Computers and Structures 110–111 (2012) 93–106

Contents lists available at SciVerse ScienceDirect

Computers and Structures

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

Improving the MITC3 shell finite element by using the Hellinger–Reissner principle
Youngyu Lee, Kyungho Yoon, Phill-Seung Lee ⇑
Division of Ocean Systems Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 305-701, Republic of Korea

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

Article history: The objective of this study is to improve the performance of the MITC3 shell finite element. The
Received 12 December 2011 Hellinger–Reissner (HR) variational principle is modified in the framework of the MITC method, and a
Accepted 5 July 2012 special approximated transverse shear strain field is proposed. The MITC3-HR shell finite element
Available online 1 August 2012
improved by using the Hellinger–Reissner functional passes all the basic tests (zero energy mode test,
patch test, and isotropic element test). Convergence studies considering a fully clamped plate problem,
Keywords: a sixty-degree skew plate problem, cylindrical shell problems, and hyperboloid shell problems demon-
Shell structures
strate the improved predictive capability of the new 3-node shell finite element.
Finite elements
Hellinger–Reissner principle
Ó 2012 Elsevier Ltd. All rights reserved.
MITC method

1. Introduction The MITC3 triangular shell finite element shows much better
predictive capability than the displacement-based 3-node triangu-
Shell structures have been widely used in many engineering lar shell finite elements and other 3-node isotropic triangular shell
applications, and the finite element method has been dominantly finite elements [15,18]. However, the locking alleviation by MITC3
adopted for the analysis of shells. Since the effectiveness of a shell is not as large as that by MITC4; that is, the accuracy of the solu-
finite element analysis depends highly on the predictive capability tions is not as good as that of the MITC4 quadrilateral shell finite
of shell finite elements, there is strong demand for the develop- elements due to locking. This provided the motivation for this
ment of more effective shell finite elements [1–3]. work.
The kinematical behavior of shells is very complicated, espe- The purpose of this paper is to improve the MITC3 shell finite
cially as the shell thickness becomes smaller. Depending on shell element. With the help of the Hellinger–Reissner (HR) principle
geometries, boundary conditions, and applied loadings, the asymp- [1], we additionally approximate the transverse shear strain fields
totic behavior can be dominated by membrane or bending actions of the MITC3 shell finite element. The Hellinger–Reissner (HR)
or a combination of both actions [3–10]. A reliable shell finite ele- functional has been used to alleviate locking in plate and shell fi-
ment should provide accurate solutions irrespective of the compli- nite elements [19,20]. The successful use of this method depends
cated asymptotic behaviors and the magnitude of the shell on how the approximated transverse shear strain fields are con-
thickness. However, it is extremely challenging to develop such structed. We first modify the Hellinger–Reissner functional and
shell finite elements due to a locking phenomenon; that is, the introduce a special approximated transverse shear strain field
shell finite element becomes too stiff when the thickness is rela- based on rotated contravariant bases.
tively small in bending situations [1,11]. In the following sections, the MITC3 formulation is briefly re-
The MITC (Mixed Interpolation of Tensorial Components) meth- viewed and the Hellinger–Reissner principle for the shell finite ele-
od has been successfully used to develop shell finite elements to ment is presented. We then propose a method that involves the use
reduce the locking effects [12–17]. The MITC4 quadrilateral shell of the Hellinger–Reissner functional to improve the MITC3 shell fi-
finite elements were first developed by Dvorkin and Bathe nite element, after which we explain how to construct the special
[12,13]. The MITC9 and MITC16 quadrilateral shell finite elements approximated transverse shear strain field. The basic test results
were later developed by Bucalem and Bathe [14]. In spite of the and well-established convergence studies numerically show that
fact that triangular elements are very useful for the mesh genera- the MITC3 shell finite element is successfully improved.
tion of arbitrary shell geometries, due to the relatively complicated
representation of triangular geometries in the element formula-
tion, the MITC method was recently applied to develop isotropic 2. Formulation of the MITC3 shell finite element
triangular shell elements, MITC3 and MITC6, by Lee and Bathe [15].
The geometry of a 3-node continuum mechanics based triangu-
⇑ Corresponding author. Tel.: +82 42 350 1512; fax: +82 42 350 1510. lar shell finite element is interpolated by Lee and Bathe [15] and
E-mail address: [email protected] (P.S. Lee). Lee et al. [18]

0045-7949/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruc.2012.07.004
94 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106

