Derivation of The Shell Element, Ahmed Element, Midlin Element in Finite Element Analysis - Hani Aziz Ameen

Derivation of the shell element , Ahmed Element , Dr.
Hani Aziz Ameen

Midlin Element in Finite Element Analysis
Derivation of the shell element ,

Ahmed Element , Midlin Element
in Finite Element Analysis
Asst. Prof. Dr. Hani Aziz Ameen

Technical College - Baghdad
Dies and Tools Eng. Dept.
E-mail:[email protected]
www.mediafire.com/haniazizameen
1- Definition of a Shell
Shell is defined as an object which, for the purpose of stress
analysis may be considered as the materialization of a curved surface [1].
This definition implies that the thickness of a shell must be small
compared with its other dimensions, but it does not require the smallness
be extreme. Most shells, of course, are made of a solid material, and
generally, it will be assumed that the material is isotropic and elastic.
1
Derivation of the shell element , Ahmed Element , Dr. Hani Aziz Ameen
In most cases, a shell is bounded by two curved surfaces, the faces. The
thickness "ts" of the shell may be assumed the same everywhere or it may
vary from point to point. The middle surface of a shell is defined as the
surface which passes midway between the two faces. If the shape of the
middle surface and thickness are known, then the shell is geometrically
fully described.
2- Types of Shell Elements:

The analysis of shells with an arbitrarily defined shape presents an
intractable analytical problem. If, in addition, the shell is a thick one in
which the shear deformation is significant, the applicability of a classical
approach becomes a question. In many engineering structures, these
difficulties might be overcome, if satisfactory (and hopefully optimized)
designs are ever to be achieved. Over the years, much has been written on
the various attempts to produce efficient, accurate, and reliable shell
elements in (FEM). Three distinct classes of shell elements have emerged
[2]:
1. Flat, plate-like elements which are sometimes called facet elements
because they approximate the curved shell by a faceted surface.
2. Curved shell elements founded on some shell theory.
3. Degenerated shell elements based on the three-dimensional continuum
theory.
In the first approach, the shell is replaced by an assemblage of flat plate
elements which are either triangular or quadrilateral in shape, as the
triangle shown in figures (2) (a) and (b) [3]. Each plate element is
connected in some fashion to those surrounding it and undergoes both in-
plane (membrane) and bending (flexural) deformations. The method has
2
the disadvantage that there is no coupling between bending and stretching

with each element. The coupling only appears indirectly through the
degrees of freedom at the nodal points linking the adjacent elements.
Consequently, a large number of elements must be used to achieve
satisfactory accuracy. Although of the certain shortcomings in the
approach, facet elements are very efficient for the approximate analysis
of many shell structures.
The second type includes curved shell elements based upon thin shell
theories of classical mechanics, consisting of the analysis of deep or
shallow shells. A commonly used theory of deep shells is based upon the
strain-displacement relationships of Novozhilov [4]. On the other hand,
specialized theories of shallow shells follow the simplified strain-
displacement relationships of Vlasov [5]. The later method is more
approximate than the former, but accurate results have been obtained,
even when shallow-shell concepts were applied to deep shells, Cowper
et.al. [6]. Figure (3) shows the geometry for an arbitrary shallow shell
element. Although the above type of shell elements are quite popular, but
they also suffer from various limitations associated with the lack of
consistency in many shell theories and also with the difficulty in finding
appropriate deformation idealization which allows truly strain-free rigid
body movement.
Finally, the curved elements (third class) for shell analysis can be
devised by specializing three dimensional solid elements to be thin in one
direction while introducing constraint conditions on nodal displacements.
As examples, the hexahedron and the pentahedron in figures (4) (a) and
(b) can be specialized to become quadrilateral and triangular shell
elements that are curved in three dimensional spaces. The characteristics
and analysis of this type is illustrated in the following section.
3
z y
3
 qi 2
i
1 u q i1
3
2
(a)
z y
3
qi 3
w qi 5
1 i
qi 4
3
2
(b)
x
Figure (2) Flat-facet element: (a) Membrane components (b) Flexural
3
components.
v
w
u
2
1
 ( , )
z
y 
3
1
 c
b
a
2
x 
Figure (3) Shallow shell element geometry and coordinate system [6].
4
3- Finite Element Formulation of Shell Element

Among all of the shell elements, the Ahmad [7] type
"degenerated" isoparametric shell element based on an independent
translational and rotational displacement interpolation, has become the
most popular in shell analyzation. In this element, the Mindlin [8] theory
is employed, where the "normal" to the middle surface of the element is
constrained to remain straight (but no longer normal) after deformation in
order to overcome the numerical difficulty associated with a large
stiffness ratio through the thickness direction. The strain energy
associated with the stress perpendicular to the middle surface is also
neglected. By adopting the isoperimetric geometric description, the
element can be used to represent thin and thick shell components with
arbitrary shapes.
Figure (5) (a), shows the original isoperimetric hexahedron element
which has a quadratic formula defining its geometry, where ui, vi, and wi
are translations in the global coordinates as demonstrated by Weaver and
Johnston [9]. In order to convert this hexahedron to a thin curved
quadrilateral element for the analysis of shell, one can first form a flat
rectangular solid by making the curvilinear coordinate's ξ, η, and ζ
orthogonal and the ζ dimension is small. The resulting element appears in
figure (5) (b) is the rectangular parent of shell element before constraints.
Note that groups of three nodes occur at the corners, while pairs of nodes
are at the mid-edge locations of the element. By invoking the former
constraints, each group and pair of nodes could be converted to a single
node on the middle surface, as shown in figure (5) (c), where αi and βi are
small rotations about two local tangential axes. The Relationships
between the nodal displacement at a corner and mid-edge with a node of
5
shell elements can be seen more clearly in figures (6) (a), (b) and (c). So,
the nine nodal, translations in figure (6) (a), can
(b)
(a)
Figure (4) Specialization of solids [6]:

