Finite Elements in Analysis and Design 69 (2013) 37–47

Contents lists available at SciVerse ScienceDirect

Finite Elements in Analysis and Design

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

The finite cell method for the J2 flow theory of plasticity

Alireza Abedian a,n, Jamshid Parvizian b, Alexander Düster c, Ernst Rank d
a
Department of Mechanical Engineering, Isfahan University of Technology, Isfahan, Iran
b
Department of Industrial Engineering, Isfahan University of Technology, Isfahan, Iran
c
Numerische Strukturanalyse mit Anwendungen in der Schiffstechnik (M-10), Technische Universität Hamburg-Harburg, Hamburg, Germany
d
Lehrstuhl für Computation in Engineering, Fakultät für Bauingenieur- und Vermessungswesen, Technische Universität München, München, Germany

a r t i c l e i n f o abstract

Article history: The finite cell method (FCM) is an extension of a high-order finite element approximation space with
Received 2 February 2012 the aim of simple meshing. In this paper, the FCM is implemented for J2 flow theory with nonlinear
Received in revised form isotropic hardening for small displacements and small strains. The Newton–Raphson iteration scheme,
8 January 2013
combined with a radial return algorithm, is applied to find approximate solutions for the underlying
Accepted 28 January 2013
Available online 6 March 2013
physically nonlinear problem. A modified quadtree integration scheme is presented for the first time to
capture the geometry accurately and overcome the high calculation cost of the standard quadtree
Keywords: integration scheme. Numerical examples in two and three dimensions demonstrate the efficiency of the
Finite cell method FCM and the proposed integration scheme at solving materially nonlinear problems.
Plasticity
& 2013 Elsevier B.V. All rights reserved.
J2 flow theory
Quadtree
Octree

1. Introduction The paper is outlined as follows: In Section 2, the basis of FCM

is explained. A modified quadtree integration scheme for
In embedded finite element methods, the mesh does not integrating discontinuous functions is discussed in Section 3.
necessarily conform to the boundaries [1–3]. The finite cell method The model problem and the numerical treatment of the under-
(FCM) [4,5] can be considered an embedding or a fictitious domain lying set of differential-algebraic equations is presented in Section
method combined with high-order finite elements [6]. The bound- 4. Section 5 presents numerical examples used to evaluate the
ary is extended to a simple domain that can be discretized using a proposed integration scheme and the efficiency of the FCM for the
simple mesh. Thus, the discretization problem is replaced by an J2 plasticity. The paper is then concluded.
integration problem. The key advantage of the method is that it
converges to acceptable results even though the mesh does not
conform to the boundaries. For linear elastic problems, the method 2. The finite cell method
results in very efficient approximations [4,5].
The efficiency of the p-version Finite Element Method (p-FEM) The finite cell method is a combination of high-order FEM and
for problems of elasto-plasticity has been well investigated fictitious domain methods; therefore, we use the well-established
[7–12], the p-version yields clearly superior accuracy in compar- terminology of the FEM here. Without loss of generality, let us
ison with its h-version counterparts. In this paper, the finite cell consider 2D problems of linear elasticity in this section. For the
method is examined for solving materially non-linear problems. physical domain of interest, O, with the boundary G, the weak
The J2 flow theory governs small strains and displacements, form of the equilibrium equation is given in terms of the
including nonlinear isotropic hardening. The accuracy of stresses displacement vector u and the test function vector v as
for nearly incompressible materials is always of interest. Incom- Bðu,vÞ ¼ F ðvÞ ð1Þ
pressibility effects will be introduced by the plastic material law
and those effects will slow down the convergence. The numerical where the bilinear functional is
Z
examples presented in this paper demonstrate that a good
Bðu,vÞ ¼ ½LuT C½Lv dO ð2Þ
pointwise convergence of stress components can be obtained, O
even if the point of interest is located within the plastic region.
where L is the standard strain-displacement operator and C is the
elasticity matrix. The linear functional is
Z Z
n
Corresponding author. Tel.: þ98 913 104 98 33. F ðvÞ ¼ vT b dO þ vT t dG ð3Þ
E-mail address: [email protected] (A. Abedian). O GN

0168-874X/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.finel.2013.01.006
38 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47

Ωe\Ω
Ωe

+ =

Fig. 1. The domain O is extended to Oe .

