This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright

Author's personal copy

Finite Elements in Analysis and Design 44 (2008) 784 -- 796

Contents lists available at ScienceDirect

Finite Elements in Analysis and Design

journal homepage: w w w . e l s e v i e r . c o m / l o c a t e / f i n e l

Nine-node shell elements with 6 dofs/node based on two-level approximations. Part I

Theory and linear tests
P. Panasz, K. Wisniewski
IFTR, Polish Academy of Sciences, Swietokrzyska 21, 00-049 Warsaw, Poland

A R T I C L E

I N F O

Article history:
Received 10 September 2007
Received in revised form 7 April 2008
Accepted 22 May 2008
Available online 16 July 2008
Keywords:
Nine-node shell elements
6 dofs/node
Drilling rotation
Two-level approximation
Assumed strain
Selective reduced integration

A B S T R A C T

The paper concerns 9-node quadrilateral shell elements derived for Reissner's kinematics. They are based
on the Green strain and potential energy, and are applicable to large (unrestricted) rotations. The characteristic features of the developed elements are as follows:
1. Drilling rotation is included via the drill rotation constraint (RC) imposed by the penalty method. Hence,
the elements have 6 dofs per node, i.e. three displacements and three rotational parameters, including
drilling rotation.
2. Transverse shear and membrane locking as well as the in-plane shear over-stiffening are avoided using
the two-level approximation applied to the strain (assumed strain method). This method does not affect
the drilling RC.
3. A modification of the two-level approximation method is proposed, consisting in treating the sampling
and the numerical integration together, which results in six sampling points being replaced by two
sampling lines. The two-level approximation is applied to components in the ortho-normal basis at the
element center, which differs our element from the MITC family of elements, which uses the covariant
strain components.
4. Selective reduced integration (SRI) approach is revised. The total functional is split into several parts,
and a suitable integration rule is found for each part, yielding an efficient element which shows very
good mesh convergence.
Two 9-node shell elements are developed and subjected to a range of benchmark tests, to establish the
sensitivity to mesh distortion, the coarse mesh accuracy, and to confirm the lack of locking. Our results
are compared with results obtained by the MITC9 element of ADINA and the S9R5 element of ABAQUS.
2008 Elsevier B.V. All rights reserved.

1. Introduction
Nine-node shell elements are more complex than 4-node elements, but are considered to be advantageous in some applications,
involving curved boundaries and dominant in-extensional bending.
1.1. Locking of 9-node shell elements
The basic 9-node isoparametric Lagrangian shell element suffers
from locking in uni-directional bending, caused by approximations
of the transverse shear strains and of the membrane strains. The
problem is analogous to the one for beams, considered e.g. in [1].
To solve this problem, bending of a 3-node 2D beam element
is studied. For the transverse shear strain, a straight element is

Corresponding author.
E-mail address: kwisn@ippt.gov.pl (K. Wisniewski).

0168-874X/$ - see front matter 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.finel.2008.05.002

considered, see e.g. [2], and, from the condition that the approximated strain should be equal to the analytical one for the linearly
distributed
bending moment, we obtain a location of Barlow's points,
 = 1/ 3. For the membrane shear, a curved element is considered,
see e.g. [3], and, from the condition that this strain should vanish in
pure bending, we obtain the same locations as for the shear strain.
The latter result, however, is approximate and valid only for shallow
elements.
The points at which the bending strains are exact are used also
for shells; in fact several methods of using them exist, and they are
discussed below.
Another problem is a too stiff behavior of 9-node shell elements
in bending by in-plane forces, caused by the in-plane shear strain
12 . Hence, this strain component is specifically treated in 9-node
elements, and the basic observation is that much more
accurate

values of 12 are obtained at four points (, ) = (1/ 3, 1/ 3).

Several methods using this observation were developed and they are
discussed below.

Author's personal copy

785

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

b
a

D
a

C
a

Fig. 1. Location of sampling points. Reduced number of points: (a) in  direction; (b) in  direction; (c) in both directions.

1.2. Methods of avoiding locking

Several methods of avoiding locking were proposed in the literature, and they all use the points at which strains are exact, found
either for a 3-node beam or for a planar 9-node element. These methods can be summarized as follows:
1. Uniform reduced integration (URI) combined with stabilization.
The URI was proposed in [4], but it yields a rank-deficient stiffness
matrix, and requires stabilization; various methods of deriving
the stabilization matrix were proposed e.g. in [57]. Note that the
2 2 Gauss integration applied to the in-plane shear strain 12
does not yield spurious zero eigenvalues (mechanisms), see e.g.
[8].
2. Selective reduced integration (SRI), which uses the Gauss points
coinciding with the points at which the strains are exact. It has
two deficiencies. First, the membrane and bending strain energy
can be decoupled only when the material properties are constant
through the thickness or symmetric w.r.t. the mid-surface. This
excludes the use of the SRI elements e.g. to plasticity with several
integration points through the thickness. The second deficiency,
according to the literature, is that the SRI elements exhibit poor
mesh convergence in some problems; e.g. for the pinched hemisphere reported in [7, p. 405], or the pinched cylinder reported in
[9, p. 2407]. This deficiency in not shared by our 9-SRI element,
which performs in these tests very well.
3. The assumed strain (AS) method in conjunction with the concept
of two-level approximation. This concept consists of sampling
the strain components at certain points, and extrapolating these
values over the element. The AS method is applied in the current paper to 9-node shell elements with drilling rotation. The AS
method was gradually developed for plates and shells in several
works, including [2,6,917], and many others. It is also covered in
two books: [18,19]. Different variants of the AS method has been
developed for:
(a) the transverse shear strains in 4-node plate and shell elements and
(b) the transverse shear strains and the membrane strains in
3-node beams and 9-node plate and shell elements. The
most often used locations of the sampling points for 9-node
elements are shown in Fig. 1.
3. Generally, the variants of the AS found in the literature differ in
the components which are sampled, in the location of sampling
points, and in the interpolation functions.
1.3. Two-level approximations of the AS method
In the two-level approximations we have to assume the position
of the sampling (tying) points, and the form of the interpolation
functions. The three sets of the sampling points, which are in use,

are shown in Fig. 1, and they are combined with the interpolation
functions
as presented below. In all the formulas presented below,

a=

1.
3

1. For the strains  and 3 ( = 1, 2), the sampling points are
shown in Fig. 1a and b, and the set of interpolation functions
proposed in [2,16] is used:
(i) For the reduced number of points in the  direction, Fig. 1a,



   2 
1
1
,

RA (, ) =
4
a
b
b

  


 2 
1
RB (  ,  ) =
1+
,

4
a
b
b

  


 2 
1
RC (, ) =
1+
,
+
4
a
b
b

  


 2 
1
RD (, ) =
1
,
+
4
a
b
b






 2
1
RE (, ) =
,
1+
1
2
a
b





2


1
RF (, ) =
1
1
.
(1)
2
a
b
This set is applied to the strains 11 and 31 .
(ii) For the reduced number of points in the  direction, Fig. 1b,
 

  2 
1
1

RA (, ) =
,
4
a
b
b
 

  2 
1
1+

RB ( ,  ) =
,
4
a
b
b
 

  2 
1
1+
+
RC (, ) =
,
4
a
b
b
 

  2 
1
1
+
RD (, ) =
,
4
a
b
b

 2


1
1+
RE (, ) =
,
1
2
a
b

 2


1
1
RF (, ) =
.
(2)
1
2
a
b
This set is applied to strains 22 and 32 .
1. In the direction in which the number of points is not reduced,
the
 points can be located in two ways, and then either b = 1 or

3 can be used in the above functions, see [16]. The first value is
5

preferred in the cited paper, because: (i) it provides continuity of

shear strains between adjacent elements, p. 53, and (ii) the shear
patch test is passed, see the remark on p. 90, and the test on
p. 59. The second value is used e.g. in [17] and in the current
paper, because it has the advantage that the sampling points and
the integration points coincide.

