Professional Documents
Culture Documents
Development of MITC Isotropic Triangular Shell Finite Elements1
Development of MITC Isotropic Triangular Shell Finite Elements1
www.elsevier.com/locate/compstruc
Abstract
We present a simple methodology to design isotropic triangular shell finite elements based on the Mixed Interpo-
lation of Tensorial Components (MITC) approach. Several mixed-interpolated isotropic triangular shell finite elements
are proposed. We perform well-established numerical tests and show the performance of the new elements.
2004 Published by Elsevier Ltd.
Keywords: Shell structures; Finite elements; Triangular elements; MITC shell elements
undertaken and more progress has also been achieved. cial attention must be given to the direction of each
Consequently, in shell finite element analyses, quadri- element in the model.
lateral elements are usually used due to their better In fact, this condition is a major obstacle in the
performance than observed using triangular elements. construction of locking-free triangular shell elements.
Indeed, there does not exist yet a ‘‘uniformly optimal’’ Usually, some ‘‘averaging’’ or ‘‘cyclic’’ treatments are
triangular shell element, and not even an element close employed to construct isotropic triangular elements
to optimal. The motivation of this research comes from [18,19]. A simple systematic way, which is mechanically
the fact that the development of optimal triangular shell clear, to construct isotropic triangular shell elements is
elements is still a great challenge [10,12,15–19]. desirable.
It is extremely difficult to obtain a shell finite element If a triangular shell element satisfies all the above
method that is uniformly optimal and a mixed formu- conditions, it is an optimal and ideal triangular shell
lation must be used. In the formulation we should aim to finite element. Such an element is very difficult to
satisfy [1,2]: reach and we can ‘‘soften’’ the requirements somewhat
for practical purposes. We summarize the practical
• Ellipticity. This condition ensures that the finite ele- requirements on triangular shell finite elements as fol-
ment discretization is solvable and physically means lows:
that there is no spurious zero energy mode. Without
supports, a single shell finite element should have–– • spatially isotropic behavior;
for any geometry––exactly six zero energy modes cor- • no spurious zero energy mode (ellipticity condition);
responding to the physical rigid body modes. This • no shear locking in plate bending problems;
condition can be easily verified by counting the num- • reliable, ideally optimal results for membrane domi-
ber of zero eigenvalues (and studying the correspond- nated shell problems;
ing eigenvectors) of the stiffness matrix of single • reliable, ideally optimal results for bending domi-
unsupported shell finite elements. nated shell problems ‘‘in the practical range of t=L’’;
• Consistency. Since the finite element discretization is • easy extension of the formulation to nonlinear anal-
based on a mathematical model, the finite element yses (simple formulation).
solutions must converge to the solution of the math-
ematical model as the element size h goes to zero. In Hence only a practical range t=L is considered in bend-
other words, the bilinear forms used in the finite ele- ing dominated shell problems (but, of course, we shall
ment discretization, which may be a function of the not abolish the aim to ultimately reach a triangular
element size h, must approach the exact bilinear element that is optimal for all analyses and for all values
forms of the mathematical model as h approaches of t=L).
zero. The objective of this paper is to develop MITC iso-
• Inf–sup condition. Ideally, a mixed finite element dis- tropic triangular shell finite elements which can be
cretization should satisfy the inf–sup condition practically used for general shell structures.
[2,11,13]. For shell finite elements, satisfying this con- In the following sections of the paper, we first review
dition implies uniform optimal convergence in bend- the MITC formulation of continuum mechanics based
ing-dominated shell problems. Then, the shell finite shell finite elements. We then propose a simple meth-
element is free from shear and membrane locking odology to design isotropic triangular shell elements
with solution accuracy being independent of the shell using the MITC technique, and demonstrate this meth-
thickness parameter. However, it is generally not pos- odology with some examples. A large number of ele-
sible to analytically prove whether a shell finite ele- ments can be constructed using our methodology. We
ment satisfies this condition and numerical tests introduce some selected MITC triangular shell finite
have been employed. elements and give the numerical test results of these
elements.