in which b is the vector of volume loads and t is the vector of

prescribed tractions on the Neumann boundary, GN  G. 300
The Dirichlet boundary conditions, applied on GD  G, are
u¼u on GD ð4Þ
To solve Eq. (1) using the high-order finite cell method, the 250
domain O, as shown in Fig. 1, is extended into a domain Oe , i.e.,
O  Oe , only with the advantage that Oe can be discretized with a
structured grid of quadrilateral cells. 200
Consequently, the bilinear functional is extended to
Z Z
Bðu,vÞ ¼ ½L vT C½L u dO þ ½L vT 0½L u dO
O Oe \O 150
Z
T
¼ ½L v aC½L u dO ð5Þ
Oe
100
in which

1:0 in O
a¼ ð6Þ
0:0 otherwise
50
In each cell, the displacement variable is approximated as
u ¼ NU ð7Þ
in which N denotes the matrix of hierarchical shape functions based 0
0 50 100 150 200 250
on integrated Legendre polynomials [13], and U is the vector
of unknowns. Based on the Bubnov–Galerkin approach, v ¼ NV, Fig. 2. A domain is partitioned using the quadtree.
inserting (7) into (5) results in the finite cell formulation as
p ,pZ ,pz p ,pZ ,pz
KU ¼ F ð8Þ S tsx ðOhst Þ, the tensor product space S psx ðOhst Þ and the space
where K is the global stiffness matrix, which is the assembly of the S p,p,q
ðOhst Þ.
c
stiffness matrix of each cell, k , given as In this paper, we will restrict our investigations to the trunk space
p ,p p ,p ,p
Z þ1 Z þ1 S tsx Z ðOqst Þ, S tsx Z z ðOhst Þ; to overcome problems of ill-conditioning,
c
k ¼ ðLNÞT aCðLNÞJJJ dx dZ ð9Þ a ¼ 0:0 is replaced by a ¼ 1007 [4].
1 1

in which JJJ is the determinant of the Jacobian matrix of mapping.

The product LN is usually denoted as the standard strain–displace-
ment matrix B. Although a may be discontinuous, the approxima-
3. An adaptive integration scheme
tion of u is based on hierarchic shape functions that are continuous
over the computational domain [4,5]. Using hierarchic bases, Ansatz
In embedded FE methods, the boundary of the physical
functions can be implemented up to any desired polynomial degree.
domain is not traced by the mesh. This task is accomplished by
Two- and three-dimensional Ansatz functions are constructed by
the integration scheme, which pays off, especially in the case of
simply building the tensor product of one-dimensional hierarchic
explicit boundary-represented problems. Adaptive integration
shape functions [6].
schemes could provide a solution to trace the boundary. Among
The two-dimensional shape functions (for quadrilaterals) can
such schemes, the quadtree is used in this paper.
be classified into three groups: nodal modes, edge modes and
Quadtree is a tree data structure used for partitioning a
internal modes. According to the definition of these shape func-
domain by recursively subdividing it into four quadrants. Fig. 2,
tions, two different types of Ansatz spaces have been implemen-
p ,pZ for instance, shows partitioning of the physical domain, Fig. 1,
ted [6,14]: the trunk space S tsx ðOqst Þ and the tensor product space using quadtree. The extension of quadtree for 3D space is called
p ,p
S psx Z ðOqst Þ.
In three dimensions, the shape functions (for hexahe- octree, in which each internal node or domain has eight children
drals) can be classified into four groups: nodal modes, edge modes, [15,16].
face modes and internal modes. Three different types of Ansatz Using quadtree or octree, the stiffness matrix of a cell (Fig. 3)
spaces have been implemented [6,14]: the trunk space can be easily computed by carrying out the composed integration
 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47 39

η η
ξ ξ s
r

x Parent cell =3 Local coordinate system

Fig. 3. Composed integration following the relationship of (x,y) and (r,s) for three levels of quadtree refinement (R).

(-1,1) (1,1)

η
η η
ξ (-1,1) (1,1)
s ξ ξ

(-1,-1) (1,-1)
(-1,-1) (1,-1)

Fig. 4. Four Gauss points are used in each direction for integration of p ¼ 3 in each Fig. 5. An adaptive integration scheme consists of a set of integration points inside
quadtree subcell. (a) and outside (b) the physical domain.

