Chapter Four

You might also like

Download as doc, pdf, or txt
Download as doc, pdf, or txt
You are on page 1of 42

54

CHAPTER FOUR
FINITE ELEMENT FORMULATION
FOR STIFFENED SHELL
4.1 Introduction
Various shell theories have been formulated by adopting simplified
hypotheses, which render them to specific cases.
The finite element method has been proven to be an efficient tool in
analyzing shell structures with arbitrary shapes, load conditions, and
support conditions.
Three distinct approaches to the finite element representation of
shell structures have been employed:
(a)

flat triangular elements.

(b)

curved elements formulated on the basis of shell theories.

(c)

elements derived from three dimensional elements.

The first numerical studies of shell problem involve facetting the


shell by plane triangular plate elements, onto which a membrane stiffness
was superposed. The results obtained were found to be satisfactory, but
depending on the problem, very fine mesh had to be employed.
Especially for shell problems with a high membrane activity the
disadvantage of the poor approximation of the membrane stiffness could
be detected, and an enormous number of elements was required to obtain
satisfactory results.

A number of difficulties and shortcomings arising

from the application of flat elements to curved shells are presented in


reference [13].
In the second approach, a classical concept is employed, whereby a
shell theory is used as the starting point of the finite element formulation.
This curved element eliminates the discontinuity of moments that plague
the flat elements [13] but leads to a complex formulation. A variety of

55
finite elements with different degrees of complexity have been formulated
for deep and shallow shells [8,20,35,36]. The inadequate membrane
representation was eliminated, but the simplicity of the element was lost.
It was necessary to introduce additional degrees of freedom namely first
and second order derivatives, which lead to difficulties at folds, and
boundaries.
An alternative simple element, which was easy to apply, was first
presented in [2]. This element was derived from a three dimensional
element by applying a so-called degeneration process. The classical
Kirchhoff hypothesis was no longer needed in this formulation since both
in-plane and shear strains were accounted for. Although this type of
element seemed very promising when it was introduced, it was only a
short time until the weakness of this model was detected. As the thickness
of the element was reduced, the results obtained worsened due to locking.
An improvement of the model was achieved by reduced and selective
integration to avoid the locking [39,28].
The above difficulties led to improve the behavior of the element
against locking and spurious mode by interpolating the strains at given
sampling points to the Gauss sampling points [5,19] which appear to be
efficient in representing the shell structure.
Finite element formulation for the stiffened shell is employed by
1.curved elements formulated on the basis of shell theories
2.elements derived from three dimensional elements.
For the shallow shell formulations, two types of elements were
used, the shallow shell element and the bar element for the formulation of
the stiffened shell structure.

The stiffeners were represented by bar

elements and then transformed to common coordinates [7,35,36]. The


shallow shell element is able to represent the arbitrary shape but is
restricted to thin shell the situation. When a shallow finite element

56
formulation is employed to represent the thick shell the situation is
restricted to specific shapes [7]. It is also important to indicate the
complex formulation and implementation of the shallow shell element for
shell structures.
It has been proved that the degenerated shell formulation is a
simple and general method to represent the stiffened and unstiffened shell
structures. When the degenerated element is used there are two methods
to represent the stiffeners. The first is by using the curved beam element
and the second is by representing the stiffeners by shell elements. When
the analysis is performed better results are obtained when the stiffeners
are represented by the curved beam elements [3,33,38]. This is due to the
crude approximation for the stiffeners when modeled by shell elements
which is in fact nothing more than using five degrees of freedom for a
case which has only two degrees at any section. Accurate results have
been obtained for both the mode and the magnitude when the stiffeners
are represented by a beam element in the finite element analysis [38].
In this chapter, a finite element formulation for stiffened shell is
presented. The modeling of the stiffened shell by the degenerated shell
element connected to the curved beam is adopted in the present work.
Compatibility must exit between the beam and the degenerated shell
element to predict the correct behavior of the stiffened shell [22].
Thannon [38] developed a nonlinear analysis based on the linear work of
Jirousek [22]and was applied to the analysis of reinforced concrete.
Layered approach for the degenerated shell element and for the curved
beam is adopted in the present work. The reinforcement is represented by
the smeared approach and by the embedded bar formulation. The
assumed strain formulation for the degenerated shell element is adopted
in the present work.

57
The consistent mass matrix for both the degenerated shell element
and for the curved beam is adopted. The displacement, velocity and
acceleration compatibility between the curved beam and the degenerated
shell element must be existed. Suitable transformation for stiffness matrix
( or B-matrix) was introduced by Jirousek[22] for linear analysis and then
developed by Thannon[38] through layers adopted in this work and it is
applied to the mass matrix. The damping matrix is obtained by
proportioning of the mass and stiffness matrices which is known as
Rayleigh damping.
4.2 Degenerated Isoparametric Elements
The degenerated shell element has been originally introduced by
Ahmed [2]. In this element Mindlin type theory is employed. The basic
idea of this element is:
(a)

The normal to the middle surface of the three dimensional

element is constrained to remain straight after deformation, by


introducing plane isoparametric element in the middle surface of the shell
with independent rotation and translation.
(b)

The three dimensional stress and strain conditions are

degenerated to shell behavior.