X3
tX 3
! The covariant strain components can be expressed by
~
xðr; s; tÞ ¼ hi ðr; sÞ~
xi þ ai hi ðr; sÞ V in ; ð1Þ
2 !
i¼1 i¼1 eij ¼ bij U ; ð5Þ
where hi(r, s) is the 2D shape function of the standard isoparametric !
in which bij is the covariant strain–displacement matrix and U is
procedure corresponding to node i, ~ xi is the position vector for node the vector of nodal displacements and rotations, which include ~ uk ,
!
i in the global Cartesian coordinate system, and ai and V in denote ak , and bk.
the shell thickness and the director vector at node i, respectively; The base vectors of the shell-aligned local Cartesian coordinate
see Fig. 1. system are defined as follows
The displacement interpolation of the element is obtained by
!
! ~
gt ! ~
g s  L t ! ! !
X3
tX 3
! ! L t ¼ ; L r ¼ ! ; L s ¼ L t  L r : ð6Þ
~
uðr; s; tÞ ¼ hi ðr; sÞ~
ui þ ai hi ðr; sÞð V i2 ai þ V i1 bi Þ; ð2Þ j~
gt j j~
g s  L t j
i¼1
2 i¼1
The strains (eij) defined in the shell-aligned local Cartesian coordi-
in which ~ ui is the nodal displacement vector in the global Cartesian nate are calculated from the covariant strain components through
! ! !
coordinate system, V i1 and V i2 are unit vectors orthogonal to V in the following relation,
and to each other, and ai and bi are the rotations of the director vec-
! ! ! ! ! ! ! ! ! ! !
tor V in about V i1 and V i2 at node i. gm  ~
eij ð L i  L j Þ ¼ emn ð~ g n Þ with L 1 ¼ L r ; L 2 ¼ L s ; L 3 ¼ L t :
The linear terms of the displacement-based covariant strain
ð7Þ
components are given by
In Eq. (7), the contravariant base vectors g are given by ~j
1
eij ¼ ð~
gi  ~
u;j þ ~
gj  ~
u;i Þ; ð3Þ
2 ~ g j ¼ dji
gi  ~ g1 ¼ ~
with ~ gr ; ~
g2 ¼ ~
gs ; ~
g3 ¼ ~
gt ; ð8Þ
where dji
where is the Kronecker delta in mixed form.
@~
x @~
u The strain vector defined in the shell-aligned local Cartesian
~
gi ¼ ; ~
u;i ¼ with r 1 ¼ r; r2 ¼ s; r3 ¼ t: ð4Þ coordinate system from the displacement-based shell formulation
@r i @r i
is
!
e ¼ BU ;
~ ð9Þ
T
where ~ e ¼ ½ err ess 2ers 2est 2ert  .
In the formulation of the MITC3 shell finite element, the covar-
iant in-plane strain field is calculated by the displacement-based
triangular shell formulation in Eq. (3) and the MITC method is only
applied to substitute the covariant transverse shear strain field
[15]. The assumed covariant transverse shear strain components,
which are spatially isotropic, are given by

~ert ¼ eð1Þ
rt þ cs;
~est ¼ eð2Þ
st  cr; ð10Þ
ð2Þ ð1Þ ð3Þ ð3Þ ðnÞ ðnÞ
where c ¼  est ert
þ est ert
and, at the tying points, and est ert
are calculated from Eq. (3), see Fig. 2.
The assumed covariant transverse shear strain components of
the MITC3 element can also be expressed by
~ !
~eij ¼ bij U : ð11Þ
The covariant strains of the MITC3 shell finite element are trans-
formed to the strains defined in the shell-aligned local Cartesian
! ! !
coordinate system ð L r ; L s ; L t Þ
!  
Fig. 1. A 3-node triangular continuum mechanics based shell finite element. eM ¼ BM U with ~eM ¼ eMrr
~ eMss 2eMrs 2eMst 2eMrt T : ð12Þ

Fig. 2. Tying positions for the transverse shear strain of the MITC3 triangular shell finite element. The constant transverse shear strain condition is imposed along its edges.
 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106 95

3. The Hellinger–Reissner principle for 3-node shell finite based on the Hellinger–Reissner principle highly depends on how
elements the interpolation function is constructed in Eq. (18).

