You are on page 1of 29

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/366516763

Strategy for accurately and efficiently modelling an internal traction-free


boundary based on the s-version finite element method: Problem clarification
and solutions verification

Article  in  Computer Methods in Applied Mechanics and Engineering · February 2023


DOI: 10.1016/j.cma.2022.115843

CITATIONS READS

0 36

6 authors, including:

Tianyu He Naoto Mitsume


The University of Tokyo University of Tsukuba
2 PUBLICATIONS   1 CITATION    58 PUBLICATIONS   271 CITATIONS   

SEE PROFILE SEE PROFILE

Naoki Morita Kazuki Shibanuma


University of Tsukuba The University of Tokyo
23 PUBLICATIONS   26 CITATIONS    131 PUBLICATIONS   721 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Multiscale model for predicting fatigue lives and limits of steels View project

All content following this page was uploaded by Kazuki Shibanuma on 23 December 2022.

The user has requested enhancement of the downloaded file.


Available online at www.sciencedirect.com
ScienceDirect

Comput. Methods Appl. Mech. Engrg. 404 (2023) 115843


www.elsevier.com/locate/cma

Strategy for accurately and efficiently modelling an internal


traction-free boundary based on the s-version finite element method:
Problem clarification and solutions verification
Tianyu Hea , Naoto Mitsumeb , Fumitaka Yasuia , Naoki Moritab , Tsutomu Fukuic ,
Kazuki Shibanumaa ,∗
a School of Engineering, The University of Tokyo, 7-3-1 Bunkyo-ku, Tokyo 113-8656, Japan
b Faculty of Engineering, Information and Systems, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki 305-8573, Japan
c Research Institute, Nippon Kaiji Kyokai, 3-3, Kioi-cho, Chiyoda-ku, Tokyo 102-8567, Japan

Received 4 October 2022; received in revised form 2 December 2022; accepted 6 December 2022
Available online xxxx

Graphical Abstract

Abstract
In this study, a strategy based on the s-version finite element method (S-method) for accurately and efficiently modelling an
internal traction-free boundary of a solid body was established utilising the Neumann boundary conditions of the local mesh.
In conventional studies, the internal boundary is modelled only considering the Neumann conditions of the local mesh; the
global mesh is defined independently of the geometry of the internal boundary. Although this conventional method is effective
in simplifying the meshing procedure for problems of a solid body with an internal boundary, its verification has not been

∗ Corresponding author.
E-mail address: shibanuma@struct.t.u-tokyo.ac.jp (K. Shibanuma).

https://doi.org/10.1016/j.cma.2022.115843
0045-7825/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.
org/licenses/by/4.0/).
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

sufficiently investigated. This study first clarifies the limitations of the conventional method based on strict verification and
discussions of the weak form of the S-method. Three methods are proposed to solve the clarified issue: (i) FCM-based, (ii)
IBM-based, and (iii) XFEM-based methods. The numerical results of the respective methods showed better accuracy than that
of the conventional method. Particularly, the XFEM-based method demonstrated significantly higher accuracy and successfully
solved the clarified issue. Consequently, the intrinsic strengths of the S-method, which are local high accuracy with low
numerical costs and simplicity in the meshing procedure, are effectively utilised for the problems of a solid body with an
internal boundary by the proposed strategy of the XFEM-based method.
© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
Keywords: s-version of the finite element method; Extended finite element method; Boundary conditions; Traction-free boundary; Internal
boundary

1. Introduction
Designing internal boundaries such as holes is an effective way to reduce the amount of the employed materials
while maintaining the strength of the entire structure. To ensure the safety of structures, it is of paramount importance
to accurately analyse the stress concentration fields generated by the presence of discontinuities. The finite element
method (FEM) is the most effective numerical method for the stress analysis of structures. However, FEM has
been developed for solving problems in continuum mechanics; hence, to solve problems with discontinuities, a
complicated meshing procedure is required and it involves high numerical costs.
For an accurate analysis of stress concentration fields, localised fine meshes are required to be compared with
other parts of the target structures. For this purpose, adaptive mesh refinement techniques have been widely studied
and applied [1–10]. Although most adaptive mesh refinement techniques utilise the bases of the existing meshes
with subdivisions, the s-version finite element method (S-method), proposed by Fish, is the method used to achieve
localised mesh refinement based on a unique strategy that uses additional locally defined meshes in the finite element
framework [11]. In the S-method, two types of meshes are generally employed: global and local meshes. A global
mesh is used to analyse the entire domain with relatively coarse elements. The local mesh is composed of relatively
fine elements that are used to analyse the local domain with a high-stress gradient. The local mesh can be inserted
into an arbitrary part of the domain independently of the global mesh, so that a complicated meshing procedure can
be avoided to analyse the stress concentration fields. Additionally, because the fine local mesh applies only to the
local domains, the S-method requires a lower numerical cost to achieve the same accuracy as that of the standard
FEM.
Considering the advantages of the S-method, many modifications have been developed in the literature [12–25].
The method has been applied for the analysis of composite materials [12,26–33]. Fracture mechanics problems
are also major applications of the S-method [34–36]. Kishi et al. recently proposed a strategy for dynamic
crack propagation analysis based on the S-method and demonstrated significant improvements in the efficiency
of calculations compared to those based on the standard FEM [37,38].
The problem of a solid body with an internal traction-free boundary, such as a hole, is typical of stress
concentration. Several researchers have applied the S-method to efficiently analyse this problem [32,39–49]. In
their approach, the internal traction-free boundary was modelled using Neumann conditions of the local mesh, and
the global mesh was defined independently of the geometry of the internal traction-free boundary. In other words,
the global mesh can be defined without considering the local structural discontinuities in the entire domain, and the
local mesh can be automatically applied to the stress concentration fields. Therefore, the approach is significantly
effective in simplifying the meshing procedure for problems of a solid body with an internal traction-free boundary.
However, strict verification and detailed discussions on the validity of this approach have not been performed.
In this study, we first verify the aforementioned approach to model an internal traction-free boundary using
Neumann boundary conditions of the local mesh by analysing the problem of a solid body with an internal traction-
free boundary and then clarify the issue of the approach. We then point out the cause of this issue based on the
derivation of the weak form discretised by the S-method. Based on this point, three strategies are proposed to
solve the clarified issue in the approach to modelling an internal traction-free boundary using Neumann boundary
conditions of the local mesh.
2
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 1. (a) Domain and boundary conditions for the problem of a solid body without any internal traction-free boundaries, (b) Schematic
of global and local meshes defined in the S-method for the problem of a solid body without any internal traction-free boundaries.

Fig. 2. Target problem of a solid body with an internal traction-free boundary; (a) Domains and boundaries of the problem of a solid body
with an internal traction-free boundary, (b) Global and local meshes and boundary conditions.

The remainder of this paper is organised as follows. Basic formulations of the S-method and the approach to
modelling an internal traction-free boundary using Neumann boundary conditions of the local mesh are presented
in Section 2. The strict verification of the conventional approach and a discussion to clarify the issue is described
in Section 3. The three proposed strategies for solving this issue using the conventional approach are introduced
in Section 4. The verification of the proposed strategies is presented in Section 5. Finally, the conclusions of this
study are summarised in Section 6.

2. Basic formulation of the s-version finite element method and its implementation

The target problem of this study is a solid body with an internal traction-free boundary such as a hole. In
this section, we first introduce basic formulations using the simplest problem of a solid body without any internal
traction-free boundaries (see Fig. 1(a)) and then present a method to model the problem of a solid body with an
internal traction-free boundary using the S-method (see Fig. 2(a)). Formulations and the concept for modelling the
internal traction-free boundary using the S-method are introduced in Section 2.1. The linear system and numerical
integration of the S-method are described in Section 2.2.
3
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

2.1. Formulations

The Cauchy momentum equation without the inertia and body forces is given as follows:
∇ · σ (u) = 0, (1)
where σ is the Cauchy stress tensor and u is the displacement vector. This study assumes infinitesimal strain,
1(
ε (u) = ∇u + (∇u)T ,
)
(2)
2
and the linear constitutive relations defined by the following fourth-order elastic modulus tensor D:
σ (u) = D : ε (u) , (3)
D = λI ⊗ I + 2µJ, (4)
where I and J are the second- and fourth-order identity tensors, respectively, λ and µ are Lamé’s constants.
Fig. 1(a) shows the domain and boundary conditions of a solid body without internal traction-free boundaries.
The Dirichlet and Neumann boundary conditions of this problem are expressed as
u=u on Γu , (5)
and
σ (u) n = t on Γt , (6)
where Γu and Γt are parts of the boundary of domain Ω as shown in Fig. 1(a), n is the outward unit normal vector
for domain Ω , u is the prescribed displacement, t is the prescribed surface traction. Γ is the boundary of the domain
Ω , which consists of Γu and Γt as Γ = Γu ∪ Γt .
The weighted residual equation for Eq. (1) subjected to the Neumann boundary condition in Eq. (6) is expressed
as follows:
∫ ∫
δu · (∇ · σ (u)) dΩ = δu · (σ (u) n) − t dΓ ,
( )
(7)
Ω Γt
where δu is the weight function (virtual displacement), and n is the outward unit normal vector of the boundary Γ .
Fig. 1(b) shows a schematic of the global and local meshes defined in the S-method for the problem of a
solid body without any internal traction-free boundaries. The entire domain Ω G that corresponds to domain Ω
is discretised by a relatively coarse global mesh. The local domain Ω L , which is assumed to be included in the
domain Ω G as Ω L ⊂ Ω G , is discretised by the finer local mesh. We assume that the boundary of the local domain
Ω L should not overlap with the boundary of the global domain Ω G . Γ GL is the boundary of the local domain Ω L .
Based on the Galerkin method, u and δu in Eq. (7) are approximated as uh and δuh using the same basis functions
as for the finite elements. u and δu can be divided into global and local parts as follows:

h + uh ,
u ≈ uh = uG L
(8)
δu ≈ δuh = δuG
h + δuLh . (9)
σG L G L
h and σh , which are the stresses with respect to uh and uh , are defined as

σ (u) ≈ σ (uh ) = σ uh + σ uh = σG h + σh = σh .
( G) ( L) L
(10)
And δεG
h and δεLh ,
which are the virtual strains with respect to δuG
h and δuLh , are defined as:

h + ε δuh = δεh + δεh .