Mindlin plate theory can be extended to cater for curved shell
structure, if local coordinates are introduced at points of the shell surface.
In the local coordinate system ( x , y , z ), z is taken to be perpendicular
to the shell midsurface and the x y coordinate plane is taken as being
tangential to the midsurface of the shell. This could make the strain
energy corresponding to stresses perpendicular to the middle surface
disregarded, i.e., the stress component normal to the shell mid-surface is
constrained to be zero in the constitutive equation.
Five degrees of freedom are specified at each point, corresponding
to its three displacements and two rotations of the normal at the node. The

58
definition of independent rotations and displacements degrees of freedom
permits transverse shear deformation to be taken into account, since
rotation are not tied to the slope of the midsurface. This approach is
equivalent to using a general shell theory and reduced to the hypotheses
of Reissner and Mindlin when applied to plates.
4.3 Coordinate System
In the formulation of the degenerated shell elements different coordinate
systems are shown in Figure (4.1):
4.3.1 Global Coordinate Set
This is a Cartesian coordinate system, freely chosen, in which the
geometry is defined in space, as well as in the global stiffness matrix and
applied force vector.
4.3.2 Curvilinear Coordinate System ( , , )
In this system

are the two curvilinear coordinates, defining the

middle plane of the shell elements.

is a linear curvilinear coordinate

approximately normal to the middle surface and relates the points through
the thickness to the middle surface.
4.3.3Local Cartesian Coordinate System ( x , y , z )
This local coordinates system is defined at the sampling Gauss points,
wherein the stresses and strains are calculated. At such point, the z
direction is taken to be normal the surface
cross product of the

=constant, being obtained by

and directions so that

x x

y y


z z

(4.1)

The direction of x can be simply taken to coincide with the


tangent to the

direction as:

Figure (5.1) Coordinate system (a) nodal and curvilinear (b) local axes
system.

59
x

y
x

z

(4.2)

The direction of y is defined by the product of the x and z


y x z

(4.3)

The local coordinate system varies throughout the shell thickness. It


is useful to define the direction cosine matrix, which enables
transformation between the local and the global coordinate systems to be
undertaken. The direction cosine matrix is defined by expression:
x , y , z .

(4.4)

4.3.4 Nodal Coordinate Set ( v1k , v 2 k , v3 k )


The nodal Cartesian system is the local Cartesian coordinate system
associated with each nodal point of the shell element and its origin is at
the midsurface. The vector

V3 k

is constructed from the nodal coordinates

of the top and bottom surface at node k , so that


xk

V3 k y k
z
k

v3k

top

xk

yk
z
k

(4.5)
bot

v3xk

3k v3yk
V3 k z
v3k

The vector

v1k

(4.6)
is perpendicular to

v3k

and parallel to the global x-

z plane, so that:
v1k j v3 k / j v3 k

Or if

v3k

is in the y-direction,

v1k v3 k i / v3 k i

where

i and

(4.7)

(4.8)

are the unit vectors along the x and y directions

respectively. The vector v 2k is normal to the plane defined by

60
(4.9)

v 2 k v1k v 3 k / v1k v3 k

A nodal coordinate system is defined at each point with origin at


the reference surface.
4.4 Geometric Definition of the Element
Figure (4.1) shows a typical thick shell element. The external faces of the
element are curved, while the cross sections across the thickness are
generated by straight lines. Pairs of points,

j top and j bot ,

each with given

Cartesian co-ordinates, prescribe the shape of the element.

and are used to be two curvilinear co-ordinates in the middle

plane of the shell and

a linear co-ordinate in the thickness direction. If

further ( , , ), are assumed to vary between 1 and 1 on the respective


faces of the element then a relationship between the Cartesian coordinates of any point of the shell and the curvilinear co-ordinates can be
written in the form:
x
xk
(1 )

y
N k ( , ) 2 y k

z
z
k

Here

N k ( , )

top

xk
(1 )

N k ( , )
yk
2

zk

(4.10)
bot

is a shape function taking a value of unity at the node k

and zero at all other nodes.


It is convenient to rewrite the relationship in a form specified by a
vector connecting the upper and the lower points (i.e a vector of length
equal to the shell thickness hk ) and the middle surface co-ordinates. Thus
x
xk

y N k yk
z
z

k

N k
mid


V3 k
2

(4.11)

61

Figure(4.1) Coordinate system: (a) nodal and curvilinear system (b) local
axes system

62
4.5 Displacement Field
The displacement through the element is defined by the three Cartesian
components of the midsurface node displacement k and two rotations of
the nodal vector

v3k

about orthogonal directions normal to it. If the two

orthogonal directions are given by vectors

v2k

and

v1k

(of unite

magnitude) with corresponding (scalar) local rotations k and k


Figure(4.1) then it is possible to write
u
uk
n
n
v
v N ( , ) hk v
N
(

k
k
1k

k
2
k 1
w
wk k 1

v2k

(4.12)

When the shell element is combined with a beam element, the local
( k , ) rotations must be expressed in terms of the global rotations (
k , k , k )

k
k

Tk k
k
k

where

(4.13)

is a transformation matrix, which can be derived from

Tk

geometric consideration as

v
z
v3k
v vy
3k

Tk

T
2k
T
1k

v3zk

v3yk

0
v3xk

v3xk
0

(4.14)

4.6 Definition of Strain


In order to apply Mindlin theory to the shell elements, a local coordinate
system should be introduced. In this local coordinates system, the basic
shell assumptions of zero normal stress in the z -direction ( z 0 ) could
be undertaken by defining the strain in this local coordinates. Then the
stresses corresponding to these strains are defined by a matrix and
they are related by the elasticity matrix D .

63
The five significant strain components are:

x

y
xy

xz
yz

u
x
v

y
u v

y x
u w

z x
v w

z y

(4.15)

where u , v , and w are the displacement component

in the local

coordinates system. The local derivatives are obtained from the global
derivatives of displacement u , v , and w by the following operation,
u
x
u

y
u

v
x
v
y
v
z

w
T

y

u
x
u
y
u
z

v
x
v
y
v
z

w
x
w

y
w

(4.16)
where is a transformation matrix
The derivatives of the displacements with respect to the global
coordinates are given by
u
x
u

y
u

v
x
v
y
v
z

w
1
J

where J is the Jacobian matrix

w

w

w

(4.17)

64
x

x
J

x

z

z

z

(4.18)

4.7 Definition of Stress


The stresses corresponding to the strains defined above are:
x

y
xy D 0

xz
yz

(4.19)

where 0 may represent any initial strain due, for instance, to thermal
expansion and,
1
1

0
0
0
0
0
0
1

0
0
E 0 0
2

D
1

1 2 0 0
0
0

2r

1
0
0
0 0

2r

(4.20)

in which E and v are Youngs modulus and Poissons ratio respectively.


