Computers and Structures 134 (2014) 128–142

Contents lists available at ScienceDirect

Computers and Structures

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

The MITC3 shell finite element enriched by interpolation covers

Hyeong-Min Jeon a, Phill-Seung Lee a,⇑, Klaus-Jürgen Bathe b
a
Department of Ocean Systems Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 305-701, Republic of Korea
b
Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we develop a scheme to enrich the 3-node triangular MITC shell finite element by interpo-
Received 1 August 2013 lation cover functions. The MITC method is used for the standard and enriched displacement interpola-
Accepted 9 December 2013 tions. The enriched 3-node shell finite element not only captures higher gradients but also decreases
Available online 28 January 2014
inter-elemental stress jumps. In particular, the enrichment scheme increases the solution accuracy with-
out any traditional local mesh refinement. Convergence studies considering a fully clamped square plate
Keywords: problem, cylindrical shell problems, and hyperboloid shell problems demonstrate the good predictive
Shell structures
capability of the enriched MITC3 shell finite element, even when distorted meshes are used. We evaluate
Enriched finite elements
Stress accuracy
the effectiveness of the method, and also illustrate the use of the enrichment scheme applied only locally
MITC shell elements through the solution of two additional shell problems: a shaft–shaft interaction problem and a monster
Interpolation cover shell problem.
Triangular elements Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction traditional nodal point movements or mesh refinements are not

used to improve the solution accuracy. The enrichment scheme
The finite element method is a popular and effective procedure not only captures higher gradients but also decreases inter-ele-
for the analysis of solid, fluid, and multi-physics problems. The ment stress jumps.
key for its success is mainly due to the fact that meshes can be Due to discontinuities in geometries (curvature or thickness),
used to span over complicated domain geometries. However, incompatibilities of boundary conditions, and irregularities in the
the solution accuracy highly depends on how the geometries loading, shell structures frequently experience stress layers. With-
are meshed. To obtain reliable solutions with desired accuracy, in stress layers, the displacements vary rapidly and induce
special mesh refinements are frequently required, in particular, concentrations of strain energies [11,12]. To achieve a desired solu-
in areas where non-smooth and near-singular solutions are tion accuracy in such stress layers, a sufficiently fine mesh and/or
sought [1]. higher order interpolation functions are required [1]. But the use of
To obtain more accurate solutions, a promising approach is to traditional mesh refinement procedures can be computationally
incorporate special enrichment functions within traditional finite demanding and tedious for complex shell geometries [13]. In sev-
element formulations. For example, Bathe and Chaudhary [2] and eral studies, enrichment functions have been employed for plates
Yoon et al. [3] successfully used enriched displacements to analyze and shells, but these studies have focused on special local
warping effects in beam problems. Benzley [4] and Dvorkin et al. enrichments near cracks [14–17] rather than on a general enrich-
[5], Belytschko and Black [6], Moes et al. [7] and Daux et al. [8] ment scheme that may be used for various areas of the shell
employed enrichment functions to account for various structure.
singularities in solid mechanics problems. Babuška and Melenk Displacement-based shell finite elements become too stiff in
[9] showed in general, mathematically, how to include Ansatz bending situations when the thickness is relatively small [1]. The
spaces containing the local properties of solutions. phenomenon is called locking. For a long time, there have been
Recently, Kim and Bathe [10] developed and studied a finite ele- numerous attempts to alleviate the locking of plate and shell
ment method in which interpolation covers are used. The scheme finite elements [1]. As well established, using the pure displace-
increases the solution accuracy of the traditional low-order finite ment-based method in general shell analyses, locking cannot
element discretization of solids without any changes in the mesh. be eliminated [18]. A major advancement has been accomplished
The major advantage of the method is that the higher order enrich- by the MITC (Mixed Interpolation of Tensorial Components)
ment is available without introducing additional nodes. That is, method for quadrilateral and triangular plate and shell elements
[18–28]. However, a particularly difficult task is to obtain a general
⇑ Corresponding author. 3-node shell element that is effective for the analysis of all shell
E-mail address: [email protected] (P.S. Lee). structures.

0045-7949/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruc.2013.12.003
 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142 129

The aim of this paper is to develop a 3-node triangular shell fi-   ðx  xi Þ ðy  yi Þ

P pi ½u ¼ u
 i þ ni gi n2i ni gi g2i  gpi u^ i with ni ¼ ; gi ¼ ; ð3Þ
nite element enriched by linear interpolation covers within the vi vi
MITC framework. The key idea is to use and treat the assumed where i
u is the standard nodal point variable, ^i ¼
u
covariant transverse shear strain fields separately for the standard g 2 g2 gp T
linear and the additional quadratic displacement interpolations. ^ ni
½u ^i
u ^ ni
u ^i
u ^ ni g
... u
u ^i lists the additional degrees of
The MITC3 shell finite element enriched by interpolation covers freedom for the cover region, p is the order of the complete polyno-
passes the patch tests and shows in example solutions good con- mial used, and vi is the diameter of the largest finite element
vergence behavior even when distorted meshes are used. Highly sharing the node i. The use of vi can improve the conditioning of
varying stresses in shells can be captured by only locally using the coefficient matrix.
interpolation covers. The enriched approximation for the solution variable is then
In the following sections, the finite element procedure to enrich given by
by interpolation covers is briefly reviewed and then the formula- !
X
q X X
q X X
tion of the enriched MITC3 shell finite element is presented. We u¼ hi P pi ½u ¼ i þ
hi u ^i
Hi u
discuss the key theoretical and numerical aspects of the scheme m¼1 i2ic ðmÞ m¼1 i2ic ðmÞ i2ic ðmÞ