ε (δuh ) = ε δuG
) ( ( L) G L
(11)
uG
h, uLh , δuG
h, δuLh , εG
h, εLh , δεG
h, δεLh , σG
h and σLh in Eqs. (8)–(11) can be expressed as
nG

uG
h = NiG (x) diG , (12)
i=1
nL

uLh = NiL (x) diL , (13)
i=1
4
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

nG

δuG
h = NiG (x) δdiG , (14)
i=1
nL

δuLh = NiL (x) δdiL , (15)
i=1
nG

εG
h = BiG (x) diG , (16)
i=1
nL

εLh = BiL (x) diL , (17)
i=1
nG

δεG
h = BiG (x) δdiG , (18)
i=1
nL

δεLh = BiL (x) δdiL , (19)
i=1
n G

σG
h = DεG
h =D BiG (x) diG , (20)
i=1
n L

σLh = DεLh =D BiL (x) diL , (21)
i=1

where NiG and NiL are the shape functions corresponding to the global and local meshes; n G and n L are the numbers
of nodes in the global and local meshes, respectively. BiG and BiL are derivatives of the shape functions NiG and
NiL , respectively. diG , diL , δdiG and δdiL are nodal degrees of freedom. x is the coordinate in physical space. Note
that NiL (x) is defined as zero in the domain outside from Ω L .
To ensure continuity between the global and local meshes, the following Dirichlet boundary condition was
imposed:
uLh = δuLh = 0 on Γ GL . (22)
G
Because Ω corresponds to Ω and δuLh L
is defined in Ω , by applying the approximations in Eqs. (8) and (9), the
discretised
∫ form of the weighted
∫ residual equation in Eq. (7) is expressed as follows:
δuGh · (∇ · σh ) dΩ + δuLh · (∇ · σh ) dΩ
ΩG ΩL
∫ (23)
δuGh · σh n − t dΓ .
( )
=
Γt

Note that there is no term corresponding to the Neumann condition for the local mesh because only the Dirichlet
boundary conditions are applied here. By applying Gauss’s divergence theorem to the terms on the left-hand side
of Eq. (23), and considering Eqs. (18), (19), and (22), we obtain the weak form of the problem as
∫ ∫ ∫
δεh : σh dΩ −
G
δuh · tdΓ +
G
δεLh : σh dΩ = 0. (24)
ΩG Γt ΩL

As stated earlier, the problem of a solid body with an internal traction-free boundary such as a hole is typical of
stress concentration, which has been analysed efficiently in the literature using the S-method [39,41,42,44,46–49].
We recall that, in this approach, the internal traction-free boundary is modelled using the Neumann conditions of the
local mesh, and the global mesh is defined independently of the geometry of the internal boundary. In other words,
the global mesh can be defined without considering the local structural discontinuities in the entire domain, and
the local mesh can be naturally applied to the stress concentration fields. Therefore, this approach is significantly
effective in simplifying the meshing procedure for problems of a solid body with an internal traction-free boundary.
5
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

The domain and boundary conditions for the problem of a solid body with an internal traction-free boundary
are shown in Fig. 2(a). Ωin denotes the inside domain of the internal traction-free boundary. Γ is the boundary
of domain Ω , which consists of three parts as Γ = Γu ∪ Γt ∪ Γin , where Γu and Γt are Dirichlet and Neumann
boundaries in the outer boundary, and Γin is the inner boundary of domain Ω , respectively. Meshes and boundary
conditions of the S-method in the conventional approach [39,41,42,44,46–49] are shown in Fig. 2(b). The domain
Ω G is defined independently of the geometry of the internal traction-free boundary so that Ω G covers both Ω and
Ωin as Ω G = Ωin ∪ Ω . The domain Ω L is defined around the internal traction-free boundary, where higher accuracy
is required because of the stress concentration.
Owing to the traction-free boundary, the following boundary conditions should be satisfied.
σ (u) n = 0 on Γin . (25)
However, the conventional approach [39,41,42,44,46–49] only imposes the Neumann boundary conditions in the
local mesh as
σLh n = 0 on Γin , (26)
Note that the traction-free boundary condition expressed in Eq. (26) is not the proposal of the present study but an
approach, which is often employed in the conventional studies [41,47].
The discretised form of the weighted residual equation in Eq. (23), and the weak form of Eq. (24) do not change
because the boundary conditions in Eq. (26) are the traction-free conditions for the local mesh approximation on
the boundary of the local domain. The aforementioned approach utilises the expected advantages of the S-method,
as follows:
• Simplicity in the meshing procedure owing to the coarse global mesh defined independently of the geometry
of the internal traction-free boundary.
• A high spatial resolution in the local domain owing to the fine local mesh defined around the internal
traction-free boundary.
However, the conventional approach has not been quantitatively verified. Therefore, we verified the conventional
method described in Section 3.

2.2. Linear system for the S-method and the recursive subdivision method to calculate stiffness matrix

By applying the finite-element approximations of the displacement, strain, and stress in Eqs. (12) and (21) into
the weak form of Eq. (24), we obtain the following linear system:
K · d = f. (27)
The matrix K is the assembled global stiffness matrix defined as follows:
[ G
KGL
]
K
K= , (28)
KLG KL
where KG and KL are the partial stiffness matrices for the global and local meshes, respectively, and KGL and KLG
are the partial stiffness matrices representing the relationship between the global and local meshes. These partial
stiffness matrices are defined as follows:

KG = BG (x)T DBG (x) dΩ , (29)
Ω G

KL = BL (x)T DBL (x) dΩ , (30)
L
∫Ω
KGL = BG (x)T DBL (x) dΩ , (31)
Ω L
∫ ( ( )
)T
KLG = BL (x)T DBG (x) dΩ = KGL . (32)
ΩL

6
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 3. Recursive subdivision method for calculating KGL and KLG (n rec = 4).

To calculate KGL and KLG accurately and efficiently, we applied a recursive subdivision method based on the
Gauss quadrature [50]. A schematic representation of this method is shown in Fig. 3. The procedure of this method
is described as follows.

(i)Specifying the number of the recursions n rec .


(ii)Selecting the local elements whose domains have common parts with the boundaries of the global elements.
(iii)Dividing the selected local elements into four equal quadrilateral sub-domains.
(iv) Treating the newly divided subdomains as new local elements and repeating the same procedures as in (ii)
and (iii) n rec times.
(v) Calculating the partial stiffness KGL , and KLG based on the Gauss quadrature on the respective sub-domains
obtained in procedures (i)–(iv).

The assembled nodal degrees-of-freedom vector d in Eq. (27) consists of dG and dL as follows:
{ G}
d
d= , (33)
dL
where
⎧ ⎫

⎪ dG 1 ⎪⎪
⎪ ⎪
⎨ dG

⎪ ⎪

2 ⎬
.. ⎪ ,
G
d = (34)
⎪ . ⎪

⎪ ⎪
⎪ ⎪
⎩dG ⎪
⎪ ⎭
n G
⎧ ⎫

⎪ dL1 ⎪ ⎪
⎪ ⎪
⎨ dL2 ⎪

⎪ ⎪

.. ⎪ .
L
d = (35)
⎪ . ⎪

⎪ ⎪
⎪ ⎪
⎩dL ⎪
⎪ ⎭
nG

7
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

In the right-hand side of Eq. (27), f is the assembled force vector defined as follows:
{ G}
f
f= L , (36)
f
where,
⎧ ⎫
G
⎪ f1 ⎪

⎪ ⎪

⎨ f2G ⎪

⎪ ⎪

fG = . , (37)

⎪ .. ⎪ ⎪

⎪ ⎪

⎩f G ⎪
⎪ ⎭
G
⎧n ⎫
fL ⎪
⎪ 1 ⎪

⎪ ⎪
⎨ f2L ⎪

⎪ ⎪

fL = . , (38)