The factor r included in the last two shear terms is taken as 1.2
4.8 Quadratic Shell Element
A brief representation of quadratic elements is given.
(i) 8-Node Serendipity
The 8-node shell element was the simplest element considered in the
original work of Ahmed[2]. When the 8-node element is used with (3x3)
integration rule the solution becomes increasingly stiff as the shell
thickness is reduced. This phenomenon is known as locking.
Improvement of the performance of the 8-node element against locking is

65
obtained when reduced integration is used [39,28] but the problem
appears again for very thin situation [29].
(ii) 9-Node Lagrangian
The performance of the 9-node element appears to be superior to the
Serendipity element when reduced or selective integration is used [29].
However, a serious problem may arise with this element when reduced
integration is employed. The stiffness matrix exhibits rank deficiency
which originates the appearance of spurious mechanisms, i.e., zero energy
modes[6,27]. Theses communicable mechanisms may result in very
erratic solutions, especially when the restraints imposed by the boundary
conditions are minimal. This drawback is not exhibited by the 8-node
Serendipity element with reduced integration.
(iii)Heterosis Element
This 9-node element was initially developed for plates [15]. The
Serendipity shape function is employed for the vertical deflection w and
Lagrangian shape functions for the two rotations. This heterosis concept
has been extended to the shell analysis using the Serendipity shape
function for the three translation degrees of freedom [18]. This element
exhibits improved characteristics when compared with the previous
quadratic elements. Its stiffness matrix possesses the correct rank when
selectively integrated and the element does not lock even in very thin
situation. Adopting uniform reduced integration, the element behaves
better than the 9-node Lagrangian element, with the appearance of
spurious mechanisms being improbable for practical application.
4.9 Numerical Integration of Shell Element
Although the shape functions of the finite element are simple
polynomials, the expressions for the stiffness become rather complex for
curved element due to the mapping between the natural coordinates
( , , )

and the Cartesian coordinates

( x, y , z ) .

For an arbitrary shaped

66
element, a numerical integration is therefore necessary. Numerical
integration is widely used for isoparametric shell element.
Two Gauss points through the thickness for linear analysis are
sufficient to capture the bending behavior, because of a linear assumption
of strain through the thickness is assumed. High order Gaussian
quadrature rules (for instance 5-7 Gauss points) have been advocated for
nonlinear analysis, but a simple and general procedure to discretize and
integrate through the thickness is offered by the layered model.
In the shell plane (surface

= constant) the normal (full)

integration rule consists of mxm Gauss points where m is the number of


nodes along each side. When a degenerated shell element is fully
integrated, shear and membrane locking arise and increase as the
thickness of the shell is reduced. This performance is elevated by using
reduced integration ((m-1)x(m-1) Gauss points), but this leads to spurious
zero energy modes for higher order elements. An alternative procedure,
selective integration is then advocated.
4.9.1 Reduced Integration
Based on the work of Dehetry et al [10] to eliminate the parasitic shear on
plane quadrilateral elements, the implementation of the reduced
integration for the degenerated shell element was firstly introduced by
Zeinkiewicz [39]. Then many papers about the reduced integration
technique have been published [16,27,28,29].
It was found that the shear strain at the normal integration points in
an element is excessive, and the point used in reduced order Gaussian
quadrature in evaluating shear strain may coincide to the point at which
the shear strain is correct [28] Figure (4.2). Therefore the solution
obtained by using a normal integrated element diverges from those
predicted by Kirchhoffs theory where the shear effect are assumed to be
negligible as the plate or the shell becomes thin.

67
Explanation based on mathematical consideration is also presented.
This explanation showed that identical terms are obtained for the
analytical and reduced approaches, but extra terms appear when the full
integration is used [17,23,27].
It was observed, however, that the reduced integrated element has
some deficiencies. The 8-node element of the Serendipity family to which
the reduced integration technique is applied may produce the locking
phenomena in problems below a certain thickness ration, even though its
performance is improved [29]. The 9-node Lagrangian and Heterosis
elements do not exhibit this drawback. Also for the 8-node element the
results may decline if the elements are distorted [29]. The 9-node
Lagrangian element displays undesirable zero or low energy modes when
the reduced integration technique is used [6,27]. This problem appears
also for the Heterosis element as shown below in the table which shows
the number of such zero energy modes for a quadratic isoparametric
element.
Although Lagrange shape functions may yield better membrane
behavior than the Serendipity pattern in highly distorted elements, the
Heterosis is preferred for thin shell application in view of its lower
number of mechanism.
Table(4.1) Number of zero energy modes
8-node S Heterosis
9-node L
No. of membrane mechanism 1(*)
1(*)
2+1(*)
No. of bending mechanism
1(*)
2+1(*)
3+1(*)
((*) not communicable in mesh of two or more elements)

xy

Excessive shear

0.6

1/ 3

Correctunder
shearlinear
Figure (4.2) Linear strain distribution in quadratic element
moment condition

68
4.9.2 Selective Integration
The concept of the selective integration was first employed by Doherty,
Wislon and Taylor [10], to obtain improved bending behavior in simple 4node elements. One-point Gauss quadrature was used on shear strain
terms, and 2x2 Gauss points quadrature was used to integrate the
remaining terms. Selective integration was first employed for the
degenerated shell element by Pawsey and Clough [28].
In the selective integration technique, bending (or bending and
membrane) energies are integrated using the normal rule, while the shear
and membrane (or just shear) terms are computed using the reduced
integration. A considerable improvement in the results by this technique
is noticed for plate and shell problems [16,17,27].
The selective integration, however, does not remove the entire
aforementioned difficulties, i.e. the selectively integrated element may
still lead to zero energy spurious modes in Lagrangian family elements.
Several modifications to the reduced or selectively integrated elements
were suggested to avoid spurious modes. The Heterosis element neither
locks nor posses unwanted zero energy modes in plate bending [15].
Using the selective integration technique, the number of mechanisms
becomes as follows:
Table(4.2) Number of zero energy modes
Heterosis
No. of membrane mechanism
1(*)
No. of bending mechanism
----((*) not communicable in mesh of two or more elements)

9-node L
2+1(*)
1

4.10 Assumed Strain Element


This method was introduced by Huang [19]. By this method the
elimination of shear and membrane locking is achieved by interpolating
new shear and membrane strain fields from the strain values at the
sampling points which are appropriately located at individual elements.