regarding the computational expense and the convergence of the  

method. The results of convergence studies considering a fully with Hi ¼ hi ni gi n2i ni gi g2i    gpi : ð4Þ
clamped square plate problem, cylindrical shell problems, and
Considering Eq. (4), the enriched cover approximation consists
hyperboloid shell problems are given. Two illustrative example
of the standard finite element interpolation plus additional
solutions, a shaft–shaft interaction problem and a ‘highly-sensitive’
higher order terms. To obtain a well-conditioned stiffness
shell problem, show the effectiveness of using the enriched MITC3
matrix, we use the local coordinate systems (ni, gi) instead of
shell finite element also only locally, that is, only in areas of high
the global coordinates (x, y) and always (although not
stress gradients.
mentioned in the example solutions) enforce not only u i ¼ 0
but also u^ i ¼ 0 when imposing the essential boundary condi-
2. The finite element method enriched by interpolation covers tions at the node i.
The basic properties of the finite element method enriched by
Enriching the finite element procedure is in principle, and the- interpolation covers were studied for general 2D and 3D finite ele-
oretically, straightforward but difficulties are encountered in ment analyses in Ref. [10]. These basic properties pertain also to
obtaining effective schemes [6–10,29–33]. To introduce the basic the finite element analysis of shells.
procedure for the enrichment scheme considered here, let us
briefly consider in this section a two-dimensional analysis
3. The enriched MITC3 shell finite element
problem.
Let Q n ¼ fxi gni¼1 be a set of n nodal point position vectors
q In this section, we present the displacement interpolation of the
xi ¼ ½ xi yi T 2 X, and let fkh g ¼ fwm gm¼1 be a family of q triangles
n MITC3 shell finite element enriched by the linear interpolation
generated by Q . The triangles correspond to the domain X in
cover. Therefore, the resulting enriched displacement interpolation
which we seek the solution variable u
can give quadratic convergence. We also present the assumed
[
q
covariant strain fields used for the enriched MITC3 shell finite
wm ¼ X; ð1Þ element.
m¼1

The triangles do not overlap, that is, wj \ wk ¼ £ for j – k. 3.1. Enriched displacement interpolation
Fig. 1(a) shows the piecewise interpolation function hi(x, y)
used in the solution. Let Ci be the support domain of hi, i.e. The geometry of the 3-node continuum mechanics based trian-
C i ¼ supp ðhi Þ; 8i ¼ 1; . . . ; N, which we call the cover region. Hence gular shell finite element is interpolated using [23,24,28]
the cover region Ci corresponds to the union of elements attached to
the node i, see Fig. 1(b). For each wm, let ic(m) be the set of cover
indices defined by
ic ðmÞ ¼ fi : C i \ wm –£g: ð2Þ
In Fig 1, the 3-node triangular element m coincides with the
overlapped region of the three cover regions Ci, Cj and Ck and
hence ic(m) = {i, j, k}. To enrich the standard finite element interpo-
lation for the solution variable u, we use interpolation cover
functions

(a) hi (b) (c)

m
i Ci i i k

Fig. 1. Description of sub-domain for enriched over interpolations; (a) usual

interpolation function, (b) cover region or elements affected by the interpolation
cover, and (c) an element. Fig. 2. A 3-node triangular continuum mechanics based shell finite element.
130 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142

Table 1
(a) Detailed information on the stiffness matrices of the 3-, 6-, and enriched 3-node shell
s s
finite elements for the meshes shown in Fig. 4.

Linear shell element Quadratic shell elements

1 : tying point 1 Standard 3-node Standard Enriched
6-node 3-node
Elements 512 (2N = 16) 128 (N = 8) 128 (N = 8)
s1 est(2) eqt(3) Nodes 289 289 81
DOFs 803 803 633
Non-zero entries 15,313 24,581 32,797
Half-bandwidth 51 100 83
r r
0
r1
1 0 ert(1) 1

(b)
s s
where r, s, and t are natural coordinates, hi is the 2D interpolation
function corresponding to node i, xi is the position vector of node
1 1 i in the global Cartesian coordinate system, and ai, Vin denote the
s2 eˆ2(2)st eˆ2(3)qt shell thickness and the director vector at node i, respectively, see
Fig. 2.
The standard displacement interpolation of the shell element is
given by
s1 eˆ1(2)st eˆ1(3)
qt
X3 X3
t  
r r ¼
u i þ
hi u ai hi Vi2 a
 i þ Vi1 b
i ; ð6Þ
0 r1 r2 1 0 eˆ1(1) eˆ2(1)rt 1 2
rt i¼1 i¼1

Fig. 3. Tying points for the covariant transverse shear strains of the enriched MITC3 in which u  i v i w
i ¼ ½ u  i T is the nodal displacement vector in the
T
shell finite element: (a) for the standard linear displacement interpolation; global Cartesian coordinate system, Vi1 ¼ ½ V i1x V i1y V i1z  and
r 1 ¼ s1 ¼ 12, and (b) for the additional quadratic displacement interpolation; i i i i T i
V2 ¼ ½ V 2x V 2y V 2z  are unit vectors orthogonal to Vn and to each
r 1 ¼ s1 ¼ 12  2p1 ffiffi3 and r 2 ¼ s2 ¼ 12 þ 2p1 ffiffi3. Note that the scheme referred to as MITC6b
other, and a  i and bi are the rotations of the director vector Vi about
in Ref. [23] is used. n
i i
V1 and V2 at node i.
X3
tX 3 To enrich the displacement interpolation in Eq. (6), we use the
xðr; s; tÞ ¼ hi ðr; sÞxi þ ai hi ðr; sÞVin linear interpolation cover, that is, the first order degree of polyno-
i¼1
2 i¼1
mial bases. The enriched displacement interpolation for the 3-node
with h1 ¼ r; h2 ¼ s; h3 ¼ 1  r  s; ð5Þ triangular shell finite element is given by

(a) (b) (c)

3-node element 6-node element Enriched 3-node element

16 16 8 8 8 8
(d) (e) (f)

803 803 633

# of non-zero entries : 15,313 # of non-zero entries : 24,581 # of non-zero entries : 32,797

Half-bandwidth : 51 Half-bandwidth : 100 Half-bandwidth : 83

Fig. 4. Meshes used and stiffness matrix structures: (a) and (d) for the 3-node shell element, (b) and (e) for the 6-node shell element, and (c) and (f) for the enriched 3-node
shell elements. Non-zero entries are colored in black.
 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142 131