over nsc subcells: again with p þ1 Gauss points in each direction, and the points
that are outside the physical domain are taken into account only
nsc Z Z
X
c sc to control ill-conditioning of the problem (Fig. 5(b)). As an
k ¼ BTc ðnðrÞÞaðxðnðrÞÞÞCBc ðnðrÞÞ det Jc det J~ c dr ds ð10Þ
sc ¼ 1 s r example, for integrating a cell of order p ¼3, instead of 400 points
(Fig. 4), only 72 Gauss points are used (Fig. 5), which is a great
sc
In Eq. (10), the mapping nðrÞ ¼ Q~ c ðr,sÞ has to be applied to saving.
establish a relationship between the coordinates of the subcell Fig. 6 presents the flowchart of the integration scheme, where
(r,s) and the cell (x, Z). In the standard p-FEM, p þ 1 Gauss points NIP, GPs, NSC, INP, hxi and hx denote the number of integration
are used in each direction for the elements of order p [8]; this points in each direction, Gauss points, the number of quadtree
usage results in a large number of Gauss points in the case of the subcells, the quadtree integration points, the size of the subcell
FCM combined with the quadtree integration method. For exam- and the edge size of the parent element (namely, 2), respectively.
ple, 400 Gauss points are used for integrating a cell of order p ¼3 Using less Gauss points in the modified quadtree saves
partitioned around the boundary line using three levels of memory, especially in materially nonlinear problems. Because
refinement (R ¼ 3) (Fig. 4). The dense number of Gauss points the integration points are fixed during calculations, they are saved
causes a high calculation cost, especially for materially nonlinear in the pre-processing part of the code. The integration points are
problems. Therefore, an integration method should capture the standard Gauss points; therefore, the program does not need to
boundary accurately and with a minimum number of Gauss save the location of the new integration points. The subcell
points appropriate to the polynomial order. The first task is anchor point ðxi , Zi Þ, its length in both directions, the number of
achieved using the quadtree method. integration points (NIPs) and an array that saves the inside or
A modification of the standard quadtree integration scheme is outside flag (Flags) are sufficient to extract the modified quadtree
proposed with two specific features (Fig. 5). First, because the integration points from standard saved Gauss quadrature points
subcells are much smaller than the main domain, the function (Fig. 7).
over the subcells could be considered a lower-order function; To numerically examine the accuracy of the first feature, let us
thus, low-order Gauss quadrature can be applied. When the order consider a plate perforated by an inclined ellipse as the integra-
of the polynomial function is decreased, quadrature with fewer tion domain (Fig. 8) for integrals of a polynomial function of order
points can be used. Therefore, we can reduce the number of Gauss 14 (Fig. 9). The polynomial P 14 ðx, ZÞ appears in the integration of
points in each direction incrementally from p þ 1 down to 1 p¼8 cells (trunk space) when used in elasticity problems. There-
following the refining of the quadtree. In addition, any integration fore, 9 to 1 Gauss points are used for the subcells in each
point outside the boundary defined by the quadtree is ignored. direction, following the refining of the quadtree. The relative
Fig. 5(a) shows the position of active Gauss points of subcells error for calculating the area, integration of P 0 ðx, ZÞ, using both
mapped from the local coordinate (r,s) to the parent cell. methods is examined as well. The exact values are presented in
To avoid ill-conditioning in fictitious domain FE methods, a Table 1.
small value should be added to the stiffness matrix of the cell. The As shown in Figs. 10 and 11, for a similar computational cost,
second feature involves avoiding the ill-conditioning inherent in the new integration scheme yields results at least 10 times more
fictitious methods, by assigning a small a to the Gauss points accurate compared with the standard quadtree. The first feature
outside the physical domain [5]. Therefore, the cell is considered of the new scheme is evident in these figures.
40 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47

Fig. 6. The flowchart of the adaptive integration scheme.

The extension of the quadtree to 3D is called octree. Reducing

the number of Gauss points in three-dimensional integration is
even more appealing, as decreasing the number of Gauss points in
each direction decreases the total number of Gauss points
cubically.

4. Classical J2 flow theory with nonlinear isotropic hardening

The physically nonlinear problem considered is the J2 flow

theory for small strains with nonlinear isotropic hardening
[17–20]. The weak formulation of the equilibrium is discretized
Fig. 7. The used arrays to save the modified quadtree scheme data.
using the FCM. The weak formulation of the equilibrium
conditions of a three-dimensional solid body with domain
1 O  R3 and boundary G ¼ GD [ GN (with GD being the Dirichlet
boundary and GN the Neumann boundary) reads as follows.
0.8 Find the displacement field u A V ¼ fvðxÞ A ½H1 ðOÞ3 :
v ¼ 0 on GD g satisfying the (homogeneous) Dirichlet boundary
0.6 conditions, such that
Z Z Z
0.4 aeðvÞ : rðuÞ dO ¼ av  b dO þ v  t dG 8 v A V ð11Þ
Oe Oe GN
0.2 1
H ðOÞ denotes the Sobolev space of functions possessing
0 square integrable derivatives. Because the nonlinear stress (r)–
strain (e) relationship depends on the strain history, the weak
−0.2 formulation is a nonlinear function that has to be solved incre-
mentally. We therefore compute a sequence of equilibrated load
−0.4 steps by applying the Newton–Raphson method for each step
½t ðnÞ ,t ðn þ 1Þ  to linearize the weak formulation:
−0.6 Z Z Z
ði þ 1Þ
aeðvÞ : CðiÞ ðn þ 1Þ : eðDu Þ dO ¼ av  b ðn þ 1Þ dO þ v  t ðn þ 1Þ dG
−0.8 Oe Oe GN
Z
−1  aeðvÞ : rðiÞ
ðn þ 1Þ ðuÞ dO ð12Þ
−1 −0.5 0 0.5 1 Oe