69
The assumed strain method was implemented to the 9-node and the 8node elements.
4.10.1 Assumed Shear Strain
The elimination of the shear locking is achieved by interpolating the
shear strain in the natural coordinate system. It was found that the
assumed shear strain

and

for the 9-node element should be taken

as polynomials of at least the following degree


b1 b2 b3 b4 b5 2 b6 2

(4.21)

c1 c 2 c3 c 4 c5 2 c6 2

(4.22)

If the assumed strain fields are expressed as polynomials of lower


degree than those in equations (4.21and 422), zero energy mode will
appear. So six sampling points are required for the representation of the
assumed shear strain. The location of the sampling points for the shear
strain must satisfy the condition that shear strain could be equal to zero.
For

the sampling should be taken at the lines

1
3

and

on which three points must be taken, Figure (4.3). One of the three

is on

0. .

In order to guarantee the continuity of shear strain

component between two adjacent elements, the other two points should
be b 1 and

b 1

at lines

and

a .

Now

and quadratic in . Similarly the six sampling points for

points

b 1, 0 and b 1 at

is linear in and quadratic in


3

are three

the two lines a and a . Thus,

. Therefore

Pi ( )Q j ( ) ij

(4.23)

i 1 j 1
3

ij
Pi ( )Q j ( )

(4.24)

i 1 j 1

where

is linear in

P1 ( z )

z z

2b b

, P2 ( z ) 1

, P3 ( z )

z z

2b b

(4.25)

70
Q1 ( z )

1
z
1
2
a

Q2 ( z )

1
z
1
2
a

(4.26)

For the 8-node element the assumed shear strains should be taken
as polynomials of at least the following degree
b1 b2 b3 b4 b5 2

(4.27)

c1 c 2 c3 c 4 c5 2

(4.28)

Consequently, five sampling points are needed to construct the


assumed shear strain formulation as shown in Figure (4.4).
The assumed shear strain field is then given by:
n

Ri ( , ) i

(4.29)

i 1
n

S i ( , ) i

(4.30)

i 1

R1

1
1 1 R5
4
a
4

(4.31)

R2

1
1 1 R5
4
a
4

(4.32)

R3

1
1 1 R5
4
a
4

(4.33)

R4

1
1 1 R5
4
a
4

(4.34)


R5 1
a

2
1

(4.35)

and
S i ( , ) Ri ( , )

(4.36)

4.10.2 Assumed Membrane Strain


It is known that the membrane stain can only be separated from the
bending strain in the local Cartesian coordinate system ( x , y , z ) in
which the x , y plane is tangent to the shell surface. Therefore, it is
expected that the use of an assumed membrane strain in orthogonal
curvilinear coordinate system ( r , s, t ) ( in which the ( r , s,0 ) surface

71
coincides with the mid-surface of the shell element) can help to eliminate
the membrane locking behavior. To ensure that a special coordinate
system is uniquely defined, symmetry between the natural coordinate
system and the orthogonal coordinate system existed [19].
In the orthogonal curvilinear coordinate system, the in-plane
displacements can be expressed as follows:
u r u r 0 t r

(4.37)

u s u so t s

(4.38)

Figure (4.3) Interpolation points for the 9-node Lagrangian element

Figure (4.4) Interpolation points for the 8-node Serendibity element

72

where

u ro

and

are displacements at the middle surface of the shell and

u so

( r , s, t ) are referred to the orthogonal coordinate system.


It is noted that

u ro

and

are quadratic in

u so

and as well

quadratic in r and s ( t 0 ). The membrane strain


em rr u r 0 ,r

(4.39)

em ss u s 0, s

(4.40)

em rs

1
(u r 0 , s u s 0 ,r )
2

(4.41)
Therefore, for

e m rr
3

e m ss

and

em rs

, the interpolation are as follows

em rr Pi ( )Q j ( )e ij m rr

(4.42)

3
2
1 (a)
e m rs P ( ) i Q( ) j (e ij m rr / 2)
2
i 1 j 1

(4.43)

i 1 j 1

em ss P ( ) i Q j ( )e ij m ss

(4.44)

3
2
1 (b)
e m rs Pi ( )Q j ( )(e ij m rs / 2)
2
i 1 j 1

(4.45)

i 1 j 1

where

Pi (z )

4.26). Thus

and

Qi (z )

e m rr

and

have the same meaning as in equations(4.25 and


1 (a)
e m rs
2

quadratic along -direction,

e m ss

are linear along the


and

direction and quadratic along the

1 (b )
em rs
2

-direction and

are linear along the -

-direction. This interpolation is

similar to that used for the assumed shear strain, Figure (4.5). Finally, the
in-plane shear strain is quadratic in both directions, that is
e m rs

1 (a)
1
e m rs em( b)rs
2
2

(4.46)

For the 8-node Serendipity shell element, the assumed membrane


strains should have at least five polynomial terms, which are the same as

73
for the assumed shear strain mentioned in section (4.10.1). Then, for
e m rr

and

1 (a)
em rs
2

the interpolation should be as follows

em rr Ri ( , )emi rr
i 1

(4.47)

5
1 (a)
em rs Ri ( , )(emi rs / 2)
2
i 1

(4.48)
and for

e m ss

and

1 (b )
em rs
2

the interpolation should be

em ss S i ( , )emi ss
i 1

5
1 (b)
em rs S i ( , )(emi rs / 2)
2
i 1

(4.50)

Figure (4.5) Interpolation points for membrane strain

(4.49)