⎪ .. ⎪⎪

⎪ ⎪ ⎪

⎩fL ⎭

nG
and the force vector for each node can be calculated by

G
fi = NiG (x) tdΓ , (39)
Γt

fiL = NiL (x) tdΓ . (40)
Γt

3. Verification of the conventional method


3.1. Target problem

As shown in Section 2.1, in conventional studies, an internal traction-free boundary is simulated using the
Neumann boundary conditions of the local mesh [39,41,42,44,46–49]. The strict verification of this approach is
performed using two problems. The exact solutions of displacement and stress fields of these two problems are
known. The two problems are as follows: (i) a 2D problem of an infinite plate with a circular hole under uniaxial
tension [51]; (ii) a 3D problem of an infinite medium a spherical hole under uniaxial tension [52,53].
Fig. 4(a) illustrates the 2D problem and domain for the analysis. The radius of the hole was a = 1 and the
remotely applied stress was σ∞ = 1. The domain for the analysis was defined as a 4 × 4 square, considering
the quarter-symmetric conditions. The boundary conditions of the problem are shown in Fig. 4(b). We applied
symmetrical boundary conditions or enforced displacements of the exact solutions at each edge of the domain.
Traction-free conditions were applied to the hole surface.
The exact solutions of the displacement field, uex , and stress field, σex , in the 2D problem are as follows [51]
⎧ ⎫
r 2a 2a 3
⎨ (κ + 1) cos θ + ((1 + κ) cos θ + cos 3θ ) − 3 cos 3θ ⎪
{ } ⎪ ⎪
ex
σ

u x ∞ a a r r

uex = = , (41)
u ex 8µ ⎪ r 2a 2a 3
y ⎩ (κ − 3) sin θ + ((1 − κ) sin θ + sin 3θ) − 3 sin 3θ ⎪
⎪ ⎪

a r r
a 2 (−3r 2 cos 2θ + (3a 2 − 2r 2 ) cos 4θ ) ⎪
⎧ ⎫
σ 1 +
⎧ ex ⎫ ⎪
⎪ ⎪
2r 4
⎪ xx ⎪ ⎪ ⎪
⎪ ⎪


⎪ ⎪ ⎪
⎪ ⎪

2 2 2 2
a (r cos 2θ + (3a − 2r ) cos 4θ )
⎨ ⎬ ⎨ ⎬
σ = σx x = σ∞
ex ex
− , (42)


⎪ ex ⎪
⎪ ⎪
⎪ 2r 4 ⎪

τx y 2 2 2 2
⎪ ⎪ ⎪
⎩ a (−r + (6a − 4r ) cos 2θ ) sin 2θ ⎪
⎩ ⎭ ⎪
⎪ ⎪

⎪ ⎭
2r 4

where E is Young’s modulus, ν is Poisson’s ratio, and µ = 2(1+ν) E


is the shear modulus; further, we assume E = 100
and ν = 0.3. x is the loading direction and y is the vertical direction against x. (r, θ) are the polar coordinates.
Considering a = 1 and σ∞ = 1, the value of maximum principal stress is well known as σmax ex
= 3 at (x, y) = (1, 0).
8
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 4. Infinite plate with a hole under uniaxial tension and domain for analysis.

Fig. 5. Finite element meshes used in verification.

For this verification, we employed three types of meshes, as shown in Fig. 5. The first is the standard FEM,
which has a single mesh considering the geometry of the hole. The second is that of the S-method, wherein the
global mesh is generated considering the geometry of the hole (matching). The last is that of the S-method whose
global mesh is generated independently of the geometry of the hole (non-matching). In both the matching and
non-matching meshes, the local mesh is defined around the hole with sufficiently fine elements, whose size is 1/4
that of the global elements. In the use of either the standard FEM or the S-method (matching), we have to take a
complicated meshing procedure considering both the geometries of the entire domain and the hole. However, the
S-method (non-matching), which corresponds to the approach in conventional studies, has a significant advantage
in simplifying the meshing procedure [39,41,42,44,46–49].
Fig. 6(a) illustrates the domain for analysis of the 3D problem. The radius of the hole was a = 1 and the
remotely applied stress was σapp = 1. The domain for analysis was defined as a 4 × 4 × 4 cube, considering the
one-eighth-symmetric condition. The boundary conditions of the problem are shown in Fig. 6(b).
We applied the symmetric boundary conditions or enforced displacements of the exact solutions at each surface of
the domain. For isotropic linear elasticity material, the exact solution of the displacement in the spherical coordinate
9
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 6. Analysis domain for the 3D problem of an infinite medium with a spherical hole under uniform tension and corresponding boundary
conditions.

(r, β, θ ) for this problem is expressed as [54–56]:


r 3A2 A3
u r = −A1 + 4

2µ 4µr 2µr 2
(
9A2 B2 (4ν − 5) cos2 β
) (43)
+ 3A1r − 4 + B1 (4ν − 2) r + ,
2r r2 2µ
B2 (1 − 2ν) sin β cos β
( )
r 3A2
u β = −3A1 − + B1r + 2 , (44)
2µ 2µr 4 r µ
u θ = 0, (45)
σapp ν σapp a 5
σapp a (6 − 5ν)
3
A1 = ; A2 = ; A3 = , (46)
1+ν 7 − 5ν 2(7 − 5ν)
σapp 5σapp a 3 E
B1 = − ; B2 = − ;µ = , (47)
2 (1 + ν) 2 (7 − 5ν) 2 (1 + ν)
where a, µ, and σapp are the radius of the spherical hole, shear modulus, and remotely applied stress in the tensile
direction along the z-axis, respectively. u r , u β and u θ are the displacement components in the spherical coordinate
system.
The exact solutions of the stress in spherical coordinates are expressed as follows [52]:
σapp
( 3
) 6a 5
)
a (
σrr = σapp cos2 β + 6 − 5(5 − ν) cos2
β + (3 cos2
β − 1) , (48)
7 − 5ν r 3 r5
10
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 7. Finite element meshes used in the verification for the 3D problem.

a3 ( ) a5
( )
3σapp
σθθ = 5ν − 2 + 5(1 − 2ν) cos β + 5 (1 − 5 cos β) ,
2 2
(49)
2(7 − 5ν) r3 r
σ
( 3
) 3a 5
)
app a
σββ = σapp sin β +
2
4 − 5ν + 5(1 − 2ν) cos β + 5 (3 − 7 cos β) ,
2 2
(
(50)
2(7 − 5ν) r 3 r
5a 3 12a 5
( ( ))
1
τrβ = σapp −1 + − 3 (1 + ν) + 5 sin β cos β. (51)
(7 − 5ν) r r
τr θ = 0; τθβ = 0 (52)

The finite element meshes that we used for verification in the 3D problem are shown in Fig. 7.
We employed (a) the relative L 2 error norm, (b) the maximum principal stress fields, and (c) the maximum stress
as parameters for verification. The relative L 2 error norm of the displacement is defined as follows:

|uh − uex |2

eL 2 = dΩ . (53)
Ω |uex |2
Accuracies of the maximum principal stress fields were evaluated using the differences between the numerical
results σmax and the exact solutions σmax
ex
. The maximum stress was the most important factor affecting the structural
11
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 8. Convergence of the relative L 2 error norm for conventional methods. (a) Results of 2D problem, (b) Results of 3D problem.

integrity. It should be found on the surface of the hole so that the S-method exhibits better accuracy than the standard
FEM.

3.2. Results

The convergence of the relative L 2 error norm evaluated against the degrees of freedom is shown in Fig. 8.
All the meshes showed optimal convergence with a slope of 1 in the 2D problem and 2/3 in the 3D problem.
However, the S-method (non-matching) showed serious errors, which were much worse than those of other meshes.
The S-method (matching) showed a slightly lower accuracy than that of the standard FEM. One possible reason for
this is the relative L 2 error norm is a parameter that represents the average error in the entire domain, so that the
S-method is not advantageous compared to the standard FEM.
Figs. 9 and 10 show the numerical results of the maximum principal stress fields and their differences from the
exact solution for the 2D and 3D problems, respectively. In the 3D problem results, we only present the result of the
local domain for the cases with the S-method. In the results of the standard FEM and the S-method (matching), the
smooth, natural stress distributions, and sufficiently good accuracies were confirmed. On the other hand, the results
of the S-method (non-matching) showed large fluctuations causing serious errors in the stress field. Although the
maximum principal stress was almost zero in most parts of the hole, non-zero values were observed in the vicinity
of the surface of the hole and even inside the hole.
The numerical results of the maximum stress for the 2D problem, which should be found at (x, y) = (0, 1),
normalised by the exact solution, are shown { in Fig. 11(a) and (b).
} For the 3D problem, the maximum stress should
be found on the boundary Γs (x, y, z) = x 2 + y 2 = 1, z = 0 , and the normalised maximum stress is shown in
Fig. 11(c) and (d). The S-method (matching) exhibited the best performance, although some fluctuations were
observed in the results. That is, the advantage of the S-method, which is the high accuracy owing to the local
mesh, could be confirmed. However, the S-method (non-matching) showed large errors and instability in the results.
Additionally, the dependence of the normalised stress on the degrees of freedom was completely different from those
of the standard FEM and S-method (matching).

3.3. Discussions

As shown in Section 3.2, using the S-method (non-matching), the values of the stress fields in most hole parts
become almost zero (see Fig. 9). However, we found large errors in the numerical results of the displacement and
stress fields, which cannot be neglected in actual use. In this section, we discuss the following two aspects of the
results based on the weak form of the S-method with Neumann boundary conditions of the local mesh: (i) the
12
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 9. Maximum principal stress fields and their differences from the exact solution for the conventional methods for the 2D problem.

Fig. 10. Maximum principal stress fields and their differences from the exact solution for the conventional methods for the 3D problem.
(For cases with the S-method, the results are only in the local domain).

13
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 11. Convergence of the normalised maximum stress, σmax /σmax


ex , against the degrees of freedom for the conventional methods. (a) Results

of 2D problem, (b) Results of 3D problem.

possibility of simulating an internal traction-free boundary and (ii) the possible cause of the non-negligible errors
in the conventional approach.