Fig. 8. The square plate perforated with an inclined ellipse (r a ¼ 0:2, r b ¼ 0:6, f ¼
30J ) is partitioned using quadtree.
uðiðnþþ1Þ1Þ ¼ uðiÞ
ðn þ 1Þ þ Du
ði þ 1Þ
 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47 41

Fig. 9. P 14 ðx, ZÞ over the square domain.

Table 1
Polynomial integration over the perforated domain.

Polynomial Exact values

2 4 2 4 2
P 14 ðx, ZÞ ¼ 33xZðZ2 1Þðx 1Þð21x 14x þ 1Þð63x 70x þ 15Þ=1024 2:217869222184484E04
P 0 ðx, ZÞ ¼ 1 3.62300888156922

10 0.1
Standard quad tree Standard quad tree
Modified quad tree Modified quad tree

1
0.01
Relative error [%]

Relative error [%]

0.1

0.001
0.01

0.0001
0.001

0.0001 1e-05
10000 100000
10000 100000
Number of integration points
Number of integration points
Fig. 10. Relative error of integration using standard and modified quadtree
R R R Fig. 11. Relative error in calculating the area using standard and modified
methods; 9ð O P 14 dO Oe P 14 dOÞ= O P 14 dO9  100. R R R
quadtree methods; 9ð O P 0 ðx, ZÞ dO Oe P 0 ðx, ZÞ dOÞ= O P 0 ðx, ZÞ dO9  100.

@rðiÞ
ðn þ 1Þ
CðiÞ
ðn þ 1Þ :¼
@eðiÞ rate-independent plasticity, t ðnÞ denotes pseudo-time, ordering
ðn þ 1Þ
the steps being applied during the incremental loading. A new
where ðÞðiÞ
ðn þ 1Þ denotes a variable at the ith Newton–Raphson equilibrated configuration is found if a proper convergence
iteration during the load step in ½t ðnÞ ,t ðn þ 1Þ . In the context of criterion is satisfied. During the Newton–Raphson iteration the
42 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47

actual stress state rðiÞ ðn þ 1Þ as well as the algorithmic tangent The strains are assumed to be small and can be decomposed
modulus CðiÞ
ðn þ 1Þ have to be recomputed at each step i. The stresses into an elastic and a plastic component:
should satisfy both the constitutive equations and the equilibrium
e ¼ ee þ ep ð13Þ
conditions. The algorithm to compute stresses is based on an
elastic-predictor/plastic-corrector scheme [20–22]. Following [20], the differential-algebraic equations of the J2 flow
theory for small strains with isotropic hardening are summarized
in Table 2.
The stress tensor r is given by a linear isotropic elastic
relationship, depending only on the elastic strains ee and the
tensor of elastic moduli C. In Table 2, the operator ‘:’ denotes the
double contracted product of C with ðeep Þ, i.e. sij ¼ C ijkl ðekl epkl Þ.
Table 2
Classic J2 flow theory with isotropic hardening. Admissible stress states are defined by the von Mises yield
criterion
1. Linear isotropic elastic stress–strain r ¼ C : ðeep Þ qffiffi
relationship f ðr, bÞ ¼ Jdev½rJ 23KðbÞ r0 ð14Þ
2. Elastic domain in stress space Es ¼ fðr, bÞ9f ðr, bÞr 0g pffiffiffiffiffiffiffiffiffiffi
3. Flow rule and hardening law
e_ p ¼ g
dev½r where J  J :¼ ð : Þ is the Euclidean norm of a tensor, dev½ :¼
Jdev½rJ ðÞ 13 tr½1 is the deviatoric part of a tensor and tr½ denotes the
rffiffiffi
2 trace operator. The elastic domain is defined as the interior of Es
b_ ¼ g
3
4. Kuhn–Tucker loading/unloading conditions g Z 0, f ðr, bÞ r 0, gf ðr, bÞ ¼ 0 (where f ðr, bÞ o 0) and the boundary of Es (where f ðr, bÞ ¼ 0) is
5. Consistency condition gf_ ðr, bÞ ¼ 0 referred to as the yield surface. States f ðr, bÞ 4 0 outside Es are
nonadmissible. g Z0 (Table 2) is called the consistency parameter
and obeys the Kuhn–Tucker conditions, which are also known as
loading/unloading conditions. The consistency condition states
that either f_ ðr, bÞ ¼ 0 or g ¼ 0. Plastic loading can only occur if
Table 3 g 40; therefore, f_ ðr, bÞ ¼ 0. The evolution of plastic strains is
Material parameters for a nonlinear model problem. given by an associated flow rule. An internal variable b being
often referred to as the equivalent plastic strain describes the
Material parameter Unit
nonlinear isotropic hardening
Bulk modulus k MPa KðbÞ ¼ s0 þhb þðs1 s0 Þð1expðobÞÞ ð15Þ
shear modulus m MPa
Initial yield stress s0 MPa which is composed of a linear and an exponential function, where
Saturation stress s1 MPa s0 is the initial yield stress, h is the linear hardening parameter,
Linear hardening h MPa
s1 is the saturation stress and o is the hardening exponent. To
Hardening exponent o –
summarize, the described material model contains six para-
meters, as listed in Table 3.