For triangular shell elements, one more requirement
exists; namely, ‘‘spatial isotropy’’. The requirement of
‘‘spatial isotropy’’ means that the element stiffness 2. MITC formulation of continuum mechanics based shell
matrices of triangular elements should not depend on finite elements
the sequence of node numbering, i.e. the element ori-
entation. Specifically, when a spatially isotropic trian- The continuum mechanics displacement-based shell
gular element has sides of equal length, the internal finite elements have been proposed as general curved
element quantities should vary in the same manner for shell finite elements [20]. While these elements offer sig-
each corner nodal displacement/rotation and each mid- nificant advantages in the modeling of arbitrary complex
side nodal displacement/rotation, respectively. If the shell geometries, they exhibit severe locking in bending
behavior of an element depends on its orientation, spe- dominated cases [1,2].
P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962 947
The basic idea of the MITC technique is to interpo- the displacement-based covariant strain components in
late displacements and strains separately and ‘‘connect’’ terms of the nodal displacements and rotations
these interpolations at ‘‘tying points’’. The displacement eij ¼ Bij U; ð7Þ
and strain interpolations are chosen so as to satisfy the
ellipticity and consistency conditions, and as closely as where B is the strain–displacement matrix and U is the
possible the inf–sup condition. nodal displacement/rotation vector, we obtain
The geometry of the q-node continuum mechanics " n #
X ij
displacement-based shell element is described by [1,2] ~eij ¼ k e ij U:
hij ðr; sÞBij jðrk ;sk ;tÞ U ¼ B ð8Þ
ij ij
X
q
t X
q k¼1
xðr; s; tÞ ¼
~ hi ðr; sÞ~
xi þ ai hi ðr; sÞV~in ; ð1Þ
i¼1
2 i¼1 Then using the proper stress–strain matrix, the element
stiffness matrix is constructed in the same manner as for
where hi is the 2D shape function of the standard iso- the displacement-based element.
parametric procedure corresponding to node i, ~ xi is the
position vector at node i in the global Cartesian coor-
dinate system, and ai and V ~i denote the shell thickness 3. Strain interpolation technique for isotropic triangular
n
and the director vector at node i, respectively. Note that shell elements
in this geometric description the vector V ~in is not nec-
essarily normal to the shell midsurface. Two important points of the successful MITC tech-
The displacement of the element is given by nique are to use appropriate assumed strain interpola-
X tions in Eq. (5) and to carefully choose the tying points.
t X
q q
uðr; s; tÞ ¼
~ hi ðr; sÞ~
ui þ ~i2 ai þ V~i1 bi Þ;
ai hi ðr; sÞðV In recent research [21], it was observed that a seemingly
i¼1
2 i¼1 small change in the tying positions can result in signifi-
ð2Þ cant differences in the predictive capability of the
MITC9 shell element.
ui is the nodal displacement vector in the global
in which ~ While the interpolation of the covariant strain com-
Cartesian coordinate system, V ~i and V ~i are unit vectors ponents is quite easily achieved for quadrilateral ele-
1 2
~ i
orthogonal to V n and to each other, and ai and bi are the ments, the interpolation is more difficult for triangular
rotations of the director vector V ~in about V~i1 and V~i2 at elements due to their shape and coordinate system. In
node i. this section we provide a systematic way to interpolate
The covariant strain components are directly calcu- the strain components to reach isotropic MITC trian-
lated by gular shell elements.