3.3.1. Possibility of simulating an internal traction-free boundary in the conventional approach


The advantage of the conventional method is that the target domain can be set as Ω G but not Ω . That is, in the
conventional method [39,41,42,44,46–49], Ω G can be discretised more easily than Ω because a global mesh can be
generated without considering the geometries of the internal boundaries. Therefore, by setting target domain as Ω G
L
instead∫ of Ω , the weak form
∫ ∫ using Ω0 , Ω ∫and Ωin as:
Eq. (24) can be rewritten
δεG G
δεG G L
δεG G
δuG
( )
h : σh dΩ + h : σh + σh dΩ + h : σh dΩ − h · tdΓ
Ω0 Ω L Ωin Γt
∫ (54)
δεLh : σG L
( )
+ h + σh dΩ = 0,
ΩL
( )
where the domain Ω0 is defined as the complement of Ω L ∪ Ωin in Ω G , i.e., Ω0 = Ω G \ Ω L ∪ Ωin .
Considering
∫ Eqs. (18) and (19)∫ and applying Gauss’s divergence ∫ theorem to Eq. (54) gives
δuh · ∇ · σh dΩ −
G G
δuh · ∇ · σh + σh dΩ −
G
δuG
( ) [ ( G L
)] ( G
)
− h · ∇ · σh dΩ
Ω ΩL Ωin
∫ 0 ∫
δuh · σh n0 − t dΓ +
G G
δuh · σh n0 + σh + σh nL dΓ
G
( ) [ G ( G L
) ]
+
Γt Γ GL
∫ ∫ (55)
δuG δuLh · ∇ · σG
[( G L
) G
] [ ( L
)]
+ h · σ h + σ h nL + σ h n in dΓ − h + σh dΩ
Γin ΩL
∫ ∫
δuLh · σG L
δuLh · σG L
[( ) ] [( ) ]
+ h + σ h nL dΓ + h + σh nL dΓ = 0,
Γ GL Γin

where n0 , nL and nin are the outward unit normal vectors for the domains Ω0 , Ω L and Ωin , respectively. These
outward unit normal vectors satisfy the following conditions:

n0 = n on Γt , (56)
n0 = −nL on Γ GL
, (57)
nL = −nin = n on Γin . (58)
14
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Substituting Eqs. (56), (57), and (58) in Eq. (55) gives


∫ ∫ ∫
δuG G
δu G
δuG
( ) [ ( G L
)] ( G
)
− h · ∇ · σ h dΩ − h · ∇ · σ h + σ h dΩ − h · ∇ · σh dΩ
L
∫ Ω0 ∫Ω ∫ Ωin

δuh · σh n − t dΓ +
G G
δuh · σh nL dΓ +
G
δuh · σLh n dΓ
( ) ( L ) G
( )
+
Γ GL
∫Γt ∫ Γin
(59)
δuLh · ∇ · σG L
δuLh · σG L
[ ( )] ( )
− h + σh dΩ + h + σh nL dΓ
L Γ GL
∫Ω
δuh · σh + σh n dΓ = 0.
L
( G L
)
+
Γin

Because we can choose arbitrary values for the weight function δuGh and δuh , except for δuh on Γ
L L GL
due to Eq. (22),
we obtain the traction-free condition on the internal boundary from the last term as
on Γin .
( G
σh + σLh n = 0
)
(60)
Similarly, from the third term, we can obtain:
∇ · σG
h =0 in Ωin . (61)
If we apply the traction-free boundary condition expressed as Eq. (26), in addition to Eqs. (60) and (61), we can
obtain:
σG
h =0 in Ωin . (62)
A similar discussion can be found in Nakasumi et al. [41]. In the aforementioned formulations, domains Ω0 and Ωin
work as the fictitious domain, and those formulations may be the reason for the possibility of simulating an internal
traction-free boundary using Neumann boundary conditions in the local mesh of the S-method (non-matching), as
shown in Fig. 9.

3.3.2. Possible cause of the non-negligible errors in the conventional approach


Although the traction-free condition in Eq. (60) derived from the weak form is apparently correct, the actual
numerical results in Section 3.2 showed non-negligible errors in the S-method (non-matching).
Because the boundary conditions in Eq. (26) is imposed directly, Eq. (60) can be rewritten as follows:
σG
hn = 0 on Γin . (63)
However, it may be difficult to satisfy Eq. (63) using the approximation in Eq. (20) based on the global elements,
whose boundaries generally do not coincide with the internal traction-free boundary Γin .
If the traction-free condition, whose discretised form expressed in Eq. (60) can be appropriately imposed, good
accuracy is expected to be obtained. Based on this consideration, the weak form of Eq. (24) is redefined by
considering the traction-free condition in Eq. (60) as an additional boundary condition as follows:
∫ ∫ ∫
δεh : σh dΩ −
G
δuh · tdΓ +
G
δεLh : σh dΩ
ΩG Γt ΩL
∫ (64)
δuG
[( G L
) ]
+ h · σ h + σ h n dΓ = 0.
Γin

The traction-free boundary condition in Eq. (26) can be imposed explicitly on the boundaries of the local mesh,
thus we obtain:
∫ ∫ ∫
δεG
h : σ h dΩ − δuG
h · tdΓ + δεLh : σh dΩ
ΩG Γt ΩL
∫ (65)
δuh · σh n dΓ = 0.
G
( G )
+
Γin

15
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

By comparing the original weak form of Eq. (24) and redefining Eq. (60), the latter has an additional term for

δuGh · σh n dΓ .
( G )
(66)
Γin

This term may not be removed using a relatively coarse global mesh because of the mismatch between the internal
boundary and the global mesh. In other words, ignoring this term in the original weak form of Eq. (24) may cause
errors in the S-method (non-matching).

4. Proposed methods
According to the discussions in Section 3.3, the weak form considering the traction-free condition on the internal
boundary, Eq. (65) has an additional term, Eq. (66) compared to the original equation, Eq. (24). This may be
the cause of serious errors in the conventional approach to model an internal traction-free boundary by Neumann
boundary conditions of the local mesh [21,26–36], as shown in Section 3.2. This is because the additional term in
Eq. (65) cannot be appropriately removed because the boundaries of the global elements and the internal boundary
are incompatible with the conventional approach. Therefore, the removal of the additional term, Eq. (66), in the
weak form of Eq. (65) is a possible solution to achieve accurate numerical simulations. In this section, we propose
three methods based on these considerations.

4.1. FCM-based method

The numerical manifold method (NMM) [57–60], or the finite cover method (FCM) [61,62], is a method
employing a dual-cover system: mathematical and physical covers. The physical cover is defined in the domain
of the target problem. Physical values, such as material properties, are defined as non-zero values only on the
physical cover and zero outside the physical cover. The mathematical cover is defined to cover all the domains of
the target problem but is not required to correspond to the geometry of the target problem domain. In the finite
element framework, an approximation using the shape function was applied to the mathematical cover. Thus, a
finite element mesh was generated based on the mathematical cover. The dual-cover system in the FCM has an
advantage in modelling arbitrary boundaries based on the geometry of the physical cover, independent of that of
the mathematical cover (the finite element mesh).
Based on the FCM, the free surface can be simulated by setting Young’s modulus to zero outside the target
domain [63]. This approach to the problem of a plate with an internal traction-free boundary, which is the target
problem in this study, as shown in Fig. 2, is expressed as
E =0 in Ωin . (67)
A schematic representation of this approach is presented in Fig. 12. Thus, we easily obtain:
σG
h =0 in Ωin . (68)
By applying Gauss’s divergence theorem to the additional term, Eq. (66), we obtain:
∫ ∫
δuG G
h − δεh : σh dΩ .
δuh · ∇ · σG
[ G ( ) G G
]
h · σ h n dΓ = (69)
Γin Ωin

By substituting Eq. (68) to the right side of Eq. (69), we find that the additional term, Eq. (66), becomes zero: This
implies that this approach can eliminate the additional term.
However, the integration of Eq. (69) may not be accurate because all the parameters in the integrand are defined
based on the global elements, but the internal boundary does not correspond to the boundaries of the global elements.
To address this problem, the set of elements m G in is defined as:

in = {e|Ωe ∩ Γin ̸ = ∅} ,
mG (70)
where Ωe is the domain of element e. The global elements included in mGhave discontinuities in their stiffness.
in
Therefore, the recursive subdivision method presented in Section 2.2 is applied to the global elements included in
mG
in (see Fig. 12). In this study, we set the number of recursions for this method to n rec = 3.
16
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 12. FCM-based method (n rec = 3).

4.2. IBM-based method

The immersed boundary method (IBM) is a numerical method mainly used in fluid dynamics to solve the moving
boundary problems [64–68]. IBM uses Lagrange points, which are basically located on the surface of boundaries
and can be independently defined from a mesh for numerical computation. Based on the Dirac delta function or
its smoothed functions, IBM can impose boundary conditions at the Lagrange points [69]. According to numerical
results and discussions presented in Sections 3.2 and 3.3, respectively, the additional term, Eq. (66), may be a vector
with non-zero values corresponding to the global nodes. In this approach, we modify the IBM formulation based
on a Dirac delta function [69] to satisfy the boundary condition in the global mesh, Eq. (63). We use the surface
nodes of the local mesh as the Lagrange points.
Using the stress tensor on an internal boundary, the equivalent local nodal force at the kth local node on Γin , f̃kL ,
is defined based on the local shape function as

f̃kL = NkL (x) (σh (x) n (x)) dΓ . (71)
Γin

Because of the traction-free boundary conditions in Eq. (26), which was explicitly applied to Eq. (71), it
becomes:

f̃kL = h (x) n (x) dΓ .
NkL (x) σG
( )
(72)
Γin

Fig. 13 shows the local nodes where the equivalent local nodal force f̃kL is calculated.
We assume that f̃kL can be approximated as a point load at xLk using the Dirac delta function, as follows
L
n in

σLh (x) n (x) ≈ f̃kL δ(xLk − x), (73)
k=1

where n Lin is the number of local nodes on Γin , and xLk is the coordinate of the kth local node. Here, we determine
the equivalent global nodal force on each global node, which cancels these point loads to satisfy the traction-free
condition (Eq. (60)).
17
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 13. IBM-based method.