Table 4 5. Numerical examples

Material parameters for the perforated
square plate. 5.1. Plane strain – cyclic loading

Material parameter Value

The first numerical example to be considered is a perforated
Bulk modulus (k) 164,206.0 (MPa) square plate under a plane strain condition with cyclic loading
shear modulus (m) 80,193.8 (MPa) (Fig. 12). This problem was defined as a benchmark in the
Initial yield stress (s0 ) 450.0 (MPa) research project ‘‘Adaptive finite element methods in applied
Saturation stress (s1 ) 715.0 (MPa)
mechanics’’ [23].
Linear hardening (h) 129.24 (MPa)
Hardening exponent (o) 16.93 Using symmetry conditions, a quarter of a perforated square
plate under uniform traction t n ¼ 100 (MPa) is considered (Fig. 12).

Fig. 12. Perforated square plate under plane strain condition with cyclic loading.
 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47 43

At the lower and right sides of the plate, the symmetry conditions would be far too fine. Deploying the modified quadtree integra-
are imposed. The load is scaled with a factor 4:5 r l r 4:5 in 205 tion scheme, the geometry of the hole is represented accurately.
load steps. The computations are performed using the h- and p-FEM code
A set of material parameters is considered, which also includes AdhoC [24], and the direct solver SPOOLES [25] is used to solve
nonlinear isotropic hardening (Table 4). The plate is discretized by the global equation system.
p ,p
100 (mesh A) and 399 (mesh B) quadrilateral cells, as depicted in All FCM computations are based on the trunk space S tsx Z ðOqst Þ
Fig. 13. Considering a p-extension for a linear elastic problem, with p ¼ px ¼ pZ ¼ 6, 8, 10 for mesh A and p ¼ px ¼ pZ ¼ 3, 5, 7 for
mesh A would lead to an efficient discretization, while mesh B mesh B. As a reference solution, we use the results presented by
WIENERS [23]. Of interest are the following physical quantities: the
displacement component ux (cm) at point a ¼ ð9,0Þ, the stress
component syy (MPa) at point a and the stress component sxx
(MPa) at point b ¼ ð6:1953,3:8047Þ (Fig. 12).
As can be seen in Figs. 14–16, increasing polynomial degree
improves the accuracy of the results. However, an accurate
approximation based on the coarse mesh A requires a quite high
polynomial degree, p ¼10. Considering the results of the finer
mesh B, it is evident that the p-version supplies a very accurate
approximation when p Z5.

5.2. Perforated thick plate

mesh A mesh A
As the second nonlinear problem, we consider a thick
Fig. 13. Perforated square plate discretized by 100 and 399 cells. perforated plate (Fig. 12), t ¼ 1 cm, under uniform load (Fig. 17).

0.002 0.002
Reference Reference
p=6 p=3
0.0015 p=8
0.0015 p=5
p = 10 p=7
0.001 0.001

0.0005 0.0005
ux atpoint a
ux atpoint a

0 0

-0.0005 -0.0005

-0.001 -0.001

-0.0015 -0.0015

-0.002 -0.002

-0.0025 -0.0025
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5
load factor load factor

Fig. 14. The displacement ux at point a using R ¼ 3.