1
eij ¼ ð~
g ~
u;j þ ~
gj ~
u;i Þ; ð3Þ 3.1. Strain interpolation methods
2 i
where Let us consider a three node isoparametric beam
element. As shown in Fig. 1(a), the displacement-based
x
@~ u
@~
gi ¼
~ ; u;i ¼
~ with r1 ¼ r; r2 ¼ s; r3 ¼ t: element has a quadratic variation of transverse shear
@ri @ri strain. In order to remove shear locking, we need to
ð4Þ linearly interpolate the transverse shear strain in the
beam element [1,2]. The linear transverse shear strain
Now we define a set of so-called tying points k ¼ field can be determined by the two transverse shear
1; . . . ; nij on the shell midsurface with coordinates (rijk , strains sampled at two different tying points (r ¼ r1 and
skij ), and define the assumed covariant strain components r ¼ r2 ¼ r1 ). Three kind of approaches shown in Fig.
~eij as 1(b)–(d) can be employed to determine the interpolation.
X
nij
~eij ðr; s; tÞ ¼ hkij ðr; sÞeij jðrk ;sk ;tÞ ; ð5Þ Method-i
ij ij
k¼1 Since we know that the resulting polynomial for ~ert is
linear, we assume
where nij is the number of tying points for the covariant
strain component ~eij and hkij are the assumed interpola- ~ert ¼ a þ br: ð9Þ
tion functions satisfying
Using the two conditions
hkij ðrijl ; slij Þ ¼ dkl ; l ¼ 1; . . . ; nij : ð6Þ
~ert ðr1 Þ ¼ eð1Þ
rt ;
Note that this tying procedure is carried out on the ð10Þ
elemental level for each individual element. Expressing ~ert ðr2 Þ ¼ eð2Þ
rt ;
948 P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962
r r
–1 r1 0 r2 1 –1 r1 0 r2 1
(c) (d) ~
e~rt ( 2)
h2 ert( 2 ) lrt
he (1) e ( 2)
1 rt
ert(1)
rt
e~rt ~
m
e rt
ert(1) rt
r r
–1 r1 0 r2 1 –1 r1 0 r2 1
the unique pair of the coefficients, a and b, can be ~ert ð1Þ ¼ m ~ rt ~lrt ;
determined, see Fig. 1(b). ~ert ð0Þ ¼ m
~ rt ; ð16Þ
Method-ii ~ rt þ ~lrt ;
~ert ð1Þ ¼ m
In this method, shown in Fig. 1(c), we use the shape
where m ~ rt is the mean value of the two tying strains and
functions of the standard isoparametric procedure ~lrt is the difference between the value at the center (r ¼ 0)
X
2 and the edge (r ¼ 1), that is,
~ert ¼ hi eðiÞ ð1Þ ð2Þ
rt ¼ h1 ert þ h2 ert ; ð11Þ ð2Þ ð1Þ
1
~ rt ¼ ðeð1Þ þ eð2Þ ~lrt ¼ ert ert :
m rt Þ; ð17Þ
i¼1
2 rt r2 r1
where h1 and h2 are the linear functions satisfying Solving Eq. (16), we obtain
with four conditions The three methods give exactly the same interpo-
lation for this example. To use ‘‘method-i’’ and ‘‘meth-
h1 ðr1 Þ ¼ 1; h1 ðr2 Þ ¼ 0;
ð14Þ od-ii’’, the interpolations start from linear polynomials,
h2 ðr1 Þ ¼ 0; h2 ðr2 Þ ¼ 1; while, in the new method, the interpolation starts from
the quadratic polynomial and the coefficient of the
we can obtain the four coefficients (a1 , b1 , a2 , b2 ). quadratic term automatically vanishes by imposition of
the linear variation. Due to this property, the method
New method can be used even when the exact space of functions for a
We here propose a simple new method, described in 2D or 3D element is not known (see sections below).
Fig. 1(d). Since the order of the transverse shear strain of Note that, for the example considered here, two un-
the displacement-based three node isoparametric beam known coefficients and two linear equations are con-
element is quadratic, we start from sidered by ‘‘method-i’’, four unknown coefficients and
four linear equations are considered by ‘‘method-ii’’ and
~ert ¼ a þ br þ cr2 : ð15Þ three unknown coefficients and three linear equations
are considered by the proposed method.