(a) Hexahedron (b) Pentahedron.
z
wi

Y 5
i x 17 13 22
vi 1
6 8
ui 9 12
 ,w 14 18
16
19
2 7
4
2a 2b
 12 15 11

3
, v (a)
wi
ts i
 ,u (b) j
vi
i
ui
k i (c)
Figure (5) Specialization of Hexahedron[9]:

(a) Isoparametric hexahedron (b) Rectangular parent as shell element
before constraints(c) Constrained nodal displacements.
6
be related to the five nodal displacements in figure (6) (c) by the

following 9×5 constraint matrix:
1 0 0 0 0 
0 1 0 0 0 
 
0 0 1 0 0 
 ts 
1 0 0 0
2 
  ts 
Gai   0 1 0
2
0 

………………………………………. (1)
0 0 1 0 0 
  ts 
1 0 0 0
2 
 ts 
0 1 0 0 
 2 
0 0 1 0 0 
Where, "ts" is the element thickness.

Similarly, the six nodal translations in figure (6) (b) are related to
the five nodal displacements in figure (6) (c) by the 6×5 constraint
matrix.
 ts 
1 0 0 0
2
 ts 
0 1 0 0
 2 
Gbi   0 0 1 0 0  …………………………………………. (2)
ts 
1 0 0 0
2
 ts 
0 1 0 0
 2 
0 0 1 0 0
If each of these constraint materials in four locations were applied, the

number of nodal displacements could be reduced from 49  46  60 to
85  40 . In the following articles, the direct formulations of shell
element in the manner described by Cook [10] will be pursued.
7
6
j
z 4 5
y ts / 2
3
x
i
2
1
ts / 2
k 8
7
9
(a)
3
j 2
k
5
4
6
(b)
1 5
i 3
2 4
k (c)
Figure (6) Nodal displacement: (a) Corner of rectangular element (b)

Mid-edge of rectangular element (c) Node of shell element, [9].
8
4- Geometric Definition of the Element

Figure (7) shows the geometric layout of the shell element, in
which the global coordinates of any point take the form,
x 8  xi  8   3i 
    ts  
 y    N i  yi    N i m3i  ………………………………………. (3)
 z  i 1  z  i 1 2 
   i  n3i 
Where, N i  ,  represents the interpolation shape functions, and they are

illustrated and explained in appendix (A1).
In addition, the terms  3i , m3i , n3i are the direction cosines of vector V3i that
is normal to the middle surface and spans the thickness "t s" of the shell at
node i . Figure (7) (b) shows this vector which is obtained as:
 x j  x k    3i 
   
V3i   y j  y k   m3i t s ……..……………………………………..... (4)
z  z  n 
 j k   3i 
Point j and k in the figure are at the upper and lower surfaces of the shell,
respectively. In a computer program, the direction cosines for V3i must be
given as data.
Generic displacements at any point in the shell elements are taken
to be in the directions of global axes. Thus, the generic displacements
vector is:
u 
 
u   v  …………………………………………………………….. (5)
 w
 
On the other hand, nodal displacements consist of these same translations

(in global directions) as well as two small rotations  i and  i about two
local tangential axes x and y  , as indicated in figure (7).
Hence,
9
4
8

7
5
6
z, w
2 
ts
z , 
(a) wi y
y, v i
vi
x, u i
z ,  ui i
i x
i
v
j v 3i
y
u
i
t v 2i
 i
2
v1i
i
ts
x
k
(b)
Figure (7) (a) Shell element (b) Nodal vectors, [9].
12
 ui 
v 
 i 
 i  wi  …………………………………………………………. (6)
 
 i
  i 
i  1,2,3,.......,8
This represents the nodal displacements vector. Generic displacements in

terms of nodal displacements are:
u  8  ui  8
      i 
 v    N i  vi    N i  i 
ts
 ……………………………….... (7)
w i 1 w  i 1 2  i 
   i
In this formula, the symbol i  denotes the following matrix:

   2i  1i 
i    m2i m1i  ……………………………………………………... (8)
  n2i n1i 
Column 1 in this matrix contains negative values of the direction cosines

of the second tangential vector V2i , and column 2 has the direction cosines
for the first tangential vector V1i (see figure (7)). These vectors are
orthogonal to the vector V3i and to each other. As infinity of vector
directions, normal to a given direction can be generated, a particular
scheme has been devised to ensure a unique definition. This is given in
appendix (A2) [11].
Figure (7) shows the local generic translations u  and v (in the
directions of V1i and V2i ) due to the nodal rotations  i and  i , respectively.
Their values are:
ts ts
u   i & v    i …………………………….………. (9)
2 2
11
Contributions of these terms to the generic displacements at any point are

given by the second summation in equation (7).
The displacement shape functions in equation (7) may be cast into
the matrix form:
 ts ts 
1 0 0   2  2i 
2
 1i 
 
N i   0 1 0   t s m2i ts
 m1i  N i .................................... (10)
 2 2 
0 0 1   t s n  n1i 
ts
 2
2i
2 
i  1,2,......8
In order to isolate terms in sub-matrix N i  that multiplied by  , let:

1 0 0 0 0
N Ai   0 1 0 0 0 Ni ………………….………………………… (11)
0 0 1 0 0
And,
0 0 0   2 i  1i 
N Bi   0 0 0  m2i m1i  s N i ……………………..…………...… (12)
t
2
0 0 0  n2i n1i 
Then,
Ni   N Ai    N Bi  ………………………………………………...… (13)
And, the shape function matrix becomes:
N   N A    NB  ……………………………………………….…… (14)
The last of these formulas will later be used to drive the consistent mass
matrix.
The 3×3 Jacobian matrix required for this element is given by:
12
 x y z 
    