Author's personal copy

786

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

2. For the shear strain 12 three approaches are used:

(a) In [6,16], the 6-point approximation schemes of Fig. 1a and
b, and of Eqs. (1) and (2) are applied to the particular parts
of 12 .
(b) In [9], the scheme also is based on the sampling points of Fig.
1a and b, but different approximation functions are used:
(i) For the reduced number of points in the  direction, Fig. 1a,



1


1

RA (, ) =
1
,
2
a
6
2



1


1
1+

,
RB (, ) =
2
a
6
2



1


1
RC (, ) =
1+
,
+
2
a
6
2



1


1
RD (, ) =
1
+
,
2
a
6
2


 2
1
RE (, ) =
1+
,
2
a 3


 2
1
1
.
(3)
RF (, ) =
2
a 3


This set of functions is applied to strains 12 .
(ii) For the reduced number of points in the  direction, Fig. 1b,


 1 
1
1
,
RA (, ) =

2
a
6
2




 1 
1
RB (, ) =
1+
,

2
a
6
2


 1 
1
RC (, ) =
1+
+
,
2
a
6
2


 1 
1
RD (, ) =
1
,
+
2
a
6
2


 2
1
1+
RE (, ) =
,
2
a 3


 2
1
1
.
(4)
RF (, ) =
2
a 3

This set of functions is applied to strains 12 .
(c) In [17], the reduced number of sampling points is used in
both directions, see Fig. 1c, and the following approximation
functions are used:


 

1
1
1
,
RA (, ) =
4
a
a





1
RB (, ) =
1+
1
,
4
a
a




1
RC (, ) =
1+
1+
,
4
a
a





1
RD (, ) =
1
1+
.
(5)
4
a
a
The use of four sampling points is more convenient for nonlinear strains.
For all the above schemes, the strain component  to which we
apply the two-level approximation, is expressed as follows:


(, ) =
Ri (, )i ,
(6)
i

where i = A, B, C, D, E, F for the schemes with six sampling points and

i = A, B, C, D for the scheme with four sampling points, see Fig. 1.
1.4. Objectives and scope of the paper
In this paper we develop and test the 9-node quadrilateral shell
elements based on Reissner's kinematics and the Green strain. Classical papers on the subject are restricted to two-parameter rotations,

which have to be defined in the local basis and transformed to the

reference basis. We define an extended configuration space and include the drilling rotation, so, in effect, we have three rotational parameters per node, and the rotation vector is directly assumed in the
reference basis, see Section 2.
The shell elements are implemented within the methodology of
either the two-level approximations or the SRI, as the means to avoid
the transverse shear and membrane locking as well as the in-plane
shear over-stiffening, see Section 3. The novel aspects of the paper
are related to them.
1. In the classical papers, the two-level approximation is applied
to strains, i.e. as the AS method or the assumed natural strain
method. We apply this methodology to strains in Section 3.1,
which does not affect the drilling rotation constraint (RC).
2. We modify the two-level approximation method, by treating the
sampling and the numerical integration together, and by replacing the six sampling points by two sampling lines, which facilitates the implementation and significantly improves efficiency of
differentiation.
A characteristic feature of all our elements is that the components
of the strain in the ortho-normal basis at the element's center are
sampled and interpolated. This feature differs our elements from
the MITC family of elements, see [17,20], which uses the covariant
strain components.
Another important issue is the selection of components of the
first order strain to which the two-level approximation is applied,
and as a result of numerical tests, we propose to modify only the
twisting component 12 .
3. The SRI approach is revised in Section 3.2, and a suitable integration rule is found for each term of the total functional. In addition
to the 2 3 and 3 2 integration schemes exploited to avoid the
transverse shear and membrane locking, we have identified several terms of the strain energy which can be integrated by the
reduced 2 2 scheme, including 212 , 11 22 and 212 , which leads
to an efficient element showing a very good mesh convergence.
Two 9-node shell elements are developed, based on the AS and
SRI methodology, and subjected to a range of benchmark tests in
Section 4. The classical five-element patch tests are passed. As a
supplement, we also suggest a one-element distortion test, with the
central node displaced along the diagonal. Our results are compared
with the results by similar elements of the commercial codes, such
as ADINA and ABAQUS, and they confirm a very good quality of our
elements.
2. Formulation of 9-node shell element
In this section we present these aspects of the formulation which
are common to all our 9-node shell elements.
2.1. Basic shell equations
Extended configuration space: The classical configuration space of
.
the non-polar Cauchy continuum is defined as: C = {v : B R3 },
where v is the deformation function defined on the reference configuration of the body B. In the present work, we consider an extended
configuration space, defined in terms of the deformation function v
and rotations Q SO(3). The rotations can be treated in two different ways: either remain unconstrained, as in the Cosserat-type continuum or are constrained, either by the polar decomposition of F
equation or the RC equation
skew(Q T F) = 0,

(7)

.
where F = v. The equivalency of these two forms of constraints
is considered e.g. in [21]. Then, the extended configuration space is

Author's personal copy

787

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

defined as follows:

Cext =. {(v, Q ) : B R3 SO(3)|v C}.

3

Note that v belongs to the classical configuration space C, i.e. it is

identical as that for the classical non-polar Cauchy continuum. The
configuration space of Eq. (8) is used in the present work. Note that
this formulation is different than in the Cosserat-type formulation
used e.g. in [22,23].
Formulation including rotations for Green strain: We use the formulation developed for a 3D continuum, see e.g. [24,25], and modify it by introducing Reissner's shell kinematics. The formulation is
based on the following functional:
.
F3 (v, Q , Ta ) =

t3

(8)

t2

2

3

t1
1

(11)

.
where S is the second PiolaKirchhoff stress tensor, and C = FT F is
1
the right CauchyGreen deformation tensor. Since 2 C = E, hence,
.
Eq. (11) implies that the Green strain E = 12 (C I) is the work
conjugate to S. Assume that the strain energy density per unit non , is a function of C, so it satisfies the objectivity
deformed volume, W
(C) = j W
(C)
requirement. The variation of the strain energy is W
C
C = jE W(E) E, and, from W = S E we obtain the constitutive
law,
(12)

Reissner's hypothesis: The initial (reference) configuration of the

shell is parameterized in terms of n = { , 2 /h},  = 1, 2, where 
[1, +1] are the natural coordinates parameterizing the reference
(middle) surface, and [h/2, +h/2] is the coordinate used in the
direction normal to this surface. h denotes the initial shell thickness.
The position vector of an arbitrary point of a shell in the initial
configuration is expressed as
y( , ) = y0 ( ) + t3 ( ),

(13)

where y0 is a position of the reference surface, and t3 is the shell

director, i.e. the vector normal to the reference surface. In the current configuration, the position vector is expressed by the Reissner
kinematical hypothesis:
x( , ) = x0 ( ) + Q0 ( ) t3 ( ),

a1
1

y0

x0
i3

i2

Fig. 2. Local bases on the reference surface of a shell.

where  (0, ) is the regularization parameter.