The Hellinger–Reissner functional (HR functional), which is de- 4. Formulation of the MITC3-HR shell finite element
rived from the Hu–Washizu functional [1], is given by
Z   Z Z We here modify the Hellinger–Reissner functional in Eq. (14) as
1 T T
PHR ¼  ~ e C~e þ ~eT C@ e~
u dV  ~ uT~
f B dV  uSf ~
~ f Sf dS follows
V 2 V Sf Z  
Z 1 MT 1 T T
T PM
HR ¼ ebm C bm~eMbm  ~cA C c~cA þ ~
~ cA C c~cM dV
 ~f Su ð~
uSu  ~
up Þ dS; ð13Þ V 2 2
Su Z Z Z
T T
 ~ uT~f B dV  uSf ~
~ f Sf dS  ~
f Su ð~
uSu  ~
up Þ dS ð19Þ
where ~ e is the strain vector, C is the stress–strain law matrix, ~ u is the V Sf Su
displacement vector, @ e is the differential operator on ~ u for strain
calculations, ~f B is the vector for body forces, ~f Sf is the vector for ap- with
0 1
plied surface tractions, ~ up is the vector for prescribed displacements, eMrr ! !
B M C M ! 2eM M! 2eAst
V is the volume of the body, Sf is the surface on which ~ f Sf is applied, M M
ebm ¼ @ ess A ¼ Bbm U ; c ¼
~ ~ st A
¼ Bc U ; c ¼
~ ¼ BAc ~
a;
Su is the surface on which ~ uSf is the displacement
up is prescribed, ~ M
2eM
rt 2eArt
2ers
vector over Sf, ~ uSu is the displacement vector over Su, and ~ f Su is the
ð20Þ
reaction vector calculated by the stresses on Su. It is important to
note that the independent variables are strains and displacements ~M ~M
in which e and c are the vector for the bending and membrane
bm
in this Hellinger–Reissner functional. strains and the vector for the transverse shear strains obtained by
For the 3-node shell finite element formulation, the Hellinger– Eq. (12) in the MITC3 shell finite element formulation.
Reissner functional is changed to separate the bending and mem- This is a very important modification of the Hellinger–Reissner
brane strains and the transverse shear strains as follows functional. As a result, the new element, MITC3-HR, is based on the
Z   formulation of the MITC3 shell finite element. It provides strong
1 T 1 AT T
PSHR ¼ ebm C bm~ebm  ~
~ c C c~ cA þ ~
cA C c~c dV potential to improve the MITC3 shell finite element. Since the MITC
V 2 2
Z Z Z formulation can be represented by the Hellinger–Reissner princi-
T T
 ~ uT~f B dV  uSf ~
~ f Sf dS  ~
f Su ð~
uSu  ~
up Þ dS; ð14Þ ple [21], the Hellinger–Reissner principle is, in effect, used twice
V Sf Su in our formulation.
Another important part for successful development is the ap-
where ~ebm is the vector for the displacement-based bending and proach taken to construct the approximated transverse strain field,
membrane strains, ~ c is the vector for the displacement-based trans- which should retain the isotropic property as well as improve the
verse shear strains, and ~cA is the vector for the approximated trans- predictive capability of the MITC3 shell finite element. For the con-
verse shear strains with unknowns,
struction of ~cA in Eq. (20), we use a special approximated trans-
0 1
err   ! verse shear strain field introduced on the rotated contravariant
B C 2est A 2eAst base vectors. The components of ~ cA are given by
ebm ¼ @ ess A; ~c ¼
~ c ¼
; ~ ; ð15Þ
2ert 2eArt ! ! ! !
2ers 2eArt ¼ 2~eAst ½ð~
0
g s  L r Þð~
0
g t  L t Þ þ ð~
0
g t  L r Þð~
0
g s  L t Þ
0 ! 0 ! 0 ! 0 !
and all the strain components are defined in the shell-aligned local þ 2~eArt ½ð~ g r  L r Þð~g t  L t Þ þ ð~ g t  L r Þð~g r  L t Þ;
Cartesian coordinate system to satisfy the material stress–strain 0 ! 0 ! 0 ! 0 !
2eAst ¼ 2~eAst ½ð~
g s  L s Þð~g t  L t Þ þ ð~ g t  L s Þð~g s  L t Þ
law for shells.
0 ! 0 ! 0 ! 0 !
In Eq. (14), Cbm is the stress–strain law matrix for the bending þ 2~eArt ½ð~ g r  L s Þð~g t  L t Þ þ ð~ g t  L s Þð~g r  L t Þ ð21Þ
and membrane strains and Cc is the stress–strain law matrix for
the transverse shear strains, with
2 3 ~eArt ¼ a2 þ a3 s; ~eAst ¼ a1  a3 r; ð22Þ
1 m 0 " #
E 6 E k 12 m 0
C bm ¼ 4m 1 0 7
5; Cc ¼ ; ð16Þ ~r0 ~s0 ~t0
in which g , g , and g are the rotated contravariant base vectors
1  m2 1  m2 0 k 12 m
0 0 1m and a1, a2, and a3 are the unknown variables for the approximated
2
covariant transverse shear strains. Here, it is important to note that
where E, m, and k are Young’ modulus, Poisson’s ratio, and a shear the type of interpolation functions in Eq. (22) is the same as that of
correction factor, respectively. In this work, we use k = 1.0. the interpolation functions used in Eq. (10), which belong to a ‘‘ro-
ebm and ~
In general, ~ c come from the 3-node displacement-based tated Raviart–Thomas space’’.
ebm and ~
shell formulation; that is, ~ c are obtained by Eq. (9) In order to calculate the rotated contravariant base vectors, as
0 1 shown in Fig. 3, we first rotate the three covariant base vectors
err  
! 2est ! about the vector (~en ) normal to the midsurface (t = 0) of the shell
ebm ¼ B
~ C
@ ess A ¼ Bbm U ; ~
c¼ ¼ Bc U : ð17Þ element
2ert
2ers
g 0i ¼ cos h~
~ g i þ ð1  cos hÞð~
gi  ~
en Þ~
en þ ð~
en  ~
g i Þ sin h ð23Þ
The approximated transverse shear strain field (c ) is constructed ~A
with
with the appropriate interpolation with unknowns (~a),
! gr  ~
~ gs
2eAst ~
en ¼ ; ð24Þ
cA ¼
~ ¼ BAc ~
a: ð18Þ j~
gr  ~
gs j
2eArt
in which ~g i are the covariant base vectors (i = r, s, and t) evaluated
Finally, the unknown variables in Eq. (14) are the nodal displace- by Eq. (4) at the integration points of the shell element, and h is
!
ments and rotations ( U ) and the approximated transverse shear the angle of rotation about ~ en . The rotated contravariant base vec-
a). The effectiveness of the shell finite element formulation
strains (~ tors are then calculated by
96 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106

!
a is given by
After obtaining the nodal displacement solution U , ~
T !
a ¼ K 1
~ ee K ue U ; ð29Þ
eMbm and ~
and the strains are then calculated by ~ cA in Eq. (20), and the
stresses are obtained using the stress–strain law in Eq. (16).

5. Basic numerical tests

In this section, the results of three basic tests of the MITC3-HR

shell finite elements are presented.