Therefore, the enriched covariant strain components are also di-

1.2E+4 (2 N 1) 2 p vided into two parts
The total number of DOFs

3-node element 1
1.0E+4 eij ¼ eij þ ^eij with eij ¼  ; j þ gj  u
ðg  u  ; i Þ and
6-node element 2 i
8.0E+3 enriched 3-node element 1
^eij ¼ ^ ; j þ gj  u
ðg  u ^ ; i Þ; ð11Þ
( N 1) 2 3 p
2 i
6.0E+3
in which eij and ^eij correspond to the standard linear and additional
4.0E+3 quadratic displacement interpolations, respectively.
To alleviate the locking phenomenon, we use the MITC meth-
2.0E+3
( N 1) 2 p od for the covariant transverse shear strains. However, different
0.0E+0 assumed covariant transverse shear strain fields are employed
for the standard and additional quadratic displacement interpo-
lations. The assumed covariant transverse shear strain fields of
0 4 8 12 16 20 24 28 32 36
the MITC3 and MITC6 shell elements are used for the strains
N eij and ^eij , respectively [23]. Note that, in the MITC6 shell ele-
ment, the covariant membrane strains are also assumed to re-
Fig. 5. The total number of degrees of freedom (DOFs) when increasing the number duce membrane locking, but this treatment is not necessary
of element layers, N, along an edge p denotes the number of degrees of freedom per
and not used for the enriched MITC3 shell element due to its flat
node, hence p = 3 for the simply supported plate problem.
geometry.
The assumed covariant transverse shear strain field used for the
standard displacement interpolation is [23]
 þu
u¼u ^ ð7Þ
eAS ð1Þ
rt ¼ ert þ cs;
eAS ð2Þ
st ¼ est  cr; ð12Þ
with
where c ¼  eð2Þ
st þ eð1Þ ð3Þ
est eð3Þ
and, at the tying points,
rt rt
ðnÞ
and est ertðnÞ
X3 X3
t   are calculated from Eq. (11), see Fig. 3(a).
^¼
u ^i þ
Hi u ai Hi Di2 a
^ i þ Di1 ^bi ; ð8Þ For the additional quadratic displacement interpolation, we use
i¼1 i¼1
2
the assumed covariant transverse shear strain field
T T
in which u ^i ¼ ½ u ^ gi j v
^ ni u ^ ni v^ gi j w ^ gi  ; a
^ ni w ^ gi  and
^ ni a
^i ¼ ½ a
^ ^g T are unknown coefficient vectors for the displace-
^n b
bi ¼ ½ b ^eAS
rt ¼ a1 þ b1 r þ c 1 s;
i i ð13Þ
ments and rotations, and the Hi are the linear cover interpolation ^eAS
st ¼ a2 þ b2 r þ c 2 s
matrices for the displacements and rotations
2 3 2 3
and we have the coefficients
V i1x 0 V i2x 0
6 7 6 7
3 2 6 0 V i1x 7 6 0 V i2x 7 ð1Þ
a1 ¼ mrt  lrt ;
ð1Þ ð1Þ
b1 ¼ 2lrt ; a2 ¼ mst  lst ; c2 ¼ 2lst ;
ð2Þ ð2Þ ð2Þ
6 7 6 7
ni gi 0 0 0 0 6 i 7 6 i 7  
6 7 6 V 1y 0 7 6V 0 7 ð3Þ ð3Þ ð3Þ ð3Þ
Hi ¼ hi 4 0 0 ni gi 0 0 5; Di1 ¼ 6 7; and Di ¼ 6 2y 7: ð9Þ c1 ¼ ða2 þ c2  a1 Þ  mst þ lst  mrt  lrt ; ð14Þ
6 i 7 2 6 i 7
6 0 V 1y 7 6 0 V 2y 7  
0 0 0 0 ni gi 6 7 6 7
6 i 7 6 i 7 ð3Þ ð3Þ ð3Þ ð3Þ
b2 ¼ ða1 þ b1  a2 Þ þ mst  lst  mrt þ lrt
4 V 1z 0 5 4 V 2z 0 5
0 V i1z 0 V i2z
with
Note that the enriched displacement interpolation for u in Eq. (7) pffiffiffi 
 and the additional
consists of two parts: the standard linear term u ðiÞ 1  ðiÞ ðiÞ

ðiÞ 3 ðiÞ 
mjt ¼ ^ ^
e1jt þ e2jt ; ljt ¼ ^e2jt  ^eðiÞ
1jt
quadratic term u^. 2 2
with j ¼ r; s for i ¼ 1; 2; 3; ð15Þ

3.2. Assumed covariant transverse shear strain fields where ^eðiÞ

and
1jt
ðiÞ
^e2jt
are calculated at the tying points in Fig. 3(b).
We finally obtain the assumed covariant transverse shear strain
The covariant strain components are directly obtained as fields for the enriched MITC3 shell element as
AS
1 eAS AS ^AS
jt ¼ ejt þ ejt ¼ Bjt U with j ¼ r; s; ð16Þ
eij ¼ ðg  u; j þ gj  u; i Þ; ð10Þ
2 i
in which BAS
is the covariant transverse shear strain–displacement
jt
 ^
@x
where gi ¼ @r i
; @u
u; i ¼ @r i
¼ @ðu@rþi uÞ with r1 = r, r2 = s, r3 = t. matrix and U is the vector that contains the degrees of freedom

Table 2
Solution times (in second) for solving the linear equations. We use 2N  2N and N  N meshes for linear and quadratic shell elements, respectively (DOFs: degrees of freedom, HB:
half-bandwidth).

N Linear shell element Quadratic shell elements

Standard 3-node Standard 6-node Enriched 3-node
DOFs HB Time DOFs HB Time DOFs HB Time
4 211 27 0.001 211 52 0.001 177 47 0.001
8 803 51 0.016 803 100 0.016 633 83 0.016
16 3,139 99 0.062 3,139 196 0.125 2,409 155 0.062
32 12,419 195 0.733 12,419 388 1.591 9,417 199 1.045
64 49,411 387 10.70 49,411 772 23.88 37,257 587 15.49
132 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142

2L

z
q t
x D C 2L

x
A B

D C D C D C

h h h

A B A B A B
(a) (b) (c)
Fig. 6. Fully clamped square plate under uniform pressure (L = 1.0, E = 1.7472  107, q = 1.0 and v = 0.3) with three different 4  4 mesh patterns: (a) and (b) triangular mesh
for the MITC3, MITC6 and enriched MITC6 shell elements, and (c) quadrilateral mesh for the MITC4 shell element.