 
J    x y z 
……………………………………………...…. (15)
   
 x y z 
 
    
The concept of Jacobian matrix is shown in appendix (A3). Derivatives in

the Jacobian matrix are found as follows:
x 8
N 8
N t
  i x i   i  s  3i
 i 1  i 1  2
x 8
N 8
N t
  i x i   i  s  3i
 i 1  i 1  2
x 8
t
  N i s l 3i and so on.
 i 1 2
The inverse of J  becomes:

    
 x x x 

J 1  J *     
 ……………………………………...….. (16)
 y y y 
    
 
 z z z 
13
5- Strain Calculations
Certain derivatives of the generic displacements (equation (7)) with
respect to local coordinates are needed. These derivatives are listed in a
column vector of nine terms as follows:
 u 
  
   N i N i N i 
 u    0 0 

 2i 

 1i 
    
 u   N i 0 0 
N i
 2i 
N i
 1i 
      
    0 0 0  N i  2i N i  1i   u i 
 v    
    0 N i N i N i
0  m2 i  m1i   vi 
 v  8      w 
    t i ...........................17 
    0 N i

N i

N i
   i 1  
0

m2 i

m1i   s  
  2 
i
 v   N i m2 i
    0 0 0 N i m1i   t s 
 w   N i N i N   i 
 0 0  n2i  i n1i   2 
    
    
 0 N i N i
n1i 
Ni
 w  0  n2i 
       
   0  N i n2i N i n1i 
  0 0
  w
  
Transformation of these derivatives to global coordinates requires that the
inverse of the Jacobian matrix be applied. Therefore,
 u   u 
 x    
 u  J  0 0   u 
*
    
 y    0 J * 0     ………………………………….…….. (18)
    0 0 J *   .....
 w   w
   
 z    
Multiplying the terms in this equation yields,
14
 u 
 x 
 u 
 
 y   ai 0 0  d i l 2i d i l1i 
 u  b
 z   i 0 0  ei l 2i ei l1i 
 v   ci 0 0  g i l 2i g i l1i   ui 
   
 x 
8 
0 ai 0  d i m2 i d i m1i   vi 
 v   
    0 bi 0  ei m2i ei m1i  wi ......................................................19
 y  i 1  0  g i m2 i

g i m1i   i 
 v   ci 0
 
 z  0 0 ai  d i n2 i d i n1i   i 
 w   
  0 0 bi  ei n2i ei n1i 
 x  0  g i n2 i g i n1i 
w  0 ci
 
 y 
 w 
 
 z 
In which,
Ni * N i
ai  J11
*
 J12
 
N i * N i
bi  J 21
*
 J 22
 
N i * N i
ci  J 31
*
 J 32 …………………………………………. (20)
 
di 
ts
2
ai  J 13
*
Ni 
ei 
ts
2
bi  J 23
*
Ni 
gi 
ts
2

ci  J 33
*
Ni 
The strain displacement vector may be written as:
15
 u 
 x 
 v 
 x   
    y 
 y   w 
    
   z    u z v  ………………………………………………. (21)
 xy    
 yz   y x 
   v w 
 xz   
z y 
 
 w  u 
 x z 
Substituting equation (19) into equation (21) gives,

  B  …………………………………………………………...… (22)
Where,  is the nodal displacement vector (equation (6)), and B is the
strain-displacement matrix. The ith part of matrix B may be written as:
ai 0 0  di  2i d i  1i 
0 bi 0  ei m2i ei m1i 

0  g i n2i 
 
Bi  
0 ci
 ei  2i  d i m2i
g i n1i
ei  1i  d i m1i 
 ………………………. (23)
 bi ai 0
0 ci bi  g i m2i  ei n2i g i m1i  ei n1i 
 
 ci 0 ai  d i n2i  g i  2i d i n1i  g i  1i 
As with the sub-matrix N i  , the terms in sub-matrix Bi  that multiply

 could be isolated to get,
Bi   BAi    BBi ……………………………………………...…….. (24)
Sub-matrices BAi  and BBi  are composed from equations (20) and (23),
but the actual details are omitted. Altogether, one can have
B  BA    BB ………………………………………………….….. (25)
which will be convenient when determining the stiffness matrix for the
shell element.
16
6- Stress Calculations
Stress-strain relationships in the local (primed) axes for isotropic

material take the form:
   D  …………………………………………………………… (26)
Where D is the stress-strain matrix or the elasticity matrix in local axes
[10] or,
1  0 0 0 0 
 x   1 0 0 0 0    x 
    
 y  0 0 0 0 0 0    y 
 z   1  
 

0 0 0 0 0    z  …………………… (27)
 xy  1     xy 
2
 2
1 
 yz  0 0 0 0 0   yz 
   2k  
 zx   1     zx 
 0 0 0 0 0
2k 
In which,  and  are Young's modulus and Poisson's ratio, respectively.

The factor k included in the last two shear terms is taken as 1.2, and its
purpose is to improve the shear displacement approximation [12].
Coordinate transformation is applied to convert matrix D to the global

matrix D  by using the 6×6 strain transformation matrix T  .
Thus,
D  T T DT  …………………………………………………….. (28)
Where,
17
  12 m12 n12  1m1 m1n1 n1 1 

 2 
 2 m22 n22  2 m2 m2 n 2 n2  2 
  23 m32 n32  3 m3 m3 n3 n3  3 
T     ….(29)
 2 1 2 2m1m2 2n1n2  1m2   2 m1 m1n2  m2 n1 n 1  2  n2  1 
2  2m2 m3 2n2 n3  2 m3   3 m2 m2 n3  m3 n2 n 2  3  n3  2 
 2 3 
 2 3  1 2m3 m1 2n3 n1  3 m1   1m3 m3 n1  m1n3 n 3  1  n1 3 
Where 1 ,  2 ,  3 , m1 ,...........etc are directional cosines of the primed axes with

respect to the global axes. The concept of coordinate transformation is
illustrated in appendix (A4) [9]. To evaluate the matrix T  at an
integration point, the directional cosines for vector V1,V2 ,V3 must be found
at that point. This may be done with the following sequence of
calculations:
e1 J1 norm. , e3  J1  J 2 norm. , e2  e3  e1 .
In these expressions, the vector J1 norm. denotes the first row of the
Jacobian matrix normalized to a unit length, and so on.
Equation (28) would be more efficient if the third row and column of
matrix D (corresponding to  z and  z ) were deleted, along with the
third row of matrix T  .
7- Stiffness Matrix [k]e for Shell Element
The stiffness matrix K e could be obtained from calculating the

strain energy U. Applying the principle of variational approach [13], the
strain energy for the element may be written as:
U
1