The virtual work (VW) of the nominal stress P can be expressed
as

S = jE W(E).

2

(9)

where the skew-symmetric Ta is the Lagrange multiplier for the RC

equation (7). By regularization of this functional in Ta we obtain a
two-field functional


.
F2 (v, Q ) = [W(FT F) + skew(Q T F) skew(Q T F)] dV + Fext , (10)
2
B

P F = 12 S C,

a2

B,Q

i1
[W(FT F) + Ta skew(Q T F)] dV + Fext ,

a3

u0

(14)

where x0 is a position of the reference surface, and Q0 SO(3) is a

rotation tensor, constant over .
Approximation of rotations: The two-field F2 (v, Q ) of Eq. (10) involves the rotations Q , and hence we also have to approximate
them over . We assume that the rotations are constant over , i.e.
Q ( , ) Q0 ( ), where Q0 ( ) is the rotation at the reference surface. Thus, the approximation of rotations in F2 (v, Q ) is the same as
in Reissner's hypothesis, Eq. (14).

Fig. 3. Numbering of nodes of 9-node element.

2.2. Features of 9-node shell element

Lagrangian shape functions: For the bi-quadratic Lagrange polynomial, we obtain the following shape functions:
N1 (, ) = 14 (2 )(2 ),
N2 (, ) = 14 (2 + )(2 ),
N3 (, ) = 14 (2 + )(2 + ),
N4 (, ) = 14 (2 )(2 + ),
N5 (, ) = 12 (2 + )(1 2 ),
N6 (, ) = 12 (2 + )(1 2 ),
N7 (, ) = 12 (2 + )(1 2 ),
N8 (, ) = 12 (2 )(1 2 ),
N9 = (1 2 )(1 2 ),

(15)

.
.
where the natural coordinates are  = 1 ,  = 2 and ,  [1, +1].
The shape functions Ni correspond to the nodes i = 1, . . . , 9, shown
in Fig. 3. Our 9-node element is iso-parametric, and the above shape
functions are used to approximate the position vectors, displacements and rotational parameters.
Canonical parametrization of rotations: Denote by w the canonical
rotation pseudovector, for which the rotation tensor Q0 is parameterized as follows:
sin

1 cos
2
.
w+
w ,
Q0 (w) = I +

2


= w = w w  0,

(16)

.
=
w I. We assume that w TI SO(3), i.e. the rotation vector
where w
belongs to the tangent space to SO(3) at Q0 = I.

Author's personal copy

788

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

g2

g2

g3

~
g
2

~t
2

t2


~
t1

g1

g1

~
g
1


t1

Fig. 4. (a) Natural basis {gk }. (b) Position of ortho-normal basis {tk }.

Local bases at the reference surface: The reference Cartesian basis is

denoted by {ik }, k = 1, 2, 3. Two local bases at the element's reference
(middle) surface are used in the element.

The displacement gradient in the local ortho-normal basis {tk } is

obtained on use of

1. The local natural bases {gk } is defined as follows:

.
where the rotation matrix R0 = [t1 |t2 |t3 ] SO(3). In order to pass
the membrane patch test, we use the modified form

g1 = y, ,

g2 = y, ,

g3 = g1 g2 ,

(17)

see Fig. 4a. In general, the vectors g1 and g2 are skew and not unit.
2. The local ortho-normal basis {tk } is defined as follows:
1
t1 = (t 1 t 2 ),
2

1
t2 = (t 1 + t 2 ),
2

.
t3 = g 3 ,

(18)

where the auxiliary vectors are

g + g 2
,
t 1 = 1
g 1 + g 2

t 2 = t3 t 1 .

(19)

The normalized vectors of the natural basis are denoted by a tilde,

i.e. g k =gk / gk . Note that the vectors t1 and t2 are equally distant
from the vectors of the natural basis, see Fig. 4b.
Drilling RC: Consider the components of the RC equation, Eq. (7),
in the local ortho-normal basis {ti }, see Eq. (18) and Fig. 2. The transverse components 3 ( = 1, 2) of this equation are typically neglected, but the components 12 and 21 at =0 provide the drilling RC
equation,
x0,1 a2 x0,2 a1 = 0,

(20)

where a =Q0 t are the vectors of the forward-rotated ortho-normal

.
basis {ai }, see Fig. 2. Denote Cd = x0,1 a2 x0,2 a1 , then, the drilling
RC can be expressed as Cd = 0, and in Eq. (10) we can use


. 
Fdrill = skew(Q0T F0 ) skew(Q0T F0 ) = Cd2 .
2
2

(21)

Displacement gradient in local basis: In this paragraph we assume

that the bold letters denote arrays of components in the reference
basis {ik }. For the position vectors of Eqs. (14) and (13), the displace.
ment of a shell is as follows u( ) = x( ) y( ) = u0 + (a3 t3 ). The
displacement gradient can be expressed as
(22)

(24)

T uR ,
(u)Lc = R0c
0c

(25)

where R0c is the rotation matrix at the element's center, n = (0, 0, 0).
Remark. Eq. (25) is a transformation of u to the local basis {tk } at
the element's center. Note that a different scheme was devised in
[16, p. 90], where components in two different bases were used for
interpolation of the membrane and transverse shear strains. For the
membrane strain interpolations, the components in the orthogonal
curvilinear coordinate system were used, while for the transverse
shear strain interpolations, the components in the natural coordinate
system were used. In our implementation we use only components
in one basis at the element's center.
Saint Venant-Kirchhoff strain energy for plane stress condition:
.
Consider Saint Venant-Kirchhoff's form of the strain energy W =
1 (tr E)2 + G tr(E2 ), where E is the Green strain, and the Lam con2
stants are:  = E/((1 + )(1 2)) and G = E/(2(1 + )), where E, 
are Young's modulus and Poisson's ratio, respectively.
For Reissner's kinematics, Eq. (14), the Green strain can be approximated linearly over the thickness, i.e. E( ) + j, and the integration of the strain energy over the thickness, see Eq. (29), yields the
shell strain energy in the following additive form: Wsh = W0 + W1 ,
where

W0 = h[ 12 (tr )2 + G tr(2 )],

W1 =

More details related to the drilling RC are given e.g. in [21,26].