 i; a
u i and the additional degrees of freedom u
i ; b ^ i; a^ i; ^
bi . Note that L4 L3 L 2 L1
we do not use assumed covariant strain fields for other covariant
D C D C
strain components and hence the scheme will not give spurious
modes in membrane strains [25,34].
Then, using the appropriate stress–strain matrix for shells,
the element stiffness is constructed in the same manner as for
the displacement-based shell element [1]. The 7-point Gauss
integration is adopted to evaluate the stiffness matrix because
the order of the enriched displacement interpolation is
quadratic.
Since the cover interpolation is based on the existing nodes, the
enriched displacement interpolation can be locally used assigning
or not assigning interpolation covers in different regions. Without
enrichment, the element is identical to the original MITC3 shell A B A B
element. (a) (b)
The enriched MITC3 shell element passes the membrane,
bending, and transverse shearing patch tests for arbitrary local Fig. 7. Distorted meshes used for the fully clamped square plate problem,
enrichments, see Refs. [1,18,23,28] for the patch tests per- cylindrical shell problems, and hyperboloid shell problems when (a) N = 4 and (b)
N = 8. The number of triangular elements for an N  N mesh is 2N2.
formed. Of course, in the tests, the nodal forces corresponding
to not only the standard degrees of freedom (u i; a i and b i )
but also the additional degrees of freedom (u ^i; a ^ i and ^ bi ) must
dard 3-node shell element is based on the linear displacement
be applied [10].
interpolation, and the 6-node and enriched 3-node shell elements
are based on quadratic displacement interpolations, we use a twice
4. Computational efficiency finer mesh for the standard 3-node shell element.
The stiffness matrix entries for the simply supported square
In this section, we study some important aspects of the compu- plate problem for some equivalent mesh patterns are plotted in
tational efficiency when using the enriched element. The standard Fig. 4(d)–(f), where the non-zero entries are colored in black. The
3- and 6-node shell elements (the MITC3 and MITC6 shell ele- size of the stiffness matrices for the meshes used is 803  803
ments) and the enriched 3-node shell element (the enriched MITC3 for the 3- and 6-node shell elements and 633  633 for the en-
shell element) are considered. riched 3-node shell element. The standard 3- and 6-node shell ele-
We first study the size and sparseness of the stiffness matrices ments and the enriched 3-node shell element give 15,313, 24,581
when using the enriched 3-node shell finite element and the stan- and 32,797 non-zero entries in the matrices, respectively. Also,
dard 3- and 6-node shell elements for the meshes shown in Fig. 4. the matrix half-bandwidths are 51, 100 and 83 for the 3- and 6-
A simply supported square plate problem is considered. The node shell elements and the enriched 3-node shell element,
meshes used are given in Fig. 4(a)–(c) when N = 8. Since the stan- respectively. Table 1 lists the information regarding these cases.
 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142 133

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 9. Convergence curves for the fully clamped square plate problem with uniform meshes. The solid and dotted lines correspond to the results obtained by the mesh
patterns in Fig. 6(a) and (b), respectively. The bold line represents the optimal convergence rate, which is 2.0 for linear elements and 4.0 for quadratic elements.

The number of non-zero entries in the stiffness matrix for the as a function of the number of elements used in the meshing of
enriched 3-node shell element is substantially larger than in the Fig. 4.
corresponding matrix for the 6-node shell element. This is due It is valuable to compare solution times required for the three
to the fact that the support of the higher order interpolation shell finite elements considered. In all the cases, of course,
functions in the enriched 3-node shell element is larger than for symmetric stiffness matrices are generated. To obtain more insight
the 3- and 6-node shell elements. However, using the enriched into the computational efforts needed in the respective solutions,
3-node shell element, all degrees of freedom are associated with we focus on the solution of the linear equations using direct Gauss
vertex nodes which are shared by several elements and the elimination, in which the factorization of the stiffness matrices
assembled system of equations is in general smaller than when represents the major expense. To check computational times, we
using the 6-node shell element where edge nodes are only shared use a quad-core machine (Intel(R) Core i7-3770 CPU @ 3.40 GHz,
by 2 elements. Therefore, the enriched 3-node shell element gives 8 GB RAM, Windows 7 64bit) for all solution cases. Table 2 shows
less equations and here also a smaller bandwidth than the 6-node the solution times for the simply supported square plate problem.
shell element. This fact shows the effectiveness of the enriched As expected, the factorization time for the enriched 3-node shell
3-node shell element from a computational point of view. Fig. 5 element is much smaller than for the standard 6-node shell
shows how the number of nodal degrees of freedom increases element.

Enriched MITC3 Enriched MITC3 Enriched MITC3

MITC3 (DISP) (MITC6a) (MITC6b)

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(1.2h) log(1.2h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 8. Convergence curves for the fully clamped square plate problem with uniform meshes. For triangular shell elements, the mesh pattern in Fig. 6(a) is used. The bold line
represents the optimal convergence rate, which is 2.0 for linear elements and 4.0 for quadratic elements.
134 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 10. Convergence curves for the fully clamped square plate problem with the distorted meshes shown in Fig. 7. The bold line represents the optimal convergence rate,
which is 2.0 for linear elements and 4.0 for quadratic elements.

5. Convergence studies
z
In this section, we perform convergence studies on well-
established problems for the enriched MITC3 shell element. The
solutions can show at most quadratic convergence in the s-norm
and the results are compared with those of the MITC3, MITC4, C
and MITC6 shell elements. We solve various problems: a fully
clamped square plate problem, cylindrical shell problems, and
hyperboloid shell problems using uniform and distorted meshes D
[23,24,27,28,35].
The s-norm proposed by Hiller and Bathe [36] is used to mea- B
sure the convergence of the finite element solutions. The s-norm
y
is suitable to check whether the finite element solutions satisfy
consistency and the inf-sup condition [36–40], and is defined as
follows A
Z x R
ku  uh k2s ¼ DeT DsdX; ð17Þ 2L
X
where u is the exact solution, uh is the solution obtained using the
finite element discretization, e and s are the strain and stress vec-
tors, and
1
De ¼ e  e h ; Ds ¼ s  sh : ð18Þ
The theoretical convergence behavior can be estimated to be
0.5
k
ku  uh k2s ffi ch ; ð19Þ
in which c is a constant and h denotes the element size. If a shell p( ) / p0 0
element is uniformly optimal, the constant is independent of the
shell thickness and k represents the optimal order of convergence,
with k = 2 for the 3-node shell finite element, and k = 4 for the 6- -0.5
node and the enriched 3-node shell elements.
Instead of the unknown exact solution, a reference finite ele-
ment solution uref calculated using a very fine mesh and a known -1
reliable element can be used, hence Eq. (17) becomes 0 30 60 90
Z
kuref  uh k2s ¼ DeT DsdXref with De ¼ eref  eh ;
Xref
Fig. 11. Cylindrical shell problem (4  4 mesh, L = R = 1.0, E = 2.0  105, v = 1/3 and
Ds ¼ sref  sh : ð20Þ p0 = 1.0).
To measure the convergence of the finite elements in the shell
problems, we use the relative error Eh
The numerical procedure to calculate the s-norm for shell finite ele-
kuref  uh k2s
Eh ¼ : ð21Þ ment solutions with general types of elements and general meshes
kuref k2s is explained in detail in Ref. [24]. In the use of Eq. (21), it is very
 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142 135

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 12. Convergence curves for the clamped cylindrical shell problem with uniform meshes. The bold line represents the optimal convergence rate, which is 2.0 for linear
elements and 4.0 for quadratic elements.

