Sie sind auf Seite 1von 16

3.

SHAPE FUNCTIONS AND DISCRETIZATION

We will discuss the element types, shape functions and discretization in this chapter.
It is important to be able to select an element type which is most suitable for the
problem of interest, and to determine the shape functions for the chosen element type.
Finally, automatic mesh generation techniques are, in practical sense, also important to
finite element analysis applications.

3.1 Element Types

i) One-dimensional Elements

linear quadratic cubic

ii) Two-dimensional Elements

linear quadratic cubic


triangular triangular triangular

Linear linear quadratic cubic


Rectangle quadrilateral quadrilateral quadrilateral

© 2005 by T. H. Kwon 90
iii) Three-dimensional Elements

Tetrahedron Right Prism Hexahedron

Element types depend on the followings:


z Shape (1-D, 2-D, 3-D, triangular, quadrilateral, tetrahedron, etc.)
z Number and type of nodes (3-node, 4-node, etc.)
∂φ ∂φ
z Type of nodal variables (φ, , , etc.)
∂x ∂y
z Type of interpolation functions

It may also be noted that degree of freedom for each node can vary depending on the
problem, for instance,

φ : 1 d.o.f. per each node


∂φ ∂φ
φ, , : 3 d.o.f. per each node
∂x ∂y
u, v : 2 d.o.f. per each node
u , v, p : 3 d.o.f. per each node
u , v, σ xx , σ xy , σ yy : 5 d.o.f. per each node

Sometimes, one may want to introduce somewhat special element type for which each
node may have different d.o.f. (e.g. 2 d.o.f. for vertex node, 1 d.o.f. for mid-side node).

© 2005 by T. H. Kwon 91
3.2 Shape Functions, Interpolation

5
6 4
1 2 3

φ e = N iφ i

i : element nodal numbering


φ e : approximation of φ over an element

φ i : nodal value of φ

N i : shape function
(xi , yi ) : nodal coordinates

3.2.1 Polynomial Approximation

1-Dimensional case:

φ = α 0 + α1 x + α 2 x 2 + L + α n x n : complete n-th order

polynomial in 1-D
n
or Pn ( x) = ∑ α i x i
i =0

Tn(1) = n + 1 : number of terms for complete

n-th order polynomial in 1-D

2-Dimensional case:

φ = α 1 + α 2 x + α 3 y + α 4 x 2 + α 5 xy + α 6 y 2 + L + α k x i y j + L

© 2005 by T. H. Kwon 92
Tn( 2 )
or Pn ( x, y ) = ∑ α k x i y j , i+ j ≤n: complete n-th order
k =1

polynomial in 2-D

(n + 1)(n + 2)
Tn( 2) = : number of terms for complete
2
n-th order polynomial in 2-D

Pascal Triangle

order name Tn( 2)

1 0 1
x y 1 linear 3
2
x xy y2 2 quadratic 6
x3 x2 y xy2 y3 3 cubic 10
x4 x3y x2y2 xy3 y4 4 quartic 15
x3y2 x2y3 5 quintic 21
x3y3 6 hexatic 28
7 septic 36

It may be noted that Pascal triangle is applied to triangular elements whereas the
diamond is applied to quadrilateral elements. The number of parameters, α i , involved

in the complete n-th order polynomial, represented by Pascal triangle, ( Tn( 2) ), is the

same as the number of nodes of the corresponding triangular type of element. The
number of parameters in the diamond is also the same as the number of nodes of the
corresponding quadrilateral element. Compare the terms in the figure above with the
nodes in the figures below.

© 2005 by T. H. Kwon 93
3-Dimensional case:

Tn( 3 )
or Pn ( x, y, z ) = ∑ α l x i y j z k , i + j + k ≤ n: complete n-th order
l =1

polynomial in 2-D

(n + 1)(n + 2)(n + 3)
Tn( 3) = : number of terms for complete
2
n-th order polynomial in 3-D

Generalized Coordinate

α i is called “generalized coordinate”, which has no physical interpretation.