ju 1
. ju
u =
J ,
=
jy jn

(u)L = R0T uR0 ,

h3
12


1
(tr j)2 + G tr(j2 ) .
2

(26)

Note that W0 consists of the membrane and transverse shear energy, while W1 of the bending and twisting energy. The constitutive equations for the cross-sectional resultant forces and couples
are obtained as follows:
.
n = j W0 = h[(tr )I + 2G],
h3
.
[(tr j)I + 2Gj].
m = jj W1 =
(27)
12

.
.
where n = {1 , 2 , 3 }, 3 = (2/h) [1, +1], and the Jacobian
matrix is

From the zero-normal-component condition for these resultants, i.e.

n33 = 0 and m33 = 0, we obtain the normal strain components:

. jy
= [g1 |g2 |g3 ].
J=

33 =

jn

(23)


( + 22 ),
 + 2G 11

33 =


( + 22 ).
 + 2G 11

(28)

Author's personal copy

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

On use of them we accordingly modify the strain energy functions

of Eq. (26).
Variation of functional F2 for shell: Let us define shell-type functionals,
.
()sh =

+h/2
h/2

() d ,

(29)

.
where () is for a 3D body, and ()sh is for a shell. Besides, = det Z,
where Z is the shifter tensor. In this way, we first define the shelltype counterparts of all components of F2 of Eq. (10), and, next,
calculate their variation:

F2sh = Wsh + (Fdrill )sh (Fext )sh ,

(30)

where the drilling term of the RC is defined by Eq. (21). Let us define
the vectors of nodal values


. u0I
.
(31)
q=

wI

Because, the rotation vector wI has three components, hence our

9-node element has 54 nodal dofs.
Below, we discuss separately the components of Eq. (30).

(i) The VW of the stress for a 3D body is W = V E S dV, where
.
the Green strain E = 12 (FT F I) and the second PiolaKirchhoff
stress S are a work conjugate pair. For a shell, due to kinematical
assumptions, the strain is a polynomial of the thickness coordinate , i.e. E( ). Therefore, the stress also is a function of , i.e.
S( ).
Separating integration over the reference surface A from inte
gration over the thickness we have: W = A Wsh dA, where
the shell VW of stress is
+h/2
Wsh =.
E( ) S( ) d .
(32)
h/2

.
On use of the tangent operator for the strain, B = jE/ jq, we can
write
+h/2
E( ) = B( )q, Wsh = q
BT( ) S( ) d .
(33)
h/2

(ii) The drilling term of the RC equation, becomes

(Fdrill )sh =.

+h/2
h/2

Fdrill d = h Cd2 = h(Cd )q bd ,

2

(34)

.
where the tangent operator bd = jCd / jq. The value of  was
selected by taking into account the results of [24] and numerical
experiments.
(iii) The VW of external loads on the upper and lower bounding surfaces, for the shell kinematics becomes
 
p
(Fext )sh = q
,
(35)

m
are shell-type external forces and moments.
where p and m
Collecting all terms together, we obtain a weak form of the equilibrium equation
 

p
q
dA = 0.
(36)
BT S + h(Cd ) bd

m
A
FE equilibrium equation for shell: Let us introduce the FE approximations in the configuration space,
y0 =

nnode

I=1

NI y0I ,

u0 =

nnode

I=1

NI u0I ,

w=

nnode

I=1

NI wI ,

(37)

789

where NI are the shape functions and I is the node index. Then, Eq.
(36) becomes

(FE)
[BT (q)S(q) + h(Cd )bd ] dA f(q),
A
 
(FE)
p
dA p,
(38)

A m
where f is the internal force vector including the drilling RC contribution and p is the vector of external forces. Hence, the residual
vector for the whole shell is
.
r = f(q) p = 0.

(39)

From this vector we can obtain the stiffness matrix K by the standard
procedure of consistent linearization.
3. Techniques improving element's performance
Below are discussed three techniques for improving the elements
performance, which subsequently are implemented in three different
shell elements.
3.1. Modified AS method
The two-level schemes of Fig. 1a and b involve 12 sampling points,
which results in a considerable number of additional evaluations.
Therefore,
we propose a modification of this scheme, taking the value


b = 35 , and considering the sampling and the numerical integration

together. This procedure is described in detail below.
Consider the Green strain components in the local basis {tk } at
the element's center. They can be obtained from the components in
the reference basis {ik } by the transformation defined in Eq. (25) for
the displacement gradient.
The transverse shear component 13 should be sampled at the
reduced number of points in -direction, see Fig. 1a and Eq. (1). The
sampling points and the points for the 3 3 Gauss
 integration are

shown in Fig. 5a. We see that for the value b = 35 , the sampling
points and the integration points are located on the same three lines
in the -direction:  =b,  =0 and  =+b. Therefore, the integration
points do sample 13 at correct -coordinates, and we do not need to
make any modification in the -direction. In consequence, it suffices
to apply the two-level approximation in the -direction only.
Therefore, instead of the 6-point formula of Eq. (6), we use a much
simpler formula for the linear approximation in the -direction,






1
1
1
1+
  (, ) =
|=a +
|=+a ,
(40)
2
a
2
a

.
where we denoted  = 13 and a = 13 . This formula involves, instead
of the sampling points, the two sampling lines shown in Fig. 5b,
where L1 is located at  = a and L2 at  = a. An analogous formula
.
can be used for the transverse shear  = 23 ,

  (, ) =



1
1
1
1+
|=a +
|=+a ,
2
a
2
a

(41)

with the reduced number of sampling points in the -direction, see

Fig. 1b and Eq. (2).
Eqs. (40) and (41) are applied also to the in-plane strains in the
local basis {tk } at the element's center, i.e. to 11 and 22 . In regard to
12 and 12 , we reduce the degree of an approximation polynomial
in both directions, and use the sampling points shown in Fig. 1c
together with the approximation functions of Eq. (5).
Remark. The proposed approximation functions of Eqs. (40)
 and (41)
can be used only if the sampling points are located at b=

3 , as then
5

Author's personal copy

790

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

b
X

b
X

L1
- sampling points
X - Gauss points

L2

L1,L2 - sampling lines

Fig. 5. Location of: (a) sampling points; (b) sampling lines.

Table 2
The integration scheme of 9-SRI shell element

Table 1
The AS interpolation of 9-AS shell element
Components

Approximation scheme

Integration rule

Component matrix K for

11 , 13
22 , 23
12

Eq. (40)
Eq. (41)
Eq. (5), Fig. 1c

33

211 , 222 , 11 22 , drill RC term

12

Eq. (5), Fig. 1c

23
32

211 , 231
222 , 232

22

212 , 11 22 , 212

they are on the same lines as the Gauss points. Our tests, suchas e.g.

of Sections 4.2.2 and 4.2.3, indicate that the elements for b = 35 are
more robust to the mesh distortion than the elements for b = 1.
The proposed forms of the two-level approximation are implemented in the shell element denoted as 9-AS, and characterized in
Table 1. The element is fully integrated, using the 3 3 Gauss points.