The following three conditions (imposing a linear vari- The proposed method is powerful, specifically when
ation) given at the nodes can be applied to evaluate a, b we construct the transverse shear strain fields for iso-
and c: tropic MITC triangular shell elements.
P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962 949
Fig. 4. Transverse shear strain tying positions for the 3-node triangular shell element with the constant transverse shear strain along its
edges.
Step 2. The conditions are direction of the right-angled triangle in the natural
coordinate system
~ert ð0; 0Þ ¼ eð1Þ
rt ; ~ert ð1; 0Þ ¼ eð1Þ
rt ; err þ ess err ess
eqq ¼ þ cosð2hÞ þ ers sinð2hÞ ð26Þ
~est ð0; 0Þ ¼ eð2Þ
st ; ~est ð0; 1Þ ¼ eð2Þ
st ; 2 2
1 ð3Þ ð23Þ
~eqt ð1; 0Þ ¼ eð3Þ
qt ¼ pffiffiffi ðest eð3Þ
rt Þ; with h ¼ 135 (cosð2hÞ ¼ 0 and sinð2hÞ ¼ 1), see Figs.
2 2 and 5.
1 The first step for the construction of the isotropic in-
~eqt ð0; 1Þ ¼ eð3Þ
qt ¼ pffiffiffi ðeð3Þ ð3Þ
st ert Þ
2 plane strain field is to independently interpolate the
three in-plane strains err , ess and eqq with the same order
and we obtain six linearly independent equations for of polynomials
the unknown coefficients a1 ; b1 ; . . . ; c2 .
Step 3. We solve the linear equations and obtain ~err ¼ a1 þ b1 r þ c1 s . . . ;
~ess ¼ a2 þ b2 r þ c2 s . . . ; ð27Þ
a1 ¼ eð1Þ
rt ; b1 ¼ 0; c1 ¼ eð2Þ ð1Þ ð3Þ ð3Þ
st ert est þ ert ;
~eqq ¼ a3 þ b3 r þ c3 s . . . ;
a2 ¼ eð2Þ
st ; b2 ¼ c1 ; c2 ¼ 0;
ð24Þ where a1 ; b1 ; . . ., a2 ; b2 ; . . ., and a3 ; b3 ; . . . denote the un-
known coefficients of the strain polynomials. Of course,
which gives the isotropic transverse shear strain field polynomials of lower order than implied by the assumed
displacements should be used.
~ert ¼ eð1Þ
rt þ cs;
ð25Þ In the second step, we select isotropic tying posi-
~est ¼ eð2Þ
st cr; tions in the triangular element and evaluate the as-
sumed strains from the displacement-based strains at
ð2Þ ð1Þ ð3Þ ð3Þ
where c ¼ est ert est þ ert .
judiciously chosen points (ri , si ) in the element, Step 1. We assume the starting polynomials
i ¼ 1; 2; . . .
~err ¼ a1 þ b1 r þ c1 s;
~err ðri ; si Þ ¼ . . . ; ~ess ¼ a2 þ b2 r þ c2 s; ð30Þ
~ess ðri ; si Þ ¼ . . . ; ð28Þ
~eqq ¼ a3 þ b3 r þ c3 ð1 r sÞ;
~eqq ðri ; si Þ ¼ . . .
where, of course, ~eqq ¼ a3 þ b3 r þ c3 s can be used
As for the transverse shear strain interpolations, the instead of ~eqq ¼ a3 þ b3 r þ c3 ð1 r sÞ.