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 13. Convergence curves for the clamped cylindrical shell problem with the distorted meshes shown in Fig. 7. The bold line represents the optimal convergence rate,
which is 2.0 for linear elements and 4.0 for quadratic elements.

important to use accurate reference solutions calculated by a pressure load. Due to symmetry, only one quarter of the plate is
reliable shell finite element. In this study, we use well-converged modeled, with u x ¼ 
hy ¼ 0 along BC, u y ¼ 
hx ¼ 0 along DC, and
reference solutions calculated using fine meshes of the MITC9 shell x ¼ u
u y ¼ u z ¼  hx ¼ 
hy ¼ 0 along AB and AD [23,24].
finite element. Of course, the MITC9 shell finite element is known to We study the convergence behavior not only using uniform
satisfy the ellipticity and consistency conditions and to show ade- meshes but also distorted meshes, as shown in Fig. 7. When the
quate convergence behavior, see Refs. [22,35,36]. N  N distorted mesh is used, each edge is divided by the ratio
In the following sections, we present the convergence curves of L1 : L2 : L3 : . . . ; LN ¼ 1 : 2 : 3 : . . . N, leading to quite distorted
the MITC3, MITC4, MITC6 shell elements and the fully enriched meshes. The reference solution is obtained with a uniform mesh
MITC3 shell element to identify the performance of the enriched of 96  96 MITC9 shell finite elements. We use N = 8, 16, 32 and
MITC3 shell element compared to other shell elements. Note that 64 for the MITC3 and MITC4 shell elements, and N = 4, 8, 16 and
some convergence curves of the MITC3, MITC4, and MITC6 shell 32 for the MITC6 shell element and the enriched MITC3 shell ele-
elements have been published before in Refs. [23,24,27]. ment. Note that in these N  N meshes N2 MITC4 elements and
2N2 triangular elements are used throughout the paper. Also, in
5.1. Fully clamped square plate problem the figures of results we consider the cases t/L = 1/100, 1/1000
and 1/10,000 and use as the ‘‘element size’’ h = L/N. To fairly com-
The plate problem shown in Fig. 6 is solved. A square plate of pare convergence behaviors among different shell elements, the
size 2L  2L and constant thickness t is subjected to a uniform equivalent element sizes 2h, h and 1.2h are used for the MITC3
136 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 14. Convergence curves for the free cylindrical shell problem with uniform meshes. The bold line represents the optimal convergence rate, which is 2.0 for linear
elements and 4.0 for quadratic elements.

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 15. Convergence curves for the free cylindrical shell problem with the distorted meshes shown in Fig. 7. The bold line represents the optimal convergence rate, which is
2.0 for linear elements and 4.0 for quadratic elements.

and MITC4 shell elements, the MITC6 shell element, and the en- Fig. 8 shows the convergence curves of the original MITC3 shell
riched MITC3 shell element. When using these equivalent element element and the enriched MITC3 shell elements based on the dif-
sizes, the numbers of degrees of freedom are similar. ferent assumptions for the transverse shear strain fields. The en-
To identify the dependency of the convergence behavior on the riched MITC3 shell element shows different solution accuracy
MITC scheme chosen for the enriched MITC3 shell element, the fol- highly depending on the assumed covariant transverse shear strain
lowing three schemes are considered for the covariant transverse field used. When the assumed covariant transverse shear strain
shear strain field of the additional quadratic displacement field of the MITC6 shell element is employed, an almost ideal con-
interpolation: vergence behavior is observed in this fully clamped square plate
problem. Note that the other enriched MITC3 shell elements show
– No MITC scheme is used (denoted by DISP in Fig. 8). an even worse convergence behavior than the original MITC3 shell
– The MITC6a scheme is used, for this scheme see Ref. [23]. element. Therefore, in the following sections, we only use the en-
– The MITC6 scheme in Eq. (13) is used (referred to as MITC6b in riched MITC3 shell element based on the MITC6 scheme in Eq. (13).
Fig. 8) Figs. 9 and 10 present the convergence curves for the fully
clamped square plate problems using uniform and distorted
In all these cases, of course, the MITC3 scheme in Eq. (12) is meshes, respectively. The performance of the enriched MITC3 shell
used for the assumed covariant transverse shear strain field of element is much better than the performance of the MITC3 and
the standard linear displacement interpolation. MITC6 shell elements. One reason is probably that the enrichments
 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142 137

z -1
0
1
1
y
0.5 x
2L
y 0 D C

-0.5

-1 B
-1 A
0 θ
1
x z
(a)

6 t

(b) (c)
Fig. 16. Hyperboloid shell problem (E = 2.0  1011, v = 1/3 and p0 = 1.0). (a) Shell geometry and boundary conditions, (b) Graded mesh for the clamped case (8  8 mesh,
t/L = 1/1000), (c) Mesh for the free case (8  8 mesh).

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 17. Convergence curves for the clamped hyperboloid shell problem with uniform meshes. The bold line represents the optimal convergence rate, which is 2.0 for linear
elements and 4.0 for quadratic elements.

span over the cover regions and distortions within the regions are pðhÞ ¼ p0 cosð2hÞ: ð22Þ
not as severe (are smoothed out) in comparison to not using cov-
ers. The MITC4 and enriched MITC3 shell finite elements show al-
The shell problem gives two different asymptotic behaviors
most optimal convergence behaviors, even though the distorted
depending on the boundary conditions at both ends: the bending-
meshes are used.
dominated behavior under free boundary conditions and the mem-
brane-dominated behavior under clamped boundary conditions.
5.2. Cylindrical shell problems Using the symmetry of the problem, the region ABCD in Fig. 11
is modeled. In the membrane-dominated case, the clamped
A cylindrical shell with uniform thickness t, length 2L, and ra- boundary condition is imposed: u  ¼ 0 along BC, u
x ¼ b y ¼ a
¼0
dius R is considered, as shown in Fig. 11. The loading is a smoothly along DC uz ¼ a
 ¼ 0 along AB, and u
x ¼ u z ¼ a
y ¼ u ¼b ¼ 0 along
varying periodic pressure p(h) normal to the shell surface AD. In the bending-dominated case, the free boundary condition
138 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 18. Convergence curves for the clamped hyperboloid shell problem with the distorted meshes shown in Fig. 7. The bold line represents the optimal convergence rate,
which is 2.0 for linear elements and 4.0 for quadratic elements.