74
4.11 Comparison Between the Assumed Strain Element and
Quadratic Shell Elements
Huang [19] performed patch test, locking test, and convergence test and
compared the results with the Heterosis and 9-node Lagrangian elements
with selective integration.
4.11.1 Patch Test
The patch test was performed with meshes of arbitrary quadrilateral
elements.
For bending patch test the assumed strain element and the 9-node
Lagrangian element passed the test but the Heterosis element did not. For
twisting patch test the assumed strain element passed the test while both
the 9-node Lagrangian element and the Heterosis element did not. For
shear patch test all elements passed this test.
4.11.2 Convergence Test
This test is a comparison between the analytical solution and the finite
element solution. Convergence studies are carried out on a uniformly
loaded, simply supported, square plate and also on a clamped circular
plate.
Comparison with the exact Reissner solution on a clamped circular
plate showed that the assumed strain element and the Heterosis element
provided a good idealization whereas the 9-node Lagrangian element
exhibited significant oscillation in the transverse displacement caused by
spurious zero energy mode [15].
Comparison was also carried out for a thin circular plate and the
results showed that for the assumed strain element a much better rate of
convergence was obtained.
For a thin skew plate of Morleys 30 0, the result of the finite
element solution was compared with the exact thin solution. The result of

75
the assumed strain element and the 9-node Lagrangian element gave good
agreement and better than the Heterosis element.
For Razzaques 600 -skew thin plate which was solved by the finite
difference solution, good convergence was obtained for all types of
elements.
More details are given in reference [19].
4.11.3 Locking Test
The element is tested to shear and membrane locking by imposing pure
bending and the results showed that no locking occurred.
4.12 Eccentric Curvilinear Beam Element
The best physical and geometric representation of the stiffened shell is by
modeling both the shell and the beam in their real behavior. The shell
element and the beam element were connected, for the analysis of
stiffened shell, displacement compatibility between and the beam element
and the shell element must be exist. This compatibility is achieved by
relating the displacement of the beam element to the axes at the
connected region between the shell and the beam (joint axes) to the
middle surface of the shell element (reference axes) [22].
4.12.1 Reference, Joint and Centroidal Axes
Figure (4.6) illustrates a typical 3-noded curvilinear beam element, with
six degrees of freedom comprising three global displacements ui, vi, wi
and three global rotations L, L,L. The points G1, G2, and G3 are the
nodes on the element centroidal axis and is a curvilinear coordinate
which varies between (-1,+1).
xG
xGL
3
y
N L ( ) y GL

G
L 1
z G
z GL

(4.51)

76
When introducing such an element to a thick shell element, two
additional axes should be introduced, the reference axis and the joint axis.
These are shown in Figure (4.6). These additional axes guarantee the
compatibility between the shell element and the beam elements.
The position of the points G L and E L with respect to the reference
axis can be specified by using vectors

e L , g L .

Consequently nodal

coordinate on the centroidal axis can be expressed in terms of the nodal


coordinate on the reference axis as:
xGL x L
x L x L
y y e g y y
L
L
GL L
L L
z GL z L
z L z L

(4.52)

The local Cartesian coordinates are x , y and z where x is normal


to the plane of the element cross section and y and z are the principal
axes of the area of the cross section, as shown in figure (4.7).
The beam cross section at a given

should ideally contain the

point E on the joint axis


xE
x EL
y
N L ( ) y EL
E
z E
z EL

(4.53)

It is assumed that the vector

coincides with the direction of the

local y so that the cross section plane is approximately normal to the


centroidal axis. For the intermediate section along the axis it holds that:
3

g ( ) N L ( ) g L ,

b( ) N L ( )bL

L 1

Using the vectors

L 1

and

, the unit vectors in the local coordinate

system can be computed:


a
i
a

j b
b

(4.55)
where

a b g

and

c a b

(4.54)

c
k
c

77
The matrix of the orthogonal transformation from the local to the
global axes will therefore be given as:

( ) i( )

j ( )

k ( )

(4.56)

4.12.2 Generalized Stress-Strain Relationship


The constitutive equation relating the generalized stress () to the
generalized strain will be of the form, Figure (4.8):
=D
N
Q
Y
Q
Z
T
MY

M Z

EA

GAY

GAZ

;D

GJ

EI Z

EI Y

u G , x
vG , x
wG , x

G , x
G , x

G , x

(4.57)
where, . , denotes local reference frame and , x for differentiation with
respect to x
A: cross sectional area for each layer
Ay Az

Equivalent shear area

E: Modulus of elasticity
G: Shear modulus
J: Torsional constant
I y , I z :

Principal moments of inertia

The displacements and the rotations in the local and global systems
are related through [22]:

u 'G

'
vG
w'G

'
uG G
G

'
T vG ; G T G
G
wG G '

(4.58)

78
Neglecting the possible small difference between the local axes x
and the tangent to the centroidal axis it follows that:
d
1 d

dx t d

(4.59)

where t is the modulus of the vector

xGL
dN
L
t( )
y GL

L 1 d
z GL
3

(4.60)

From the previous expression it is easy to derive the strain-displacement


matrix [22]
1 dN L T
t d

BGL

0
N L .k T
N L j T

1 dN L T

t d

(4.61)

Standard finite element procedure can be followed for computing


the stiffness matrix [38]:
KG

where

Ln

Ln

T
BGL
DBGL td

(4.62)

is the number of layers through thickness

4.13 Beam Element Combined with the Degenerated Shell Element


To meet the requirement of displacement compatibility along the joint
axis, it is assumed that the displacement field of the curved beam element
is generated by linking each cross-section to the reference axis by means
of a rigid rod vector e L ( ). It can be shown that[22]:
L
u GL u L
v v A
RL
L
GL L
L
wGL wL

(4.63)

79

ARL

z L
y L

z L
0
x L

y L
x L
0

(4.64)

The strain-displacement matrix in equation (4.61) associated with


the centroidal axis nodes can be transformed to be associated with the
reference axis nodes by:

where

BRL BGL TL

(4.65)

I
TRL
0

(4.66)

0
ARL

Finally the total stiffness matrix can be assembled following the


standard finite element procedure.