ðri ; si Þ do not need to be tying positions. The number of Step 2. The conditions used are
linearly independent equations reached should be equal ~ð1Þ
to the number of unknown coefficients in Eq. (27). ~ ð1Þ
~err ð0; 0Þ ¼ m rr lrr ; ~ ð1Þ
~err ð1=2; 0Þ ¼ m rr ;
s s
1 1
s2
(1)
ecrr
s1
r r
0 r1 r2 1 0 e(1) (1)
e 1
1rr 2rr
e~rr
s s
( 3)
1 1 e2qq
( 2)
e2ss
( 3) ( 3)
ecqq e1qq
( 2) ( 2)
e1ss e
css
r r
0 1 0 1
e~ss e~qq
Fig. 6. Strain tying positions for the 6-node triangular shell element with linear normal strain along edges; r1 ¼ s1 ¼ 12 2p1 ffiffi3 and
r2 ¼ s2 ¼ 12 þ 2p1 ffiffi3.
952 P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962
(a) s s
1 1 ( 3)
s2 ( 2) e2qt
e2st
s3 ( 3)
s1 ( 2)
e1st ecrt , ecst e1qt
r r
0 r1 r3 r2 1 0 (1)
e1rt (1)
e2rt 1
(b) s s
1 1 ( 3)
s2 ( 2) e2qt
e2st
( 3)
s1 ( 2)
e1st e1qt
r r
0 r1 r2 1 0 (1)
e1rt (1)
e2rt 1
Fig. 7. Tying points for the transverse shear strain interpolation of the 6-node MITC triangular shell elements: (a) for MITC6-a and
(b) for MITC6-b.
where ~ ð1Þ
a1 ¼ m ~ð1Þ b1 ¼ 2~lð1Þ
rt lrt ; rt ;
pffiffiffi
1 ðiÞ 3 ðiÞ ~ ð2Þ
a2 ¼ m ~ð2Þ c2 ¼ 2~lð2Þ
ðiÞ ðiÞ ~ðiÞ
~ jt ¼ ðe1jt þ e2jt Þ; ljt ¼
m
ðiÞ
ðe e1jt Þ st lst ; st ;
2 2 2jt ð41Þ
~ ð3Þ
c1 ¼ ða2 þ c2 a1 Þ ðm ~ð3Þ ~ ð3Þ ~lð3Þ Þ;
with j ¼ r; s for i ¼ 1; 2; 3: ð38Þ st þ lst mrt rt
~ ð3Þ
b2 ¼ ða1 þ b1 a2 Þ þ ðm ~ð3Þ ~ ð3Þ þ ~lð3Þ Þ:
st lst m rt rt
It is interesting to note that this scheme is very similar to
the interpolation scheme of the MITC7 plate bending
element in reference [9]. For this element similar tying The same in-plane strain interpolation scheme as for the
points are used and the interpolation functions, which MITC6-a element is employed.
belong to a ‘‘rotated Raviart-Thomas space’’, are given
as Fig. 8 summarizes the strain interpolation schemes
of the selected MITC triangular shell elements. We
~ert ¼ a1 þ b1 r þ c1 s þ sðdr þ esÞ; may note the geometric relationships between the
ð39Þ
~est ¼ a2 þ b2 r þ c2 s rðdr þ esÞ: tying points used for the triangular MITC3 and MITC6
elements and the MITC4 [6] and MITC9 quadri-
We use the in-plane strain interpolation scheme given by lateral elements [21]. Fig. 9 shows these relation-
Eqs. (30)–(34). ships.
MITC6-b
Considering the transverse shear strains, this 6-node
5. Numerical results
MITC triangular shell element has the same edge tying
points as the MITC6-a element. Also ‘‘linear transverse
In this section, we report upon various numeri-
shear strains along edges’’ are assumed. However, we do
cal tests of the MITC triangular shell finite elements,
not have any internal tying point and the strain varia-
MITC3, MITC6-a and MITC6-b. Selected basic tests
tions are linear inside the element. Fig. 7(b) shows the
show whether the elements satisfy the minimum require-
tying positions.
ments, see Section 5.1. To investigate in detail the
The transverse shear strain interpolations used are
predictive capability of the proposed elements, we per-
~ert ¼ a1 þ b1 r þ c1 s; formed convergence studies for various shell problems
ð40Þ [2], see Sections 5.2–5.4.