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 19. Convergence curves for the free hyperboloid shell problem with uniform meshes. The bold line represents the optimal convergence rate, which is 2.0 for linear
elements and 4.0 for quadratic elements.

is imposed: u  ¼ 0 along BC, u

x ¼ b y ¼ a
 ¼ 0 along DC, and meshes, respectively. When distorted meshes are used, the solu-
z ¼ a
u  ¼ 0 along AB [23]. tions of the MITC3, MITC4, and MITC6 shell elements deteriorate
The reference solution is calculated using a mesh of 96  96 as the shell thickness decreases, due to some locking. However,
MITC9 shell finite elements for both cases. The solutions using the enriched MITC3 shell element presents good convergence
the MITC3, MITC4, MITC6 and enriched MITC3 shell elements are behavior even when using the distorted meshes.
obtained with N  N meshes (N = 8, 16, 32 and 64 for the MITC3
and MITC4 shell elements and N = 4, 8, 16, and 32 for the MITC6 5.3. Hyperboloid shell problems
and enriched MITC3 shell elements). The element size used in
the figures is h = L/N. The distorted meshes used are generated as The hyperboloid shell shown in Fig. 16 is considered, where the
shown in Fig. 7. midsurface of the shell structure is given by
Figs. 12 and 13 present the convergence behaviors for the
clamped cylindrical shell problems with uniform and distorted x2 þ y2 ¼ 1 þ y2 ; y 2 ½1; 1: ð23Þ
meshes, respectively. All the shell finite elements considered show
excellent convergence behavior with uniform and distorted A smoothly varying periodic pressure is applied normal to the sur-
meshes. face, as in Fig. 11,
Figs. 14 and 15 present the convergence curves for the free
cylindrical shell problems obtained with uniform and distorted pðhÞ ¼ p0 cosð2hÞ; ð24Þ
 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142 139

MITC3 MITC4 MITC6 Enriched MITC3

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )

l o g ( re l a t i v e e r ro r )
-1.0 -1.0 -1.0 -1.0

-1.5 -1.5 -1.5 -1.5

-2.0 -2.0 -2.0 -2.0

-2.5 -2.5 -2.5 -2.5

-3.0 -3.0 -3.0 -3.0

-3.5 -3.5 -3.5 -3.5

-4.0 -4.0 -4.0 -4.0

-1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3 -1.8 -1.0 -0.3
log(2h) log(2h) log(h) log(1.2h)
t/L = 1/100 t/L = 1/1000 t/L = 1/10000

Fig. 20. Convergence curves for the free hyperboloid shell problem with the distorted meshes shown in Fig. 7. The bold line represents the optimal convergence rate, which is
2.0 for linear elements and 4.0 for quadratic elements.

When both ends are clamped, a membrane-dominated problem is

0.01 radius
obtained, and when the ends are free, a bending-dominated prob-
lem is obtained. The bending-dominated hyperboloid shell problem
is known to be difficult to solve [23,24,41].
Due to symmetry, the analyses are carried out using one-eighth
of the structure, corresponding to the shaded region ABCD in 0.08
Fig. 16(a). For the membrane-dominated case, the clamped bound-
ary condition is imposed using: u z ¼ b ¼ 0 along BC, u ¼0
x ¼ b
along AD, and uy ¼ a ¼ 0 along DC, and ux ¼ uy ¼ uz ¼ a ¼ b
      ¼0
along AB. For the bending- dominated case, the free boundary con- Line load of 1000 N/m
dition is imposed using: u  ¼ 0 along BC, u
z ¼ b  ¼ 0 along AD,
x ¼ b 0.0075 radius
and uy ¼ a ¼ 0 along DC [23,24].
  0.08
In common with previous problems, we use the reference solu-
tions calculated with a mesh of 96  96 MITC9 shell elements. The 0.002 radius fillet
solutions using the MITC3, MITC4, MITC6 and enriched MITC3 shell 0.04
elements are obtained with N  N meshes (N = 8, 16, 32 and 64 for
the MITC3 and MITC4 shell elements and N = 4, 8, 16, and 32 for the
MITC6 and enriched MITC3 shell elements). The element size used Clamped
in the convergence curves is h = L/N. In the p clamped
ffiffi hyperboloid
shell problem, a boundary layer of width 6 t is used for half of Fig. 21. Shaft–shaft interaction problem with fillets (E = 2.07  1011, v = 0.29).
the mesh, see Fig. 16(b). In the free hyperboloid shell problem,
the thin boundary layer is not specially meshed.
Figs. 17 and 18 show the convergence curves for both uniform using conventional standard shell finite elements with mesh
and distorted meshes in the membrane dominated case (that is, refinements. We consider two numerical examples: a shaft–shaft
the clamped hyperboloid shell problem). The performance of all interaction problem and a ‘monster’ shell problem. The maximum
shell elements is good. effective stress, the strain energy and the deformed shape will be
For the bending-dominated case (that is, the free hyperboloid evaluated with and without using local enrichments..
shell problem), the convergence curves are shown in Figs. 19 and
20. The enriched MITC3 shell element shows the best convergence 6.1. Shaft–shaft interaction problem
behavior among the shell elements considered. Even in the use of
distorted meshes, the performance of the enriched MITC3 shell ele- Consider the two cylindrical shafts connected with fillets of
ment is excellent while the other shell elements show some degree radius 0.002 m, in which the horizontal shaft is subjected to a line
of locking. load of 1000 N/m, as shown in Fig. 21, and the vertical shaft is fully
clamped at its lower end [42]. Fig. 22 presents the distribution of
6. Local use of cover interpolations the effective stress (von Mises stress) obtained using the MITC4
and MITC3 shell elements, and using the enriched MITC3 shell
In the convergence studies given in Section 5, we showed a element throughout the mesh or only locally. The reference solu-
good performance when the enriched MITC3 shell element is used tion is given by a fine mesh of the MITC4 shell element, in which
throughout the mesh. In this section, we illustrate the local use of 2,150 elements and 10,805 DOFs are used, see Fig. 22(a).
cover interpolation functions over the solution domains. This Fig. 22(b) presents a finer mesh of the MITC3 shell element and
scheme of increasing the solution accuracy is quite different from Fig. 22(c)–(e) show the same coarse mesh used for the MITC3 shell
140 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142

(a) (b)

Effective stress

MITC4 MITC3
(DOFs = 10,805, Reference stress) (DOFs = 12,750, Error = 5.68%)

(c) (d) (e)

MITC3 Fully enriched MITC3 Locally enriched MITC3

(DOFs = 3,125, Error = 27.54%) (DOFs = 9,375, Error = 5.28%) (DOFs = 3,845, Error = 6.75%)

Fig. 22. Distributions of effective stress for the shaft–shaft interaction problem: for (a) the 2,193 node model of the MITC4 shell elements, (b) the 2,582 node model of the
MITC3 shell elements, (c) the 641 node model of the MITC3 shell elements, (d) the  641 node  model fully enriched, and (e) the 641 node model locally enriched. The red dot
represents enriched nodes (DOFs: total number of degrees of freedom used, Error = rref ref
v  rv =rv  100). (For interpretation of the references to color in this figure legend,
h

the reader is referred to the web version of this article.)