5.1. Strain energy mode test

We count how many zero eigenvalues exist in the stiffness ma-

trix of a single unsupported element. The single shell finite ele-
ment should pose exactly six zero eigenvalues, corresponding to
Fig. 3. The rotated covariant base vectors for the MITC3-HR shell finite element. the six physical rigid body modes. When a shell finite element
The dotted and solid vectors correspond to the original and rotated covariant base has additional zero eigenvalues corresponding to spurious energy
vectors, respectively. modes, the stability of the solutions cannot be guaranteed. This
0 test is called the zero energy mode test. The MITC3-HR shell finite
~ g j ¼ dji :
g 0i  ~ ð25Þ element passes the zero energy mode test except for the case
Note that ~
en is a constant vector, because the midsurface of the 3- where the angle h in Eq. (23) is 90 deg and the element has a uni-
node shell element is flat. form thickness.
Invoking dPM Tables 1 and 2 present the eigenvalues of the strain energy
HR ¼ 0 in Eq. (19), we obtain
  !! !! modes of the MITC3-HR shell finite element according to the rota-
K uu K ue U R tion angle (h) of the contravariant base vector for the plate bending
¼ ð26Þ
K eu K ee a
~ ~
0 problem shown in Fig. 4(a) when t/L is 0.001 and 0.0001 (uniform
thickness). Because the eigenvalues change with the order of (t/L)3
a, the stiffness matrix of the
and, using the static condensation on ~ in the modes 7, 8, and 9, these modes are easily identified as bend-
MITC3-HR shell finite element is finally given as ing modes. After investigating the strain distributions by the
K ¼ K uu  K ue K 1 T modes 11 and 13, it is found that these two modes include both
ee K ue ð27Þ
transverse shear and bending modes.
with When the angle h = 0, all of the eigenvalues of the MITC3-HR
Z shell finite element are identical to those of the MITC3 shell finite
T
K uu ¼ BM M
bm C bm Bbm dV; element. As the angle h increases to 90 deg, the 10th eigenvalue de-
ZV creases. Fig. 5 presents the 10th mode shapes (h = 89 deg) of the
T
K ue ¼ BM A
c C c Bc dV; displacement based 3-node, MITC3 and MITC3-HR shell finite ele-
V
Z ments. It is very interesting to note that, while the 10th strain en-
T
K ee ¼  BAc C c BAc dV: ð28Þ ergy mode of the displacement based 3-node shell element is the
V pure in-plane twisting mode (Fig. 5(a)), the 10th modes of MITC3

Table 1
The eigenvalues of the strain energy modes of the MITC3-HR shell finite element for the plate bending problem shown in Fig. 4(a) when t/L = 0.001. Modes 1–6 are rigid body
modes. Modes 12, 14 and 15 are membrane modes.

Mode Angle (h)

0 deg (MITC3) 30 deg 60 deg 89 deg 90 deg
7 6.6764E04 6.6764E04 6.6764E04 6.6764E04 6.6764E04
8 8.1454E04 8.1454E04 8.1454E04 8.1265E04 1.0752E03
9 2.4924E03 2.4924E03 2.4924E03 2.4924E03 2.4924E03
10 3.6928E+02 2.7702E+02 9.2381E+01 1.1284E01 0.0000E+00
11 4.6707E+03 4.6696E+03 4.6676E+03 4.6667E+03 4.6667E+03
13 1.1760E+04 1.1760E+04 1.1760E+04 1.1760E+04 1.1760E+04

Table 2
The eigenvalues of the strain energy modes of the MITC3-HR shell finite element for the plate bending problem shown in Fig. 4(a) when t/L = 0.0001. Modes 1–6 are rigid body
modes. Modes 12, 14 and 15 are membrane modes.

Mode Angle (h)

0 deg (MITC3) 30 deg 60 deg 89 deg 90 deg
7 6.6764E07 6.6764E07 6.6764E07 6.6764E07 6.6764E07
8 8.1455E07 8.1455E07 8.1455E07 8.1453E07 1.0752E06
9 2.4924E06 2.4924E06 2.4924E06 2.4924E06 2.4924E06
10 3.6928E+01 2.7702E+01 9.2381E+00 1.1258E02 0.0000E+00
11 4.6707E+02 4.6696E+02 4.6676E+02 4.6667E+02 4.6667E+02
13 1.1760E+03 1.1760E+03 1.1760E+03 1.1760E+03 1.1760E+03
 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106 97

(a) (b)
Fig. 4. Strain energy mode test (L = 1.0, E = 1.7472  107 and m = 0.3). (a) A single triangular shell finite element. (b) In-plane twisting mode for the single shell finite element.
The dotted and solid lines correspond to the top surfaces before and after deformation, respectively.

(a) (b) (c)

Fig. 5. Shapes of the normalized 10th strain energy mode. The dotted and solid lines correspond to the top surfaces before and after deformation, respectively. (a)
Displacement based 3-node shell finite element. (b) MITC3 shell finite element (MITC3-HR shell finite element when h = 0 deg). (c) MITC3-HR shell finite element when
h = 89 deg.

and MITC3-HR shell finite elements includes a bending mode as Fig. 6 shows the ratio of the strain energy of the MITC3-HR shell
well as the pure in-plane twisting mode (Fig. 5(b) and (c)). Also, finite element to the strain energy of the MITC3 shell finite element
we note that the 10th mode shape of the MITC3-HR shell finite ele- depending on the rotation angle of the contravariant base vector
ments varies depending on the angle h, but the change is very (0 6 h 6 89 deg) when the in-plane twisting mode is applied. Note
small, see Fig. 5(b) and (c). that less strain energy stored in the element means that the ele-
We then investigate how the strain energy of the MITC3-HR ment is more flexible. The MITC3-HR shell finite element has the
shell finite element varies according to the rotation angle (h) of same flexibility as the MITC3 shell finite element when the rotation
the contravariant base vector in the in-plane twisting mode. Note angle h = 0 deg. However, as the angle increases, the MITC3-HR
that the in-plane twisting mode results in only non-zero transverse shell finite element becomes increasingly flexible. As the angle ap-
shear strains. When the single triangular elements shown in proaches 90 deg, the MITC3-HR shell finite element becomes rap-
Fig. 4(a) are twisted at the angle u about the z-axis located at idly flexible and the zero strain energy is calculated when
the barycenter, the in-plane twisting
pffiffiffiffiffiffi mode shown pinffiffiffiffiffiffiFig. 4(b) is h = 90 deg. Therefore, it can be expected that the 10th mode of
obtained: hAx ¼ hCx ¼ hAy ¼ hBy ¼ 1= 12, hBx ¼ hCy ¼ 2= 12. The sin- the MITC3-HR shell finite element with h = 89 deg becomes the
gle displacement based 3-node and MITC3 shell finite elements spurious in-plane twisting mode when h = 90 deg.
give the exact strain energy in the in-plane twisting mode, which In order to understand why the behaviors of the MITC3-HR shell
results in only non-zero transverse shear strains. finite element with h = 0 deg are identical to the MITC3 shell finite
98 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106