2 V
 
 eT  e dV …………………………………………...…….. (30)
18
Substituting equations (22) and (26) in equation (30) yields,
1  
U   eT  B DBdV  e …………………………………………. (31)
T
2 V 
Or
U   eT K e  e …………………………………………………….. (32)
1
2
Where, K e is the stiffness matrix and takes the form;
K e   BT DBdV ……………………………………………...... (33)

V
But, dV according to appendix (A5) could be written as
dV  dxdydz  J ddd …………………………………………...… (34)
Where, J is the determinate of the Jacobian matrix. Substituting

equations (25) and (34) into equation (33) yields,
1 1 1
K e     Ba    Bb T DBa    Bb  J ddd …………………… (35)
1 1 1
In calculating the Jacobian matrix J  using equation (3), the symbol  to

the first power appears in certain terms. Zienkiewic [14] suggested that
these terms might be neglected in comparison with terms to which they
were added for the purpose of simplifying integration through the
thickness. Thus, the matrix [J] becomes independent of  , and the explicit
integration through the thickness is possible. The products  Ba T DBb 
and  Bb T DBa  are linear in  and therefore, vanish on integration
19
through the thickness. While, the products Ba T DBa and  2 Bb DBb 
may be integrated with respect to  at once. Thus, equation (35) is
reduced to
K e     2Ba T DBa   2 Bb DBb  J dd ………………………… (36)

1 1
1 1  3 
Hence, the first part of matrix K e in equation (36) is due to the transverse
shearing deformations, whereas the second part is associated with flexural
deformations. To evaluate the integrals in equation (36) numerically,
Gauss-quadrature technique [15] is adopted using two integration points
in each of  and  coordinates. This method is explained in appendix
(A5).
8- Consistent Mass Matrix [M]e for Shell Element
The mass matrix M e might be obtained by calculation the kinetic

energy KE as follows:
For any body of infinitesimal mass dm and velocity vector qe , the kinetic
energy is:
1 T
2
KE  qe qe dm ………………………………………………....….. (37)
Since, dm  dV …………………………………………………..... (38)
Where,  is the mass density. Then,
1
KE  
2 V
qeT qe dV ……………………………………………..…… (39)
But, qe  N e …………………….………………………………… (40)
22
d
Where, e   e ………………………………...……………...….. (41)
dt
Substituting equation (40) gives
1  
KE  eT    N  N dV e ……………………………………...…. (42)
T
2  V 
Or,
KE  eT M e e …………………………………………………….. (43)

1
2
Where, M e is the consistent mass matrix [16] and is defined as:
M e    N T N dV …………………………………………...…… (44)

V
Where, N  is the shape function matrix (equation (14)).
Substituting equations (14) and (34) into equation (44) yields,
1 1 1
M e      N A    N B T N A    N B  J ddd ……………... (45)
1 1 1
Employing the same techniques used in simplifying equation (35) gives,
M e      2N A T N A   2 N B T N B  J dd ……………………….... (46)

1 1
1 1  3 
Hence, the first part of matrix M e consists of the translational inertias,

and the second part gives a rotational (or rotary) inertia. The integrals in
equation (46) are evaluated in the same manner as in integrals equation
(36).
21
9- Equation of Motion for Finite Element
The equation of motion (or dynamic equation) can be derived, using the
energy balance principle which involves that "the summation of the
structure energies is stationary", i.e., the summation of kinetic energy,
dissipation energy, strain energy and potential energy is stationary, or
KE  DE  U  PE  Stationary ……………………………………… (47)
If these energies are defined in terms of a nodal displacement vector  ,

then,

KE  DE  U  PE   0 ………………………….………………. (48)

The first and third terms of equation (47) are obtained by equations (43)
and (32), respectively. Now, the second and the fourth terms will be
created. The dissipation energy DE depends upon the nature of damping,
and for the case of viscous damping, a damping matrix ce can be defined
such that:
DE  eT C e e ……………………………………………………. (49)

1
2
Finally, the potential energy PE (with the absence of body forces) can be
written as:
PE  W   eT Fe t  ………………………………………………… (50)
Where, Fe t  is the nodal forces vector.
Substituting equations (32), (43), (49), and (50) in equation (48) gives,
22
 1 T 
  e M e e  e C e e   e K e  e   e Fe (t )  =0……………. (51)
1 T 1 T T 1 T
e  2 2 2 2 
The derivation of the first term of the upper equation is obtained as

follows:
  1 T  d  e   1  T
  e M e e  
 e  2
 d
  e M e e  
 dt  e e  2  dt
 
M e e  M e e
The other terms can be easily derived to get the final form of the dynamic
equation of finite element.
M e e  C e e  K e e  Fe (t ) ……………………………….……….. (52)
Equation (52) with zero damping becomes,
M e e  K e  e  Fe (t ) ……………………………………………… (53)
23
Appendices
Appendix (A1): The Shape Functions
Shell Element Shape Functions.
In the finite element analysis, the region of interest is subdivided
into a number of sub-regions known as elements, which are defined by
the locations of their nodal points. The main concept here is that the
geometry of the element is defined using the nodal coordinates and the
shape functions, which are used to interpolate the main unknowns (i.e.,
displacement) with an isoparametric formulations in terms of a non-
dimensional element coordinates  , ,  which varies from -1 to +1 over
the element called natural coordinates. This coordinate system is
particularly useful when the adoption of numerical integrations is
considered to evaluate any integrals which are required during the
stiffness matrix calculations for example. Figure (A1.1) shows the
rectangular parent element (a) of the isoparametric quadrilateral element
(b) which is geometrically similar to the shell element used. Since 8-node
elements have been employed, and according to Pascal's triangle, the 8
terms polynomials are assumed for the displacement function as follows
u  c1  c2  c3  c4 2  c5  c6 2  c7 2  c8 2 …………………… (A.1.1)
(And similar polynomials for other displacements)