800 800
Reference Reference
p=6 p= 3
600 p=8 600 p= 5
p = 10 p= 7
400 400
σyy at point a

200
σyy at point a

200

0 0

-200 -200

-400 -400

-600 -600

-800 -800
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5
load factor load factor

Fig. 15. The stress component syy at point a using R ¼ 3.

44 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47

50 50
Reference Reference
40 p =6 40 p =3
p =8 p =5
30 p =10 30 p =7
20 20
10 10
xx at point b

xx at point b
0 0
-10 -10
-20 -20
-30 -30
-40 -40
-50 -50
-60 -60
-5 -4 -3 -2 -1 0 1 2 3 4 5 -5 -4 -3 -2 -1 0 1 2 3 4 5
load factor load factor

Fig. 16. The stress sxx at point b using R ¼ 3.

450 λ t¯n
400
350
load (MPa)
300
250
200
h
z 150
y
x c r 100
t 10 20 30 40 50 60
w load step

Fig. 17. Thick perforated plate under a uniform load.

0.0012
Reference
p= 3
0.0011 p= 5
p= 7
0.001

0.0009
uxat pointc

0.0008

0.0007
Fig. 18. Thick perforated plate discretized by 399 cells.
0.0006
The corresponding material parameters are listed in Table 4.
0.0005
This problem was also defined as a benchmark in [23].
Symmetry conditions are applied, and the plate is discretized
0.0004
using 399 equally spaced hexahedral cells (Fig. 18) in conjunction 1 1.5 2 2.5 3 3.5 4 4.5
p ,p ,p
with the trunk space S tsx Z z ðOhst Þ. The solution is provided for loadfactor
p ¼ px ¼ pZ ¼ pz ¼ 3,5,7. The load is raised monotonously within
61 load steps up to a factor l ¼ 4:15. Fig. 19. The displacement ux at point c using R ¼ 3.
The first result to be considered is the displacement compo-
nent ux (mm) at point c ¼ ð9,0,0Þ for 1:0 r l r 4:15 (Fig. 19). As a same problem by the p-version method using a coarse mesh [14],
reference solution, we use the results provided in [23] with a fine while the curved boundary has been represented exactly with the
mesh of 1,048,576 hexahedral cells. The results of the FCM blending functions [6]. Therefore, accurate results are achievable
approach for p ¼7 conform well to the reference solution (Fig. 19). using FCM in comparison to the p-version FEM. The results can be
Fig. 20 shows syy (MPa) at point c for p ¼ 3, 5, 7 at all load improved even further by higher refinement levels (Fig. 21).
steps. The source of oscillation is the plastic zone crossing the As shown in Fig. 21, by refining the octree, the accuracy of the
interface of the two cells. As shown in this figure, the results results improves significantly while p¼7.
improve significantly by increasing the degrees of freedom (Dofs) A three-dimensional view of the von Mises stress (MPa) at the
while R ¼ 3. Errors smaller than 1:2% have been reported for the Gaussian points for the plastic region is shown in Fig. 22 for the
 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47 45

14 the element level without any further communication. This

increases the efficiency of the overall parallel approach which
12 takes advantage of the fact that the computation of element
matrices of high order constitutes a significant part of the overall
Relative error at point c [%]

10 numerical effort.
It is difficult to achieve a fair comparison for the total computa-
8 tional time of the FCM and p-FEM, since two different approaches
are used for mesh generation in these methods. It is, however, safe
to conclude that the FCM pre-processing is faster than p-FEM pre-
6
processing due to the simple and fast generation of Cartesian
meshes which do not need to account for complex geometries. The
4
hexahedral and tetrahedral meshes require much more efforts to
be built compared to the Cartesian mesh since they should trace
2 the boundary. In addition, hexahedral and tetrahedral meshes
should satisfy especial aspect ratios to avoid close to zero Jacobian
0 which may require some iterations to be satisfied. Moreover, in the
1 1.5 2 2.5 3 3.5 4 4.5
p-FEM the mesh should account for curved geometries, requiring
load factor special element mapping techniques.
Fig. 20. Relative error of syy at point c using R ¼ 3; 9ðsyy,ref syy,FCM Þ=syy,ref 9  100. To evaluate the solution time of the FCM compared to the
p-FEM without losing generality, this problem is considered again
with t y ¼ 400 MPa which is applied in 10 load steps. The solution
time consists of reading the mesh, computing and assembling the
2.5 element matrices, the time needed to solve the overall equation
=3
=5 system and also the post-processing. A grid of 196 equally spaced
cells and 199 elements (Fig. 23) is considered for calculations. The
2 problem is solved using AdhoC and the polynomial degree for
both methods is increased from 2 to 9. Four cores are used in
Relative error [%]