Fig. 6. Ratio of the strain energy of the MITC3-HR shell finite element to the strain
energy of the MITC3 shell finite element depending on the rotation angle (h) of the
contravariant base vectors in the in-plane twisting mode.
Fig. 7. A single 3-node triangular shell finite element for the isotropic test. The
!
input values of the geometry are presented in Table 3, P ¼ ½ F x F y F z M a Mb T .

element and why the MITC3-HR element with h = 90 deg has a spu-
rious mode, we calculate the approximated transverse shear
5.2. Isotropic element test
strains in Eq. (21) for the single triangular element with an uniform
thickness (node numbering: 1:A ? 2:B ? 3:C) in Fig. 4(a) and then
A single 3-node triangular shell finite element in Fig. 7 is ana-
separate the covariant transverse shear strain fields corresponding
lyzed with three different node numberings. All the input values
to the original contravariant base vectors ~ gj ,
are the same except for the node numbering. Different sequences
eArt ¼ a1 sin h þ a2 cos h þ a3 ðr sin h þ s cos hÞ; of node numberings are used as follows: 1 ? 2 ? 3, 2 ? 3 ? 1,
and 3 ? 1 ? 2. The results should be the same for all possible tip
eAst ¼ a1 cos h þ a2 sin h þ a3 ðr cos h þ s sin hÞ: ð30Þ
forces and moments. The input values of the geometry of the 3-
When h = 0 deg, Eq. (30) has the same form as Eq. (10), this results node triangular shell element are presented in Table 3. Each node
cA ¼ ~
in ~ cM in Eq. (19). In this case, the MITC3-HR shell finite element has different thickness and director vectors, which are not normal
becomes identical to the MITC3 shell finite element. In the case of to the shell midsurface in order to consider a general case.
h = 90 deg, Eq. (30) is as follows: The MITC3-HR shell finite element gives identical results
regardless of the numbering sequences; that is, the element passes
eArt ¼ a1 þ a3 r; the isotropic element test as the MITC3 shell finite element does.
eAst ¼ a2 þ a3 s: ð31Þ Also, the isotropic behavior is independent of the rotation angle
of the contravariant base vectors. Of course, the MITC3-HR element
In addition, a comparison of Eqs. (31) and (10) shows that the linear shows isotropic behavior for other arbitrary element geometries.
terms in each covariant transverse shear strain component are
switched. This results in a3 = 0; thus, the constant transverse shear
5.3. Patch test
strain field is obtained. As pointed out in the literature [18], the 3-
node plate/shell finite elements with a constant transverse shear
We use the mesh shown in Fig. 8 to perform the patch test [1].
strain field display a spurious zero energy mode.
The minimum number of degrees of freedom is constrained so as to
Through the studies in this section, we can conclude that
prevent rigid body motions. Nodal forces that should result in con-
the special approximated transverse shear strain field introduced
stant stress conditions are applied. The constant stress should be
by the modified Hellinger–Reissner functional selectively increases
calculated to pass the patch test. The MITC3-HR shell finite
the flexibility of the MITC3 shell finite element corresponding to
element passes the membrane, bending, and transverse shearing
the 10th strain energy mode. In other words, the transverse shear
patch tests.
strains by the 10th strain energy mode that contains both the
in-plane twisting and bending modes becomes smaller in the
MITC3-HR shell finite element. 6. Convergence studies
In all the following numerical examples, we use the rotation an-
gle of the contravariant base vectors, h = 89 deg, to make the In this section, well-established convergence studies are per-
MITC3-HR shell finite element flexible enough in the 10th strain formed for the MITC3-HR shell finite element developed in this
energy mode without the spurious energy mode. study and the results are compared with those of the MITC3 shell

Table 3
The input values for the geometry shown in Fig. 7 for the isotropic test.

