A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 1

A nine-node displacement-based nite element for

Reissner-Mindlin plates based on an improved formulation of the
NIPE approach

G. Castellazzi
1
, P. Krysl
2
1
University of Bologna, Viale Risorgimento 2, 40136 Bologna, Italy [email protected]
2
University of California, San Diego, 9500 Gilman Dr #0085, La Jolla, CA 92093-0085, [email protected]
SUMMARY
We consider a reformulation of the Nodally Integrated Plate Element formulation (NIPE) and present a new
9-node element conguration.
The nodally integrated plate element formulation is an assumed strain nite element technique for shear
deformable plates that enforce weakly the balance and the kinematic equations. The strain-displacement
operators are derived via nodal integration satisfying a priori the kinematic weighted residual statement.
The new 9-node element uses bi-quadratic interpolations of the transverse displacement and rotations and
is computed by means of a nine-node quadrature rule. The present work analyzes the NIPE technique,
including the new element, by testing thoughtfully the sensitivity of the elements to severe geometry
distortions. We propose an improvement that confers robustness to all element shapes developed by the
NIPE formulation.
A brief review of the improved triangular and quadrangular NIPE elements is reported for elastic plate
analyses. A few challenging benchmark carried out on extreme distorted meshes illustrate the performance
of the introduced NIP-Q9 element. We detail that the new NIPE formulation confers insensitivity to extreme
distortions for the quadratic quadrilateral element and allows to solve for severe distortion with the NIPE
element family. Copyright 2011 John Wiley & Sons, Ltd.
Received . . .
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
Int. J. Numer. Meth. Engng 2011; 00:233
Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nme
KEY WORDS: Reissner-Mindlin plate, nite element, shear locking, assumed strain, weighted residual,
nodal quadrature, distortion insensitivity, NIPE element
1. INTRODUCTION
As well known, standard nite element formulation, developed for plate structures, suffers from
shear locking in the thin plate limit. Moreover things turn for the worse when geometric distortions
arise, see for instance [1] and the reference therein.
In this work a nine-node locking-free nite element, based on the Nodal Integrated Plate Element
formulation, is presented and employed for linear nite element analyses.
The NIPE formulation is an assumed-strain nite element technique derived from a weighted
residual statement that weakly enforces both the balance equation and the kinematic equation. As
shown in [2] the NIPE formulation has a number of desirable attributes: the degrees of freedom
are only the primitive displacement variables, it is free of shear-locking, provide nodal quantities
without needing any post-processing operation. Despite these nice features the NIPE elements,
similarly to classical nite element techniques, it is sensitive to a severe geometrical distortions
of the nite element mesh.
In general simplex nite elements give reliable solutions when the discretizing mesh is regular, but
exhibit poor performance when the element geometry is distorted. Also it is a common knowledge
that the interpolation precision of quadrilateral nite elements deteriorates if the element geometry
is not a parallelogram.
In this sense the particular effect of this loss of performance is commonly associated with a
progressive increasing of the stiffness, resulting in a sort of locking for the elements response, that
causes the inability of the element to nd a good approximation of the solution [1][3].
Among effective examples of low order nite element formulations weakly sensitive to geometry

Correspondence to: University of Bologna, Viale Risorgimento 2, 40136 Bologna, Italy

[email protected]

Please ensure that you use the most up to date class le, available from the NME Home Page at
http://www3.interscience.wiley.com/journal/1430/home
Copyright 2011 John Wiley & Sons, Ltd.
Prepared using nmeauth.cls [Version: 2010/05/13 v3.00]
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 3
distortions mixed/hybrid stress approach, see for instance [4][5] and the references therein, and
Smoothed Finite Element Method [6] can be listed as examples. The rst owes its property to
the particular description of the transverse displacement approximation that is linked to the nodal
rotations through higher order interpolation functions plus the co-ordinate invariant description of
the assumed stress approximation. The second avoids the evaluation of the derivatives of the mesh-
free shape functions at the nodes, where they vanish, and thus eliminates defective modes.
Nevertheless, in some cases, when severe distortions arise, standard elements fail completely the
solution because of when the nite element is highly distorted in the physical domain, the tangent
vectors to the parametric curves may become almost linearly dependent. The Jacobian matrix
invertability is thus compromised.
Given the above reported desirable attributes of the NIPE formulation it is of interest to ascertain
how to confer upon this technique the insensitivity to geometrical distortion of the nite elements.
Therefore, the NIPE technique is here analyzed from this viewpoint, in order identify the source of
the sensitivity to the shape of the elements, and propose an improvement of the NIPE formulation.
Overview
The paper is organized as follows: in Section 2 the governing equations and the basic notation
concerning the plate structures are given. In Section 3 the nodally integrated plate element, NIPE
procedure is presented for the new 9-node element. The procedure is then analyzed in Section 4 with
respect to geometry distortion along with the proposed remedy which consist in the reformulation
of the present averaging technique. Finally in Section 5 some benchmark problems are employed to
test the robustness of the improved NIPE formulation. Some concluding remarks end the paper.
2. GOVERNING EQUATIONS AND BASIC NOTATION
Consider a plate referred to the following Cartesian coordinate frame:
V = {(x, y, z) R
3
|z [t/2; t/2], (x, y) A R
2
} , (1)
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
4 G. CASTELLAZZI, P. KRYSL
(a) (b)
Figure 1. Reference geometry with the indication of positive direction
where A is the area of the plate, and t is the thickness of the plate. The boundary of the plate
is C = A. The rst-order shear deformable theory, often called ReissnerMindlin theory, is
employed. Thus the 3-D displacement is denoted u and its Cartesian components are
u
x
= z
y
, u
y
= z
x
, u
z
= w , (2)
where
x
=
x
(x, y) and
y
=
y
(x, y) are the rotations of the transverse normal about the
Cartesian axes x and y, and w = w(x, y) is the deection. The functions
x
,
y
, and w are the
unknown elds in the plate bending problem, see Figure 1(a) and we can conveniently express the
three-dimensional displacement vector in terms of the mixed-component vector of the generalized
displacements u
[ u] = [w,
x
,
y
]
T
, (3)
as
u = S u , (4)
where we introduce the shifter
S =
_