80

Figure (4.6) Geometry of the eccentric curved beam element.

81

Figure(4.7)Local coordinate system in the nodal section of the eccentric


curved beam

Figure(4.8 )Generalized stresses of the curved beam element.

82
4.14 Dynamic Equilibrium Equation
The dynamic equilibrium equations are obtained by using the principle of
virtual work [4,32] which states that for any arbitrary kinematically
consistent set of displacements, the virtual work done must be equal to
that done by external forces irrespective of the material behavior, as

dv

st

tds u

b u cu dv

(4.67)

where u is a vector of virtual displacements, is the vector of


associated virtual strains and is the vector of stresses. The term t is a
vector of surface tractions acting on the portion s t of the boundary S .
Vectors

and
b, u

are
cu

the body, inertial and damping forces

respectively. The symbol (.) denotes differentiation with respect to time.


is the mass density and c is the damping parameter.

For the finite element representation, the displacements, velocities


and accelerations
variables

d , d

and

u , u and
by
d

can
u

be defined in terms of the nodal

the expressions

u N i ( , , )d i N d , u N d

(4.68)

i 1
m

u N i ( , , )d i N d

(4.69)

i 1
m

u N i ( , , )di N d

(4.70)

i 1

where u N i ( , , )d i N d is the matrix of the shape functions.


i 1

With standard strain matrix B , the virtual strain vector can be


related to the nodal variables as:
m

Bi d i Bd

(4.71)

i 1

Upon substitution of equations (4.68-4.71) into equation(4.67) then

d T Md Cd p ( d ) d T f

(4.72)

83
in which the mass matrix M , the damping matrix C , the internal force
vector

and the external applied vector

p (d )

f e have

the following

element contributions

Me

Ndv

(4.73)

ve

Ce

cN

dv

Ndv

(4.74)

ve

pe

(4.75)

ve

fe

tdS

se

where

Nbdv

(4.76)

ve

s e and v e

denote the surface and volume of the element under

consideration. Equation (4.72) may be written as:


(4.77)

Md Cd p ( d ) f e

For linear elastic situation, the stresses are related to the strain

as follows:
D DBd

(4.78)

Therefore, the internal forces

p (d ) can

be written as

p (d ) K s d
Ks

(4.79)

DBdv

ve

4.15 Modeling of Mass


The mass matrix

Me

in equation (4.73) is called consistent mass matrix.

The resulting element matrix is a full matrix and the global mass matrix
has the same structure as the stiffness matrix. A very convenient
simplification is achieved by lumping the element mass to nodes. This is
done in order to avoid the factorization process and to solve the
associated problem by explicit algorithm. Several lumping schemes are
proposed and in general depending on intuitive reasoning [14].

84
The sub-matrix of the element mass matrix linking node ( i ) and (
j

)can be expressed as:


M e ij

N j dv

(4.80)

ve

which does not lead to a diagonal matrix, when the adopted shape
functions are identical to those used in the evaluation of the element
stiffness matrix.
In the present work a consistent mass matrix is adopted for both the
degenerated shell element and the beam element.
Acceleration compatibility between the degenerated shell element
and the beam element is achieved by the following transformation:
T
M RL TRL
M GLTRL

where
TRL :

M GL is

(4.81)

the mass matrix of the beam element at the centroidal axis

transformation matrix defined in equation(4.66).

4.16 Modeling of Damping


In the present study, Rayleigh damping is adopted for modeling the
damping matrix. In this approach, the damping matrix is assumed to be
proportional to the mass and stiffness matrix and leads to a banded
damping matrix[32,14].
C d M d K

In the present study the approximation that

(4.82)
d 0

is adopted so

that[14,32]:
cij d mij

where d

(4.83)
2 r r

in which r and r are the damping ratio and the circular frequency for
the r th mode.

85
4.17 Layered Model
In reinforced concrete shell problems, the nonlinear behavior of
compressive concrete, concrete cracking and reinforcement response need
a convenient representation across the thickness. Layers of a different
thickness can be employed, Figure (4.9). The natural coordinate
varies from

which

-1 to 1, is determined at middle point of each layer, as

illustrated in Figure(4.9). Strain and stress components and stiffness


contribution are all computed at the midsection of each layer. Layers are
numbered sequentially, starting at the bottom surface of the shell. Each
layer contains stress points and are assumed to be constant over the
thickness of each layer, so that the actual stress distribution of the shell is
modeled by piecewise constant approximation.
The stress resultants are obtained by integrating the corresponding
stress components with respect to the thickness coordinate:
-normal forces
N

x
( y)

h/2

h / 2

x
( y)

dz

h n i
x i
2 i 1 ( y )

(4.84)

-bending moments,
M

x
( y)
( xy )

h/2

h / 2

zdz

x
( y)

xy

h2
4

i i

i
x

i 1

( y)

xy

(4.85)

-shear force
Q

x
( y)

h/2

h / 2

xz
( yz )

dz

h n i
xz i
2 i 1 ( yz )

(4.86)

where n( i=1,n) is the number of layers.


The volume integral is split into integrals over the area of the shell
midsurface and through the thickness ( h ). Thus, the stiffness may be
written as:
Using an isoparametric mapping, then

86
1 1 1

Ke

DB J ( , , ) ddd

1 1 1

(4.87)
1 1

h j

1 1

where

DB J ( , , )

2h j

d d

j 1

J ( , ,

(4.88)

is the determinant of the Jacobian matrix for layer

is the thickness of the

jth

layer and n is the total number of layers.

Figure (4.9) Layered model for plate, shell and beam.