Node (i) Cartesian coordinate (~

xi ) Director vectors (V in Þ Thicknesses

xi yi zi V inx V iyx V inz

1 2.0 0.0 0.0 0.43775 0.87549 0.20467 0.1

2 0.0 1.0 0.0 0.18273 0.69520 0.69520 0.01
3 0.0 0.0 1.0 0.43775 0.20467 0.87549 0.2
 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106 99

e ¼ ½ exx
~ eyy ezz 2exy 2eyz 2ezx T ;
s ¼ ½ sxx
~ syy szz 2sxy 2syz 2szx T ; ð33Þ

and
* *
e ¼ ~e  ~eh ;
D~ Ds ¼ s ~
sh : ð34Þ
The position vectors ~
x and ~
xh correspond to the continuum domain
and the discretized domain, respectively, and the relationship with
a one-to-one mapping P is given as
* *
x ¼ Pðx h Þ: ð35Þ
The theoretical convergence behavior can be estimated by
k
k~ uh k2s ffi ch ;
u ~ ð36Þ
in which c is a constant and h denotes the element size. When the 3-
node shell element is uniformly optimal, the constant is indepen-
dent of the shell thickness and k = 2.
Instead of the exact solution, the finite element solution calcu-
Fig. 8. Triangular mesh used for the patch tests.
lated by a very fine mesh can be practically used,
finite element. We consider various shell problems: a clamped
Z
plate problem, sixty-degree skew plate problem, cylindrical shell k~ uh k2s ¼
uref  ~ eT D~
D~ s dXref ð37Þ
Xref
problems, and hyperboloid shell problems [2,15,18,22].
The s-norm proposed by Hiller and Bathe [23] is used to mea- with
sure the convergence of the finite element solutions. The s-norm * * * *
is suitable to check whether the finite element solutions satisfy e ¼ ~eref  ~eh ;
D~ Ds ¼ s ~
sh ; x ref ¼ Pref ðx h Þ:
ref ð38Þ
consistency and inf–sup conditions [23–27]. The s-norm is defined To measure the convergence of the finite elements in the shell prob-
as follows lems with various shell thicknesses, the relative error Eh is used,
Z
k~ uh k2s ¼
u ~ eT D~
D~ s dX; ð32Þ k~ uh k2s
uref  ~
X Eh ¼ * : ð39Þ
ku ref k2s
where ~u is the exact solution, ~
uh is the solution of the finite element
s are the strain vector and the stress vector
e and ~
discretization, and ~ The numerical procedure to calculate the s-norm for shell finite
in the global Cartesian coordinate system, respectively, defined by element solutions with general types of elements and general
meshes is explained in detail in Ref. [18]. In the use of Eq. (39), it

Fig. 9. Fully clamped square plate problem under uniform pressure (L = 1.0, E = 1.7472  107, q = 1.0 and m = 0.3) with two different 4  4 mesh patterns in (a) and (b).
100 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106

Fig. 10. Convergence curves for the fully clamped square plate problem. The bold line represents the optimal convergence rate. The solid and dotted lines correspond to the
results obtained by the mesh patterns in Fig. 9(a) and (b), respectively.

(a) (b)
Fig. 11. Simply supported sixty-degree skew plate problem (L = 1.0, E = 1.7472  107 and m = 0.3). (a) Problem solved. (b) Mesh pattern used for N = 4.

Fig. 12. Convergence curves for the simply supported sixty-degree skew plate problem. The bold line represents the optimal convergence rate.
 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106 101

In the following sections, we present the convergence curves of

the MITC3 and MITC3-HR shell finite elements to show the im-
proved performance of the MITC3-HR shell finite element com-
pared to the MITC3 shell finite element. Note that some of the
convergence curves of the MITC3 shell finite element have been
published before in Refs. [15,18].

6.1. Fully clamped square plate problem

The plate bending problem shown in Fig. 9 is considered. A

square plate of dimensions 2L  2L with uniform thickness t is
subjected to uniform pressure normal to the flat surface and all
the edges are fully clamped. Due to symmetry, only a one-quarter
model is considered, with the following boundary conditions
imposed: ux = hy = 0 along BC, uy = hx = 0 along DC and ux = uy = uz =
hx = hy = 0 along AB and AD [15,18].
The convergence curves of the MITC3 and MITC3-HR shell
elements calculated by Eq. (39) are shown in Fig. 10. The reference
solution is obtained by a mesh of 96  96 MITC9 shell finite
elements. The solutions of the MITC3 and MITC3-HR shell finite
elements are calculated by N  N meshes (N = 4, 8, 16, 32, and
64). The element size is h = L/N. For the range of t/L (1/100, 1/
1000, and 1/10,000) considered, the performance of the MITC3-
HR shell finite element is much better than that of the MITC3 shell
finite element.

6.2. Simply supported sixty-degree skew plate problem

The simply supported skew plate with edges of 2L and uniform

thickness t is subjected to uniform pressure normal to the flat sur-
face, as shown in Fig. 11. The boundary condition uz = 0 is imposed
Fig. 13. Cylindrical shell problem (4  4 mesh, L = R = 1.0, E = 2.0  105, m = 1/3 and along all edges [22].
p0 = 1.0). Fig. 12 shows the convergence curves of the MITC3 and
MITC3-HR shell finite elements in the simply supported skew
is very important to use accurate reference solutions calculated by plate problem. A mesh of 128  128 MITC9 shell finite elements
reliable shell finite elements. In this study, we use well-converged is used for the calculation of the reference solution. N  N
reference solutions calculated by fine meshes of MITC9 shell finite meshes (N = 8, 16, 32, 64, and 128) are used to evaluate the
elements. Of course, the MITC9 shell finite element is known to solutions of the MITC3 and MITC3-HR shell finite elements.
satisfy ellipticity and consistency conditions and to show adequate The element size is h = 2L/N. The graphs indicate the improved
convergences for the shell problems studied in this paper, see Refs. capability of the MITC3-HR shell finite element compared with
[16,22,23]. the original MITC3 element.

Fig. 14. Convergence curves for the clamped cylindrical shell problem. The bold line represents the optimal convergence rate.
102 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106

Fig. 15. Convergence curves for the free cylindrical shell problem. The bold line represents the optimal convergence rate.