parallel for computations. The ratio of real times (T ) in terms of

1.5 Dofs is shown in Fig. 24. As can be seen, the solution times for
both methods converge by increasing Dofs. While in the FCM the
numerical integration is more expensive, it is the mapping and
1 calculation of Jacobian which are more expensive in the p-FEM.
As shown in Fig. 25, both methods yield a sufficient accuracy
while an error of 5% is considered acceptable for engineering
0.5 problems. Since the FCM and p-FEM have almost similar real
solution times and the FCM is faster for mesh generation one may
conclude that the computational time required for the FCM is in
0 total less.
1 1.5 2 2.5 3 3.5 4 4.5
load factor

Fig. 21. Relative error of stress component syy at point c using different levels of 5.3. Cube with spherical hole
refinement.
A 3D numerical example is presented to demonstrate the
accuracy of the FCM with the implemented adaptive integration.
A cube with a spherical hole (Fig. 26) is considered. The cube is
under compression, t z ¼ 350 MPa. Symmetric boundary
conditions are applied at y ¼ 0, z ¼ 0 and x ¼ 10 planes. In order
to compute the error of the FCM approximation, an overkill FCM
solution with 340,485 Dofs based on either h- or p-extensions and
a high level of octree refinement are used. The computations are
performed using the h- and p-FEM code AdhoC [24]. The direct
solver Pardiso [26] based on an LU factorization is used to solve
the global equation system.
The domain is discretized using 512 equally spaced cells while
8 cells inside the hole, which are completely outside the physical
domain, are ignored from computations (Fig. 27). To show the
accuracy of the method through the body of interest, the von
Mises stress along the cut-line AB passing the diagonal of the
Fig. 22. The von Mises stress and Gaussian points where yielding occurs. cube is shown in Fig. 28 while p ¼5, 7. As shown, the considerable
differences appear in the plastic zone front (s ¼ 450) and the hole
boundary (X AB ¼ 5:83,11:48), where stress concentration occurs
last load step l ¼ 4:15 using 100 cells and R ¼ 3. The results are especially in the elastic part. Fig. 29 shows that the maximum
visualized by Paraview [27] using a fine post-processing mesh. error is at the interface of the hole which is about 1:4% for p ¼7
The FCM enjoys all the advantages of the p-version FEM, such (89,841Dofs) and 2% for p ¼5 (38,337 Dofs). A three-dimensional
as condensing interior Dofs which are confined to the element. cut-plane view of the von Mises stress (MPa) with the cut-line
The elimination of interior Dofs can be performed in parallel to AB for p ¼5 is shown in Fig. 30.
46 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47

Fig. 23. Thick perforated plate discretized by 199 hexahedral elements and 196 cells.

10

FCM
p FEM
Ratio

0.1
1000 10000 100000
Dofs

Fig. 24. The ratio of real time T for 196 cells (FCM) and 199 elements (p-FEM). Fig. 26. A cube with a spherical cavity under simple compression.

5
Relative error [%]

0
1000 10000 100000
Dofs

Fig. 25. Relative error of stress component syy at point c using the FCM and
p-FEM. Fig. 27. The perforated cube is discretized using 504 cells.

6. Conclusions method depends very much on the integration scheme. In the

present study, the classic quadtree and octree methods with a
The main advantage of the finite cell method is its extremely quadrature scheme are modified to improve the convergence
fast meshing; the mesh is simply a Cartesian grid that does not behavior of the method for a lower cost. For the investigated
necessarily conform to the boundaries. In this paper, a FCM examples, the proposed integration scheme yields for the same
approach for the J2 flow theory with nonlinear isotropic numerical effort at least ten times better accuracy than the
hardening is presented. This approach can stimulate the research standard quadtree.
into the simulation of different materially nonlinear problems The capabilities of the FCM in simulating materially nonlinear
using a simple background mesh. The accuracy of the finite cell problems are inspected numerically. The numerical results show
 A. Abedian et al. / Finite Elements in Analysis and Design 69 (2013) 37–47 47