The equivalent global nodal force at the ith global node, f̃iG , can be approximated using Eqs. (60) and (73), as
follows

G
NkG (x) σGh (x) n (x) dΓ
( )
f̃i =
Γin

NkG (x) −σLh (x) n (x) dΓ
( )
=
Γin
n in L (74)

NiG xk f̃k .
( L) L
≈−
k=1

By substituting Eq. (72) in Eq. (74), multiplying the weight δdiG on both sides of the equation, and arranging
the equation, we obtain:
L
n in
∫ ∑
δdiG f̃iG −δdiG h (x) n (x) dΓ .
NiG xLk NkL (x) σG
( ) ( )
· = · (75)
Γin k=1

Owing to
L
n in

NiG (x) ≈ NiG xLk NkL (x) on Γin ,
( )
(76)
k=1

we obtain:

δdiG · f̃iG ≈ −δdiG · h (x) n (x) dΓ .
NiG (x) σG
( )
(77)
Γin

The summation of Eq. (77) for all the global nodes is


n G n ( G ∫ )
∑ ∑
δdiG f̃iG δdiG NiG (x) σG (x) n (x) dΓ .
( )
· ≈− · h (78)
i=1 i=1 Γin

18
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

The additional term Eq. (66) can be written as:


∫ ∫ ∑ nG
δuh · σh n dΓ =
G G
NiG (x) δdiG · σG
h n dΓ . (79)
Γin Γin i=1

By arranging the term on the right side of Eq. (79), we obtain:


∫ ∑nG ( ∫ )
δuh · σh n dΓ =
G G
δdiG · NiG (x) σG
h n dΓ . (80)
Γin i=1 Γin

By comparing Eqs. (78) and (80), the additional term, Eq. (66), can be approximated by the equivalent global nodal
force f̃iG as
∫ n G

δuG
h · σG
hn dΓ ≈ − δdiG · f̃iG . (81)
Γin i=1

4.3. XFEM-based method

The extended finite element method (XFEM) extends the finite element approximation by adding enrichment
functions and relating degrees of freedom to simulate discontinuities or singularities [70,71]. Using the XFEM,
we can avoid the complicated procedure of mesh generation considering the geometries of the discontinuities
[70–75].
In this proposed method, we employ the Heaviside function H (x), which is defined as
{
−1 in Ωin
H (x) = , (82)
1 in Ω
to the global elements to simulate traction-free boundaries. Using Heaviside enrichment and approximating
Eqs. (12), (14), and (20), they are modified as
∑ ∑
uG
h = NiG (x) diG + H (x)NiG (x) diGX , (83)
i∈n G i∈n GX
∑ ∑
δuG
h = NiG (x) δdiG + H (x)NiG (x) δdiGX , (84)
i∈n G i∈n GX
∑ ∑
σG
h =D BiG (x) diG + D H (x) BiG (x) diGX , (85)
i∈n G i∈n GX

where diGX and δdiGX are the additional nodal degrees of freedom corresponding to Heaviside enrichment, and n GX
is a set of enriched nodes defined as
n GX = i|ΩiG ∩ Γin ̸= ∅ ,
{ }
(86)
where ΩiG is the support of global nodes i, which is defined as
ΩiG = x|NiG (x) > 0 .
{ }
(87)
GX
A schematic of the set of enriched nodes n is shown in Fig. 14.
Because the Heaviside function can introduce strong discontinuities to the finite-element approximation, we can
assume that the global mesh is separated into two domains by Heaviside enrichment. One domain corresponds to
the domain Ω , and the other domain corresponds to the inside domain of the internal traction-free boundary Ωin .
Therefore,
∫ the redefined∫ weak form, Eq. (65),
∫ can be rewritten
∫ as follows:
δεh : σh dΩ +
G
δεh : σh dΩ −
G
δuh · tdΓ +
G
δεLh : σh dΩ
Ω Ωin Γt ΩL
∫ (88)
+ δuGh · σG
h n dΓ = 0.
Γin
19
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 14. XFEM-based method (n rec = 3).

By applying Gauss’s divergence theorem to the first, second, and third terms in Eq. (88), and considering Eqs. (10)
and (26),∫ we obtain: ∫ ∫
− δuh · (∇ · σh ) dΩ + 2
G
δuh · σh n dΓ +
G G
δuG G
h · σh n dΓ
ΩG Γin Γin−
∫ ∫ (89)
− δuLh · (∇ · σh ) dΩ + δuLh · σG
h n dΓ = 0,
ΩL Γin

where Γin− is the outer boundary of Ωin , which is different from Γin , the inner boundary of Ω .
As δuGh and δuh are arbitrary weight functions, if we assume δuh = 0 on Γin ∪ Γin− and δuh = 0 on Ω ∪ Γin ,
L G L L

we obtain:

δuG
h · (∇ · σh ) dΩ = 0. (90)
ΩG

Then, if we assume δuG h = 0 on Γin ∪ Γin− and δuh = 0 on Γin , we obtain:


L

δuLh · (∇ · σh ) dΩ = 0. (91)
ΩL

Then, if we assume δuLh = 0 on Γin , we obtain:


∫ ∫
2 δuGh · σG
h n dΓ + δuG G
h · σh n dΓ = 0. (92)
Γin Γin−

Because δdiG and δdiGX are arbitrary weights, if we assume δdiG = −δdiGX , we obtain δuG
h = 0 on Γin using Eqs. (82),
(84), (86), and (87). Therefore, the second term in Eq. (92) becomes:

δuG G
h · σh n dΓ = 0. (93)
Γin−

Substituting Eq. (93) in Eq. (92), we obtain:



δuG G
h · σh n dΓ = 0. (94)
Γin

According to the aforementioned derivation, the additional term Eq. (66) becomes zero: Therefore, by applying
Heaviside enrichment to the approximation in the global elements, the additional term in the redefined weak form,
Eq. (65), is successfully vanished for the problem of a solid body with an internal traction-free boundary.
20
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

The elements included in m Gin , which are defined in Section 4.1, have discontinuities in their domains owing to
Heaviside enrichment. Therefore, the recursive subdivision method is applied to the elements included in m Gin in the
same manner as the procedure in Section 4.1 (see Fig. 14).
By employing Heaviside enrichment, Eqs. (34) and (37) become
⎧ G ⎫

⎪ d1 ⎪ ⎪
G ⎪
⎪ ⎪
d

⎪ ⎪

⎪ 2 ⎪

⎪ .. ⎪

⎪ ⎪

.
⎪ ⎪

⎪ ⎪


⎪ ⎪

⎨ dG ⎪
⎪ ⎬
nG
G
d = , (95)
⎪ dGX
⎪ 1 ⎪

⎪ ⎪
⎪ dGX ⎪

⎪ ⎪

⎪ 2 ⎪

⎪ ⎪
.

⎪ .. ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪


⎩ GX ⎪
dn GX

⎧ ⎫
fG ⎪
⎪ 1 ⎪

⎪ ⎪
⎪ G ⎪
⎪ f2 ⎪

⎪ ⎪

⎪ .. ⎪

⎪ ⎪


.
⎪ ⎪


⎪ ⎪

⎪ ⎪
⎨ fG ⎪
⎪ ⎬
G
f = n G
, (96)
⎪ f GX ⎪ ⎪


⎪ 1 ⎪
⎪ GX ⎪
f ⎪
⎪ ⎪
⎪ 2 ⎪

⎪ ⎪
.. ⎪
⎪ ⎪


. ⎪

⎪ ⎪

⎪ ⎪

⎩f GX ⎪
⎪ ⎭
n GX

where

fiGX = H (x)NiG (x) tdΓ . (97)
Γt

5. Verification of proposed methods


5.1. Target problems

To strictly verify the proposed methods, they are applied to analyse the same problems represented in Section 3.1.
The first one is a 2D problem of an infinite plate with a hole, which is a typical internal traction-free boundary under
uniaxial tension (see Fig. 4). The second problem is a classic 3D stress concentration problem: a spherical hole in an
infinite medium subjected to uniform tension at infinity (see Fig. 6). Meshes employed for the proposed methods are
the same as those for the S-method (non-matching) presented in Section 3.1 (see Figs. 5 and 7). In other words, the
global mesh is generated independently of the geometry of the hole, and the local mesh is defined around the hole
with sufficiently fine elements. Meshes for the proposed methods have a great advantage for simplification in the
meshing procedure compared to those for the standard FEM and the S-method (matching) presented in Section 3.1.
The numerical results of the proposed methods were verified in addition to those of the conventional methods
presented in Section 3.1. We employed (a) the relative L 2 error norm, (b) maximum principal stress fields, and (c)
maximum stress as parameters for the verification, as described in Section 3.

5.2. Results

The convergence of the relative L 2 error norm defined in Eq. (53) was evaluated for the proposed methods.
The numerical results for the convergence of the relative L 2 error norm against the degrees of freedom are shown
in Fig. 15. In these results, all the proposed methods exhibited better accuracy compared to the original S-method
21
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 15. Convergence of the relative L 2 error norm for the proposed methods compared with those for the conventional methods. (a) Results
of 2D problem, (b) Results of 3D problem.