In the subsequent section, it will be explained that the shape function is
better to be used than this generalized coordinate for the interpolation
over an element.

At this point, one should think about how one selects the element type for a certain
problem, in particular, with respect to the order of polynomial in interpolation, in other
words, with respect to the number of nodes within each element. The higher the order
is, more accurate finite element analysis result is expected, but at the cost of
computation time. Certainly, there is a trade-off between the accuracy and the
computational cost. We will discuss this matter in more detail later.

Geometric Isotropy

Field variable representation should remain unchanged under a coordinate


transformation. This fact is called “Geometric Isotropy” This concept is applied to,
for instance, temperature, displacement, strain field, etc.
The guideline to ensure the geometry isotropy is as follows:
1. complete polynomial
2. incomplete polynomial with appropriate terms to preserve “symmetry”
e.g.

P ( x, y ) = α 1 + α 2 x + α 3 y + α 4 x 2 + α 5 xy + α 6 y 2 + α 7 x 3 + α 10 y 3
P ( x, y ) = α 1 + α 2 x + α 3 y + α 4 x 2 + α 5 xy + α 6 y 2 + α 8 x 2 y + α 9 xy 2

© 2005 by T. H. Kwon 94
3.2.2 Relationship between Generalized Coordinate and Shape Function

Field variable was represented in terms of polynomial series with “Generalized


Coordinate” having no physical meaning. From this representation, one can derive a
representation of a field variable in terms of interpolation functions and physical degree
of freedom.

y y

x x

Tn( 2 ) ( n +1) 2
φ = ∑α k x y , i + j ≤ n
i j
φ= ∑α k x i y j , i ≤ n, j ≤ n
k =1 k =1
Tn( 2 ) ( n +1) 2
= ∑ N i ( x, y )φ i
k =1
= ∑ N ( x, y)φ
k =1
i i

⎛ e.g. φ = α 1 + α 2 x + α 3 y + α 4 xy ⎞
φ = ⎣P ⎦{α } ⎜⎜ ⎟⎟
⎝ ⎣P ⎦ = ⎣1 x y xy ⎦ ⎠
φ i at (xi , y i ) is the nodal value at node i :

⎛ ⎡1 x1 y1 x1 y1 ⎤ ⎞
⎜ ⎢ ⎟
{φ } = [G ]{α } ⎜ e.g. [G ] = ⎢M M M M ⎥⎥ ⎟
⎜ ⎢⎣1 x 4 y4 x 4 y 4 ⎥⎦ ⎟⎠

{α } can be determined by the inverse:

{α } = [G ]−1 {φ }

∴ φ = ⎣P ⎦[G ]−1 {φ }

© 2005 by T. H. Kwon 95
i.e. φ = ⎣N ⎦{φ } (= N iφ i )

⎣N ⎦ = ⎣P ⎦[G ]
−1
with

where ⎣N ⎦ is called “Shape Function” or “Interpolation function”

There is a more systematic procedure to obtain ⎣N ⎦ without having to find [G ] .


−1

We will cover this method extensively later. In any case, there are important
characteristics of shape functions based on which one can generate formulae for various
shape functions.

3.2.3 Characteristics of Shape Functions

⎧1 at node i, i.e. at ( xi , y i )
1) N i ( x, y ) = ⎨
⎩0 at any other node

(Q φ = N jφ j , φ i = N j ( xi , yi )φ j = φ i )

5 5
6 6
1 4 1 4
1 1 3
2 3 2

N 1 ( x, y ) N 2 ( x, y )

2) ∑N i
i =1 (for C0 continuity, piecewise continuous element)

For a constant field variable φ, i.e. φ = constant, φ k = constant for any k .