References

[1] V. Saul’ev, On solving boundary value problems with high performance

computers by a fictitious domain method, Siberian Math. J. 4 (4) (1963) 912.
[2] P. Neittaanmäki, D. Tiba, An embedding of domains approach in free
boundary problems and optimal design, SIAM J. Control Optim. 33 (5)
(1995) 1587–1602.
[3] T. Rusten, P. Vassilevski, R. Winther, Domain embedding preconditioners for
mixed systems, Numer. Linear Algebra Appl. 5 (5) (1998) 321–345.
[4] J. Parvizian, A. Düster, E. Rank, Finite cell method: h- and p-extension for
embedded domain problems in solid mechanics, Comput. Mech. 41 (1)
(2007) 121–133.
[5] A. Düster, J. Parvizian, Z. Yang, E. Rank, The finite cell method for 3D problems
of solid mechanics, Comput. Meth. Appl. Mech. Eng. 197 (2008) 3768–3782.
[6] B. Szabó, I. Babuška, Finite Element Analysis, John Wiley & Sons, 1991.
[7] A. Düster, E. Rank, A p-version finite element approach for two-and three-
dimensional problems of the J2 flow theory with non-linear isotropic
hardening, Int. J. Numer. Methods Eng. 53 (1) (2001) 49–63.
[8] A. Düster, E. Rank, The p-version of the finite element method compared to an
adaptive h-version for the deformation theory of plasticity, Comput. Methods
Appl. Mech. Eng. 190 (15–17) (2001) 1925–1935.
[9] A. Düster, A. Niggl, V. Nübel, E. Rank, A numerical investigation of high-order
finite elements for problems of elastoplasticity, J. Sci. Comput. 17 (1) (2002)
Fig. 28. The von Mises stress along the cut-line AB. 397–404.
[10] S. Holzer, Z. Yosibash, The p-version of the finite element method in
incremental elasto-plastic analysis 39 (11) (1996) 1859–1878.
[11] B. Jeremic, C. Xenophontos, Application of the p-version of the finite element
2 method to elasto-plasticity with localization of deformation 15 (12) (1999)
867–876.
[12] B. Szabó, R. Actis, S. Holzer, Solution of Elastic–Plastic Stress Analysis
Problems by the p-Version of the Finite Element Method, Tech. Rep.
relative error invon Mises stress [%]

p=5 WU/CCM-93/3, Center for Computational Mechanics, Washington University

1.5 p=7 1993.
[13] B. Szabó, A. Düster, E. Rank, The p-version of the finite element method, in:
E. Stein, R. de Borst, T.J.R. Hughes (Eds.), Encyclopedia of Computational
Mechanics, vol. 1, John Wiley & Sons, 2004, pp. 119–139. (Chapter 5).
[14] A. Düster, H. Bröker, E. Rank, The p-version of the finite element method for
1 three-dimensional curved thin walled structures, Int. J. Numer. Methods Eng.
52 (2001) 673–703.
[15] M. De Berg, O. Cheong, M. Van Kreveld, M. Overmars, Computational
Geometry: Algorithms and Applications, Springer-Verlag, New York Inc.,
2008.
0.5 [16] H. Samet, Applications of Spatial Data Structures: Computer Graphics, Image
Processing, and GIS. Addison-Wesley.
[17] W. Chen, D. Han, Plasticity for Engineers, Springer-Verlag, New York, 1988.
[18] J. Lemaitre, J.-L. Chaboche, Mechanics of Solid Materials, Cambridge
University Press, Cambridge, 1990.
0 [19] J. Lubliner, Plasticity Theory, Macmillan Publishing Company, 1990.
0 2 4 6 8 10 12 14 16
[20] J.C. Simo, T.J.R. Hughes, Computational Inelasticity, Springer-Verlag, New
XA−B in the direction of the cut-line York, 1998.
[21] S. Hartmann, G. Lührs, P. Haupt, An efficient stress algorithm with applica-
Fig. 29. Relative error in von Mises stress (9ðsRef sFCM Þ=sRef 9  100) along the tions in viscoplasticity and plasticity 40 (1997) 991–1013.
cut-line AB. [22] M. Wilkins, Calculation of elastic–plastic flow, in: B. Alder et al. (Ed.),
Methods of Computational Physics 3, Academic Press, New York, 1964.
[23] E. Stein, Error-controlled Adaptive Finite Elements in Solid Mechanics, Wiley,
2003.
[24] A. Düster, H. Bröker, H. Heidkamp, U. Heißerer, S. Kollmannsberger, R. Krause,
A. Muthler, A. Niggl, V. Nübel, M. Rücker, D. Scholz, AdhoC4—User’s Guide,
Lehrstuhl für Bauinformatik, Technische Universität München 2004.
[25] /http://www.netlib.org/linalg/spooles/spooles.2.2.htmlS.
[26] /http://www.pardiso-project.orgS.
[27] /http://www.paraview.org/S.

Fig. 30. A three-dimensional cut-plane view of the von Mises stress (MPa).

that a relative error o5% is achievable using the FCM. Therefore,

the FCM offers an accurate and efficient approximation for the
investigated physically nonlinear problems.