~est ¼ a2 þ b2 r þ c2 s;
For convergence studies, we use the s-norm [14]
where using Eq. (38), defined as
954 P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962
Fig. 8. Strain interpolation schemes and tying points of the MITC triangular shell finite elements.
Z
discretized domain, respectively, and the relationship
k~ uh k2s ¼
u ~ T D~
D~ r dX; ð42Þ
X between them is
¼~
D~ h ¼ Bð~
~ xÞU Bh ð~
xh ÞUh ;
ð44Þ with
r ¼~
D~ r ~rh ¼ Cð~ xÞU Ch ð~
xÞBð~ xh ÞBh ð~
xh ÞUh ;
D~ ref ~
¼~ h ¼ Bref ð~
xÞUref Bh ð~
xh ÞUh ;
where C denotes the material stress–strain matrix and B
is the strain–displacement operator. The position vectors r ¼~
D~ rref ~
rh ¼ Cref ð~
xÞBref ð~
xÞUref ð47Þ
x and ~
~ xh correspond to the continuum domain and the Ch ð~
xh ÞBh ð~
xh ÞUh :
P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962 955
(a)
(b)
(c)
Fig. 9. Geometric relationships between tying points used for triangular and quadrilateral MITC shell elements. For the MITC9
element, see Ref. [21]: (a) selection of tying points for the transverse shear strains of the MITC3 element; (b) selection of edge tying
points for the transverse shear strains of the MITC6 element and (c) selection of tying points for the in-plane strains of the MITC6
element.
To consider the convergence of the discretization dal numbering sequences as shown in Fig. 10. The r-
schemes with various thicknesses, we use the relative axis and s-axis always run from nodes 1 to 2 and
error given by nodes 1 to 3, respectively. To pass the test, exactly
the same response should be obtained for all possible
k~ uh k2s
uref ~ (three-dimensional) tip forces and moments.
relative error ¼ : ð48Þ • Zero energy mode test. This test is performed by
uref k2s
k~
counting the number of zero eigenvalues of the stiff-
ness matrix of one unsupported shell finite element,
5.1. Basic tests which should be exactly six, and the corresponding
eigenvectors should of course be physical rigid body
The following basic tests are performed as basic modes. We recommend that, when performing this
requirements for the triangular shell elements. test, various possible geometries be taken because
an element might pass the test for a certain geometry
• Isotropic element test. Although the theory shows but not for other geometries.
that the proposed elements are isotropic, we include • Patch test. The patch test has been widely used to test
this numerical test to illustrate the isotropy that tri- elements, despite its limitations for mixed formula-
angular elements, in general, should satisfy. Consid- tions, see Ref. [1]. We use the test here in numerical
ering any geometrical triangular element, this test form to merely assess the sensitivity of our elements
should be passed. The test is performed by analyzing to geometric distortions. The mesh used for the patch
the three same triangular elements with different no- test is taken from Ref. [1] and shown in Fig. 11. The
956 P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962
P i ¼ ffx ; fy ; fz ; ma ; mb gT .
Fig. 10. Isotropic element test of the 6-node triangular shell element. ~
Table 1
Basic test results of the MITC triangular shell finite elements
Element Isotropic element test Zero energy mode test Membrane patch test Bending patch test
MITC3 Pass Pass Pass Pass
MITC6-a Pass Pass Pass Pass
MITC6-b Pass Pass Pass Pass
P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962 957
z
q t
x
y y
2L 2L
D C 2L D C 2L
x x
A B A B
(a) (b)
Fig. 12. Clamped plate under uniform pressure load with uniform 4 · 4 meshes of triangular elements (L ¼ 1:0, E ¼ 1:7472 107 and
m ¼ 0:3).