_
0 0 z
0 z 0
1 0 0
_

_
. (5)
.
The compatibility equations can be written as follows:

b
= zD
b
u ,
b
= D
b
u ,
s
= D
s
u , (6)
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 5
where the vectors u,
b
,
b
and
s
collect, respectively, the generalized displacements, the bending
strains, the curvatures, and the shear strains. Here the operators D
b
and D
s
are, respectively, the
bending and the shear symmetric gradient operator, dened as
D
b
=
_

_
0 0 /x
0 /y 0
0 /x /y
_

_
, D
s
=
_

_
/x 0 1
/y 1 0
_

_
, (7)
The bending and the shear stresses are generated by the constitutive equation

b
= C
b

b
= zC
b

b
,
s
= C
s

s
= kG
s
. (8)
where C
b
and C
s
are the matrices of bending and transverse shear moduli that describe a generic
anisotropic material. The balance equation may be written using the above operators as
zD
b
T

b
+D
sT

s
+b = 0 . (9)
We assume [b] = [0, 0, b
z
(x, y)]
T
for simplicity. Integrating the balance equation through the
thickness of the plate obtains
D
b
T
m+D
sT
s + tb = 0 , (10)
and the resultants
m =
_
t/2
t/2
z
b
dz =
t
3
12
C
b

b
, s =
_
t/2
t/2

s
dz = tkG
s
. (11)
whose components are depicted in Figure 1(b).
3. SUMMARY OF THE NIPE FORMULATION
In this section we briey recall the derivation of the weak form of the Reissner-Mindlin governing
equations by using the original NIPE technique.
In Reference [2] the NIPE formulation was derived from a weighted residual statement that weakly
enforced both the balance equation and the kinematic equation.
Here the formulation of the NIPE technique will be elaborated in terms of stationarity of a multi-eld
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
6 G. CASTELLAZZI, P. KRYSL
functionals, as weak formulation of the model problem described above [3]. In particular, consider
the following multi-eld variational form for a thick plate problem:
R( u, ) =
_
A
_

1
2

T
C +S
T
( D u) b
T
u
_
dA +
_
t

t
T
udC = 0, (12)
where =
_

b
,
s

T
are the assumed strain, S = [m, s]
T
collects the resultants and
C =
_

_
C
b
0
0 C
s
_

_
.
Variation of the (12) provides the formulation of the associated saddle-point problem: variation with
respect to the displacement u gives the equilibrium equation
R
BC
= R( u, ) u =
_
A
_
(D u)
T
S + t u
T
b
_
dA = 0, (13)
and, variation with respect to gives the discrete kinematic equation
R
K
= R( u, ) =
_
A
_

T
C( )

dA = 0. (14)
Here we will consider methods where the strains are assumed in the form
= D u, = D u (15)
where
D =
_

_
D
b
0
0 D
s
_

_
,
is the (as yet undetermined) assumed strain-displacement operator. Hence we can equivalently write
the weak problem of nding u, that satisfy the balance residual
R
BC
=
_
A
_
(D u)
T
S + t u
T
b
_
dA +R
C
= 0 . (16)
and the kinematic residual
R
K
=
_
A

T
C( D u) dA = 0 , (17)
that represent the balance equation and the kinematic equation residuals respectively, where R
C
represents the natural boundary condition residual,
R
C
=
_
C,1
u
1
(s
nz
s
nz
) dC +
_
C,2
u
2
(m
nn
m
nn
) dC +
_
C,3
u
3
(m
ns
m
ns
) dC = 0 ,
(18)
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 7
Here u is the test function (generalized displacement, as indicated by the tilda), which is assumed
to vanish along the portions of boundary where essential boundary conditions are prescribed.
[ u]
i
= 0 where i
th
component of the generalized displacement is prescribed on C. (19)
Now we turn our attention to the kinematic residual (17). we intend to use this residual equation as
the denition of the assumed gradient operator D. Highlighting in the kinematic equation residual
(17) the curvature and the shear strain components we obtain
R
Kb
=
_
A

bT
C
b
_

b
D
b
u
_
dA = 0 , R
Ks
=
_
A

sT
C
s
(
s
D
s
u) dA = 0. (20)
The test functions
b
and
s
have the meaning of generalized strains (curvatures and shear
deformations). The actual (generalized) strains
b
,
s
and the test strains
b
,
s

b
= D
b
u ,
b
= D
b
u ,
s
= D
s
u ,
s
= D
s
u , (21)
Substituting for the stress resultants from the constitutive equations and for the generalized strains
from (21), the balance residual results
R
BC
=
_
A
_

_
D
b
u
_
T
t
3
12
C
b
D
b
u (D
s
u)
T
tkGD
s
u + t u
T
b
_
dA +R
C
= 0 , (22)
and the kinematic residuals
R
Kb
=
_
A
(D
b
u)
T
C
b
_
D
b
u D
b
u
_
dA = 0 ,
R
Ks
=
_
A
(D
s
u)
T
C
s
_
D
s
u D
s
u
_
dA = 0 .
(23)
The resulting formulation is therefore based on the generalized displacement eld only. The
assumed-strain operators D
b
, D
s
are then derived from the kinematic residuals (23), see [2] . The
process will be the same for both D
b
and D
s
. Consequently we will describe the construction of
D
b
, with the adjustments necessary to construct D
s
implied. It is interesting to note that Equation
(21) recalls the B-bar technique [7] in the denition of the actual strains. Nevertheless a substantial
difference with respect to this technique lie in the denition of the derivation of the assumed
strain operator that, here, is derived directly from the equations (23) by means of simple algebraic
manipulation of the discretized expression.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
8 G. CASTELLAZZI, P. KRYSL
3.1. Patch averaged strain matrix
The kinematic residual statement (23) is discretized with the usual nite element approximation
u =