(non-matching). Particularly, the XFEM-based method showed significant improvements, the accuracy of which was
even better than that of the original S-method (matching). Although the accuracy of the XFEM-based method was
slightly lower than that of the standard FEM, this may be considered as a natural result because the relative L 2
error norm represents the average error in the displacement in the entire domain.
Similar to the verification in Section 3.2, the accuracies of the maximum principal stress fields were evaluated
using the differences between the numerical results σmax and the exact solutions σmax ex
. Figs. 16 and 17 show the
distributions of σmax and σmax − σmax for the 2D and 3D problems, respectively. In these results, all the proposed
ex

methods showed improvements in accuracy compared to the original S-method (non-matching). Particularly, the
XFEM-based method showed very smooth stress distributions, and the errors were significantly lower than those
of the original S-method (non-matching), although the same global mesh was employed.
Numerical results of the maximum stress for the 2D problem, which should be found at (x, y) = (1, 0),
normalised by the exact solution, are shown { in Fig. 18(a) and (b).
} For the 3D problem, the maximum stress should
be found on the boundary Γs (x, y, z) = x 2 + y 2 = 1, z = 0 , and the normalised maximum stress is shown in
Fig. 18(c) and (d). Large errors and fluctuations can be found in the results of the FCM-based method, which were
even worse than the original S-method (non-matching), and the tendency of the dependence of the normalised stress
on the degrees of freedom was different from that of the other methods. Conversely, the IBM-based and XFEM-
based methods showed better accuracies than the original S-method (non-matching). Particularly, in the results of
the XFEM-based method, the degrees of freedom that required to achieve the same accuracy were significantly
less than those of the standard FEM. Although the IBM-based method sometimes showed better accuracy than the
XFEM-based method, the results of the XFEM-based method showed significantly more stable convergence than
those of the IBM-based method.
The possible reasons for the lower accuracy of FCM- and IBM-based methods compared to the XFEM-based
method are as follows. In the FCM-based method, Young’s modulus for the domain of the hole is set to be zero,
and for the other domain, Young’s modulus has the normal value. The difference between Young’s modulus in the
two domains will cause discontinuities in the stiffness of the global elements included in m Gin (Eq. (61)); however,
the shape functions used for stiffness are continuous. The discrepancy of continuity between Young’s modulus
and shape functions might be the reason for the unsatisfactory results. In the IBM-based method, the traction-free
condition is satisfied by cancelling the additional nodal force for the local nodes on the boundary of the hole
with equivalent global nodal force. To calculate the equivalent global nodal force, two approximations are applied
(Eqs. (64) and (67)), implying that the additional local nodal force on the boundary of the hole might not be
completely cancelled. On the other hand, the XFEM-based method applied the Heaviside function to simulate the
traction-free boundary condition. Because the Heaviside function completely coincides with the boundary of the hole
22
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 16. Maximum principal stress fields and their difference from the exact solution for 2D problem evaluated by proposed methods
compared to those by FEM and standard S-method.

Fig. 17. Maximum principal stress fields and their difference from the exact solution for the 3D problem in local domain evaluated by the
proposed methods compared with those by the FEM and standard S-method.

23
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Fig. 18. Convergence of normalised maximum stress, σmax /σmax ex , evaluated by proposed methods compared with those by FEM and

standard S-method: (a) Results of the 2D problem in the range 0.80 ≤ σmax /σmaxex ≤ 1.20, (b) Results of the 2D problem in the range

0.96 ≤ σmax /σmax


ex ≤ 1.04, (c) Results of the 3D problem in the range 0.50 ≤ σ
max /σmax ≤ 1.20, (d) Results of the 3D problem in the range
ex

0.96 ≤ σmax /σmax ≤ 1.04.


ex

and strong discontinuities are introduced to the finite-element approximation by the Heaviside function to simulate
the traction-free boundary, this proposed method enables us to simulate discontinuities consummately. Therefore,
the XFEM-based method may produce the best accuracy in the proposed methods.
The degrees of freedom for the standard FEM and the XFEM-based method to achieve the maximum stress error
of 2%, which are evaluated by applying the linear interpolation to the results presented in Fig. 18, are summarised
in Table 1. The results clearly showed that, compared to the standard FEM in both the 2D and 3D problems, the
proposed XFEM-based method requires much lower numerical costs to achieve the same accuracy.
The aforementioned verification results demonstrated that the proposed XFEM-based method makes it possible
to utilise the intrinsic strengths of the S-method, which are (i) local high accuracy with low numerical costs and (ii)
simplicity in the meshing procedure, even in the problem of a solid body with an internal traction-free boundary.

6. Conclusions
In this study, we established a strategy based on the S-method for accurately and efficiently modelling an internal
traction-free boundary using Neumann boundary conditions of the local mesh with the Heaviside enrichment of the
24
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Table 1
Required degrees of freedom to achieve the maximum stress error of 2% for the standard FEM
and XFEM-based methods.
Degrees of freedom Efficiency improvements
FEM (2D) 4181 –
XFEM-based method (2D) 1899 54.6%
FEM (3D) 266,109 –
XFEM-based method (3D) 93,793 64.8%

XFEM, which makes it possible to utilise the intrinsic strengths of the S-method, that is, (i) local high accuracy
with low numerical costs, and (ii) simplicity of the meshing procedure.
First, we performed strict verification of the approach used in conventional studies to model an internal traction-
free boundary using Neumann boundary conditions of the local mesh by analysing a problem whose exact solution
is known. In the numerical results obtained using the global mesh generated independent of the geometry of the
internal traction-free boundary, non-negligible serious errors in the displacement and stress fields were clarified.
Through discussions based on the weak form of the S-method, we found a possible cause of the serious errors due
to the existence of the additional term in the weak form caused by mismatching between the internal traction-free
boundary and the global mesh.
Based on this discussion, three strategies were proposed to solve the clarified issue by eliminating the additional
term in the weak form: (i) FCM-based, (ii) IBM-based, and (iii) XFEM-based methods. In the FCM-based method,
the additional term was eliminated by setting Young’s modulus to zero inside the internal traction-free boundary.
In the IBM-based method, the additional term is directly computed as forces on the surface local nodes. In the
XFEM-based method, the additional term was eliminated by employing Heaviside enrichment in the finite element
approximation of the global elements.
Similar to the conventional studies, the proposed methods were verified through a classical 3D stress concen-
tration problem: a spherical hole in an infinite medium subjected to uniform tension at infinity. Numerical results
obtained using the proposed methods showed better accuracies than those obtained using the conventional approach.
Particularly, the XFEM-based method showed significantly high accuracy in numerical results of the displacement
and stress fields. That is, the clarified issue in the S-method to model an internal traction-free boundary using the
Neumann boundary conditions of the local mesh was successfully solved by the proposed XFEM-based method.
Consequently, the intrinsic strengths of the S-method are effectively utilised using the proposed XFEM-based
method, even for the problems of a solid body with an internal traction-free boundary. Compared to the traditional
FEM and X-FEM methods, the proposed XFEM-based method has two major advantages: (i) The advantages of the
S-method can be directly utilised. For example, local fine meshes can be superposed arbitrary parts of the domain,
thus eliminating the need of mesh refinement procedures for the global mesh. That is, the entire domain can be
simply modelled using a coarse global mesh without any consideration of the mesh refinement to the stress/strain
concentration regions due to the internal boundaries. (ii) Because the local mesh is modelled considering the
geometry of discontinuities, the Heaviside function can be employed easily along the edge of the local mesh. The
internal traction-free boundaries are likely to generate stress concentration. Because the local mesh automatically
covers the stress concentration domain with a fine mesh resolution, the procedure to generate the numerical model
is significantly simplified. The present study showed applications to the 2D and 3D problems, and the proposed
method performed perfectly in both the problems. Therefore, the established strategy based on a combination of
the S-method and the Heaviside enrichment in the XFEM has the potential to be used as a basis for analysing the
engineering problems of complicated structures.

Declaration of competing interest


The authors declare the following financial interests/personal relationships which may be considered as potential
competing interests: Kazuki Shibanuma reports financial support was provided by Japan Society for the Promotion
of Science (JSPS). Naoki Morita reports computational resources was provided by Joint Usage/Research Center for
Interdisciplinary Large-scale Information Infrastructures (Project ID: jh210046-NAH and jh220047).
25
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

Data availability
No data was used for the research described in the article.

Funding
This study was supported by JSPS KAKENHI [grant number 22H00242].