Table 3
Our 9-node shell elements with 6 dofs/node
Element

Characteristics

Integration over lamina

Described in

9
9-AS
9-SRI

Basic
AS

FI
FI
SRI

Section 2.2
Section 3.1
Section 3.2

3.2. Selective reduced integration

Consider the case of the material properties being constant
through the thickness or symmetric w.r.t. the mid-surface. Then, the
shell strain energy, modified by the zero-normal stress assumption
of Eq. (28) and integrated over the shell thickness, is decoupled:
Wsh = W0 + W1 , where W0 and W1 are defined by Eq. (26), and
are sums of the following strain terms:

W0 = W0 (211 , 222 , 11 22 , 212 , 231 , 232 ),

(42)

W1 = W1 (211 , 222 , 11 22 , 212 ).

(43)

Each term is multiplied by the respective shell-type constitutive coefficient, and yields the corresponding stiffness matrix,
K0 = K(211 ) + K(222 ) + K(11 22 ) + K(212 ) + K(231 ) + K(232 ),

(44)

K1 = K(211 ) + K(222 ) + K(11 22 ) + K(212 ).

(45)

A suitable Gauss integration rule can be selected for each of these

matrices separately.

Remark. It was remarked in [16] that if the value b= 35 is used, see
Fig. 1 or 5, then the SRI can be used instead of the AS method, pro-

vided the Gauss points coincide with the points at which the strains
are exact. This observation was applied to the strain components
responsible for locking, but under-integration of other components
can be considered as well. However, in general, the SRI technique is
not fully equivalent to the AS technique.
Our tests yield the integration scheme shown in Table 2, which is
implemented in the shell element denoted as 9-SRI. As shown in this
table, the matrix for the drilling RC, Fdrill of Eq. (21), is also included
and fully integrated. Note that the reduced 2 2 rule is applied to
several components, but this does not lower the rank of the total
stiffness matrix.
4. Numerical tests
In this section we describe numerical tests of our new isoparametric 9-node elements presented in Table 3. All the elements are
analytically integrated over the thickness, while the integration over
the lamina is characterized in Table 3, where FI denotes the full integration (3 3 Gauss points), and SRI denotes the selective reduced
integration. Note that when we discuss accuracy of our elements, we
do not refer to the basic element, denoted by 9, which is included
only for comparison.

Author's personal copy

791

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796
Table 4
Reference 9-node shell elements with 5 dofs/node

8
19

Characteristics

Integration over lamina

Ref.

MITC9

AS

FI

S9R5

Stress/displ.
formulation

SRI

ADINA [35],
our input files
ABAQUS [36],
our input files
[7]

URI + stabilization

12

20

17
3

23

10

21
14
9

5
3

16

22

25

24
1

13

18
4

0.12

Element



11

15
6

0.24

Fig. 7. The patch of elements, E = 1. 0 106 ,  = 0. 25, h = 0. 001.

Fig. 6. Shapes of elements used in the eigenvalue analysis.

9
8
The reference solutions were obtained by using the elements
listed in Table 4, where URI denotes the uniform reduced integration.
Besides, for comparison, we also use the 4-node shell element
with 6 dofs/node and unrestricted rotations, designated as 4-EADG4.
Its membrane part is enhanced by the enhanced assumed displacement gradient (EADG) method, see [27], with the gradient of compatible displacements uc enhanced additively as follows:
.
, ) ,
F(, ) = [I + uc (, ) ] + H(
  
  
  
assumed

compatible

(46)

enhancing

where at a Gauss point g we compute



 
q  q3  1 jc
.
g =
Jc 1
Jc
.
H
q2  q4 
jg

(47)

Here Jc denotes a Jacobian matrix for the transformation between the

reference basis {ik } and the local natural basis {gk } at the element's
.
center (c). Besides, j = det J. The assumed matrix G depends on four
parameters qi . The transverse shear part of the 4-EADG4 element is
treated by the ANS method of [14], with the sampling points located
at the mid-sides of edges.
For the 4-EADG4 element, we use meshes with the same number
of nodes as for 9-node elements, so the number of elements is 4
times bigger.
Our elements were derived using the AceGen program developed
by Korelc [28], and were tested in the FEAP program developed by
Taylor [29]. The use of these programs is gratefully acknowledged.
4.1. Eigenvalues
The eigenvalues are computed for a single not supported element.
The data are as follows: E = 106 ,  = 0. 3 and the thickness h = 0. 1.
Two element's geometries shown in Fig. 6 are tested. The side of the
square element is equal to 1, while the distorted element is obtained
by shifting one corner node of the square element by the vector
(0. 5, 0. 5). The position of the other nodes is as yielded by the bilinear shape functions, i.e. edges are straight and non-corner nodes
are in the mid-side positions. All our 9-node elements, including
the basic element denoted as 9, have a correct number of zero
eigenvalues (6).

dY = n*0.01

dX = n*0.02
1

Fig. 8. One-element patch test. Diagonal shift of the internal node.

Table 5
Results of the membrane/bending one-element patch tests
Element

n=0

n=1

n=2

n=3

n=4

9
9-AS
9-SRI

p/p
p/p
p/p

p/
p/
p/

p/
p/
p/

p/
p/
p/

/
/
/

MITC9
S9R5
4-EADG4

p/p
p/p
p/p

/
p/
p/p

/
p/
p/p

N/N
p/
p/p

N/N
/
/p

4.2. Patch tests

4.2.1. Five-element patch test
The five-element patch of elements shown in Fig. 7 was proposed
in [30], and later included into the set of problems used to evaluate
the elements' accuracy in [31].
For shells, we have to verify whether the elements can represent
the constant membrane and bending strains.
For the membrane test, we assume:
ux = 0. 001(x + 12 y),

x = 0,

uy = 0. 001( 12 x + y),

y = 0,

uz = 0,

z = 0.

(48)

The values of ux , uy , uz are imposed at the external nodes, while

x , y , z remain unconstrained for all the nodes. The computed
values for the internal nodes must agree with these computed by
use of Eq. (48).
For the bending test, we assume:
ux = 0,
uy = 0,

x = 0. 0005(x + 2y),
y = 0. 0005(2x + y),

uz = 0. 0005(x2 + xy + y2 ),

z = 0.

(49)

Author's personal copy

792

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

membrane patch test

0.00014

analytical

0.00012

9-AS

9
9-SRI
MITC9

0.0001

ux

S9R5
8e-05
6e-05
4e-05
2e-05
0
0

3
n

bending patch test

1.2e-05

1e-05

uz

8e-06

6e-06

4e-06

2e-06

0
0

3
n

Fig. 9. One-element patch tests. Displacements at node 9 for the distortion n.