I
N
I
u
I
, u =

I
N
I
u
I
.
We shall assume, as is usual, that the nite element nodes carry the degrees of freedom, w
i
,
xi
,
yi
associated with node i. It is well known that in the limit of the thickness approaching zero the
deection w and the rotations
x
and
y
become tied together by the Kirchhoff constraint. In
the discrete case, this is equivalent to 2N
q
algebraic constraints between the 3N
n
parameters
w
i
,
xi
,
yi
, where N
q
is the number of quadrature points and N
n
is the number of nodes.
Quadrature at integration points which correspond one-to-one to the nodes (that is N
q
= N
n
) may
yield exactly the right number of constraints, provided these are linearly independent.
The discrete kinematic weighted residual equation becomes

I,J
u
T
I
__
A
B
b
I
T
D
b
(B
b
J
B
b
J
) dA
_
u
J
= 0 ,
(24)
where we have introduced matrices B
b
J
= D
b
(N
J
), and the nodal strain-displacement matrices B
b
I
,
as not yet unspecied, that are used to produce the assumed strains as
=

I
B
b
I
u
I
. (25)
Evidently, the u
I
are entirely arbitrary, and furthermore we would not wish the matrices B
b
I
to
depend on the solution u
J
. Consequently, the vanishing of the integrals
_
A
B
b
I
T
C
b
(B
b
J
B
b
J
) dA = 0 , J , (26)
is implied for any xed I. Refer to Figure 2 for an illustration that provide a better interpretation
of the indices. For the index I xed, we note that in order to formulate strictly local operations,
we should only consider the strain-displacement matrices B
b
J
dened within the elements e
r
, r =
1, ..., 4 connected to node I. The index J then ranges over J = J
q
, q = 1, ..., 8 and J = I. Therefore,
we replace J in (26) with the limited range J nodes(elems(I)); the term nodes(e) refers to the
nodes of the element e, the term elems(I) refers to the elements connected to the node I, and the
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 9
I
I
J
e
1
J
2
J
3 J
4
J
5
J
8
J
7
J
6
1
e
2
e
3
e
4
Figure 2. Nodal patch illustrated on a four-node quadrilateral mesh
term nodes(elems(I)) refers to the union of the nodes of the elements connected to the node I.
Thus, we will work with the integrals
_
A
B
b
I
T
C
b
(B
b
J
B
b
J
) dA = 0 , J nodes(elems(I)) , (27)
for any xed node I.
Next, numerical quadrature at the nodes is introduced in the area integrals. For the quadrilateral
element types that we consider in this section, the nodal quadrature rules are easily specied as
_
A
()(x) dA

Knodes(e)
()(x
K
)J(x
K
)W
K
, (28)
where e ranges over all the elements in the mesh, K ranges over all the quadrature points in the
element (the quadrature points coincide with the nodes), x
K
is the location of the quadrature point
(node), W
K
is the weight of the quadrature point and J(x
K
) is the Jacobian of the isoparametric
mapping
J(x
K
) = det[J],
and, Table I denes the quadrature rules for the 9-node element formulations proposed in this paper.
Refer to [2] for the quadrature rules used with other NIPE elements.
In the following developments, we assume that I is one particular xed node. Using numerical
quadrature, the integral (27) is replaced with this double sum

Knodes(e)
J(x
K
)W
K
B
b
I
T
C
b
(x
K
)
_
B
b
J
B
b
J
(x
K
)
_
= 0 , J nodes(elems(I)) , (29)
where e ranges over all the elements in the mesh. The sum over all the elements implies a mesh-
wide operation. Quite reasonably, we will try to avoid this. Reversing the order in which we apply
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
10 G. CASTELLAZZI, P. KRYSL
the double summation, with K ranging over all the nodes in the mesh, we obtain

eelems(K)
J(x
K
)W
K
B
b
I
T
C
b
(x
K
)
_
B
b
J
B
b
J
(x
K
)
_
= 0 , J nodes(elems(I)) . (30)
Finally, for efciency reasons the integration points should be independent. Therefore, we will
require the vanishing of each of the terms in the above sum separately

eelems(K)
J(x
K
)W
K
B
b
I
T
C
b
(x
K
)
_
B
b
J
B
b
J
(x
K
)
_
= 0 , J nodes(elems(I)) . (31)
for all integration points K. The matrices B
b
J
(x
K
) and the Jacobian J(x
K
) are multi-valued at
node K, depending on which element these quantities are evaluated at the node K. The constitutive
matrix C
b
(x
K
) is assumed to be single-valued at x
K
. This would not be the case for points x
K
located at a multi-material interface, and such possibility is discussed in [2].
Note that B
b
I
and B
b
J
are element-independent. Therefore, by rewriting (31) as

eelems(K)
J(x
K
)W
K
B
b
I
T
C
b
(x
K
)
_
B
b
J
B
b
J
(x
K
)
_
=
= B
b
I
T
C
b
(x
K
)

eelems(K)
J(x
K
)W
K
_
B
b
J
B
b
J
(x
K
)
_
=
= B
b
I
T
C
b
(x
K
)
_
_
B
b
J

eelems(K)
J(x
K
)W
K

eelems(K)
J(x
K
)W
K
B
b
J
(x
K
)
_
_
= 0 ,
(32)
we can conclude that the expression in the brackets must vanish, yielding
B
b
J

eelems(K)
J(x
K
)W
K

eelems(K)
J(x
K
)W
K
B
b
J
(x
K
) = 0 .
(33)
This nally leads to the denition of the assumed-strain nodal matrix as a weighted average of the
elemental strain-displacement matrices
B
b
J
=

eelems(K)
J(x
K
)W
K
B
b
J
(x
K
)