Several methods could be used in obtaining the displacement shape
function. Hence, a direct substituting method will be used by applying the
above equation to each node in the element. Thus,
u1  c1  c21  c31  c412  c511  c612  c7121  c8112
u2  c1  c2 2  c32  c4 22  c512  c622  c7 222  c8 222
And so on, substituting the values of  i ,i (where i the node number,
i  1,2,…….8) which are listed in table (A.1.1) into the above equations
24
and solve them simultaneously, the values of the constants c1 , c2 ,…..etc.

can be calculated. Substituting these constants into equation (A.1.1), the
displacement shape functions are obtained as follows:
 1

7
4 7 3
  1  1
8 6

y 7
x 1 5 2
  1
(a)

3
4 7
6 
8
y
2
x 1 5
(b)
Figure (A1.1) (a) Rectangular parent element (b) isoparametric element.
25
u  u1 N1  u2 N 2  u3 N 3  u4 N 4  u5 N 5  u6 N 6  u7 N 7  u8 N8 …………… (A.1.2)
Or, it can be written in the form:

8
u   N i ui , and similar for other displacements.
i 1
Where, the shape functions could be written as follows:
N1 
1
1   1        1
4
N2 
1
1   1       1
4
N3 
1
1   1       1
4
N4 
1
1   1        1
4
N5 
1
2
 
1   2 1   
N6 
1
2

1    1   2 
N7 
1
2
 
1   2 1   
N8 
1
2

1    1   2 
These shape functions must satisfy two conditions:
8
1-  N ( , )  1
i 1
i
1 if i j
2- N i ( i , j )  
0 if i j
Table (A.1.1) Nodal coordinates for shell element

i 1 2 3 4 5 6 7 8
i -1 1 1 -1 0 1 0 -1
i -1 -1 1 1 -1 0 1 0
26
The geometric interpolation functions are taken to be the same as the

displacement shape functions obtained. Physically, this means that the
natural coordinates  , ,  are curvilinear, and all sides of the element
become quadratic curves.
Thus,
8 8 8
x   N i xi , y   N i yi , z   N i zi
i 1 i 1 i 1
Fluid Element Shape Functions

Figure (5) (a) in chapter three shows the isoparametric hexahedron
element used in the fluid finite element formulation. Both types of shape
functions for the fluid element could be obtained in the same manner as
for the shell element. Thus, the velocity shape functions N vi  could be
written as :
N vi 
1
1  0 1  0 1   0 0  0   0  2 i  1,2,......8
8
N vi 
1
4
 
1   2 1  0 1   0  i  9,11,17,19
N vi 
1
4
 
1   2 1   0 1   0  i  10,12,18,20
N vi 
1
4
 
1   2 1   0 1  0  i  13,14,15,16
Where,
 0   i ,  0   i ,  0   i
The values of  i ,  i , and  i required in these formulas are given in table

(A1.2).
27
Table (A1.2) Nodal coordinates for fluid element.

i i i i i i i i
1 -1 -1 -1 11 0 1 -1
2 1 -1 -1 12 -1 0 -1
3 1 1 -1 13 -1 -1 0
4 -1 1 -1 14 1 -1 0
5 -1 -1 1 15 1 1 0
6 1 -1 1 16 -1 1 0
7 1 1 1 17 0 -1 1
8 -1 1 1 18 1 0 1
9 0 -1 -1 19 0 1 1
10 1 0 -1 20 -1 0 1
Also, the pressure shape functions N pi  could be written as :
N pi 
1
1  0 1  0 1   0  i  1,2,......8
8
Where,
 0   i ,  0   i ,  0   i
The values of  i ,  i , and  i required in these formulas are given in table

(A1.2).
28
Appendix (A2): Unique Definition of Directions Normal to a

Reactor [56]
If a vector V3 is defined (by its three Cartesian components for
instance), it is possible to erect an infinity of mutually perpendicular
vectors orthogonal to it. Some scheme therefore has to be adopted to
eliminate this choice, and indeed quite arbitrary decisions can be made
here. A convenient scheme adopted in the present work related the choice
to the global x and y axis.
If i for instance is the unit vector along the x axis,
V1  i  V3
makes the vector V1 perpendicular to the plane defined by the direction

V3 and the x axis. As V2 has to be orthogonal to both V1 and V3 , one can
have,
V2  V3  V1
To obtain unit vectors in the three directions, V1 , V2 , and V3 are simply

divided by their scalar lengths, giving the unit vectors:
v1 , v2 , and v3 .
29
Appendix (A3): The Jacobian Matrix [J]

In calculating the element strain, certain derivatives of the generic
displacement u, v, w with respect to the global coordinates x, y, z  are
needed. But, since the shape functions are expressed in terms of the local
coordinates  ,,  , it is useful to use a convenient transformation as
follows:
The chain rule of partial differential calculus for differentiation of shape
functions N  ,,  with respect to  ,  , and  produces:
N N x N y N z
  
 x  y  z 
N N x N y N z
  
 x  y  z 
N N x N y N z
  
 x  y  z 
In matrix form:
 N   x y z   N 
         x 
  
 N    x y z   N 
         y 
 N   x y z   N 
    
         z 
For this arrangement, the terms in the coefficient matrix are easily
obtained by differentiating equation (3) . This array is called the Jacobian
matrix [J] which contains the derivatives of the global coordinates with
respect to the local coordinates. Thus,
 N i   N i 
    x 
   
