Professional Documents
Culture Documents
S.S. Bhavikatti - Finite Element Analysis-New Age Publications (Academic) (2004) - 232-241
S.S. Bhavikatti - Finite Element Analysis-New Age Publications (Academic) (2004) - 232-241
13
Isoparametric Formulation
13.1 INTRODUCTION
The various elements so far we have seen are having straight edges. To take care of curved boundaries refined
meshes are to be used when straight edged elements are employed. Even with refined meshes analysts were
not happy with the results since unnecessary stress concentrations are introduced. Higher order elements also
do not overcome the problem of suitably approximating curved boundaries. The isoparametric concept brought
out by Taig [1] and latter on generalized by B.M. Irons [2] revolutionized the finite elements analysis and it
also helped in properly mapping the curved boundaries. They brought out the concept of mapping regular
triangular and rectangular elements in natural coordinate system, to arbitrary shapes in global system as
shown in Fig. 13.1. In this chapter method of coordinate transformation of natural coordinates to global
coordinate system is presented. The terms isoparametric, super parametric and subparametrics are defined.
The basic theorems on which isoparametric concept is based are explained and need for satisfying uniqueness
theorem of mapping is presented. Assembling of stiffness matrix is illustrated. For assembling stiffness matrix
integration is to be carried out numerically. The Gaussion integration technique which is commonly employed
is explained briefly. To make the procedure clear few small numerical problems are illustrated and lastly
application to structural engineering problems is presented.
Parent elements in natural Mapped element if global
coordinate system system
x
0
y
h 3 (x3, y3)
4 (–1, 1) 4 (x4, y4)
3 (1, 1)
P (,x) y
P (x,h )
2 (x2, y2)
1 (– 1, –1) 2 (1, –1) 1 (x1, y1)
x
0
y
h 7 3
4 7 3 4 P (,x) y Quadratic
curve
P ( x,h ) 6
8 x 8
6
5 2
1 5 2 1
x
0
y
h
4 10 9 4 10
3 9 3
P ( x,h )
11 8 11 8
P (,x) y
x
12 7 12 7 Cubic curves
1 2 1 5 6 2
5 6
x
0
x
0
x
0
where N are shape functions and (x)e are the coordinates of nodal points of the element. The shape functions
are to be expressed in natural coordinate system.
For example consider mapping of a rectangular parent element into a quadrilateral element:
y
3 (x3, y 3)
4 (–1, 1) 4 (x4, y 4)
3 (1, 1)
P (,x) y
P ( )
2 (x2, y2)
1 (– 1, –1) 2 (1, –1) 1 (x1, y1)
x
0
(a) Parent element (b) Mapped element
The parent rectangular element shown in Fig. 13.2 (a) has nodes 1, 2, 3 and 4 and their coordinates are
(–1, –1), (–1, 1), (1, 1) and (1, –1). The shape functions of this element are
N1 =
b1 − ξ g b1 − ηg , N = b1 + ξg b1 − ηg
2
4 4
N3 =
b1 + ξg b1 + ηg and N = b1 − ξg b1 + ηg
4
4 4
P is a point with coordinate (ξ , η ). In global system the coordinates of the nodal points are
(x1, y1), (x2, y2), (x3, y3) and (x4, y4)
To get this mapping we define the coordinate of point P as
x = N1 x1 + N2 x2 + N3 x3 + N4 x4
and y = N1 y1 + N2 y2 + N3 y3 + N4 y4
Noting that shape functions are such that at node i, Ni = 1 and all others are zero, it satisfy the coordinate
value at all the nodes. Thus any point in the quadrilateral is defined in terms of nodal coordinates.
Similarly other parent elements are mapped suitably when we do coordinate transformation.
A A
B
x
0
Fig. 13.3
It may be observed that in the parent element, for any point on edge AB, shape functions Ni = 0 for nodes
not on the edge and Ni exists for nodes on the edge. Hence the final function is the same for the common edge
AB in any two adjacent elements, when we give the same coordinate values for the nodes on common edge.
Hence edge AB is contiguous in the adjacent elements.
Isoparametric Formulation 223
Theorem II: It states, if the shape functions used are such that continuity of displacement is represented in
the parent coordinates, then the continuity requirement, will be satisfied in the isoparametric elements also.
The proof is same as for theorem 1.
Theorem III: The constant derivative conditions and condition for rigid body are satisfied for all isoparametric
elements if,
∑N = 1 i
u= ∑N u i i
∴u = ∑ N bα
i 1+ α 2 xi + α 3 yi + α 4 zi g
= α1 ∑N + α ∑N x + α ∑N y + α ∑N z
i 2 i i 3 i i 4 i i
∑N i xi = x
∑N y = yi i
∑N z = z i i
∴u = αi ∑N + α i 2x + α 3y + α 4z …(13.2)
Hence if equation 13.2 has to represent equation 13.1 uniquely, then
∑N = 1 i
The shape functions developed in natural coordinate systems satisfy this requirement. Hence they can be
safely used for isoparametric representation. This theorem is known as convergence criteria for isoparametric
elements.
u = [ N ] {δ }e , x = [N] {x}e and y = [N] {y}e where [N] is quadratic shape function of serendipity family.
The element in which more number of nodes are used to define geometry compared to the number of nodes
used to define displacement are known as superparametric element. One such element is shown in Fig. 13.5
(b) in which 8 nodes are used to define the geometry and displacement is defined using only 4 nodes. In the
stress analysis where boundary is highly curved but stress gradient is not high, one can use these elements
advantageously.
Figure 13.5 (c) shows a subparametric element in which less number of nodes are used to define geometry
compared to the number of nodes used for defining the displacements. Such elements can be used
advantageously in case of geometry being simple but stress gradient high.