4.18 Embedded Bar Simulation
A major advantage of the isoparametric element is that relatively fewer
elements are required to represent a particular structure. In order to retain
this efficiency, special line elements were developed to represent steel
reinforcement [1,9,11,25,30,34].
This line element (or embedded bar) that lies somewhere inside the
basic element (concrete element) is not like the bar element on which its
nodes had to coincide with the nodes of the basic element (concrete

87
element). This means less number of elements are required to represent a
particular structure.
In the present study simple implementation is formulated for
reinforcement with arbitrary location in the degenerated shell element.
Perfect bond is assumed between the reinforcement and the
concrete, which means that the strain of concrete and reinforcement has
the same values at the common points.
The basic idea is only to introduce the correct direction cosines
relating the uniaxial strain in steel in the local coordinate system of the
bar to the global coordinate system. Then the same strain-displacement
matrix of the shell element will be used with the corresponding direction
cosines of the embedded bar.
4.18.1 Geometric Definition of Bar Element
Figure (4.10) shows the representation of the reinforcement by a bar
element in the degenerated three dimensional Ahmed element. In the
present study 3-noded curved element is used to represent the geometry
of the reinforcement. These elements are used to calculate the direction
cosines at the Guass points and to calculate the length of the bar using a
numerical integration method in order to evaluate the stiffness matrix and
stresses.
It is only sufficient to give the natural coordinate of the bar element
within the parent element (shell element). Then the Cartesian coordinates
of the bar element are calculated using equation (4.10). By using these
coordinates and the shape functions below, one can calculate the Jacobian
matrix, which is used to compute the length of the bar at the Gauss points
as follows:
The global coordinate of the bar is related to the natural coordinate
using the three-noded elements:

88

x i 1 N i x i
n

(4.89)
y i 1 N i y i
n

(4.90)
z i 1 N i z i
n

where

(4.91)

is one dimensional shape function

Ni

N1

1
( 1)
2

(4.92)

N 2 (1 2 )
N3

where

(4.93)

1
( 1)
2

(4.94)

is dimensionless coordinate along the bar axis having values

from 1 to 1. Then the Jacobian matrix of the bar element is calculated

Jb

(4.95)

The determinant of Jacobian matrix can be evaluated as:


Jb

(4.96)

Embedded bar
2

Mid surface

89

4.18.2 Strain and Stress Definition of the Embedded Bar Element


It was pointed out that only longitudinal strain x in reinforcement is
considered, with perfect bond assumption between reinforcement and
concrete. The longitudinal strain can easily be evaluated by using the
global derivatives of displacements u , v , and w by standard operation:
u
x

u
T u
x
b
x
y
u

v
x
v
y
v
z

w
x
w
b
y
w

(4.97)Embedded bar in the shell element.


Figure(4.10)
where
x

y

z


b
Jb

where

Jb

(4.98)

is the determinant of the Jacobian matrix of the bar element.

Or

b T

x
x

y
x

z
x

(4.99)

So the B-matrix will be the same as that computed by the shell


element, but with different direction cosines and one dimensional
elasticity matrix is needed at the Guass point located on the bar element.
It should be noted that the location of the Gauss point on the bar element
should be calculated on the shell element in terms of the natural
coordinates which represent the Gauss point on the shell element wherein

90
the B-matrix is calculated.

This natural coordinate system could be

computed by two methods. The first is by assuming that the coordinates


of the nodes on reinforcement in global coordinate system are available
and the using of the inverse mapping is utilized to obtain the natural
coordinates in shell element [11]. The second is by assuming that the
natural coordinates on the shell element (which coincide with the nodes
of reinforcement) are available and using the shape functions of the bar
element to interpolate the location of the Gauss points [34]. The second
approach will be adopted in this study.
The strain at the Gauss points in the reinforcement bar can be
related to the nodal displacements of the shell element as follows:
x B b d

(4.100)

d is the displacement vector of the shell element.

The stresses corresponding to these strains are defined by a matrix

x and related by the elasticity matrix D b .


x D b x
D b

E
0

0
0

0
0
0
0
0

0
0
0
0
0

0
0
0
0
0

0
0
0

0
0

(4.94)

(4.101)

The stiffness matrix is computed for the bar element which


contributes to the concrete:
Kb

Db Bb dv

(4.102)

91
References
1. Abdul-Razzk, A. A. (1996). "Nonlinear finite element analysis of
fibrous reinforced concrete structural members." Ph.D. Thesis, Mosul
University, Iraq, 234 pps.
2. Ahmed, S., Irons, B. M., and Zeinkiewicz, O. C. (1970). " Analysis
of thick and thin shell structure by curved finite elements."
International Journal for Numerical Methods in Engineering, Vol. 2,
pp. 419-451.
3. Arnesen, A. (1979). "Analysis of reinforced concrete shells
considering material and geometric nonlinearities." Division of
Structural mechanics, Norwegian Institute of Technology, University
of Trondheim, Norway, Report No. 79-1.
4. Bathe, K. J. (1996). "Finite element procedure." Prentice-Hall Inc.
5. Bathe, K. J., Iosilevich, A., and Chapelle, D. (2000)."An evaluation
of the MITC shell elements." Computers and Structures, Vol.75, pp. 130.
6. Bicanic, N. (1979). "Spurious modes in two dimensional
isoparametric elements." International Journal for Numerical Methods
in Engineering, Vol. 14, pp. 1545-1557.
7. Bouabdalla, M. S., Batoz, J. L. (1996). "Formulation and evaluation
of a finite element model for the linear analysis of stiffened composite
cylindrical panels." Finite Element in Analysis and Design, Vol. 21,
pp. 265-189.
8. Barut, A., Madenci, E., Tessler, A., and Starnes, J. H. (2000). "A
new stiffened shell element for geometrically nonlinear analysis of
composite laminates." Computers and Structures, 77, pp. 11-40.