Table 3  
Relative errors in maximum effective stress in the shaft–shaft interaction problem for the five different shell models in Fig. 22. Relative error (%) ¼ rref ref
max  rmax =rmax  100.
h

Fine mesh Coarse mesh

MITC4 (reference) MITC3 MITC3 Fully enriched MITC3 Locally enriched MITC3
Elements 2,150 5,078 1,240 1,240 1,240
Nodes 2,193 2,582 641 641 641
Enriched nodes – – – 641 72
Free DOFs 10,805 12,750 3,125 9,375 3,845
Max. effective stress (rmax) 1.78E+09 1.68E+09 1.29E+09 1.69E+09 1.66E+09
Relative error (%) – 5.68 27.54 5.28 6.75

element and the fully and locally enriched shell models. The red is no exact solution to the problem, we use the reference solution
dots in Fig. 22(e) represent the selected nodes carrying interpola- given by a fine mesh of 48 (axial)  192 (circumferential) MITC4
tion covers around the fillet area where stress concentration is shell elements; see Fig. 24(a). This is a sufficiently fine mesh to
expected. identify and reasonably resolve the boundary layer on the free
Table 3 gives the numbers of elements, nodes, degrees of free- edge. Fig. 24(b) presents a finer mesh of MITC3 shell elements,
dom used, and the relative errors in the maximum effective stress and Fig. 24(c)–(e) show a coarser mesh for the MITC3 shell element
obtained when using the shell models in Fig. 22. In the shaft–shaft and the fully and locally enriched cases.
interaction problem, the maximum effective stress is obtained Fig. 24 shows the calculated deformed shapes of the shell when
around the fillet area. Using the local enrichments, the maximum the shell thickness is 0.001 (t/L = 1/10,000). For visualization, the
effective stress is well predicted with a much smaller number of displacements are normalized so that the maximum outward total
degrees of freedom. displacement value is equal to 2.0. We note that the displacements
are dominant in the immediate vicinity of the free boundary,
6.2. A ‘‘highly-sensitive’’ shell problem namely within the boundary layer. Fig. 24(b) and (c) show that
the MITC3 shell element meshes are not effective in predicting the
Fig. 23 shows the problem considered (referred to also as ‘‘the displacement oscillations1 in the circumferential direction ithin the
monster shell problem’’) [12]. The shell geometry corresponds to
a half-sphere with the top sliced off. The shell is clamped around 1
In the monster shell problem, the number of displacement oscillations in the
its entire lower boundary. A smoothly distributed pressure is boundary layer increases as the shell thickness decreases. The number is given by
applied over a small part of the interior of the shell. Since there log (L/t).
 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142 141

Z boundary layer. However, when the coarse MITC3 shell element mesh
is fully enriched, the displacements in the boundary layer are
calculated accurately as shown in Fig. 24(d). Also, the local use of
the cover interpolations within the boundary layer results in excellent
overall accuracy with a significantly reduced number of degrees of
18 freedom, see Fig. 24(e). Table 4 shows the number of elements, nodes
and degrees of freedom used, the number of displacement oscilla-
tions, and the relative errors in the strain energies.

7. Conclusions

In this paper, we have proposed a 3-node shell finite element

enriched by interpolation covers based on the MITC method. The
enriched MITC3 shell finite element is obtained by applying linear
36 displacement interpolation covers to the standard 3-node shell ele-
ment, and using the MITC procedure. Good convergence behavior
in the analysis of various shell problems has been seen, even when
Clamped distorted meshes are used. The fact that, in the solutions consid-
ered, the enriched 3-node element performs sometimes signifi-
Y cantly better than the MITC4 and MITC6 shell elements when
X distorted meshes are used is particularly noteworthy. The effec-
tiveness of using the enrichment scheme only locally was also
Fig. 23. A ‘‘highly-sensitive’’ shell problem. (L = R = 10, E = 6.285  107, v = 0.3, and illustrated by using the cover interpolation functions only in criti-
2
pðrÞ ¼ e4r ). cal areas of an analysis domain.

Fig. 24. Deformed shapes for the monster shell (t/L = 1/10, 000): for (a) the 48 (axial)  192 (circumferential) mesh of the MITC4 shell elements, (b) the 48  192 mesh of the
MITC3 shell elements, (c) the 16  64 mesh of the MITC3 shell elements, (d) the 16  64 mesh model fully enriched, and (e) the 16  64 mesh model locally enriched. In the
figure (e), the red dot represents enriched nodes (DOFs: the total number of degrees of freedom used, Error = (Eref  Eh)/Eref  100). (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

Table 4
Relative errors in strain energy in the monster shell problem for the five different shell models in Fig. 24. Relative error (%) = (Eref  Eh)/Eref  100.

Fine mesh 48 (axial)  192 (circumferential) Coarse mesh 16 (axial)  64 (circumferential)

MITC4 (reference) MITC3 MITC3 Fully enriched MITC3 Locally enriched MITC3
Elements 9,216 18,432 2,048 2,048 2,048
Nodes 9,408 9408 1,088 1,088 1,088
Enriched nodes – – – 1,088 409
Free DOFs 46,080 46,080 5,120 15,360 9,120
Oscillations 4 3 2 4 4
Strain energy (E) 5.21E04 2.83E04 1.26E04 5.37E04 4.87E04
Relative error (%) – 45.66 75.81 3.11 6.49
142 H.M. Jeon et al. / Computers and Structures 134 (2014) 128–142