(∴ φ ( x, y ) = ∑ N iφ i = φ k ∑ N i = φ k = constant )
i i

© 2005 by T. H. Kwon 96
3) Note that φ along one side of an element should depend only on the nodal
values associated with the side due to the continuity and compatibility.

j
i j k
i

In general, nodes along a side of an element should be shared with a side of an


adjacent element. However, for some reasons, one may introduce incompatible
element which, for instance, looks like the following figure. [Hughes P.159]

This kind of mesh is somehow useful with a special care for transient region
from a coarse one to a finer one.

We will discuss a special coordinate system, namely natural coordinates, which is


more suitable for representing shape functions than a global coordinate system, before
explaining the method of finding shape functions in detail.

© 2005 by T. H. Kwon 97
3.2.4 Natural Coordinates

Note that N i ( x, y ) in our current form is represented in terms of the nodal


coordinates (xi , y i ) and a global coordinate ( x, y ) . One can have a better form in
terms of so-called “Natural Coordinate”, in particular for triangular type of elements
(or “Normalized Coordinate” for a quadrilateral type of elements )

1) one-dimensional case

l2 l1

x x1 xp x2
l

l1 l
xp = x1 + 2 x 2
l l → Natural coordinate L1 , L2
= L1 x1 + L2 x 2
“Length coordinate”
L1 + L2 = 1

L1 L2

1 1 (looks like a linear interpolation)

x1 x2

2) two-dimensional case
3
A2

L2 A1
L1 Ai
Li = : “Area coordinate”
A
L3
1
2
A3

© 2005 by T. H. Kwon 98
(L1 , L2 , L3 ) ↔ ( x,y )

⎧ x = Li xi

⎩ y = Li y i

The relationship between the natural coordinate and global coordinate system for 2-
dimensional case is as follows:

x = L1 x1 + L2 x 2 + L3 x3 ⎡ x1 x2 x3 ⎤ ⎧ L1 ⎫ ⎧ x ⎫
⎢y ⎪ ⎪ ⎪ ⎪
y = L1 y1 + L2 y 2 + L3 y 3 → ⎢ 1 y2 y 3 ⎥⎥ ⎨ L2 ⎬ = ⎨ y ⎬
1 = L1 + L2 + L3 ⎢⎣ 1 1 1 ⎥⎦ ⎪⎩ L3 ⎪⎭ ⎪⎩1 ⎪⎭

solve for Li :

⎧ L1 ⎫ ⎡ b1 c1 a1 ⎤ ⎧ x ⎫
⎪ ⎪ 1 ⎢ ⎪ ⎪
⎨ L2 ⎬ = ⎢b2 c2 a 2 ⎥⎥ ⎨ y ⎬
⎪ L ⎪ 2∆ ⎢b a 3 ⎥⎦ ⎪⎩ 1 ⎪⎭
⎩ 3⎭ ⎣ 3 c3

where ai, bi, ci’s have already been defined when φ = N iφ i for a triangular element
in plane elasticity.

3) three-dimensional case

4
V1 Vi
V2 Li = : “Volume coordinate”
V

3 x = Li xi
y = Li y i
1
z = Li z i
V4 4
2 1 = ∑ Li
V3 i =1

© 2005 by T. H. Kwon 99
Normalized coordinate for a rectangular element:

(1,1)
ξ (0,1)

ξ=0 ξ=1 ξ
(0,0) (1,0)

or
η

ξ (-1,1) (1,1)

ξ
ξ=-1 ξ=1
(-1,-1) (1,-1)

Similar normalized coordinate system can be defined for 3-Dimensional case.

3.2.5 Classical Interpolation Functions

A) Lagrange Polynomials (Lagrange Interpolation Functions)


B) Hermitian Polynomials

A) Lagrange Polynomials

0 1 2 k n

x x0 x1 x2 xk xn

xi being nodal coordinates.

n-th order polynomial needs (n+1) points.