The values of ux , uy , uz , x , y are imposed at the external nodes,

and the drilling rotation z remains unconstrained at all the nodes.
The computed values for the internal nodes must agree with these
computed by use of Eq. (49).
All our shell elements pass the above described membrane and
bending patch tests.
Remark. Note that the value z = 0 in the membrane patch test is
implied by the linearized drill RC equation, for which the drilling
rotation is

z (x, y) =. 12 (ux,y uy,x ) = 0,

(50)

for ux and uy of Eq. (48). The drilling rotation at the boundary nodes
can be prescribed in several ways, and e.g. in [32] the patch test was

performed for the following three forms of boundary conditions for

the drilling rotation: (1) it was restrained at all boundary nodes, (2)
it was restrained at one boundary node, and (3) it remained free
(unconstrained) at all boundary nodes. The third form is the most
demanding and is used here for the 9-node elements.
4.2.2. One-element patch test
In this test only one 9-node element is used, but the boundary
conditions and other data are the same as in the five-element patch
test of Section 4.2.1.
The mesh distortion is caused by a shift of the internal node 9
along a diagonal to the position 9
, see Fig. 8, which yields a parabolic
internal line 79
5. The magnitude of the shift is: dx = 0. 02n and
dy = 0. 01n, where n is the distortion parameter.

Author's personal copy

793

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

4
2

13

8
y
1

14
10

12

9
5

15
11

Rectangular
x

6
2

Trapezoidal

45

45

10
Fig. 10. Two-element distortion test. E = 102 ,  = 0. 3, h = 10.

0.2

45

Parallelogram
Table 6
Two-element distortion test
Element

d=0

d=1

9
9-AS
9-SRI

0.75225
0.75225
0.75225

0.75225
0.75225
0.75225

MITC9
S9R5
4-EADG4

0.75225
0.72770
0.75225

0.74568
0.74040
0.73550

Exact

0.75225

Vertical displacement at node 2.

The displacements and rotations are defined in Eqs. (48) and (49).
The results are presented in Table 5, where p means that the test
is passed, i.e. the computed results agree with the analytical ones,
while minus that it failed. N means that the solution is not
available because the program was stopped. Besides, the selected
displacement components at node 9 are shown in Fig. 9a and b as
functions of the distortion parameter n.
All the tested elements pass both patch tests for node 9 located
at the center, i.e. for n = 0. But for the shifted node 9, when n > 0, the
situation is different.
1. The membrane patch test is passed by our elements for up to n=3.
For n=4, the angles between iso-parametric lines are too large and
a correct solution cannot be obtained. In [6], the elements failed
the patch test (weakly) when the element sides were parabolically
curved in-plane, so our elements perform better.
2. More demanding is the bending patch test, which for n > 0 is not
passed by any of the elements. However, from Fig. 9b, we see that
error is small for n  3. A similar behavior of the reference 
element is reported in [7].
The reference elements MITC9 and S9R5 were used in this test with 5
dofs/node, because the drilling rotation is included as a variable into
the analysis only when a boundary value is prescribed for it. This,
however, would be a violation of the boundary conditions accepted
for this test, see the discussion following Eq. (50).
Summarizing this test, we see that our elements perform similarly
as the S9R5 element, and better than the MITC9 element.
4.2.3. Two-element distortion test
The two-element mesh is shown in Fig. 10, and the tilt of the
boundary between the elements (line 597) is described by the
distortion parameter d. The position of the nodes is as produced
by the bi-linear shape functions, i.e. edges are straight and noncorner nodes are in the mid-side positions. This test was proposed
as a higher order patch test by [29], and we use the data suggested
therein. The bending is caused by two opposite forces P = 5.
The results are presented in Table 6, where for d = 0 the tilt of
the line 597 is 0 and the elements are rectangular, while for
d = 1 the tilt is 45 and the elements are trapezoidal. We see that
all our elements yield exact values, and are more accurate than the
reference elements.

6
Fig. 11. Straight cantilever beam. E = 1. 0 107 ,  = 0. 3, h = 0. 1.

Table 7
Straight cantilever beam
Element

Rectangular

Trapezoidal

Parallelogram

In-plane shear load (uy 10)

9
1.0703
9-AS
1.0748
9-SRI
1.0748

1.0606
1.0715
1.0715

1.0617
1.0749
1.0746

MITC9
S9R5
4-EADG4

1.0748
0.9344
1.0737

1.0646
0.9261
0.7803

1.0746
0.9854
1.0146

Ref.

1.0810

Out-of-plane shear load (uz 10)

9
4.2719
9-AS
4.2958
9-SRI
4.2958

4.2329
4.2812
4.2812

4.2397
4.2929
4.2929

MITC9
S9R5
4-EADG4

4.2955
4.3100
4.2850

4.2823
4.3110
4.2886

4.2930
4.3110
4.2886

Ref.

4.3210

Twisting by pair of forces (x 100)

9
3.0176
9-AS
3.0307
9-SRI
3.0277

3.0209
3.0334
3.0342

2.9836
3.0229
3.0212

MITC9
S9R5
4-EADG4

2.9128
3.0400
3.0313

2.9130
3.0400
3.0345

2.9033
3.0300
2.9150

Ref.

3.2080

Bending by moment (y 100)

9
3.5927
9-AS
3.5929
9-SRI
3.5929

3.5927
3.5930
3.5930

3.4612
3.5344
3.5340

MITC9
S9R5
4-EADG4

3.5930
3.5995
3.5919

3.5928
3.5994
3.5927

3.5178
3.6000
3.5923

Ref.

3.6000

Results for three meshes and four load cases.

4.3. Straight cantilever beam

In this test we check the element's ability to reproduce several
basic deformation modes for: (a) large aspect ratios (about 5), and
(b) distorted shapes, as proposed in [31]. The beam is clamped at
one end and external loads are applied at the other end, see Fig. 11.
The mesh consists of only six elements of three shapes: rectangular,
trapezoidal and parallelogram.

Author's personal copy

794

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

Py
x

16

44

44

Fig. 12. Geometry of slender cantilever, and in-plane shear load.

Table 8
Slender cantilever
Element

ux 102

uy

z 102

9
9-AS
9-SRI

2.9992
2.9993
2.9993

3.9987
3.9989
3.9989

5.9986
5.9988
5.9988

MITC9
S9R5
4-EADG4

2.9993
2.9999
2.9994

3.9989
4.0000
3.9991

5.9988
5.9997
5.9989

Ref.

3.0000

4.0000

6.0000

Results at the middle node of the tip.

Besides the types of loading proposed in [31], additionally, we

compute the out-of-plane bending by a unit moment. The reference
solution for this load case is calculated by the formulas from [33].
The obtained results are presented in Table 7, and they indicate
that the element's shape does not affect the results for 9-node elements much, differently than for 4-node elements. Besides, accuracy
of our elements is as follows:
for the in-plane load: similar as of MITC9, and better than of S9R5,
for the out-of-plane load: similar as of MITC9, and both worse than
of S9R5,
for the twist load: better than of MITC9, and similar as of S9R5,
for the bending by moment: better than of MITC9, and worse than
of S9R5.
Note, however, that the differences between results for particular
elements are small.

y
x
48
Fig. 13. Cook's membrane. E = 1,  =

Table 9
Cook's membrane
Element

11

22

9
9-AS
9-SRI

19.643
21.799
21.799

23.288
23.576
23.576

MITC9
S9R5
4-EADG4

22.208
26.540
21.037

23.613
23.980
23.014

Ref.