(x4, y 4)
0
(x2, y2)
1 (– 1, –1) 2 (1, –1) (x1, y 1)
x
0
Ni =
b1 + ξξ g b1 + ηη g
i i
…(13.3)
4
i.e., N1 =
b1 − ξ g b1 − ηg , N = b1 + ξ g b1 − ηg
2
4 4
N3 =
b1 + ξg b1 + ηg and N = b1 − ξg b1 + ηg
4
4 4
We use the above functions for defining the displacement as well as for defining the geometry of any
point within the element in terms of nodal values.
226 Finite Element Analysis
R| x U|
1
|| y ||
1
TyW N 0 N1 0 N2 0 N3 0 N Q| |
2 …(13.4)
||xM ||
4
|T y |W
4
The above relation helps to determine the (x, y) coordinates of any point in the element when the
corresponding natural coordinates ξ and η are given.
We are also using the same functions for defining the displacement at any point in the element
R|u U|1
||v ||
1
RSuUV = LMN
1 0 N2 0 N3 0 N4 0O| |
P Sv V
u2
TvW N 0 N1 0 N2 0 N3 0 N Q| |
2 …(13.5)
||uM ||
4
|Tv |W
4
In assembling the stiffness matrix we need the derivatives of displacements with respect to global x, y
system. It is easy to find derivatives with respect to local coordinates ξ and η but it needs suitable assembly
to get the derivatives w.r.t. to global Cartesian system.
The relationship between the coordinates can be computed using chain rule of partial differentiation.
Thus,
∂ ∂x ∂ ∂y ∂
= +
∂ξ ∂ξ ∂x ∂ξ ∂y
∂ ∂x ∂ ∂y ∂
= +
∂η ∂η ∂x ∂η ∂y
R| ∂ U| R| ∂x ∂y U| R ∂ U R| ∂ U|
| |
i.e., S| ∂ξ∂ V| = S| ∂ξ∂x ∂ξ
∂y V| S| ∂∂x V| = J S|∂∂x V| …(13.6)
|T∂η |W |T∂η ∂η |W |T ∂y |W |T ∂y |W
R| ∂x ∂y U
∂ξ |
J =S
∂y V
∂ξ
||∂η |
where …(13.7)
∂x
T ∂η |W
Isoparametric Formulation 227
The matrix [J] shown above is called Jacobian matrix. It relates derivative of the function in local coordinate
system to derivative in global coordinate system. In case of three dimensional problem it is given by
LM ∂x ∂y ∂z OP
MM ∂ξ∂x ∂ξ
∂y
∂ξ
∂z
PP
J =
MM ∂η ∂η ∂η PP
MM ∂ς∂x ∂y ∂z
PP
N ∂ς ∂ς Q
Now going back to isoparametric quadrilateral element,
Let
J =
LMJ
11 J12 OP
NJ 21 J 22 Q
∂x ∂y
Where J11 = J12 =
∂ξ ∂ξ
∂x ∂y
J 21 = J 22 =
∂η ∂η
4
we know, x= ∑N x i i = N1 x1 + N2 x 2 + N3 x 3 + N4 x 4
i =1
∂x ∂N1 ∂N 2 ∂N 3 ∂N 4
∴ J11 = = x1 + x2 + x3 + x4
∂ξ ∂ξ ∂ξ ∂ξ ∂ξ
Similarly J12, J21 and J22 can be assembled.
Then we get
LM ∂N x ∂N y OP
4 4
M∑ ∂ξ ∑ ∂ξ PP
i i
i i
J=M
i =1 i =1
MM∑ ∂N x ∑ ∂N y PP
4 4
…(13.8)
i i
N ∂η ∂η
Q
i i
i =1 i =1
For any specified point the above matrix can be assembled. Now,
R| ∂ U| R| ∂ U|
S| ∂ξ∂ V| = J S|∂∂x V|
|T∂η |W |T ∂y |W
228 Finite Element Analysis
R| ∂ U| R| ∂ U|
∴ S ∂∂x V = S| ∂ξ∂ V|
−1
|| ∂y ||
J
T W |T ∂η |W
R∂ U
LJ
=M
*
J12* OP |S ∂ξ |V
MNJ PQ | ∂ |
11
* …(13.9)
J *22
T|∂η W|
21
* *
where J11 , J12 , J *21 and J 22
*
are the elements of Jacobian inverse matrix. Since for a given point Jacobian
matrix is known its inverse can be calculated and Jacobian inverse matrix is assembled. With this transformation
relation known, we can expresses derivatives of the displacements as shown below:
R| ∂u U| LM J *
11
*
J12 0 0OP R| ∂u U|
|| ∂∂ux || MM PP || ∂ξ∂u ||
| ∂y | M J PP |S∂η |V
* * 0 0
J 22
=S V=M
21
|| ∂∂vx || M 0 J P | ∂ξ |
∂v
|| ∂v || MM PP || ∂v ||
0 * * …(13.10)
J11 12
T ∂y W MN 0 0 *
J 21 J PQ |
*
22 T∂η |W
The strain displacement relation is given by
R|∂u U|
R| ε U| L1 |
0O | u |
∂x
∂ |
V| = MM0 P | ∂y |
0 0
lεq = S ε 1P S V
x
|Tγ 0PQ | |
0 0
W MN0
y
∂v
xy 1 1
||∂∂xv ||
|T ∂y |W
R| ∂u U|
LMJ | ∂ξ
|
0 O | ∂u |
P | ∂η |
* *
J12 0
J PS V
11
=M0 * *
MNJ J PQ | |
0 J 21 22
∂v
|| ∂ξ || …(13.11)
*
21 J *22 *
J11 *
12
|T ∂η
∂v
|W