eng. Comput. 32
(2015) 406–423.
[15] S.A. Silling, Reformulation of elasticity theory for discontinuities and long-range forces, J. Mech. Phys. Solids 48 (2000) 175–209.
[16] E. Madenci, E. Oterkus, Peridynamic Theory and its Applications, Springer, 2014. http://dx.doi.org/10.1007/978-1-4614-8465-3.
[17] A. Shojaei, T. Mudric, M. Zaccariotto, U. Galvanetto, A coupled meshless finite point/Peridynamic method for2D dynamic fracture analysis,
Int. J. Mech. Sci. 119 (2016) 419–431.
[18] S.A. Silling, E. Askari, A meshfree method based on the peridynamic model of solid mechanics, Comput. Struct. 83 (2005) 1526–1535.
[19] P. Demmie, S. Silling, An approach to modeling extreme loading of structures using peridynamics, J. Mech. Mater. Struct. 2 (2007)
1921–1945.
[20] X. Lai, B. Ren, H. Fan, S. Li, C.T. Wu, R.A. Regueiro, L. Liu, Peridynamics simulations of geomaterial fragmentation by impulse loads, Int.
J. Numer. Anal. Methods Geomech. 39 (2015) 1304–1330.
[21] W. Gerstle, N. Sau, S. Silling, Peridynamic modeling of concrete structures, Nucl. Eng. Des. 237 (2007) 1250–1258.
[22] Y.D. Ha, F. Bobaru, Studies of dynamic crack propagation and crack branching with peridynamics, Int. J. Fract. 162 (2010) 229–244.
[23] Y.D. Ha, F. Bobaru, Characteristics of dynamic brittle fracture captured with peridynamics, Eng. Fract. Mech. 78 (2011) 1156–1168.
[24] D. Dipasquale, M. Zaccariotto, U. Galvanetto, Crack propagation with adaptive grid refinement in 2D peridynamics, Int. J. Fract. 190 (2014)
1–22.
 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700 699

[25] D. Dipasquale, G. Sarego, M. Zaccariotto, U. Galvanetto, Dependence of crack paths on the orientation of regular 2D peridynamic grids, Eng.
Fract. Mech. 160 (2016) 248–263.
[26] S.A. Silling, O. Weckner, E. Askari, F. Bobaru, Crack nucleation in a peridynamic solid, Int. J. Fract. 162 (2010) 219–227.
[27] S. Oterkus, E. Madenci, A. Agwai, Fully coupled peridynamic thermomechanics, J. Mech. Phys. Solids 64 (2014) 1–23.
[28] S. Oterkus, E. Madenci, A. Agwai, Peridynamic thermal diffusion, J. Comput. Phys. 265 (2014) 71–96.
[29] F. Bobaru, M. Duangpanya, The peridynamic formulation for transient heat conduction, Int. J. Heat Mass Transfer 53 (2010) 4047–4059.
[30] E. Askari, J. Xu, Peridynamic analysis of damage and failure in composites, in: 44th AIAA Aerospace Sciences Meeting and Exhibit, 2006,
pp. 1–12.
[31] B. Kilic, A. Agwai, E. Madenci, Peridynamic theory for progressive damage prediction in center-cracked composite laminates, Compos.
Struct. 90 (2009) 141–151.
[32] D. Huang, G.D. Lu, P.Z. Qiao, An improved peridynamic approach for quasi-static elastic deformation and brittle fracture analysis, Int. J.
Mech. Sci. 94–95 (2015) 111–122.
[33] D. Huang, G.D. Lu, C.W. Wang, P.Z. Qiao, An extended peridynamic approach for deformation and fracture analysis, Eng. Fract. Mech. 141
(2015) 196–211.
[34] S.A. Silling, M. Epton, O. Weckner, J. Xu, E. Askari, Peridynamic states and constitutive modeling, J. Elasticity 88 (2007) 151–184.
[35] Q.V. Le, W.K. Chan, J. Schwartz, A two-dimensional, ordinary state-based peridynamic model for linearly elastic solids, Internat. J. Numer.
Methods Engrg. 98 (2014) 547–561.
[36] S.A. Silling, Linearized theory of peridynamic states, J. Elasticity 99 (2010) 85–111.
[37] G. Sarego, Q.V. Le, F. Bobaru, M. Zaccariotto, U. Galvanetto, Linearized state-based peridynamics for 2-D problems, Internat. J. Numer.
Methods Engrg. 108 (2016) 1174–1197.