23.810

Vertical displacement at node A.

sym.

m
sy
.

4.4. In-plane bending of slender cantilever

4.5. Cook's membrane

This test was proposed in [34]. The trapezoidal membrane is
clamped at one end and subjected to a unit distributed in-plane shear
load on the other end, as shown in Fig. 13. This test involves shear

(u diap
hra
x ,u
gm
y ,
z -c
on
str
.)

m
.

sy

The in-plane bending caused by the force Py is used to evaluate

accuracy of the drilling rotation z yielded by the developed elements. The numerical values are compared with the reference analytical solution for the Bernoulli beam.
The cantilever is shown in Fig. 12, and the data are as follows: E =
106 ,  =0. 3, L=100, b=1, and the thickness h=1. The shell thickness
is perpendicular to the X0Y plane. One boundary is clamped, while
at the other the in-plane force Py = 1 is applied.
The cantilever is modelled by one layer of elements in the
y-direction, while in the x-direction, 100 elements are used to avoid
the questions of mesh convergence or aspect ratio.
The results of computations for the top node of the tip are presented in Table 8, and we see that all our 9-node elements perform
almost identically. The accuracy of the drilling rotation obtained by
our elements is very good, and similar to this by the reference elements. For the MITC9 element, we obtained the drilling rotation
using the RIGIDNODES command.

1
3 , h = 1.

Fig. 14. Pinched cylinder. E = 3 106 ,  = 0. 3, h = 3, R = 300, L = 300, P = 0. 25.

deformation, and the elements are skew. Two meshes are used; a
1 1-element mesh and a 2 2-element mesh.
The computed vertical displacement at node A is given in Table 9,
and we see that the results obtained by our elements are less accurate
than the solution by the MITC9 element. Note that the results for
the S9R5 element are over-estimated.

Author's personal copy

795

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796
Table 10
Pinched cylinder

Table 11
Pinched hemispherical shell with a hole

Element

22

55

Element

44

88

9
9-AS
9-SRI

0.0918
1.4535
1.4547

0.4701
1.8194
1.8197

9
9-AS
9-SRI

0.2066
9.3306
9.3250

2.4160
9.3473
9.3473

MITC9
S9R5

1.7979
1.8040

1.7548

MITC9
S9R5
4-EADG4

8.1762
9.3365
9.1566

8.5687
9.3513
9.3076

4-EADG4

1.3180
1.3870
1.3449
1.3855

Ref.

1.8249



Ref.

9.4000
2

Displacement (10 ).

Vertical displacement (105 ) under the force.

18

y
R = 10
Pin-plane

z
x

y
P=1
w=0

P=1
free

Pout-of-plane

Fig. 16. Twisted beam. E = 2. 9 107 ,  = 0. 22, h = 0. 0032. Length = 12. 0, width = 1. 1,
twist = 90 (from root to tip), P = 1 106 .

Fig. 15. Pinched hemispherical shell with hole. E = 6. 825 107 ,  = 0. 3, h = 0. 04.

4.6. Pinched cylinder

In this example, transverse shear locking is much greater than
the membrane locking, see the comments in [7].
A cylindrical shell is pinched by two concentrated loads, and its
ends are supported by rigid diaphragms. Because of the symmetries,
only one-eight of the cylinder is modelled, see Fig. 14, where also
the data are defined. The 2 2 and 5 5-element meshes are used.
The vertical displacement at the point where the force is applied
are presented in Table 10. We see that our elements are more accurate than the reference elements.
We note that in [9, p. 2407], Fig. 7, the SRI element shows very
slow mesh convergence, and for 17 nodes/side the error is about 20%.
In our computations, the 5 5 mesh involves only 11 nodes/side,
but the error for our 9-SRI element is < 1%. Hence, the opinion that
the SRI elements are characterized by slow mesh convergence is not
confirmed.

4.7. Pinched hemispherical shell with hole

A hemispherical shell with an 18 hole is loaded by two pairs of
equal but opposite external forces, applied in the plane z = 0, along
the 0X- and 0Y-axes, see Fig. 15, where the data are defined. The
shell undergoes an almost in-extensional deformation, and in this
example membrane locking can strongly manifest itself, see [7].
Due to a double symmetry, only a quarter of the shell is modelled,
and two meshes, of 4 4 and 8 8 elements, are used.
The displacement at the point where the load is applied and in
the direction of the load is shown in Table 11. The reference value is
taken from [31]. We see that our elements are more accurate than
the reference element MITC9.

Note that a good behavior of our 9-AS and 9-SRI elements is not
caused by additional drilling rotations, because:
1. The basic 9 element also has drilling rotations, but its response
is locked for both meshes.
2. In this example, we can obtain the solution for the drilling RC not
being enforced, as for  =0 the tangent matrix still is non-singular.
Then, we have only two tangent rotational parameters at Gauss
points, which corresponds to the formulation with 5 dofs/node.
For the 9 element and the 4 4 mesh, using  = 0 we obtain
0.2669, instead of 0.2066, i.e. the response is slightly softer.
We conclude that not the drilling rotations but the AS and SRI techniques improve the accuracy.

4.8. Twisted beam

This example was used in the set of tests of [31], and was also used
in [7], to illustrate the importance of accounting for the variation of
the Jacobian through the thickness.
The initial geometry of the beam is twisted, see Fig. 16. The beam
is clamped at one end, and loaded by a unit force at the other. The
force is applied either along the Z-axis (in-plane shear) or the Y-axis
(out-of-plane shear).
The mesh of 212-elements was used in computations. The shell
thickness h = 0. 0032, i.e. the shell is very thin, and the membrane
locking can strongly manifest itself.
The results are presented in Table 12; our elements turn out to be
more accurate for the out-of-plane load, while the MITC9 element is
more accurate for the in-plane load. The results for the S9R5 element
are greater than the reference values for both cases.

Author's personal copy

796

P. Panasz, K. Wisniewski / Finite Elements in Analysis and Design 44 (2008) 784 -- 796

Table 12
Twisted beam
Element

In-plane load

Out-of-plane load

9
9-AS
9-SRI

0.4754
5.2283
5.2280

0.1163
1.2935
1.2934

MITC9
S9R5
4-EADG4

5.2468
5.2683
5.2297
5.1888

1.2920
1.2958
1.3069
1.2861

Ref.

5.2560

1.2940



Displacement (103 ) in direction of load.