eelems(K)
J(x
K
)W
K
. (34)
Thus, constructing the nodal strain-displacement matrices as averages of the strain-displacement
matrices from the connected elements will satisfy the kinematic residual statement, enabling nodal
quadrature in the process.
The size of B
b
J
depends by the numbers of nodes involved in the construction of the patch. The
structure of B
b
J
has the same non-zero structure of the element-wise strain-displacement matrix. In
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 11
Table I. Nodal quadrature rule for the nine-node quadrilateral.
Type Parametric coordinates Weights W
K
Nine-node quadrilateral - NIP-Q9 :
1
9
, 2 :
4
9
, :
16
9
particular the assumed-strain matrix B
b
J
may be constructed from assumed-strain basis functions
gradients as discussed in detail in [8]. In general, it will be illuminating to dene the assumed
gradient operator as
[gradN
J
(x
K
)] =

eelems(K)
J(x
K
)W
K
[gradN
J
(x
K
)]

eelems(K)
J(x
K
)W
K
. (35)
where
[gradN
J
(x
K
)] =
_
N
J
(x
K
)
x
,
N
J
(x
K
)
y
_
in two spatial dimensions. Equation (14) suggests that the NIPE formulation my also be referred to
as the assumed-basis function gradient technique.
4. ANALYSIS OF THE ORIGINAL IMPLEMENTATION AND PROPOSED IMPROVEMENT
Parametric nite elements can lose their performance when distortions of the element geometry
arise. Geometry distortions cause global coordinates to be non-linear functions of local coordinates
when described through the coordinate transformation. This transformation is ruled by the inverse
of the Jacobian matrix.
Focusing on simplex elements such as triangular 3-node or quadrangular 4-node the offending
features of poorly shaped element are small interior angles (close to zero). This conguration
cause the condition number of the nite element matrices to abnormally increase. Moreover large
interior angles (close to ) degrade the interpolation accuracy causing an erroneous computation of
the stiffness matrix resulting in an overestimate the stiffness matrix and thus stiffness properties.
The elements show in this sense a sort of locking phenomenon. Increasing the interpolation order
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
12 G. CASTELLAZZI, P. KRYSL
describing the map from the standard shape to the physical shape the above distortion scenario
widens to include quadratic, quasi-cubic and extreme distortions [9]. In the latter case higher order
nite elements can produce spatial arrangements of the nodes that lead to locally inverted geometries
with negative Jacobians.
Jacobian matrices that are close to being singular (Jacobian determined being close to zero), or that
correspond to inverted geometries (Jacobian negative) are generally to be avoided in classical nite
element methods.
Consider now the effect of geometrical distortion on the original NIPE formulation. As evident
from Equation (35), the technique requires the computation of the gradients of the isoparametric
basis functions. These are evaluated from the gradients of the basis functions with respect to the
parametric coordinates (as indicated by the () subscript) [gradN
J
((x
K
))] using the well-known
formula
[grad
(x)
N
J
(x
K
)] = [grad
()
N
J
((x
K
))][J(x
K
)]
1
(36)
where we may note the use of the inverse of the Jacobian matrix, dened in three dimensions as
[J] =
_

_
x
1
x
2
y
1
y
2
_

_
The Jacobian matrix expresses the mapping from the parametric coordinates to the physical
Cartesian coordinate system, and consequently its columns are tangent vectors to the parametric
curves.
The key to the proposed improvement of the NIPE formulation lies in the analytical formula for the
inverse of a square matrix that uses the adjugate matrix [10, 11]. The cofactor expansion (Laplace
formula) for the determinant of a square matrix [J] yields the expression
[J]adj[J] = det[J][Id] (37)
where [Id] is the identity, and adj[J] is the adjugate matrix of [J] whose elements are dened as
adj[J]
km
= (1)
k+m
minor
mk
[J]
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 13
The adjugate matrix consists of the elements of the transpose of the cofactor matrix of [J]; the
numbers minor
mk
[J] are the minors of the k, m elements of [J] (i.e. the determinants of the
submatrices of [J] obtained by deleting row k and column m). When the matrix [J] is invertible,
J = det[J] = 0 and it follows that the inverse may be obtained from (37) as
[J]
1
=
1
det[J]
adj[J] = J
1
adj[J]
Let us assume for the moment that [J] is invertible, and substitute (4) into (36), and then the resulting
expression into (35) for the assumed-strain basis function gradient. The formula
[gradN
J
(x
K
)] =

eelems(K)
J(x
K
)W
K
[gradN
J
(x
K
)]

eelems(K)
J(x
K
)W
K
, (38)
=

eelems(K)
J(x
K
)W
K
[gradN
J
((x
K
)
K
)][J(x
K
)]
1

eelems(K)
J(x
K
)W
K
(39)
=

eelems(K)
J(x
K
)W
K
[gradN
J
(
K
)]J
1
(x
K
)adj[J(x
K
)]

eelems(K)
J(x
K
)W
K
(40)
=

eelems(K)
(x
K
)W
K
[gradN
J
(
K
)]adj[J]

eelems(K)
J(x
K
)W
K
. (41)
Simplifying J(x
K
) and J
1
(x
K
) in the sum in the numerator we nally obtain
[gradN
J
(x
K
)] =

eelems(K)
(x
K
)W
K
[gradN
J
(
K
)]adj[J]

eelems(K)
J(x
K
)W
K
. (42)
The new expression (42) for the assumed-gradient of the basis function can now be used to
evaluate Equation 34 robustly for highly distorted elements. This new formulation of B
b
J
and B
s
J
confers considerable improvement upon the NIPE methodology: we may now safely relinquish
the requirement that the Jacobian matrix be invertible. The adjugate matrix always exists, and its
computation presents no numerical problems. The only condition required for the application of
(42) is that the denominator must be different from zero

eelems(K)
J(x
K
)W
K
= 0 (43)
The meaning is evident: Equation(43) is the nodal-quadrature approximation of the integral
_