[38] J.T. Foster, S.A. Silling, W.W. Chen, Viscoplasticity using peridynamics, Internat. J. Numer. Methods Engrg. 81 (2010) 1242–1258.
[39] J.T. Foster, S.A. Silling, W. Chen, An energy based failure criterion for use with peridynamic states, Int. J. Multiscale Comput. Eng. 9 (2011)
675–688.
[40] S. Sun, V. Sundararaghavan, A peridynamic implementation of crystal plasticity, Int. J. Solids Struct. 51 (2014) 3350–3360.
[41] T.L. Warren, S.A. Silling, A. Askari, O. Weckner, M.A. Epton, J. Xu, A non-ordinary state-based peridynamic method to model solid material
deformation and fracture, Int. J. Solids Struct. 46 (2009) 1186–1195.
[42] M.R. Tupek, J.J. Rimoli, R. Radovitzky, An approach for incorporating classical continuum damage models in state-based peridynamics,
Comput. Methods Appl. Mech. Engrg. 263 (2013) 20–26.
[43] R.W. Macek, S.A. Silling, Peridynamics via finite element analysis, Finite Elem. Anal. Des. 43 (2007) 1169–1178.
[44] B. Kilic, E. Madenci, Coupling of peridynamic theory and the finite element method, J. Mech. Mater. Struct. 5 (2010) 707–733.
[45] A. Agwai, I. Guven, E. Madenci, Damage prediction for electronic package drop test using finite element method and peridynamic theory, in:
Proc. - Electron. Components Technol. Conf., vol. 59, 2009, pp. 565–569.
[46] E. Oterkus, E. Madenci, O. Weckner, S. Silling, P. Bogert, A. Tessler, Combined finite element and peridynamic analyses for predicting failure
in a stiffened composite curved panel with a central slot, Compos. Struct. 94 (2012) 839–850.
[47] W.Y. Liu, J.W. Hong, A coupling approach of discretized peridynamics with finite element method, Comput. Methods Appl. Mech. Engrg.
245 (2012) 163–175.
[48] G. Lubineau, Y. Azdoud, F. Han, C. Rey, A. Askari, A morphing strategy to couple non-local to local continuum mechanics, J. Mech. Phys.
Solids 60 (2012) 1088–1102.
[49] Y. Azdoud, F. Han, G. Lubineau, The morphing method as a flexible tool for adaptive local/non-local simulation of static fracture, Comput.
Mech. 54 (2014) 711–722.
[50] P. Seleson, S. Beneddine, S. Prudhomme, A force-based coupling scheme for peridynamics and classical elasticity, Comput. Mater. Sci. 66
(2013) 34–49.
[51] F. Han, G. Lubineau, Y. Azdoud, Adaptive coupling between damage mechanics and peridynamics: A route for objective simulation of
material degradation up to complete failure, J. Mech. Phys. Solids 94 (2016) 453–472.
[52] U. Galvanetto, T. Mudric, A. Shojaei, M. Zaccariotto, An effective way to couple FEM meshes and Peridynamics grids for the solution of
static equilibrium problems, Mech. Res. Commun. 76 (2016) 41–47.
[53] H.F. Fan, S.F. Li, A Peridynamics-SPH modeling and simulation of blast fragmentation of soil under buried explosive loads, Comput. Methods
Appl. Mech. Engrg. 318 (2017) 349–381.
[54] H.F. Fan, G.L. Bergel, S.F. Li, A hybrid peridynamics-SPH simulation of soil fragmentation by blast loads of buried explosive, Int. J. Impact
Eng. 87 (2016) 14–27.
[55] A. Shojaei, M. Zaccariotto, U. Galvanetto, Coupling of 2D discretized Peridynamics with a meshless method based on classical elasticity
using switching of nodal behaviour, Eng. Comput. 34 (2017) 1334–1366.
[56] M. Zaccariotto, D. Tomasi, U. Galvanetto, An enhanced coupling of PD grids to FE meshes, Mech. Res. Commun. 84 (2017) 125–135.
[57] F. Han, G. Lubineau, Y. Azdoud, A. Askari, A morphing approach to couple state-based peridynamics with classical continuum mechanics,
Comput. Methods Appl. Mech. Engrg. 301 (2016) 336–358.
[58] H. Feng, X.Y. Cui, G.Y. Li, S.Z. Feng, A temporal stable node-based smoothed finite element method for three-dimensional elasticity
problems, Comput. Mech. 53 (2014) 859–876.
[59] H. Feng, X.Y. Cui, G.Y. Li, A stable nodal integration method with strain gradient for static and dynamic analysis of solid mechanics, Eng.