5. Final remarks
Two 9-node quadrilateral shell elements are derived and tested in
the paper, all based on Reissner's shell hypothesis, the Green strain
and the potential energy functional. The variational formulation includes the drilling rotation, so the elements have 6 dofs per node. To
eliminate the transverse shear and membrane locking as well as the
in-plane shear over-stiffening, either the two-level approximation
technique or the selective reduced integration are used.
From the computational point of view, the most novel aspect of
this paper is the proposed modification of the two-level approximation method, by treating the sampling and the numerical integration together, which results that six sampling points are replaced by
two sampling lines. This modification simplifies the derivation of the
element, and improves the efficiency of automatic differentiation.
The modified two-level approximation is applied to the strain,
the assumed strain method is implemented in the 9-AS element.
Besides, the selective reduced integration is implemented in the
9-SRI element. A characteristic feature of both our elements is that
the components in the ortho-normal basis at the element's center
are sampled and interpolated, which differs them from the MITC
family of elements, which uses the covariant strain components.
The developed shell elements are subjected to a range of benchmark tests, and compared with the results by two commercial
elements; more indicative are comparisons with the MITC9 element,
because it is based on a similar methodology.
Summarizing, these tests indicate that the methodology of sampling and interpolating the components in the ortho-normal basis
at the element's center certainly is not inferior to the methodology
based on the covariant strain components.
Acknowledgment
This research was partially supported by the Polish Committee
for Scientific Research (KBN) under Grant no. N501-290234.
References
[1] H. Stolarski, T. Belytschko, Membrane locking and reduced integration for
curved elements, J. Appl. Mech. ASME 49 (1982) 172176.
[2] H.C. Huang, E. Hinton, A nine node Lagrangian Mindlin plate element with
enhanced shear interpolation, Eng. Comput. 1 (1984) 369379.
[3] K.C. Park, Improved strain interpolation for curved C 0 elements, Int. J. Numer.
Methods Eng. 22 (1986) 281288.

[4] O.C. Zienkiewicz, R.L. Taylor, J.M. Too, Reduced integration technique in general
analysis of plates and shells, Int. J. Numer. Methods Eng. 3 (1971) 275290.
[5] T. Belytschko, J.S. Ong, W.K. Liu, A consistent control of spurious singular modes
in the 9-node Lagrangian element for the Laplace and Mindlin plate equations,
Comput. Methods Appl. Mech. Eng. 44 (1985) 269295.
[6] K.C. Park, G.M. Stanley, A curved C 0 shell element based on assumed naturalcoordinate strains, Trans. ASME 53 (1986) 278290.
[7] T. Belytschko, B.L. Wong, H. Stolarski, Assumed strain stabilization procedure
for the 9-node Lagrange shell element, Int. J. Numer. Methods Eng. 28 (1989)
385414.
[8] T. Belytschko, W.K. Liu, J.S. Ong, Mixed variational principles and stabilization
of spurious modes in the 9-node element, Comput. Methods Appl. Mech. Eng.
62 (1987) 275292.
[9] J. Jang, P.M. Pinsky, An assumed covariant strain based 9-node shell element,
Int. J. Numer. Methods Eng. 24 (1987) 23892411.
[10] R.H. MacNeal, A simple quadrilateral shell element, Comput. Struct. 8 (2) (1978)
175183.
[11] T.J.R. Hughes, T.E. Tezduyar, Finite elements based upon Mindlin plate theory
with particular reference to the four-node isoparametric element, J. Appl. Mech.
48 (1981) 587596.
[12] R.H. MacNeal, Derivation of element stiffness matrices by assumed strain
distributions, Nucl. Eng. Des. 70 (1982) 312.
[13] E.N. Dvorkin, K.-J. Bathe, A continuum mechanics based four-node shell element
for general nonlinear analysis, Eng. Comput. 1 (1984) 7788.
[14] K.-J. Bathe, E.N. Dvorkin, A four-node plate bending element based on
MindlinReissner plate theory and mixed interpolation, Int. J. Numer. Methods
Eng. 21 (1985) 367383.
[15] K.-J. Bathe, E.N. Dvorkin, A formulation of general shell elementsthe use of
mixed interpolations of tensorial components, Int. J. Numer. Methods Eng. 22
(1986) 697722.
[16] H.C. Huang, E. Hinton, A new nine node degenerated shell element with
enhanced membrane and shear interpolation, Int. J. Numer. Methods Eng. 22
(1986) 7392.
[17] L. Bucalem, K.-J. Bathe, Higher-order MITC general shell elements, Int. J. Numer.
Methods Eng. 36 (1993) 37293754.
[18] H.-Ch. Huang, Static and Dynamic Analyses of Plates and Shells, Springer, Berlin,
1989.
[19] D. Chapelle, K.J. Bathe, The Finite Element Analysis of ShellsFundamentals,
Springer, Berlin, 2003.
[20] K.-J. Bathe, F. Brezzi, M. Fortin, Mixed interpolated elements for
ReissnerMindlin plate, Int. J. Numer. Methods Eng. 28 (1989) 17871801.
[21] K. Wisniewski, E. Turska, Second order shell kinematics implied by rotation
constraint equation, J. Elasticity 67 (2002) 229246.
[22] J. Chroscielewski, J. Makowski, H. Stumpf, Genuinely resultant shell finite
elements accounting for geometric and material nonlinearity, Int. J. Numer.
Methods Eng. 35 (1992) 6394.
[23] J. Chroscielewski, W. Witkowski, Four-node semi-EAS element in six-field
nonlinear theory of shells, Int. J. Numer. Methods Eng. 68 (2006) 11371179.
[24] T.J.R. Hughes, F. Brezzi, On drilling degrees of freedom, Comput. Methods Appl.
Mech. Eng. 72 (1989) 105121.
[25] J.C. Simo, D.D. Fox, T.J.R. Hughes, Formulations of finite elasticity with
independent rotations, Int. J. Numer. Methods Eng. 95 (1992) 227288.
[26] K. Wisniewski, E. Turska, Kinematics of finite rotation shells with in-plane twist
parameter, Comput. Methods Appl. Mech. Eng. 190 (810) (2000) 11171135.
[27] J.C. Simo, F. Armero, Geometrically non-linear enhanced strain mixed methods
and the method of incompatible modes, Int. J. Numer. Methods Eng. 33 (1992)
14131449.
[28] J. Korelc, Multi-language and multi-environment generation of nonlinear finite
element codes, Eng. Comput. 18 (2002) 312327.
[29] O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method. Vol. 1. Basic
Formulation and Linear Problems, fourth ed., McGraw-Hill, New York, 1989.
[30] J. Robinson, S. Blackham, An Evaluation of Lower Order Membranes as Contained
in MSC/NASTRAN, ASA and PAFEC FEM Systems, Robinson and Associates,
Dorset, England, 1979.
[31] R.H. MacNeal, R.L. Harder, A proposed standard set of problems to test finite
element accuracy, Finite Elem. Anal. Des. 1 (1985) 320.
[32] K. Wisniewski, E. Turska, Enhanced Allman quadrilateral for finite drilling
rotations, Comput. Methods Appl. Mech. Eng. 195 (4447) (2006) 60866109.
[33] C.Y. Warren, Roark's Formulas for Stress and Strain, sixth ed., McGraw-Hill,
New York, 1989.
[34] R.D. Cook, Improved two dimensional finite element, J. Struct. Div. ASCE 100
(1976) 18511863.
[35] ADINA. Ver.8.3.1.
[36] ABAQUS. Ver.6.6-2.