dV
over the patch of elements around node J. A nonzero value indicates that node J has nonzero area
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
14 G. CASTELLAZZI, P. KRYSL
associated with it. As long as this condition is satised, the expression for the assumed-strain basis
function gradient (42) is well dened, and can be evaluated robustly.
5. NUMERICAL STUDY OF THE REFORMULATED NIPE TECHNIQUE
In this section, we give the numerical results of some simple representative plate bending problems
using the improved version of the NIPE element formulation. We consider two linear problems for
plates with square and circular geometries. For these benchmarks the reference solutions are known
[12, 13, 14, 15]. Numerical comparisons are carried out using the MITC4 and MITC9 elements
[16, 17, 18] and the Q4-SRI element [7, 19]
5.1. Test 1a: Square plate with uniform load: convergence
A square plate of side length L, under uniformly distributed transverse load b
z
= 1, is herein
considered for clamped and simply supported boundary conditions of hard (SS2) type [19]. The
analysis is carried out for two different aspect ratio: L/t = 10 and L/t = 1000. Regular mesh
obtained by four uniform renements of the starting mesh shown in Figures 3(a) and 4(a) are
used. The following properties are assumed: E = 10.92, = 0.3 and k =
5
6
. For symmetry, only
one quarter of the plate is modeled. As regards the clamped condition, Figure 3(b) presents the
displacement results at the centre of the plate.
The NIP-Q9 plate nite element gives uniform optimal convergence curves for all the cases of
thickness t considered. The MITC4 element is used to compare the apparent ultra-convergence ratio
of the NIP-Q4 element. In fact the convergence tends, after some further renements, to an order
equal to 2 similarly to the MITC4 element [2]. As regards the simple supported boundary condition
Figure 4(b) presents the displacement results at the centre of the plate.
The NIP-Q9 plate nite element gives also uniform optimal (almost ideal) convergence curves for
t = 10 and tends to the reference solution with a slightly slower ratio for t = 1000. The MITC9
element is used for this example to compare the convergence ratio of the NIP-Q9 element.
As it can be noted by comparing this results with those reported in [2] the introduced improvement
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 15
x
y
L
/
2
L/2
x
w=0, =0, =0
y
x
w
=
0
,
=
0
,
=
0
y
(a)
10
1
10
0
10
1
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
log(element size)
l
o
g
(
e
r
r
o
r

%
)


NIPQ4
NIPQ9
MITC4
MITC9
10
1
10
0
10
1
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
log(element size)
l
o
g
(
e
r
r
o
r

%
)


NIPQ4
NIPQ9
MITC4
MITC9
clamped L/h = 10 clamped L/h = 1000
1
4
1
2
1
4
1
2
(b)
Figure 3. Test 1a - Clamped square plate uniformly loaded: convergence of the transverse displacement at
the centre using regular meshes for L/t = 10 and L/t = 1000.
x
y
y
w=0, =0 ( ) SS2
x
w
=
0
,
=
0
(
)
S
S
2
(a)
10
1
10
0
10
1
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
log(element size)
l
o
g
(
e
r
r
o
r

%
)


NIPQ4
NIPQ9
MITC4
MITC9
10
1
10
0
10
1
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
log(element size)
l
o
g
(
e
r
r
o
r

%
)


NIPQ4
NIPQ9
MITC4
MITC9
SS2 L/h = 1000 SS2 L/h = 10
1
4
1
2
1
4
1
2
(b)
Figure 4. Test 1a - Simply supported (SS2) square plate uniformly loaded: convergence of the transverse
displacement at the centre using regular meshes for L/t = 10 and L/t = 1000.
of the NIPE technique does not modify the excellent performance of the NIPE elements when
geometry is regular.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
16 G. CASTELLAZZI, P. KRYSL
x
y
y
w=0, =0 ( ) SS2 j
c
d
a
b
x
w
=
0
,
=
0
(
)
S
S
2
j
(a)
x
y
y
w=0, =0 ( ) SS2
( x , y )
x
w
=
0
,
=
0
(
)
S
S
2
c c
(b)
x
y
y
w=0, =0 ( ) SS2
x
w
=
0
,
=
0
(
)
S
S
2
f
e
( x , y ) c c
(c)
Figure 5. Square plate under uniform load: SS2 simply supported plate with the meaning of the distortion
parameters used in the test: (a) Test 1b, (b) Test 2a and (c) Test 2b.
5.2. Test 1b: Square plate with uniform load: distortion sensitivity
The effect of mesh distortion is rstly studied on the simply supported square plate with uniform
load (Youngs modulus E = 30 10
3
, Poisson ratio = 0.3, side length L = 10, thickness t =
L/100, transverse uniform load b
z
= 0.4 all data in consistent units) [20]. Six parameters sets
are used (see Table II, and Figure 5(a)), only one quadrant is modeled with appropriate symmetry
boundary conditions. The comparison is carried out using the elements of the NIPE family with the
Q4-SRI element that is known to be fairly insensitive to distortions.
Figure 6 shows the results as the plot of normalized error in the center deection [(reference-
computed)/reference] for the six distortion parameter sets (set 1 gives the undistorted mesh).
The poorer elements NIP-T3 and NIP-Q4 are comparatively more sensitive to the distortion of
the mesh. The quadratic NIP-T7 and the new NIP-Q9 elements show the best performance on
distorted meshes. Note the excellent behavior of the NIP-Q9 element as also illustrated by Figure
7. Moreover, by comparing this results with those reported in [2], the reader may note that the
introduced improvement of the NIPE technique does not modify the excellent performance of the
NIPE elements also when geometry is lightly to moderately distorted.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 17
Table II. Test 1b - Distortion parameters for the simply supported square plate study.
Set a; b; c; d
1 0.5; 0.5; 0.5; 0.5
2 0.25; 0.85; 0.75; 0.25
3 0.95; 0.25; 0.25; 0.04
4 0.05; 0.85; 0.05; 0.84
5 0.15; 0.25; 0.05; 0.7
6 0.85; 0.05; 0.98; 0.98
1 2 3 4 5 6
0.1
0.05
0
0.05
0.1
0.15
Distortion parameter set [ND]
N
o
r
m
a
l
i
z
e
d

e
r
r
o
r

o
f

c
e
n
t
e
r

d
e
f
l
e
c
t
i
o
n

[
N
D
]