References
[1] P. Hennig, M. Kästner, P. Morgenstern, D. Peterseim, Adaptive mesh refinement strategies in isogeometric analysis— A computational
comparison, Comput. Methods Appl. Mech. Engrg. 316 (2017) 424–448, http://dx.doi.org/10.1016/j.cma.2016.07.029.
[2] S. Zhang, A.L. Gain, J.A. Norato, Adaptive mesh refinement for topology optimization with discrete geometric components, Comput.
Methods Appl. Mech. Engrg. 364 (2020) 112930, http://dx.doi.org/10.1016/j.cma.2020.112930.
[3] J. Zhang, S. Natarajan, E.T. Ooi, C. Song, Adaptive analysis using scaled boundary finite element method in 3D, Comput. Methods
Appl. Mech. Engrg. 372 (2020) 113374, http://dx.doi.org/10.1016/j.cma.2020.113374.
[4] W. Li, N. Nguyen-Thanh, J. Huang, K. Zhou, Adaptive analysis of crack propagation in thin-shell structures via an isogeometric-meshfree
moving least-squares approach, Comput. Methods Appl. Mech. Engrg. 358 (2020) 112613, http://dx.doi.org/10.1016/j.cma.2019.112613.
[5] S. Sarkar, I.V. Singh, B.K. Mishra, Adaptive mesh refinement schemes for the localizing gradient damage method based on
biquadratic-bilinear coupled-field elements, Eng. Fract. Mech. 223 (2020) 106790, http://dx.doi.org/10.1016/j.engfracmech.2019.106790.
[6] J. Gu, T. Yu, L. Van Lich, S. Tanaka, H. Yuan, T.Q. Bui, Crack growth adaptive XIGA simulation in isotropic and orthotropic materials,
Comput. Methods Appl. Mech. Engrg. 365 (2020) 113016, http://dx.doi.org/10.1016/j.cma.2020.113016.
[7] D. Koliesnikova, I. Ramière, F. Lebon, A unified framework for the computational comparison of adaptive mesh refinement strategies
for all-quadrilateral and all-hexahedral meshes: Locally adaptive multigrid methods versus h-adaptive methods, J. Comput. Phys. 437
(2021) 110310, http://dx.doi.org/10.1016/j.jcp.2021.110310.
[8] D. Manta, R. Gonçalves, D. Camotim, Combining shell and GBT-based finite elements: Vibration and dynamic analysis, Thin-Walled
Struct. 167 (2021) 108187, http://dx.doi.org/10.1016/j.tws.2021.108187.
[9] D. van Huyssteen, F.L. Rivarola, G. Etse, P. Steinmann, On mesh refinement procedures for the virtual element method for
two-dimensional elastic problems, Comput. Methods Appl. Mech. Engrg. 393 (2022) 114849, http://dx.doi.org/10.1016/j.cma.2022.
114849.
[10] M. Dinachandra, A. Alankar, Adaptive finite element modeling of phase-field fracture driven by hydrogen embrittlement, Comput.
Methods Appl. Mech. Engrg. 391 (2022) 114509, http://dx.doi.org/10.1016/j.cma.2021.114509.
[11] J. Fish, The s-version of the finite element method, Comput. Struct. 43 (1992) 539–547, http://dx.doi.org/10.1016/0045-7949(92)90287-
A.
[12] S.L. Angioni, A. Visrolia, M. Meo, Combining X-FEM and a multilevel mesh superposition method for the analysis of thick composite
structures, Composites B 43 (2012) 559–568, http://dx.doi.org/10.1016/j.compositesb.2011.07.005.
[13] S.-H. Lee, J.-H. Song, Y.-C. Yoon, G. Zi, T. Belytschko, Combined extended and superimposed finite element method for cracks,
Internat. J. Numer. Methods Engrg. 59 (2004) 1119–1136, http://dx.doi.org/10.1002/nme.908.
[14] J. Fish, V. Belsky, M. Pandheeradi, Composite grid method for hybrid systems, Comput. Methods Appl. Mech. Engrg. 135 (1996)
307–325, http://dx.doi.org/10.1016/0045-7825(95)00965-5.
[15] J. Fish, A. Wagiman, Multiscale finite element method for a locally nonperiodic heterogeneous medium, Comput. Mech. 12 (1993)
164–180, http://dx.doi.org/10.1007/BF00371991.
[16] S. Ye, C.-C. Zhang, P.-Y. Zhang, X.-C. Zhang, S.-T. Tu, R.-Z. Wang, Fatigue life prediction of nickel-based GH4169 alloy on the basis
of a multi-scale crack propagation approach, Eng. Fract. Mech. 199 (2018) 29–40, http://dx.doi.org/10.1016/j.engfracmech.2018.05.023.
[17] P. Cheng, H. Zhu, Z. Yan, Y. Shen, J. Fish, Multiscale modeling for fire induced spalling in concrete tunnel linings based on the
superposition-based phase field fracture model, Comput. Geotech. 148 (2022) 104832, http://dx.doi.org/10.1016/j.compgeo.2022.104832.
[18] N. Shogo, K. Suzuki, O. Hideomi, Numerical Meth Engineering - 2007 - Nakasumi - Crack growth analysis using mesh superposition
technique and X-FEM.pdf, Internat. J. Numer. Methods Engrg. 75 (2008) 291–304.
[19] M. Kikuchi, Y. Wada, Y. Shimizu, Y. Li, Crack growth analysis in a weld-heat-affected zone using S-version FEM, Int. J. Press.
Vessel. Pip. 90–91 (2012) 2–8, http://dx.doi.org/10.1016/j.ijpvp.2011.10.001.
[20] M. Kikuchi, Y. Wada, Y. Shintaku, K. Suga, Y. Li, Fatigue crack growth simulation in heterogeneous material using s-version FEM,
Int. J. Fatigue. 58 (2014) 47–55, http://dx.doi.org/10.1016/j.ijfatigue.2013.04.022.
[21] M. Kikuchi, Y. Wada, Y. Li, Crack growth simulation in heterogeneous material by S-FEM and comparison with experiments, Eng.
Fract. Mech. 167 (2016) 239–247, http://dx.doi.org/10.1016/j.engfracmech.2016.03.038.
[22] K. Suga, M. Kikuchi, Y. Wada, H. Kawai, Study on fatigue growth of multi-surface flaws in shaft under rotary bending by S-FEM,
Eng. Fract. Mech. 174 (2017) 207–214, http://dx.doi.org/10.1016/j.engfracmech.2016.11.001.
[23] Z. Yue, D.H. Robbins, Adaptive superposition of finite element meshes in elastodynamic problems, Internat. J. Numer. Methods Engrg.
63 (2005) 1604–1635, http://dx.doi.org/10.1002/nme.1331.
[24] X. Qiang, C. Jianyun, Y. Hongyuan, L. Jing, A study on the S-version FEM for a dynamic damage model, Internat. J. Numer. Methods
Engrg. 115 (2018) 427–444.
[25] W. Sun, J. Fish, Superposition-based coupling of peridynamics and finite element method, Comput. Mech. 64 (2019) 231–248,
http://dx.doi.org/10.1007/s00466-019-01668-5.
26
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