We can conclude that cover interpolations and the MITC meth- [17] Zhuo Z, BinBin C. A novel enriched CB shell element method for simulating
arbitrary crack growth in pipes. Sci China Phys Mech Astron 2011;54:
od are promising schemes for enriching shell element behaviors.
1520–31.
Based on the observed good behavior of the enriched MITC3 shell [18] Dvorkin EN, Bathe KJ. A continuum mechanics based four-node shell element
finite element, we expect that an enriched MITC4 shell element for general nonlinear analysis. Eng Comput 1984;1:77–88.
will likely also be effective when distorted meshes must be used [19] Bathe KJ, Dvorkin EN. A formulation of general shell elements – the use of
mixed interpolation of tensorial components. Int J Numer Methods Eng
in areas of shell bending. In future studies, it would be valuable 1986;22:697–722.
to mathematically analyze the method, to further test the scheme, [20] Brezzi F, Bathe KJ, Fortin M. Mixed-interpolated elements for Reissner/Mindlin
and to develop the method for nonlinear analysis [1,18,43] and for plates. Int J Numer Methods Eng 1989;28:1787–801.
[21] Bucalem ML, Bathe KJ. Higher-order MITC general shell elements. Int J Numer
solving problems containing very high stress gradients, like shell Methods Eng 1993;36:3729–54.
problems containing cracks [32,33,44]. In all of these develop- [22] Bathe KJ, Lee PS, Hiller JF. Towards improving the MITC9 shell element.
ments, the ultimate aim should probably be to establish automatic Comput Struct 2003;81:477–89.
[23] Lee PS, Bathe KJ. Development of MITC isotropic triangular shell finite
procedures for improving stress solutions [45]. elements. Comput Struct 2004;82:945–62.
[24] Lee PS, Noh HC, Bathe KJ. Insight into 3-node triangular shell finite elements:
Acknowledgments the effects of element isotropy and mesh patterns. Comput Struct
2007;85:404–18.
[25] Beirão da Veiga L, Chapelle D, Paris Suarez I. Towards improving the MITC6
This research was supported by the Basic Science Research Pro- triangular shell element. Comput Struct 2007;85:1589–610.
gram through the National Research Foundation (NRF) of Korea [26] Bucalem ML, Bathe KJ. The mechanics of solids and structures – hierarchical
modeling and the finite element solution. Springer; 2011.
funded by the Ministry of Science, ICT and Future Planning (No.
[27] Chapelle D, Bathe KJ. The finite element analysis of shells – fundamentals. 2nd
2011-0014387). ed. Berlin: Springer; 2011.
[28] Lee Y, Yoon K, Lee PS. Improving the MITC3 shell finite element by using the
References Hellinger-Ressner principle. Comput Struct 2012;110–111:93–106.
[29] Shi GH. Manifold method of material analysis. In: Trans. 9th Army Conf. on
Applied Mathematics and Computing; 1991. p. 57–76.
[1] Bathe KJ. Finite element procedures. Cambridge, MA: Klaus-Jürgen Bathe; [30] Strouboulis T, Copps K, Babuška I. The generalized finite element method.
2006. Comput Methods Appl Mech Eng 2001;190:4081–193.
[2] Bathe KJ, Chaudhary A. On the displacement formulation of torsion of shafts [31] Tian R, Yagawa G, Terasaka H. Linear dependence problems of partition of
with rectangular cross-sections. Int J Numer Methods Eng. 1982;18:1565–8. unity-based generalized FEMs. Comput. Methods Appl. Mech. Eng
[3] Yoon KH, Lee YG, Lee PS. A continuum mechanics based 3-D beam finite 2006;195:4768–82.
element with warping displacements and its modeling capabilities. Struct Eng [32] Hong WT, Lee PS. Coupling flat-top partition of unity method and finite
Mech 2012;43–4:411–37. element method. Finite Elem Anal Des 2013;67:43–55.
[4] Benzley S. Representation of singularities with isoparametric finite elements. [33] Hong WT, Lee PS. Mesh based construction of flat-top partition of unity. App
Int J Numer Methods Eng 1974;8:537–45. Math Comput 2013;219:8687–704.
[5] Dvorkin EN, Cuitiño AM, Gioia G. Finite elements with displacement [34] Kim DN, Bathe KJ. A triangular six-node shell element. Comput Struct
interpolated embedded localization lines insensitive to mesh size and 2009;87:1451–60.
distortions. Int J Numer Methods Eng 1990;30(3):541–64. [35] Lee PS, Bathe KJ. The quadratic MITC plate and MITC shell elements in plate
[6] Belytschko T, Black T. Elastic crack growth in finite elements with minimal bending. Adv Eng Software 2010;41:712–28.
remeshing. Int J Numer Methods Eng 1999;45:601–20. [36] Hiller JF, Bathe KJ. Measuring convergence of mixed finite element
[7] Moes N, Dolbow J, Belytschko T. A finite element method for crack growth discretizations: an application to shell structures. Comput Struct
without remeshing. Int J Numer Methods Eng 1999;46:131–50. 2003;81:639–54.
[8] Daux C, Moes N, Dolbow J, Sukumar N, Belytschko T. Arbitrary branched and [37] Bathe KJ, Iosilevich A, Chapelle D. An inf-sup test for shell finite elements.
intersecting cracks with the extended finite element method. Int J Numer Comput Struct 2000;75:439–56.
Methods Eng 2000;48:1741–60. [38] Chapelle D, Bathe KJ. The mathematical shell model underlying general shell
[9] Babuška I, Melenk JM. The partition of unity method. Int J Numer Methods Eng elements. Int J Numer Methods Eng 2000;48:289–313.
1997;40:727–58. [39] Bathe KJ. The inf-sup condition and its evaluation for mixed finite element
[10] Kim J, Bathe KJ. The finite element method enriched by interpolation covers. methods. Comput Struct 2001;79:243–52.
Comput Struct 2013;116:35–49. [40] Bathe KJ, Lee PS. Measuring the convergence behavior of shell analysis
[11] Lee PS, Bathe KJ. On the asymptotic behavior of shell structures and the schemes. Comput Struct 2011;89:285–301.
evaluation in finite element solutions. Comput Struct 2002;80(3–4):235–55. [41] Lee PS, Bathe KJ. Insight into finite element shell discretizations by use of the
[12] Bathe KJ, Chapelle D, Lee PS. A shell problem ‘highly-sensitive’ to thickness ‘‘basic shell mathematical model’’. Comput Struct 2005;83:69–90.
changes. Int J Numer Methods Eng 2003;57(8):1039–52. [42] ADINA Primer: ARD 12-7, ADINA R & D Inc., Watertown, MA; December
[13] Bathe KJ, Zhang H. A mesh adaptivity procedure for CFD & fluid–structure 2012.
interactions. Comput Struct 2009;87:604–17. [43] Sussman T, Bathe KJ. 3D-shell elements for structures in large strains. Comput
[14] Areias PM, Belytschko T. Non-linear analysis of shells with arbitrary evolving Struct 2013;122:2–12.
cracks using XFEM. Int J Numer Methods Eng 2005;62:384–415. [44] Thanh CD, Zi G, Lee PS, Rabczuk T, Song JH. Phantom-node method for shell
[15] Areias PM, Song JH, Belytschko T. Analysis of fracture in thin shells by models with arbitrary cracks. Comput Struct 2012;92–93:242–56.
overlapping paired elements. Comput Method Appl Mech 2006;195:5343–60. [45] Kim J, Bathe KJ. Towards a procedure to automatically improve finite element
[16] Wyart E, Coulon D, Duflot M, Pardoen T, Remacle J-F, Lani F. A substructured solutions by interpolation covers. Comput Struct 2014;131:81–97.
FE-shell/XFE-3D method for crack analysis in thin-walled structures. Int J
Numer Methods Eng 2007;72:757–79.