NIPQ9
NIPQ4
NIPT7
NIPT3
Q4SRI
Q4SRI
NIPT7
NIPQ4
NIPQ9
NIPT3
Figure 6. Test 1b - Normalized error in the center deection for the simply supported square plate with a
uniform load.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
18 G. CASTELLAZZI, P. KRYSL
Figure 7. Test 1b - NIP-Q9 element: distorted meshes for the simply supported square plate with a uniform
load.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 19
5.3. Test 2a - Square plate with uniform load: severe distortion test for non-convex shape element
The test aims at check the effectiveness of the introduced improved formulation, checking the
ability of quadrilateral NIPE elements to resist problematic geometry such as non-convex shapes. In
particular, the effect of severe mesh distortion is studied on the simply supported square plate with
uniform load (Youngs modulus E = 30 10
3
, Poisson ratio = 0.3, side length L = 10, thickness
t = L/10000, transverse uniform load b
z
= 0.4 all data in consistent units) [20]. Two parameter
sets are used to generate the severe distortion, see Figure 5(b). The distortion progressively increases
by moving the central node of the mesh by multiplying its coordinates by the parameters and as
illustrated in Figure 5(b).
Only one quadrant is modeled with appropriate symmetry boundary conditions. Figures 8
and 9 show the results as the plot of normalized error in the center deection [(reference-
computed)/reference] for NIP-Q4 and for NIP-Q9 elements respectively. The values of parameters
and are reported below each plot.
Note that even though the mesh posses some not convex elements (with pointwise negative Jacobian
when internal angles are > ) these elements contribute to the stiffness matrix, see Figures 9(c), 9(f)
and 9(i) for NIP-Q4 element and Figures 9(c), 9(f) and 9(i) for NIP-Q9 element. The adjugate matrix
is not zero for these elements, and therefore each element contributes to the assumed-strain basis
function gradients, and therefore to the patchwise stiffness matrix, and hence also to the overall
global stiffness matrix.
Figure 10 shows the bending stress m
x
as a plot over the square plate calculated at the element
nodes using NIP-Q4 elements. Normalized value of the maximum moment is reported above each
plot (only one value owing to the symmetry of the problem) [13]. Note that these results are not post-
processed (brown scale map eld): the nodal moments are computed directly using the curvatures
from Equation (6). Except for the solutions obtained for the undistorted conguration, small
oscillations in the predicted stress are observed (colored contour lines are used to highlight values
transitions). Nevertheless the resulting plot of the recovered stress quantity appear to be smooth over
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
20 G. CASTELLAZZI, P. KRYSL
the whole domain also near the area described by element affected by extreme distortions or with
non-convex shapes.
Figure 11 shows the dependence of the condition number of the stiffness matrix on the number of
equations. Each line represents the computation of the condition number for a uniform renement
starting from an imposed distortion. Red lines and dashed black lines are used to describe the old
and the new versions of the NIPE formulation respectively. As expected the condition number of
progressively distorted mesh is progressively increasing with a certain dependence from the used
example, but, as reported in Figure 11 using the old NIPE formulation it is possible to compute the
condition number only for light distortion. In fact the old NIPE formulation, red solid curves, fails
when the central node is moved by means of coefcients that causes edges to be aligned or concave
(for instance consider = 1.5 and = 1.5 or less. Note that the 180interior plate angle implies that
at the nodal quadrature point the Jacobian vanishes. For larger shifts the Jacobian becomes negative
and the old NIPE formulation fails. For the new NIPE formulation the condition number increases
with increasing distortions but the displacements and the resultants remain close to the reference
solutions.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 21
(a) = = 0.9 (b) = = 0.75 (c) = = 0.45
(d) = = 1.1 (e) = = 1.25 (f) = = 1.55
(g) = 1.1, = 0.9 (h) = 1.25, = 0.75 (i) = 1.55, = 0.45
Figure 8. Test 2a - NIP-Q4 quadrilateral 4-node element: distorted meshes for the simply-supported square
plate with a uniform load. Center of meshes is moved progressively away from its original position by using
the scheme reported in Figure 5(b)
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
22 G. CASTELLAZZI, P. KRYSL
(a) = = 0.9 (b) = = 0.5 (c) = = 0.125
(d) = = 1.1 (e) = = 1.5 (f) = = 1.875
(g) = 1.1, = 0.9 (h) = 1.5, = 0.5 (i) = 1.875, = 0.125
Figure 9. Test 2a - Central normalized displacement using NIP-Q9 quadrilateral 9-node element: distorted
meshes for the simply-supported square plate with a uniform load. Center of meshes is moved progressively
away from its original position by using the scheme reported in Figure 5(b)
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 23
(a) = = 0.0 (b) = = 0.9
(c) = = 0.75 (d) = = 0.45
Figure 10. Test 2a - mx moment distribution obtained using NIP-Q4 quadrilateral 4-node elements:
progressively distorted meshes for the simply-supported square plate with a uniform load. Center of meshes
is displaced progressively away from its original position by using the scheme reported in Figure 5(b).
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
24 G. CASTELLAZZI, P. KRYSL
10
2
10
3
10
9
10
10
10
11
10
12
10
13
Number of unknowns [ND]
C
o
n
d
i
t
i
o
n