-3 -3 -3
t/L = 1/100
t/L = 1/1000
t/L = 1/10000
Fig. 13. Convergence curves for the clamped plate problem. The bold line shows the optimal convergence rate, which is 2.0 for linear
elements and 4.0 for quadratic elements. For the MITC3 element, the solid and dotted lines correspond, respectively, to the results
obtained using the meshes in Fig. 12(a) (solid line results) and (b) (dotted line results).
Fig. 13 reports the convergence of the MITC trian- use the solution using the MITC9 element with a mesh
gular shell elements in the relative error of Eq. (48). We of 96 · 96 elements as reference. As the plate thickness
958 P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962
decreases, the MITC3 element locks but still good ux ¼ b ¼ 0 along BC;
accuracy characteristics are seen for t=L up to about uy ¼ a ¼ 0 along DC; and ð51Þ
1/1000. The MITC6-a element shows almost optimal
uz ¼ a ¼ 0 along AB:
convergence.
For the clamped case, the boundary conditions are
C x 2 þ z2 ¼ 1 þ y 2 ; y 2 ½1; 1
: ð53Þ
β
D The loading imposed is the smoothly varying periodic
pressure normal to the surface,
θ B
y pðhÞ ¼ p0 cosð2hÞ; ð54Þ
A
x R which is the same distribution as shown in Fig. 14.
2L
A bending dominated problem is obtained when both
α ends are free and a membrane dominated problem is
obtained when the ends are clamped.
Using symmetry, the analyses are performed using
1
one eighth of the structure, the shaded region ABCD in
Fig. 17(a). Considering the boundary conditions, we
0.5 have for the free case
p ( θ ) / p0 0
uz ¼ b ¼ 0 along BC;
ux ¼ b ¼ 0 along AD; and ð55Þ
-0.5 uy ¼ a ¼ 0 along DC
log ( relativeerror )
-3 -3 -3
t/L =1/100
t/L =1/1000
t/L =1/10000
Fig. 15. Convergence curves for the clamped cylindrical shell problem. The bold line shows the optimal convergence rate, which is 2.0
for linear elements and 4.0 for quadratic elements.
-3 -3 -3
t/L = 1/100
t/L = 1/1000
t/L = 1/10000
Fig. 16. Convergence curves for the free cylindrical shell problem. The bold line shows the optimal convergence rate, which is 2.0 for
linear elements and 4.0 for quadratic elements.
960 P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962
z
0
1
1
0.5
y
α 2L x
y 0 D C
-0.5
β
-1 B
-1 A
0 θ
1
(a) x z
(b)
Fig. 17. (a) Hyperboloid shell problem (E ¼ 2:0 1011 , m ¼ 1=3 and p0 ¼ 1:0) and (b) graded mesh (8 · 8, t=L ¼ 1=1000, clamped case).
-3 -3 -3
t/L = 1/100
t/L = 1/1000
t/L = 1/10000
Fig. 18. Convergence curves for the clamped hyperboloid shell problem. The bold line shows the optimal convergence rate, which is
2.0 for linear elements and 4.0 for quadratic elements.
For both cases, we use the reference solution cal- free case, the very thin boundary layer was not specially
culated using a mesh of 96 · 96 MITC9 shell elements. meshed.
For the clamped case, halfpffi the mesh is used in the Fig. 18 shows the convergence curves of the
boundary layer of width 6 t, see Fig. 17(b). For the MITC triangular shell elements in the clamped case.
P.S. Lee, K.J. Bathe / Computers and Structures 82 (2004) 945–962 961
-3 -3 -3
t/L = 1/100
t/L = 1/1000
t/L = 1/10000
Fig. 19. Convergence curves for the free hyperboloid shell problem. The bold line shows the optimal convergence rate, which is 2.0 for
linear elements and 4.0 for quadratic elements.
The MITC3 and MITC6-a elements show quite good Chapelle and Bathe [2,22]), the same interpolation ap-
convergence for this membrane dominated shell proach is of course also applicable to 3D-shell elements
problem. [23].