[26] J. Fish, S. Markolefas, The s-version of the finite element method for multilayer laminates, Internat. J. Numer. Methods Engrg. 33
(1992) 1081–1105, http://dx.doi.org/10.1002/nme.1620330512.
[27] J. Fish, Rav Guttal, The s-version of finite element method for laminated composites, 1996.
[28] J. Fish, A. Suvorov, V. Belsky, Hierarchical composite grid method for global-local analysis of laminated composite shells, Appl.
Numer. Math. 23 (1997) 241–258, http://dx.doi.org/10.1016/S0168-9274(96)00068-2.
[29] S.L. Angioni, A. Visrolia, M. Meo, A hierarchical multiple plate models theory for laminated composites including delamination and
geometrical nonlinear effects, Compos. Struct. 93 (2011) 780–791, http://dx.doi.org/10.1016/j.compstruct.2010.08.003.
[30] Y. Jiao, J. Fish, On the equivalence between the s-method, the XFEM and the ply-by-ply discretization for delamination analyses of
laminated composites, Int. J. Fract. 191 (2015) 107–129, http://dx.doi.org/10.1007/s10704-015-9996-2.
[31] Y. Kumagai, S. Onodera, Y. Nagumo, T. Okabe, K. Yoshioka, Multiscale modeling of free-surface effect on crack formation in
unidirectional off-axis laminates, Composites A 98 (2017) 136–146, http://dx.doi.org/10.1016/j.compositesa.2017.03.016.
[32] S. Sakata, Y. Chan, Y. Arai, On accuracy improvement of microscopic stress/stress sensitivity analysis with the mesh superposition
method for heterogeneous materials considering geometrical variation of inclusions, Internat. J. Numer. Methods Engrg. 121 (2020)
534–559, http://dx.doi.org/10.1002/nme.6233.
[33] S. Sakata, S. Tanimasu, Mesh superposition-based multiscale stress analysis of composites using homogenization theory and re-
localization technique considering fiber location variation, Internat. J. Numer. Methods Engrg. 123 (2022) 505–529, http://dx.doi.
org/10.1002/nme.6865.
[34] J. Fish, Hierarchical modelling of discontinuous fields, Commun. Appl. Numer. Methods 8 (1992) 443–453, http://dx.doi.org/10.1002/
cnm.1630080704.
[35] R. Fan, J. Fish, The rs-method for material failure simulations, Internat. J. Numer. Methods Engrg. 73 (2008) 1607–1623.
[36] H. Okada, S. Endoh, M. Kikuchi, On fracture analysis using an element overlay technique, Eng. Fract. Mech. 72 (2005) 773–789,
http://dx.doi.org/10.1016/j.engfracmech.2004.05.003.
[37] K. Kishi, Y. Takeoka, T. Fukui, T. Matsumoto, K. Suzuki, K. Shibanuma, Dynamic crack propagation analysis based on the s-version
of the finite element method, Comput. Methods Appl. Mech. Engrg. 366 (2020) 113091, http://dx.doi.org/10.1016/j.cma.2020.113091.
[38] K. Shibanuma, K. Kishi, T. He, N. Morita, N. Mitsume, T. Fukui, S-version finite element strategy for accurately evaluating local
stress in the vicinity of dynamically propagating crack front in 3D solid, Comput. Methods Appl. Mech. Engrg. 399 (2022) 115374,
http://dx.doi.org/10.1016/j.cma.2022.115374.
[39] Y. Yumoto, Y. Yusa, H. Okada, Element subdivision technique for coupling-matrix-free iterative s-version FEM and investigation of
sufficient element subdivision, Mech. Eng. J. 3 (2016) 16–00361, http://dx.doi.org/10.1299/mej.16-00361.
[40] S. Katsuyuki, O. Hideomi, S. Takushiro, M. Seungjae, Improvement of the accuracy of zooming analysis using overlaying mesh method,
Nav. Arch. Japan (1999) 197–201, http://dx.doi.org/10.2534/jjasnaoe1968.1999.197.
[41] N. Shogo, S. Katsu, F. Daiji, O. Hideomi, A study on the analysis of the plate model with a hole by overlaying mesh method, Trans.
Japan Soc. Comput. Eng. Sci. 08 (2015) 8–11, http://dx.doi.org/10.11421/jsces.2001.20010016.
[42] X. Chen, Z. Li, H. Wang, Progressive failure analysis of an open-hole composite laminate by using the S-version finite-element method,
Mech. Compos. Mater. 50 (2014) 279–294, http://dx.doi.org/10.1007/s11029-014-9414-2.
[43] N. Takano, Y. Okuno, Three-scale finite element analysis of heterogeneous media by asymptotic homogenization and mesh superposition
methods, Int. J. Solids Struct. 41 (2004) 4121–4135, http://dx.doi.org/10.1016/j.ijsolstr.2004.02.049.
[44] Y. Yumoto, Y. Yusa, H. Okada, An s-version finite element method without generation of coupling stiffness matrix by using iterative
technique, Mech. Eng. J. 3 (2016) 16-00001, http://dx.doi.org/10.1299/mej.16-00001.
[45] N. Takano, M. Zako, Y. Okuno, Multi-scale finite element analysis of porous materials and components by asymptotic homogenization
theory and enhanced mesh superposition method, Model. Simul. Mater. Sci. Eng. 11 (2003) 137–156, http://dx.doi.org/10.1088/0965-
0393/11/2/303.
[46] S. Wang, M.Y. Wang, A moving superimposed finite element method for structural topology optimization, Internat. J. Numer. Methods
Engrg. 65 (2006) 1892–1922, http://dx.doi.org/10.1002/nme.1527.
[47] M. Kawagai, A. Sando, N. Takano, Image-based multi-scale modelling strategy for complex and heterogeneous porous microstructures
by mesh superposition method, Model. Simul. Mater. Sci. Eng. 14 (2006) 53–69, http://dx.doi.org/10.1088/0965-0393/14/1/005.
[48] Y. Yusa, J. Okamoto, D. Toyama, H. Okada, Analysis of a many-hole problem using coupling-matrix-free iterative s-version FEM with
multiple local meshes, Mech. Eng. J. 5 (2018) 18–00264, http://dx.doi.org/10.1299/mej.18-00264.
[49] K. Suzuki, O. Hideomi, M. Seungjae, S. Takushiro, Multi scale analysis of ship structure using overlaying mesh method, Trans. Japan
Soc. Comput. Eng. Sci. (1999) http://dx.doi.org/10.11421/jsces.1999.19990020.
[50] H. Okada, C.T. Liu, T. Ninomiya, Y. Fukui, N. Kumazawa, Analysis of particulate composite materials using an element overlay
technique, Comput. Model. Eng. Sci. 6 (2004) 333–347, http://dx.doi.org/10.3970/cmes.2004.006.333.
[51] Z.E. Stowell, Stress and strain concentration at a circular, Natl. Advis. Comm. Aeronaut. (1950).
[52] R.R. Hiemstra, G. Sangalli, M. Tani, F. Calabrò, T.J.R. Hughes, Fast formation and assembly of finite element matrices with application
to isogeometric linear elasticity, Comput. Methods Appl. Mech. Engrg. 355 (2019) 234–260, http://dx.doi.org/10.1016/j.cma.2019.06.020.
[53] J.N. Goodier, P.G. Hodge Jr., Elasticity and Plasticity: The Mathematical Theory of Elasticity and the Mathematical Theory of Plasticity,
Courier Dover Publications, 2016.
[54] A.F. Bower, Applied Mechanics of Solids, CRC Press, 2009.
[55] J.R. Barber, Elasticity, Springer, 2002.
[56] M.A. Scott, R.N. Simpson, J.A. Evans, S. Lipton, S.P.A. Bordas, T.J.R. Hughes, T.W. Sederberg, Isogeometric boundary element analysis
using unstructured T-splines, Comput. Methods Appl. Mech. Engrg. 254 (2013) 197–221, http://dx.doi.org/10.1016/j.cma.2012.11.001.
[57] S. Shi, A.G. Dawson, D.E. Smith, Geomorphological impact of the flores tsunami of 12th december, 1992, 1995, pp. 187–195,
http://dx.doi.org/10.1007/978-94-015-8565-1_13.
27
T. He, N. Mitsume, F. Yasui et al. Computer Methods in Applied Mechanics and Engineering 404 (2023) 115843

[58] G.H. Shi, Modeling rock joints and blocks by manifold method, in: 33rd U.S. Symp. Rock Mech. USRMS 1992, 1992, pp. 639–648,
http://dx.doi.org/10.1016/0148-9062(93)91974-n.
[59] X. Li, J. Zhao, An overview of particle-based numerical manifold method and its application to dynamic rock fracturing, J. Rock
Mech. Geotech. Eng. 11 (2019) 684–700, http://dx.doi.org/10.1016/j.jrmge.2019.02.003.
[60] G. Ma, X. An, L. He, The numerical manifold method: A review, Int. J. Comput. Methods 7 (2010) 1–32, http://dx.doi.org/10.1142/
S0219876210002040.
[61] B. Havard, S. Katsuyuki, O. Hideomi, Finite Cover Method on Contact Problem in 3D Solid, (n.d.).
[62] K. Terada, M. Asai, M. Yamagishi, Finite cover method for linear and non-linear analyses of heterogeneous solids, Internat. J. Numer.
Methods Engrg. 58 (2003) 1321–1346, http://dx.doi.org/10.1002/nme.820.
[63] S. Katsuyuki, O. Hideomi, J. Chuanrong, N. Katsuyoshi, Three Dimensional Solid Analysis Using Finite Cover Method, (1998)
505–512.
[64] R.B. Jr, A computational model of the cochlea using the immersed boundary method, J. Comput. Phys. 162 (1992) 145–162,
http://www.sciencedirect.com/science/article/pii/0021999192901807.
[65] L.J. Fauci, Interaction of oscillating filaments: A computational study, J. Comput. Phys. 86 (1990) 294–313, http://dx.doi.org/10.1016/
0021-9991(90)90103-8.
[66] A.L. Fogelson, A mathematical model and numerical method for studying platelet adhesion and aggregation during blood clotting, J.
Comput. Phys. 56 (1984) 111–134, http://dx.doi.org/10.1016/0021-9991(84)90086-X.
[67] C.S. Peskin, Flow patterns around heart valves: A numerical method, J. Comput. Phys. 10 (1972) 252–271, http://dx.doi.org/10.1016/
0021-9991(72)90065-4.
[68] A.M. Roma, C.S. Peskin, M.J. Berger, An adaptive version of the immersed boundary method, J. Comput. Phys. 153 (1999) 509–534,
http://dx.doi.org/10.1006/jcph.1999.6293.
[69] C.S. Peskin, The immersed boundary method, Acta Numer. 11 (2002) 479–517, http://dx.doi.org/10.1017/S0962492902000077.
[70] N. Moës, A. Gravouil, T. Belytschko, Non-planar 3D crack growth by the extended finite element and level sets-Part I: Mechanical
model, Internat. J. Numer. Methods Engrg. 53 (2002) 2549–2568, http://dx.doi.org/10.1002/nme.429.
[71] N. Sukumar, D.L. Chopp, N. Moës, T. Belytschko, Modeling holes and inclusions by level sets in the extended finite-element method,
Comput. Methods Appl. Mech. Engrg. 190 (2001) 6183–6200, http://dx.doi.org/10.1016/S0045-7825(01)00215-8.
[72] N. Moës, J. Dolbow, T. Belytschko, A finite element method for crack growth without remeshing, Internat. J. Numer. Methods Engrg.
46 (1999) 131–150, http://dx.doi.org/10.1002/(SICI)1097-0207(19990910)46:1<131::AID-NME726>3.0.CO;2-J.
[73] T. Belytschko, T. Black, Elastic crack growth in finite elements with minimal remeshing, Internat. J. Numer. Methods Engrg. 45 (1999)
601–620, http://dx.doi.org/10.1002/(SICI)1097-0207(19990620)45:5<601::AID-NME598>3.0.CO;2-S.
[74] C. Daux, N. Moës, J. Dolbow, N. Sukumar, T. Belytschko, Arbitrary branched and intersecting cracks with the extended finite element
method, Internat. J. Numer. Methods Engrg. 48 (2000) 1741–1760, http://dx.doi.org/10.1002/1097-0207(20000830)48:12<1741::AID-
NME956>3.0.CO;2-L.
[75] A. Gravouil, N. Moës, T. Belytschko, Non-planar 3D crack growth by the extended finite element and level sets-Part II: Level set
update, Internat. J. Numer. Methods Engrg. 53 (2002) 2569–2586, http://dx.doi.org/10.1002/nme.430.

28

View publication stats

You might also like