n
u
m
b
e
r

[
N
D
]
new NIPE
old NIPE
new NIPE only
new NIPE only
Figure 11. Test 2a - Condition numbers for the simply-supported square plate with a uniform load for a
progressive distortion introduced using the resulting most severe set of coefcient for this example(see
Figures 9(d), 9(e), 9(f). The thicker the line the more severe the distortion (note that some of the lines
overlap). Black dashed lines for the new NIPE-Q4 element, red lines marked with asterisk () for original
NIPE. Original NIPE conguration fails for mesh conguration obtained using = = 1.5 and = =
1.55.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 25
5.4. Test 2b - Square plate with uniform load: severe distortion test for almost zero area element
The test aims to check the effectiveness of the introduced improved formulation, checking the ability
of quadrilateral NIPE elements to resist problematic geometry such as almost zero area elements.
In particular, the effect of severe mesh distortion is studied on the simply supported square plate with
uniform load (Youngs modulus E = 30 10
3
, Poisson ratio = 0.3, side length L = 10, thickness
t = L/10000, transverse uniform load b
z
= 0.4 all data in consistent units) [20]. With respect
to Figures 5(c) and 12(a) an initial mesh formed by merging together 4 sub-meshes of 64 elements
each is used. Four parameters are used to progressively shift the node marked by a black asterisk, see
Figure 12(a), from its initial position to the corner node. A shift equal to 1 makes the black-asterisk
node to coincide with one at the upper-right corner and at the same time the red-asterisk node to be
coincident with the node at the lower-left corner. The following intermediate ratios are use to test the
NIP-Q4 and the NIP-Q9 elements: [0.5, 0.75 0.9 0.9999], see Table III for a proper indication of the
used combination. Only one quadrant is modeled with appropriate symmetry boundary conditions.
Table III. Test 2b - Distortion parameters for the simply supported square plate study.
Set , e; f;
1 0;
7L
16
;
7L
16
;
2 0.5;
L
4
;
L
4
;
3 0.75;
3L
16
;
3L
16
;
4 0.9;
L
20
;
L
20
;
5 0.999;
L
2000
;
L
2000
;
Figure 12(b) shows the results for the NIP-Q4 element as the plot of normalized error in the center
deection [(reference-computed)/reference] for the undistorted conguration. Figure 13 shows the
results for the NIP-Q4 element as the plot of normalized error in the center deection [(reference-
computed)/reference] for the four distortion parameters. As shown by Figure 13 the NIP-Q4 element
pass the test showing an excellent agreement with the reference solution for the rst three values
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
26 G. CASTELLAZZI, P. KRYSL
*
* *
*
(a) Initial Geometry (b) Norm. Deection. 99.73%
Figure 12. Test 2b - Reference meshes for the simply-supported square plate with a uniform load. Center
(red asterisk mark) and mid point edge (black asterisk mark) of meshes are moved progressively away from
its original position.
of the used ratio. As expected, when extreme ratio is enforced, the NIP-Q4 shows a stable solution
but with a considerable degradation of the performance. Figure 14 shows the results for the NIP-Q9
element as the plot of normalized error in the center deection [(reference-computed)/reference] for
the four distortion parameter. As shown by Figure 14 also the NIP-Q9 element pass the test showing
an excellent agreement with the reference solution even if the most severe distortion is enforced.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 27
(a) Set 2 (b) Set 3
(c) Set 4 (d) Set 5
Figure 13. Test 2b - Distorted meshes for the simply-supported square plate with a uniform load and NIP-
Q4 elements. Center (red asterisk mark) and mid point edge (black asterisk mark) of meshes are moved
progressively away from its original position.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
28 G. CASTELLAZZI, P. KRYSL
(a) Set 2 (b) Set 3
(c) Set 4 (d) Set 5
Figure 14. Test 2b - Distorted meshes for the simply-supported square plate with a uniform load and NIP-
Q9 elements. Center (red asterisk mark) and mid point edge (black asterisk mark) of meshes are moved
progressively away from its original position.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 29
5.5. Test-03 - Simply-supported circular plate with uniform load: severe distortion test
Response of the present model is nally studied for the conguration of a simply supported circular
plate of radius R with uniform load. Only NIP-Q4 element is studied for the sake of brevity. Three
structured quadrilateral mesh with 4, 8 and 16 elements per radius are generated on a quadrant (using
two orthogonal symmetry planes). The vertices are optimally located to produce well-shaped
elements. Then element vertices are displaced by an initial random perturbation of rp = 5R/100
magnitude and then a random perturbation of rp = 15R/100. The obtained meshes are illustrated
in Figure 15. Note that the element shapes, in the perturbed mesh conguration, tend to be in some
cases extremely poor.
The parameters are: Youngs modulus E = 1092, Poisson ratio = 0.3, radius R = 1, and
transverse uniform load b
z
= 1 (in consistent units). The results are summarized in Figure 15 where
the normalized deection is plotted versus the thickness ratio t/h (that is thickness/(mesh size)).
Next the convergence performance are studied with respect to geometry distortions: the span to
thickness ratio is taken variously as R/t = [10, 100, 1000, 2000, 5000, 10000, 100000] and element
vertices are displaced by an initial random perturbation of rp = 5R/100 magnitude and then of
rp = 10R/100. The error in energy curves for the different R/t ratios are overlaid in the graphs,
with the thickness of the line indicating the thickness of the plate (the thicker the line, the thicker
the plate).
The element displays essentially the same rate of convergence and absolute error for all thicknesses
except the thickest plate. Such performance is an indicator of an essentially uniform convergence as
the thickness of the plate decreases.
Clearly, all models run into numerical difculties at about t/h = 10
4
(which indeed is a very thin
element), and numerical blowup follows at t/h = 10
5
. Up to t/h = 10
3
the NIP-Q4 model show
the hoped-for insensitivity to the thickness ratio, Figure 16. A little deterioration of the convergence
rate appear in the range of plate thicknesses investigated. Presumably, this is mainly due to the effect
of the high distortions introduced. Nevertheless, the overall numerical response of the element is
satisfying, conrming the good performance of this element even in presence of severe distortions.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
30 G. CASTELLAZZI, P. KRYSL
Figure 15. Test-03 - Distorted meshes (0R/100, 3R/100, 5R/100) for the simply-supported square plate
with a uniform load.
6. CONCLUSION
The reformulation of the Nodally Integrated Plate Element formulation (NIPE) has been presented
along with the introduction of a new 9-node element conguration.
The new 9-node element uses bi-quadratic interpolations of the transverse displacement and
rotations and it is computed by means of a nine-node nodal quadrature rule. The present work has
analyzed the NIPE technique, by inspecting thoughtfully the sensitivity of the elements to severe
geometry distortions. A successful improvement that confers robustness to all element shapes has
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 31
10
2
10
3
10
4
10
3
10
2
Elements [ND]
l
o
g
(
N
o
r
m
a
l
i
z
e
d