6.3. Cylindrical shell problems

A cylindrical shell of uniform thickness t, length 2L, and radius R

is considered, as seen in Fig. 13. The loading is smoothly varying
periodic pressure p(h) normal to the shell surface

pðhÞ ¼ p0 cosð2hÞ: ð40Þ

The shell presents two different asymptotic behaviors depending on

the boundary conditions at both ends: the bending dominated
behavior under a free boundary condition and the membrane dom-
inated behavior under a clamped boundary condition.
Considering the symmetry of the problems, the region ABCD in
Fig. 13 is modeled. In the membrane dominated case, the
(a) (b) clamped boundary condition is imposed: ux = b = 0 along BC,
uy = a = 0 along DC uz = a = 0 along AB, and ux = uy = uz = a = b = 0
Fig. 16. Distorted meshes for the cylindrical shell problem. (a) Mesh pattern used along AD. In the bending dominated case, the free boundary condi-
for N = 4. (b) Mesh pattern used for N = 8. tion is imposed: ux = b = 0 along BC, uy = a = 0 along DC, and
uz = a = 0 along AB [15].

Fig. 17. Convergence curves for the clamped cylindrical shell problem with the distorted meshes shown in Fig. 16. The bold line represents the optimal convergence rate.
 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106 103

Fig. 18. Convergence curves for the free cylindrical shell problem with the distorted meshes shown in Fig. 16. The bold line represents the optimal convergence rate.

(a)

(b)
Fig. 19. Hyperboloid shell problem (E = 2.0  1011, m = 1/3 and p0 = 1.0). (a) Problem solved. (b) Graded mesh for the clamped case (8  8 mesh, t/L = 1/1000).

Figs. 14 and 15 present the convergence behaviors of the MITC3 and 18 present the convergence curves for the clamped and free
and MITC3-HR shell finite elements for the clamped and free cylindrical shell problems, respectively. Although the extremely
cylindrical shell problems, respectively. The reference solution is distorted mesh is used, the MITC3 and MITC3-HR shell finite ele-
calculated by a mesh of 96  96 MITC9 shell finite elements for ments still present good convergence behavior.
both cases. The solutions of the MITC3 and MITC3-HR shell finite
elements are obtained by N  N meshes (N = 4, 8, 16, 32, and 64). 6.4. Hyperboloid shell problems
The element size is h = L/N. Excellent convergence behavior of the
MITC3 element in the cylindrical shell problems has been reported The hyperboloid shell is considered as in Fig. 19 and the midsur-
in Ref. [15]. The MITC3-HR shell finite element similarly shows face of the shell structure is given by
excellent convergence behavior.
x2 þ z2 ¼ 1 þ y2 ; y 2 ½1; 1: ð41Þ
It is valuable to study the convergence behaviors of the MITC3
and MITC3-HR shell finite elements for the distorted meshes Smoothly varying periodic pressure normal to the surface is applied,
shown in Fig. 16. When the N  N mesh is used, each edge is
pðhÞ ¼ p0 cosð2hÞ; ð42Þ
divided by the following ratio: L1 : L2 : . . . : LN = 1 : 2 : . . . : N. Figs. 17
104 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106

Fig. 20. Convergence curves for the clamped hyperboloid shell problem. The bold line represents the optimal convergence rate.

Fig. 21. Convergence curves for the free hyperboloid shell problem. The bold line represents the optimal convergence rate.

which has the same distribution as shown in Fig. 13. For both the membrane and bending dominated cases, the con-
When both ends are clamped, a membrane dominated problem vergence curves are shown in Figs. 20 and 21. In the membrane
is obtained, and when the ends are free, a bending dominated dominated case (that is, the clamped hyperboloid shell problem),
problem is obtained. The bending dominated hyperboloid shell the performance of the MITC3-HR shell finite element is
problem is known to be very difficult to solve [11,15,18]. almost the same as that of the MITC3 shell finite element. Fig. 21
Due to symmetry, the analyses are carried out in one-eighth of displays the substantially improved performance of the MITC3-HR
the structure, corresponding with the shaded region ABCD in shell finite element compared to the MITC3 shell finite element.
Fig. 19(a). For the membrane dominated case, the clamped bound- Finally, Fig. 22 presents the convergence curves of the MITC3-
ary condition is imposed: uz = b = 0 along BC, ux = b = 0 along AD, HR shell finite element for the free hyperboloid shell problem
and uy = a = 0 along DC, and ux = uy = uz = a = b = 0 along AB. For depending on the rotation angle (h) of the contravariant base vec-
the bending dominated case, the free boundary condition is tors. As the angle approaches to 90 deg, the performance of the
imposed: uz = b = 0 along BC, ux = b = 0 along AD, and uy = a = 0 MITC3-HR shell finite element becomes better.
along DC [15,18].
In both cases, we use the reference solutions calculated with a
mesh of 96  96 MITC9 shell finite elements. The solutions of the 7. Conclusions
MITC3 and MITC3-HR shell finite elements are evaluated by
N  N meshes (N = 4, 8, 16, 32, and 64). The element size is h = L/N. The objective of this paper is to improve the MITC3 shell finite
In the
pclamped
ffiffi hyperboloid shell problem, a boundary layer of element without losing the desirable characteristics of the MITC3
width 6 t is used for half of the mesh; see Fig. 19(b). In the free shell finite element. The MITC3-HR shell finite element was devel-
hyperboloid shell problem, the thin boundary layer is not specially oped by using the modified Hellinger–Reissner (HR) functional and
meshed. applying the rotated approximated transverse shear strain field.
 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106 105

(a) (b)

(c) (d)
Fig. 22. Convergence curves for the free hyperboloid shell problem depending on the rotation angle (h) of the contravariant base vectors. The bold line represents the optimal
convergence rate. (a) h = 75 deg, (b) h = 80 deg, (c) h = 85 deg, and (d) h = 90 deg.