 N i   J  N i 
    y 
 N   N i 
 i  
    z 
32
Where,
 x y z 
    

x y z 
Jacobian matrix [J] = 
    
 x y z 
 
    
Finally, to find the global derivatives, [J] must be inverted as:

 N i   N i 
 x    
 N   
 i   [ J ]1  N i 
 y    
 N i   N 
   i
 z    
Where, [J]-1 is the inverse of the Jacobian matrix.
31
Appendix (A4): Coordinate Transformation[6]

The major purpose of the coordinate transformation is to permit the
material or element properties, known with reference to one coordinate
system, to be used in another coordinate system.
Let two Cartesian reference frames x y z and x  y  z be arbitrarily
oriented with respect to one another, figure (A4.1). Also, the directional
cosines of the angles between the various axes are tabulated in the same
figure. For example,  1 is the directional cosine of axis x  with respect to
x axis and so on.
y
z
x x
x y z
x 1 m1 n1
y 2 m2 n2
z 3 m3 n3
Figure (A4.1) Cartesian coordinate systems and table of directional

cosines.
32
Any state of stress and strain may be expressed in either coordinate

system as  and  in x y z coordinates or as   and   in x  y  z
coordinates. Stresses  and   are arranged in the order:
 x    x 
   
 y  y 
    
   z  …………... (a)     z  ……………… (b)
 xy   xy 
 yz   yz 
   
 zx   zx 
Also the strains  and   :

 x    x 
   
 y  y 
     
   z  …………..(c)     z  ……………….(d)
 xy   xy 
 yz   yz 
   
 zx   zx 
Stress-strain relationships may be written in either coordinate system, as

  [D] ……………………………………………………………. (A4.1)
Or,
   [D]  ………………………………………………………….. (A4.2)
Where, D  and D  are the stress-strain matrix (see sec.3.3.4) in the either
coordinate system, respectively. Now, the transformation of D  to D 

and vice versa can be implemented through the following approach [13].
For the convince in rotation of axes, the stress vector  may be recast
into the form of a symmetric 3×3 matrix as follows:
 x  xy  xz 
    yx  y  yz  ………………………………………….……… (A4.3)
 zx  zy  z 
 
Then, the rotation-of-axes transformation for stress can be stated as:
33
   R RT ……………………………………………………… (A4.4)

Where, [R] is the rotation matrix and has the form
1 m1 n1 
R   2 m2 n2  ………………………………………………….. (A4.5)
 3 m3 n3 
In this matrix, the terms  1 , m1 and so on, are the directional cosines.
Similarly, the strain vector  may be recast as the symmetric 3×3 matrix:
x  xy  xz 
    yx 
 y  yz  ………………………………………….……… (A4.6)
 zx  zy  z 

For which the rotation transformation is:

   R RT ………………………………………………………. (A4.7)
Now, rewrite the expanded result of equation (A4.7) as:
   T  …………………………………………………………..... (A4.8)
In this equation, the strains are in the forms of equation (A4.c) and (A4.d)
instead of equation (A4.6). The 6×6 strain transformation matrix T  in
equation (A4.8) is as follows:
  12 m12 n12  1m1 m1n1 n1 1 
 2 
 2 m22 n22  2 m2 m2 n 2 n2  2 
  23 m32 n32  3 m3 n3  3 
T    m3 n3
 … (A4.9)
 2 1  2 2m1m2 2n1n2  1m1   2 m1 m1n2  n2 m1 n1 2  n2  1 
2  2m2 m3 2n2 n3  2 m3   3 m2 m2 n3  n2 m3 n2  3  n3  2 
 2 3 
 2 3 1 2m3 m1 2n3 n1  3 m1   1m3 m3 n1  n3 m1 n3  1  n1 3 
The form of the stress transformation matrix T  is derived from the
argument that during any virtual displacement, the resulting increment in
strain energy density U o must be the same regardless of the coordinate
system in which it is computed. Thus,
U o   T     T  ……………………………………………. (A4.10)
34
Then, substituting the transposed incremental from of equation (A4.8)

into equation (A4.10) to obtain:
 T T T     T  ………………………………………………. (A4.11)
Hence, one can conclude that,
   T  ………………………………………………………… (A4.12)
Where,
T   T T ………………………………………………………. (A4.13)
Thus, the stress transformation matrix T  is proven to be the transposed
inverse of the strain transformation matrix T  .
Now, to transform the stress-strain relationships from one set of
coordinates to another, substitute equation (A4.8) and equation (A4.12)
into equation (A4.1) to obtain:
T   DT  ………………………………………………….… (A4.14)
Then, premultiply equation (A4.13) by T 1 and use equation (A4.12) to
find:
  T T DT  …………………………………………………... (A4.15)
Or,
  D …………………………………………………………... (A4.16)
Where,
D  T T DT  ………………………………………………….... (A4.17)
which represents the transformation of D to D  .
The reverse transformation is:
D  T DT T …………………………………………………... (A4.18)
35
Appendix (A5): Gaussian Quadrature[6]

The process of computing the value of a definite integral (see
figure A5.1 (a)) from a set of numerical values of the integral is called
numerical integration.
x2
I x   f ( x)dx ……………………………………………………….. (A5.1)

x1
The problem is solved by representing the integrand by an interpolation

formula and then integrating this formula between specified limits. When
applied to the integration of a function of a single variable, the method is
referred to as mechanical quadrature. The most accurate quadrature
formula in common usage is that of Gauss, which involves unequally
spaced points that are symmetrically placed. To apply Gauss's method,
the variable is changed from x to the dimensionless coordinate  with its
origin at the center of the range of integration, as shown in figure (A5.1
(b)). The expression for x in term of  is
x
1
1   x1  1   x2  ……………………………………………. (A5.2)
2
Substitution of equation (A5.2) into the function in equation (A5.1) gives,
f ( x)   ( ) ………………………………………………………..... (A5.3)
Also,
1
dx  ( x 2  x1 )d ………………………………………………….... (A5.4)
2
Then, substituting equations (A5.3) and (A5.4) into equation (1) and
changing the limits of integration yields,
1
1
I x  ( x 2  x1 )   ( )d …………………………………………...… (A5.5)
2 1
Gauss's formula for determining the integral in equation (A5.5) consists

of summing the weighted values of  ( ) at n specified points as follows:
36
f (x )
f (x )
0 x1 x2 x
(a)
 ( )
 ( )
(b)