92
9. Chang T. Y., Taniguchi H., Chen W.F (1987)."Nonlinear finite
element analysis of reinforced concrete panel. " ASCE, Vol. 113, No.1.
10. Doherty, W. P., Wilson, E. L., and Taylor; R. L. (1969). "Stress
analysis of axi-symmetric solids using higher order quadrilateral finite
elements. " Structural Engineering Laboratory, Report. No. SESM 693, University of California, Berkeley.
11. Elwi, A. E., Hrudy, T. M. (1989)."Finite element model for curved
embedded reinforcement." ASCE, Vol. 115, No.4.
12. Fezans, G., and Verchery, G. (1982)."Some results on the behavior
of degenerated shell element." Nuclear Engineering Design, Vol. 70,
pp. 27-35.
13. Gallagher; R. H. (1976). "Problem and progress in thin shell finite
elements analysis, in finite elements for thin shells and curved
members." D. G. Ashwell and R.H Gallagher (eds.), John Wiley &
Sons, London.
14. Hinton, E. (1988). "Numerical methods and software for dynamic
analysis of plates and shells." Pineridge Press, Swansea UK.
15. Hughes, J. R., and Cohen, M. (1979). "The heterosis finite element
plate bending. " Computers and structures, Vol.9, pp. 445-450.
16. Hughes, J. R., Cohen, M., and Haroun, M. (1978). "Reduced and
selective integration technique in finite element analysis of plates."
Nuclear Engineering Design, Vol.46, pp. 203 222.
17. Hughes, T. J. R., Taylor, R. L., and Kanoknuchulchai, W. (1977). "A
simple and efficient finite element for plate bending. " International
Journal for Numerical Methods in Engineering, Vol. 11, pp. 15291543.

93
18. Hughes, J. R. and Liu, W. K. (1981). "Nonlinear finite element
analysis of shells. Part 1, three dimensional shells." Computer
Methods in Applied Mechanics and Engineering, Vol.26, pp. 331-362.
19. Huang, H. C. (1989). "Static and dynamic analyses of plates and
shells." Springer-Verlag, Berlin Heidelberg.
20. Idelsuhn, S. (1981). "On the use of deep, shallow or flat shell finite
elements for the analysis of thin shell structure." Computer Methods in
Applied Mechanics and Engineering, Vol.26, pp. 321-330.
21. Jiang, J., and Olson, M. D. (1991). "Nonlinear dynamic analysis of
blast loaded cylindrical shell structures." Computers and Structures,
Vol.41, No.1, pp. 41-52.
22. Jirousek, J. (1981). "A family of variable section curved beam and
thick

shell

or

membrane

stiffening

isoparametric

element."

International Journal for Numerical Method in Engineering, Vol. 17,


pp. 171-186.
23. Kreja, I., and Cywinski, Z. (1988)."Is reduce integration just a
numerical trick." Computers and structures." Vol.29, No.3, pp. 491496.
24. Lio, C. L. and Cheng, C. R. (1994). "Dynamic stability of stiffened
laminated composite plate and shell subjected to in-plane pulsating
forces. " International Journal for Numerical Methods in Engineering,
Vol.37, pp. 4167-4183.
25. Mahmood, M. N., (1994). "Investigation of post-cracking behavior
of reinforced concrete girder bridge" Ph.D. Thesis, University of
Roorkee, India, 247 pps.

94
26. Morris, I. R., and Dawe, D. J. (1980). "Free vibration of curved
plate assemblies with diaphragm ends." Journal of Sound and
Vibration, Vol.73, No.1, pp. 1-17.
27. Parich, H. (1979). "A critical survey of the 9-node degenerated shell
element with special emphasis on thin shell application and reduced
integration."

Computer

Method

in

Applied

Mechanics

and

Engineering, Vol.20, pp. 323-350.


28. Pawsey, S. F. and Clough, R. W. (1971). "Improved numerical
integration of thick shell finite element." International Journal for
Numerical Methods in Engineering, Vol.3, pp. 575-586.
29. Pugh, E. D. L., Hinton, E. and Zienkiewicz, O. C. (1978) "A study
of quadrilateral plate bending elements with reduced integration."
International Journal for Numerical Methods in Engineering, Vol.12,
pp. 1059-1097.
30. Phillips, D. V., and Zienkiwevicz, O. C. (1976). "Finite elemets
nonlinear analysis of concrete structures. " Proc. Instn. Civ. 61, pp. 5988.
31. Olson, M. D. (1991). "Efficient modeling of blast loaded stiffened
plate an d cylindrical shell structures." Computers and structures
Vol.40, No.5, pp. 1139-1149.
32. Owen, D. R. J., and Hinton, E. (1980). "Finite element in plasticity.
" Pineridge Press, Swansea, U.K.
33. Ram, E. and Kompfner, T. A. "Reinforced concrete shell analysis
using an inelastic large deformation finite element formulation. " In
Proceedings of the International Conference on Computer Aided
Analysis and Design of Concrete Structures (edited by F. Dmamjanic
et al 1984) 581-597, Pineridge Press, Swansea.

95
34. Ranjbaran, A. (1991). "Embedding of reinforcement in reinforced
concrete elements implemented in DENA." Computers and Structures,
Vol. 40, No. 4, pp. 925-930.
35. Sinha, G. and Mukhopadhyay, M. (1997). "Static, free and forced
vibration analysis of arbitrary non-uniform shells with tapered
stiffeners. Computers and structures." Vol.62, No.5, pp. 919-933.
36. Sinha, G. and Mukhopadhyay, M. (1995)."Transient dynamic
response of arbitrary stiffened shells by the finite element methods
Journal of Vibration and Acoustics." Vol. 117, pp. 11-16.
37. Stolartski, H. and Blytschko, T. (1982). "Membrane locking and
reduced integration for curved elements." Journal of Applied
Mechanics, Vol.49, pp. 172-176.
38. Thannon, A. Y.

(1988). "Ultimate load analysis of reinforced

concrete stiffened shells and folded slabs used in architectural


structures. " Ph.D. Thesis, University of Wales, Swansea, U.K.
39. Zienkiewicz, O. C., Taylor; R. L., and Too, J. M. (1971). "Reduced
integration technique in general analysis of plates and shells."
International Journal for Numerical Methods in Engineering, Vol.3,
pp. 275-290.

You might also like