Anal. Bound. Elem. 62 (2016) 78–92.
[60] G.R. Liu, T. Nguyen-Thoi, H. Nguyen-Xuan, K.Y. Lam, A node-based smoothed finite element method (NS-FEM) for upper bound solutions
to solid mechanics problems, Comput. Struct. 87 (2009) 14–26.
700 Y.H. Bie et al. / Comput. Methods Appl. Mech. Engrg. 331 (2018) 675–700

[61] X. Hu, X.Y. Cui, Q.Y. Zhang, G. Wang, G.Y. Li, The stable node-based smoothed finite element method for analyzing acoustic radiation
problems, Eng. Anal. Bound. Elem. 80 (2017) 142–151.
[62] G. Wang, X.Y. Cui, H. Feng, G.Y. Li, A stable node-based smoothed finite element method for acoustic problems, Comput. Methods Appl.
Mech. Engrg. 297 (2015) 348–370.
[63] S. Li, X.Y. Cui, H. Feng, G. Wang, An electromagnetic forming analysis modelling using nodal integration axisymmetric thin shell, J. Mater.
Process. Tech. 244 (2017) 62–72.
[64] H. Feng, X.Y. Cui, G.Y. Li, A stable nodal integration method for static and quasi-static electromagnetic field computation, J. Comput. Phys.
336 (2017) 580–594.
[65] H. Feng, X.Y. Cui, G.Y. Li, Coupled-field simulation of electromagnetic tube forming process using a stable nodal integration method, Int. J.
Mech. Sci. 128 (2017) 332–344.
[66] X.Y. Cui, Z.C. Li, H. Feng, S.Z. Feng, Steady and transient heat transfer analysis using a stable node-based smoothed finite element method,
Int. J. Therm. Sci. 110 (2016) 12–25.
[67] X.B. Hu, X.Y. Cui, H. Feng, G.Y. Li, Stochastic analysis using the generalized perturbation stable node-based smoothed finite element method,
Eng. Anal. Bound. Elem. 70 (2016) 40–55.
[68] G.R. Liu, L. Chen, T. Nguyen-Thoi, K.Y. Zeng, G.Y. Zhang, A novel singular node-based smoothed finite element method (NS-FEM) for
upper bound solutions of fracture problems, Internat. J. Numer. Methods Engrg. 83 (2010) 1466–1497.
[69] N. Vu-Bac, H. Nguyen-Xuan, L. Chen, S. Bordas, P. Kerfriden, R.N. Simpson, G.R. Liu, T. Rabczuk, A node-based smoothed extended finite
element method (NS-XFEM) for fracture analysis, Comput. Model. Eng. Sci. 73 (2011) 331–355.
[70] F. Bobaru, J.T. Foster, P.H. Geubelle, S.A. Silling, Handbook of Peridynamic Modeling, CRC Press, 2016.
[71] H. Ren, X. Zhuang, T. Rabczuk, Dual-horizon peridynamics: A stable solution to varying horizons, Comput. Methods Appl. Mech. Engrg.
318 (2017) 762–782.
[72] H. Ren, X. Zhuang, Y. Cai, T. Rabczuk, Dual-horizon peridynamics, Internat. J. Numer. Methods Engrg. 108 (2016) 1451–1476.
[73] S.P. Timoshenko, J.N. Goodier, Theory of Elasticity, 3rd ed., McGraw Hills, NewYork, 1970.
[74] B. Kilic, E. Madenci, An adaptive dynamic relaxation method for quasi-static simulations using the peridynamic theory, Theor. Appl. Fract.
Mech. 53 (2010) 194–204.
[75] F.P. Bowden, J.H. Brunton, J.E. Field, a.D. Heyes, Controlled Fracture of Brittle Solids and Interruption of Electrical Current, Nature 216
(1967) 38–42.
[76] J.E. Field, Brittle fracture: Its study and application, Contemp. Phys. 12 (1971) 1–32.
[77] J.F. Kalthoff, S. Winkler, Failure mode transition at high rates of shear loading, DGM Informationsgesellschaft mbH, in: Impact Loading and
Dynamic Behavior of Materials. Vol. 1, 1988, pp. 185–195.
[78] K.F. Graff, Wave Motion in Elastic Solids, Clarendon Oxford, 1975.
[79] T. Rabczuk, G. Zi, S. Bordas, H. Nguyen-xuan, A simple and robust three-dimensional cracking-particle method without enrichment, Comput.
Methods Appl. Mech. Engrg. 199 (2010) 2437–2455.