  1  0  1
Figure (A5.1) Gaussian quadrature.
x  r
d d
 
y
dA
xb y
d
 r
d

r
r d
 
y
d
x
r
xb

x
xb
d
y r 
xb r d
j 
xb
k x
xb i xb
xb
z
xb
Figure (A5.2) Infinitesimal area in natural coordinates.
37
1 n
I     ( )d   R j  ( j )
1 j 1
Or,
I  R1 (1 )  R2 (2 )  ............... Rn (n ) …………………………..… (A5.6)
In this expression,  j is the location of integration point j relative to the
center, R j is a weighting factor for point j , and n is the number of points

at which  ( ) is to be calculated. The values of these parameters are listed
in table (A5.1).
For quadrilaterals in Cartesian coordinates, the type of integration to be
performed is:
y2 x2
I   f ( x, y)dxdy ………………………………………………… (A5.7)

y1 x1
However, this integral is more easily evaluated if it is first transformed to

the natural coordinates for a quadrilateral. One can accomplish this by
expressing the function f in terms of  ,  and using the limits -1 to 1
for each of the integrals. In addition, the infinitesimal area dA  dxdy must
be replaced by an appropriate expression in terms of d and d . For this
purpose, figure (A5.2) shows an infinitesimal area dA in the natural
coordinates. Vector r locates a generic point in the Cartesian coordinates
x and y , as follows:
r  x  y  xi  yi …………………………………………………..... (A5.8)
The rate of change of r with respect to  is :

r x y
 i j …………………………………………………...… (A5.9)
  
Also, the rate of change of r with respect to  is:

r x y
 i j ………………………………………………….... (A5.10)
  
38
Table (A5.1) Coefficients for Gaussian quadrature.

n ± i Ri
1 0.0 2.0
2 0.5773502692 1.0
3 0.7745966692 0.5555555556
0.0 0.8888888889
4 0.8611363116 0.3478548451
0.3399810436 0.6521451549
5 0.9061798459 0.2369268851
0.53884693101 0.4786286705
0.0 0.5688888889
6 0.9324695142 0.1713244924
0.6612093865 0.3607615730
0.2386191861 0.4679139346
7 0.9491079123 0.1294849662
0.7415311856 0.2797053915
0.4058451514 0.3818300505
0.0 0.4179591837
8 0.9602898565 0.1012285363
0.7966664774 0.2223810345
0.5255324099 0.3137066459
0.1834346425 0.3626837834
39
When multiplied by d  and d , the derivatives in equations (A5.9) and

(A5.10) form two adjacent sides of the infinitesimal parallelogram of area
dA in the figure. This area may be determined from the following vector
triple product:
 r r 
dA   d  d   k ……………………………………………. (A5.11)
   
Substitution of equations (A5.9) and (A5.10) into equation (A5.11)

produces,
 x y x y 
dA    dd ………………………………………… (A5.12)
     
The expression in the parentheses of equation (A5.12) may be written as

a 2×2 determinate. That is,
x y
 
dA  dd  J dd ……………………………………….. (A5.13)
x y
 
In which J is the determinate of the 2 2 Jacobian matrix. Thus, the new

form of the integral in equation (A5.7) becomes,
1 1
I    f ( , ) J dd ……………………………………………….. (A5.14)
1 1
Two successive applications of Gaussian quadrature result in,

n n
I   R j Rk f ( j , k ) J ( j , k ) …………………………………..… (A5.15)
k 1 j 1
Where, R j and Rk are weighting factors for the function evaluated at the
point  j , k  . Integration points for n  1,2,3, and 4 (each way) on a

quadrilateral are illustrated in figure (A5.3).
42
 
 
(a) (b)




(c) (d)
Figure (A5.3) Integration points for quadrilateral (a) n  1 (b) n  2

(c) n  3 (d) n  4 (each way).
For hexahedral in Cartesian coordinates, the type of integral to be

evaluated has the form:
I   f ( x, y, z )dxdydz …………………………………………… (A5.16)
Before integrating, one can rewrite the functions in terms of the natural
coordinates  , , and  and using the limits -1 to 1 for each of the
integrals. In addition, the infinitesimal volume dV  dxdydz must be
41
replaced by an appropriate expression of d , d , and d . By employing

the same procedure used for the quadrilaterals, dV can be written as:
x y z
  
x y z
dV  ddd  J ddd ………………………...… (A5.17)
  
x y z
  
In which, J is the determinate of the 3 3 Jacobian matrix. Hence, the

revised form of the integral in equation (A5.16) becomes,
1 1 1
I     f ( , ,  ) J ddd …………………………………….… (A5.18)
1 1 1
Three successive applications of Gaussian quadrature yield,