S
t
r
a
i
n

E
n
e
r
g
y
)

[
N
D
]


t = 0.1
t = 0.01
t = 0.001
t = 0.0005
t = 0.0002
t = 0.0001
t = 1e05
(a) rp = 5R/100
10
2
10
3
10
4
10
3
10
2
Elements [ND]
l
o
g
(
N
o
r
m
a
l
i
z
e
d

S
t
r
a
i
n

E
n
e
r
g
y
)

[
N
D
]


t = 0.1
t = 0.01
t = 0.001
t = 0.0005
t = 0.0002
t = 0.0001
t = 1e05
(b) rp = 10R/100
Figure 16. Test-03 - Simply supported circular plate with uniform load: convergence of the strain energy
for various thicknesses (R/t =100,1000,2000,5000,10000,100000). The thicker the line, the thicker the plate
(note that some of the lines overlap). Results are shown for model NIP-Q4 for two imposed distortions:
rp = 5R/100 (a) and rp = 10R/100 (b).
been developed for the NIPE formulation.
A brief review of the improved triangular and quadrangular linear NIPE elements has been reported
for linear plate analyses. Challenging benchmarks carried out on regular and extreme distorted
meshes have been used to illustrate the excellent performance of the introduced NIP-Q9 element. We
detail that the new NIPE formulation confers insensitivity to extreme distortions for the quadratic
quadrilateral element and allows to solve for any severe distortion that produces concave shape or
almost zero area element with the NIPE element family.
ACKNOWLEDGEMENT
The facilities were provided by the Laboratory of Computational Mechanics (LAMC) - University of
Bologna.
REFERENCES
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
32 G. CASTELLAZZI, P. KRYSL
1. Castellazzi G. On the performances of parametric nite elements when geometry distortions occur.
Finite Elements in Analysis and Design Dec 2011; 47(12):13061314, doi:16/j.nel.2011.07.004. URL
http://www.sciencedirect.com/science/article/pii/S0168874X11001417.
2. Castellazzi G, Krysl P. Displacement-based nite elements with nodal integration for Reissner-Mindlin plates.
International Journal for Numerical Methods in Engineering Oct 2009; 80(2):135162, doi:10.1002/nme.2622.
URL http://onlinelibrary.wiley.com/doi/10.1002/nme.2622/abstract.
3. Castellazzi G, Krysl P. Patch-averaged assumed strain nite elements for stress analysis. International Journal for
Numerical Methods in Engineering In press; xx(x):xxxxxx, doi:10.1002/nme.4264.
4. de Miranda S, Ubertini F. A simple hybrid stress element for shear deformable plates. International Journal for
Numerical Methods in Engineering 2006; 65(6):808 833.
5. Castellazzi G, Miranda S, Ubertini F. Patch based stress recovery for plate structures.
Computational Mechanics 2010; 47(4):379394, doi:10.1007/s00466-010-0548-3. URL
http://www.springerlink.com/content/g621h37188334575/.
6. Nguyen-Xuan H, Rabczuk T, Bordas S, Debongnie J. A smoothed nite element method for plate analysis.
Computer Methods in Applied Mechanics and Engineering 2008; 197(13-16):1184 1203.
7. Hughes T. Generalization of selective integration procedures to anisotropic and non-linear media. International
Journal for Numerical Methods in Engineering 1980; 15(9):1413 1418.
8. Broccardo M, Micheloni M, Krysl P. Assumed-deformation gradient nite elements with nodal integration for
nearly incompressible large deformation analysis. International Journal for Numerical Methods in Engineering
2009; 78(9):11131134, doi:10.1002/nme.2521.
9. Barlow J. More on optimal stress points-reduced integration, element distortions and error estimation. Int. J. Numer.
Methods Engrg. 1989; 28:14871504.
10. Stewart GW. On the adjugate matrix. Linear Algebra and its Applications 1998; 283(1-3):151 164, doi:DOI:
10.1016/S0024-3795(98)10098-8.
11. Krysl P, Kagey H. Reformulation of nodally integrated continuum elements to attain insensitivity to distortion.
International Journal for Numerical Methods in Engineering 2011; :n/an/adoi:10.1002/nme.3342. URL
http://dx.doi.org/10.1002/nme.3342.
12. Taylor R, Govindjee S. Solution of clamped rectangular plate problems. Communications in Numerical Methods in
Engineering 2004; 20(10):757 765.
13. Timoshenko S, Woinowsky-Krieger S. Theory of plates and shells. McGraw-Hill, 1959.
14. Xu Z. A thick thin triangular plate element. International Journal for Numerical Methods in Engineering 1992;
33(5):963 973.
15. Lee P, Bathe K. The quadratic MITC plate and MITC shell elements in plate bending.
Advances in Engineering Software May 2010; 41(5):712728, doi:16/j.advengsoft.2009.12.011. URL
http://www.sciencedirect.com/science/article/pii/S0965997809002634.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme
A NINE-NODE FINITE ELEMENT FOR REISSNER-MINDLIN PLATES 33
16. Bathe KJ, Dvorkin EN. A four-node plate bending element based on mindlin/reissner plate theory and a mixed
interpolation. International Journal for Numerical Methods in Engineering 1985; 21(2):367383, doi:10.1002/
nme.1620210213. URL http://dx.doi.org/10.1002/nme.1620210213.
17. Bathe K, Brezzi F, Cho S. The MITC7 and MITC9 plate bending elements. Computers and Structures 1989; 32(3-
4):797 814.
18. Bathe K. ADINA System Online Manuals. ADINA R&D, Inc.: 71 Elton Ave, Watertown, MA 02472, USA, 2005.
19. Hughes TJR. The nite element method. Linear static and dynamic nite element analysis. Dover publications, New
York, 2000.
20. Zienkiewicz OC, Taylor RL. Finite Element Method: Volume 2, Fifth Edition. 5 edn., Butterworth-Heinemann,
2000.
Copyright 2011 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2011)
Prepared using nmeauth.cls DOI: 10.1002/nme