The results of the basic numerical tests show that the MITC3-HR Technology Evaluation and Planning (KETEP) grant funded by the
shell finite element satisfies all the basic requirements. The results Korea government Ministry of Knowledge Economy.
of the convergence studies proved that the MITC3 shell finite ele-
ment is successfully improved; that is, the MITC3-HR shell finite
element shows much better convergence behavior than the MITC3 References
shell finite element, especially in bending-dominated problems.
[1] Bathe KJ. Finite element procedures. New York: Prentice Hall; 1996.
Interestingly, although the single MITC3-HR shell finite element [2] Chapelle D, Bathe KJ. The finite element analysis of shells-fundamentals. 2nd
is very flexible in the in-plane twisting mode, any loss of solution ed. Berlin: Springer; 2011.
accuracy was not observed in the convergence studies performed [3] Chapelle D, Bathe KJ. Fundamental considerations for finite element analysis of
shell structures. Comput Struct 1998;66:19–36. 711–712.
in this paper. The MITC3 shell finite element is improved signifi-
[4] Lee PS, Bathe KJ. On the asymptotic behavior of shell structures and the
cantly but locking still exists. Hence, further research efforts are re- evaluation in finite element solutions. Comput Struct 2002;80:235–55.
quired to develop a uniformly optimal 3-node triangular shell [5] Baiocchi C, Lovadina C. A shell classification by interpolation. Math Models
Meth Appl Sci 2002;12(10):1359–80.
finite element.
[6] Beirão da Veiga L. Asymptotic energy behavior of two classical intermediate
benchmark shell problems. Math Models Meth Appl Sci 2003;13:1279–302.
Acknowledgement [7] Beirão da Veiga L. Uniform error estimates for a class of intermediate
cylindrical shell problems. Numer Math 2004;96:661–89.
[8] Pitkäranta J, Leino Y, Ovaskainen O, Piila J. Shell deformation states and finite
This work was supported by the Human Resources Develop- element method: a benchmark study of cylindrical shells. Comput Meth Appl
ment (No. 20114030200040) of the Korea Institute of Energy Mech Eng 1995;128:81–121.
106 Y. Lee et al. / Computers and Structures 110–111 (2012) 93–106

[9] Hakula H, Leino Y, Pitkäranta J. Scale resolution, locking, and high-order finite [19] To CWS, Liu ML. Hybrid strain based three-node flat triangular shell finite
element modeling of shells. Comput Meth Appl Mech Eng 1996;133:157–82. elements. Finite Elem Anal Des 1994;17:169–203.
[10] Bathe KJ, Chapelle D, Lee PS. A shell problem ‘highly sensitive’ to thickness [20] Rhiu JJ, Lee SW. A new efficient mixed formulation for thin shell finite element
changes. Int J Numer Meth Eng 2003;57:1039–52. models. Int J Numer Meth Eng 1987;24:581–604.
[11] Lee PS, Bathe KJ. Insight into finite element shell discretizations by use of the [21] Bathe KJ, Iosilevich A, Chapelle D. An evaluation of the MITC shell elements.
‘‘basic shell mathematical model’’. Comput Struct 2005;83:69–90. Comput Struct 2000;75:1–30.
[12] Dvorkin EN, Bathe KJ. A continuum mechanics based four-node shell element [22] Lee PS, Bathe KJ. The quadratic MITC plate and MITC shell elements in plate
for general nonlinear analysis. Eng Comput 1984;1:77–88. bending. Adv Eng Software 2010;41:712–28.
[13] Bathe KJ, Dvorkin EN. A formulation of general shell elements – the use of [23] Hiller JF, Bathe KJ. Measuring convergence of mixed finite element
mixed interpolation of tensorial components. Int J Numer Meth Eng discretizations: an application to shell structures. Comput Struct
1986;22:697–722. 2003;81:639–54.
[14] Bucalem ML, Bathe KJ. Higher-order MITC general shell elements. Int J Numer [24] Bathe KJ, Iosilevich A, Chapelle D. An inf-sup test for shell finite elements.
Meth Eng 1993;36:3729–54. Comput Struct 2000;75:439–56.
[15] Lee PS, Bathe KJ. Development of MITC isotropic triangular shell finite [25] Bathe KJ. The inf-sup condition and its evaluation for mixed finite element
elements. Comput Struct 2004;82:945–62. methods. Comput Struct 2001;79:243–52. 971.
[16] Bathe KJ, Lee PS, Hiller JF. Towards improving the MITC9 shell element. [26] Bathe KJ, Lee PS. Measuring the convergence behavior of shell analysis
Comput Struct 2003;81:477–89. schemes. Comput Struct 2011;89:285–301.
[17] Beirão da Veiga L, Chapelle D, Paris Suarez I. Towards improving the MITC6 [27] Chapelle D, Bathe KJ. The mathematical shell model underlying general shell
triangular shell element. Comput Struct 2007;85:1589–610. elements. Int J Numer Meth Eng 2000;48:289–313.
[18] Lee PS, Noh HC, Bathe KJ. Insight into 3-node triangular shell finite elements:
the effects of element isotropy and mesh patterns. Comput Struct
2007;85:404–18.