The convergence curves when the edges of the The three elements considered are good candidates
structure are free are shown in Fig. 19. This is a difficult for the analysis of general shell structures in engineering
problem to solve when the thickness is small [21], but the practice in which the range of t=L is usually from 1/10 to
problem is an excellent test case because of the negative about 1/1000. The elements show good behavior in the
Gaussian curvature of the shell surface. The elements chosen test problems for that range of thickness values.
show all some locking but in fact good accuracy char- However, it is still necessary to study these elements
acteristics for the practical range of t=L up to about 1/ further and to obtain uniformly optimal triangular shell
1000. finite elements that behave equally well in all types of
shell problems.
6. Conclusions
[5] Bathe KJ, Chapelle D, Lee PS. A shell problem ‘highly [15] Bletzinger KU, Bischoff M, Ramm E. A unified approach
sensitive’ to thickness changes. Int J Numer Methods Eng for shear-locking-free triangular and rectangular shell finite
2003;57:1039–52. elements. Comput Struct 2000;75:321–34.
[6] Dvorkin EN, Bathe KJ. A continuum mechanics based [16] Argyris JH, Papadrakakis M, Apostolopoulou C, Kout-
four-node shell element for general nonlinear analysis. Eng sourelakis S. The TRIC shell element: theoretical and
Comput 1984;1:77–88. numerical investigation. Comput Methods Appl Mech Eng
[7] Bathe KJ, Dvorkin EN. A formulation of general shell 2000;182:217–45.
elements––the use of mixed interpolation of tensorial [17] Bernadou M, Eiroa PM, Trouve P. On the convergence of
components. Int J Numer Methods Eng 1986;22:697– a discrete Kirchhoff triangle method valid for shells of
722. arbitrary shape. Comput Methods Appl Mech Eng 1994;
[8] Bucalem ML, Bathe KJ. Higher-order MITC general shell 118:373–91.
elements. Int J Numer Methods Eng 1993;36:3729–54. [18] Sze KY, Zhu D. A quadratic assumed natural strain
[9] Bathe KJ, Brezzi F, Cho SW. The MITC7 and MITC9 curved triangular shell element. Comput Methods Appl
plate bending elements. Comput Struct 1989;32:797–814. Mech Eng 1999;174:57–71.
[10] Bucalem ML, N obrega SHS. A mixed formulation for [19] Kim JH, Kim YH. Three-node macro triangular shell
general triangular isoparametric shell elements based on element based on the assumed natural strains. Comput
the degenerated solid approach. Comput Struct 2000;78: Mech 2002;29:441–58.
35–44. [20] Ahmad S, Irons BM, Zienkiewicz OC. Analysis of thick
[11] Bathe KJ, Iosilevich A, Chapelle D. An inf–sup test and thin shell structures by curved finite elements. Int J
for shell finite elements. Comput Struct 2000;75:439–56. Numer Methods Eng 1970;2:419–51.
[12] Bathe KJ, Iosilevich A, Chapelle D. An evaluation of the [21] Bathe KJ, Lee PS, Hiller JF. Towards improving the
MITC shell elements. Comput Struct 2000;75:1–30. MITC9 shell element. Comput Struct 2003;81:477–89.
[13] Bathe KJ. The inf–sup condition and its evaluation for [22] Chapelle D, Bathe KJ. The mathematical shell model
mixed finite element methods. Comput Struct 2001;79:243– underlying general shell elements. Int J Numer Methods
52, 971. Eng 2000;48:289–313.
[14] Hiller JF, Bathe KJ. Measuring convergence of mixed [23] Chapelle D, Ferent A, Bathe KJ. 3D-shell elements and
finite element discretizations: an application to shell their underlying mathematical model. Math Models Meth-
structures. Comput Struct 2003;81:639–54. ods Appl Sci 2004;14:105–42.