n n n
I   R j Rk Rl f ( j , k ,  l ) J ( j , k ,  l ) ……………………...… (A5.19)
l 1 k 1 j 1
Where R j , Rk , Rl are weighting factors for the function evaluated at the
point ( j ,k ,  l ) . Integration points for n  1,2,3 and 4 (each way) are:
1,8,27, and 64, respectively.
42
References
[1] Flugg W., "Stress in shell", Springer-Verlag, 4th ed., New York, 1967.
[2] Hou-Cheng H., "Static and Dynamic Analysis of Plate and Shells",
Springer-Verlag, Berlin, 1989.
[3] Zienkiewiz O.C. and Taylor R.L., "The Finite Element Method",
Butterworth-Heinemann, 5th ed., Oxford, 2000.
[4] Novozhlov V.V., "The Theory of Thin Shells", 2 nd ed., Noordhoff
Ltd., Groningen, Netherlands, 1964.
[5] Vlasov V.Z., "General Theory of shells and its Applications in
Engineering", NASA TTF-99, 1964.
[6] Cowper G.R., Lindberg G.M., and Olson M.D., "A shallow Shell
Finite Element of Triangular Shape", International Journal of
Solution Structures, Vol., 6, No.8, pp. 1133-1156, 1970.
[7] Ahmed S., Itons B.M. and Zienkiewic O.C., "Analysis of Thick and
Thin Shell Structure by Curved Finite Elements", International
Journal of Numerical and Mechanical Engineering, Vol.2, No.3, pp.
419-451, 1970.
[8] Mindlin R.O., "Influence of Rotary Inertia and Shear in Flexural
Motion of Isotropic and Elastic Plates", Journal of Applied
Mechanics, Vol.73, pp.31-38, 1951.
[9] Weaver W.Jr. and Johnston P.R., "Finite Element for Structural
Analysis", Prentice-Hall, Englewood Cliffs, N.J. 1984.
[10] Cook R.D., "Concepts and Applications of Finite Element Analysis",
2nd ed., Wiley, New York, 1981.
[11] Ahmad S., Irons B.M., and Zienkiewic O.C, "A Simple Matrix-
Vector Handling Scheme for Three-Dimensional and Shell
Analysis", IJNME, Vol.2, No.4, pp.509-522, 1970.
43
[12] Timoshenko S.P. and Woinowsky-Krieger S., "Theory of Plates and

Shells", 2nd ed., McGraw-Hill, New York, 1959.
[13] Washizu K., "Variational Methods in Elasticity and Plasticity", 2 nd
ed., Pergamon Press, New York, 1975.
[14] Zienkiewiz O.C., Taylor R.L., and Too J.M., "Reduced Integration
Technique in General Analysis of Plate and Shells", IJNME, Vol.3,
No.2, pp.275-290, 1971.
[15] Scarborough J.B., "Numerical Mathematical Analysis", 9th ed., Johns
Hopkins Press, Baltimore, 1986.
[16] Weaver W. Jr. and Johnston P.R., "Structural Dynamic by Finite

Elements", Prentic-Hall, Englewood Cliffs, N.J., 1987.
44

Derivation of The Shell Element, Ahmed Element, Midlin Element in Finite Element Analysis - Hani Aziz Ameen

Uploaded by

Copyright:

Available Formats

Derivation of The Shell Element, Ahmed Element, Midlin Element in Finite Element Analysis - Hani Aziz Ameen

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Derivation of The Shell Element, Ahmed Element, Midlin Element in Finite Element Analysis - Hani Aziz Ameen

Uploaded by

Copyright:

Available Formats

Derivation of the shell element , Ahmed Element , Dr.

Hani Aziz Ameen

Derivation of the shell element ,

Asst. Prof. Dr. Hani Aziz Ameen

2- Types of Shell Elements:

the disadvantage that there is no coupling between bending and stretching

3- Finite Element Formulation of Shell Element

Figure (4) Specialization of solids [6]:

Figure (5) Specialization of Hexahedron[9]:

be related to the five nodal displacements in figure (6) (c) by the

Where, "ts" is the element thickness.

If each of these constraint materials in four locations were applied, the

Figure (6) Nodal displacement: (a) Corner of rectangular element (b)

4- Geometric Definition of the Element

Where, N i  ,  represents the interpolation shape functions, and they are

On the other hand, nodal displacements consist of these same translations

This represents the nodal displacements vector. Generic displacements in

In this formula, the symbol i  denotes the following matrix:

Column 1 in this matrix contains negative values of the direction cosines

Contributions of these terms to the generic displacements at any point are

In order to isolate terms in sub-matrix N i  that multiplied by  , let:

The concept of Jacobian matrix is shown in appendix (A3). Derivatives in

The inverse of J  becomes:

Substituting equation (19) into equation (21) gives,

As with the sub-matrix N i  , the terms in sub-matrix Bi  that multiply

Bi   BAi    BBi ……………………………………………...…….. (24)

B  BA    BB ………………………………………………….….. (25)

Stress-strain relationships in the local (primed) axes for isotropic

   D  …………………………………………………………… (26)

In which,  and  are Young's modulus and Poisson's ratio, respectively.

Coordinate transformation is applied to convert matrix D to the global

D  T T DT  …………………………………………………….. (28)

  12 m12 n12  1m1 m1n1 n1 1 

Where 1 ,  2 ,  3 , m1 ,...........etc are directional cosines of the primed axes with

e1 J1 norm. , e3  J1  J 2 norm. , e2  e3  e1 .

7- Stiffness Matrix [k]e for Shell Element

The stiffness matrix K e could be obtained from calculating the

Substituting equations (22) and (26) in equation (30) yields,

Where, K e is the stiffness matrix and takes the form;

K e   BT DBdV ……………………………………………...... (33)

But, dV according to appendix (A5) could be written as

dV  dxdydz  J ddd …………………………………………...… (34)

Where, J is the determinate of the Jacobian matrix. Substituting

In calculating the Jacobian matrix J  using equation (3), the symbol  to

and  Bb T DBa  are linear in  and therefore, vanish on integration

K e     2Ba T DBa   2 Bb DBb  J dd ………………………… (36)

The mass matrix M e might be obtained by calculation the kinetic

Since, dm  dV …………………………………………………..... (38)

Where,  is the mass density. Then,

But, qe  N e …………………….………………………………… (40)

Substituting equation (40) gives

KE  eT M e e …………………………………………………….. (43)

Where, M e is the consistent mass matrix [16] and is defined as:

M e    N T N dV …………………………………………...…… (44)

Where, N  is the shape function matrix (equation (14)).

Substituting equations (14) and (34) into equation (44) yields,

Employing the same techniques used in simplifying equation (35) gives,

M e      2N A T N A   2 N B T N B  J dd ……………………….... (46)

Hence, the first part of matrix M e consists of the translational inertias,