© 2005 by T. H. Kwon 100


n
φ ( x) = ∑ Lk ( x)φ k
k =0

One should find Lk (x) such that

⎧0 when i ≠ k
Lk ( x i ) = ⎨ , satisfying the characteristics of shape function.
⎩1 when i = k

It is easily proved that the formula below satisfies the requirement described by the
equation above.

n
x − xm ( x − x 0 )( x − x1 ) L ( x − x k −1 )( x − x k +1 ) L ( x − x n )
Lk ( x ) = ∏ = (3.1)
m =0 x k − x m ( x k − x0 )( x k − x1 ) L ( x k − x k −1 )( x k − x k +1 ) L ( x k − x n )
m≠ k

It may be noted that Lk (x) is an n-th order polynomial.

n
Question: Is ∑L
k =0
k ( x) = 1 ? i.e. ∑N i
i =1

Answer: yes.
n

∑L
k =0
k ( x) is an n-th order polynomial and is equal to unity at (n+1) points.

The only n-th order polynomial being equal to unity at (n+1) points is just
one over the whole domain of x.

The interpolation via Lagrange polynomials is demonstrated below:

φ0 φ1 φ2 φ3

x0 x1 x2 x3

© 2005 by T. H. Kwon 101


3
φ ( x) ≈ φ ( x) = ∑ Li ( x)φ i = ⎣L ⎦{φ } = ⎣N ⎦{φ }
~
i =0

∴ N i ( x) = Li ( x)

Li(x) look like the following figures.

L0(x)

L1(x)

L2(x)

L3(x)

© 2005 by T. H. Kwon 102


It may be noted that, instead of global coordinate x, one can use normalized (local)
coordinate ξ in the Lagrange interpolation function.

0 1 2 k n

ξ ξ0 ξ1 ξ2 ξk ξn

The Lagrange interpolation for 1-dimensional case can be easily extended to 2-


dimensional case:

J=n i=14
(I,J)=(3,2)
N i = N IJ = LI (ξ ) L J (η )
14
6
J=0
1 2 3 4 5
I=0 I=n

3-dimensional extension:

N i = N IJK = LI (ξ ) LJ (η ) LK (ς )

B) Hermitian Interpolation

Basic type of interpolation used to represent a function in terms of values of both


the function itself and derivatives of the function.
One method uses end values at only (2 nodes) (e.g. beam bending analysis).
General form for two end nodes and derivatives up to order m-1:

φ = N 1φ1 + N 2φ1′ + N 3φ1′′ + L + N mφ1( m −1)


(3.2)
+ N m +1φ 2 + N m + 2φ 2′ + N m + 3φ 2′′ + L + N 2 mφ 2( m −1)

which is good for Cm-1-interelement continuity. Note that this interpolation function is

© 2005 by T. H. Kwon 103


(2m-1)-th order polynomials.
Alternatively, one can introduce m points and make use of function and its first
order derivatives at each node:

φ = N 1φ1 + N 2φ1′ + N 3φ 2 + N 4φ 2′ + L + N 2 m −1φ m + N 2 mφ m′ (3.3)

which is good for C1-interelement continuity.


More generally, one can further introduce n points and make use of function and its
derivatives up to (m-1)-th order at each node as depicted below:

φ1 , φ1′,L , φ1( m −1) φ n , φ n′ , L, φ n( m −1)

1 2 n

In this case, the interpolation can be expressed by

φ = N 1φ1 + N 2φ1′ + L + N mφ1( m −1) + N m +1φ 2 + N m + 2φ 2′ + L + N 2 mφ 2( m −1)


(3.4)
+ LL + N ( n −1) m +1φ n + N ( n −1) m + 2φ n′ + L + N nmφ n( m −1)

In this case, there are total degree of freedom of nm and there are nm number of shape
functions. Each shape function should satisfy nm requirements1 for each degree of
freedom. In this regard, each shape function is described in terms of (nm-1)-th order
polynomial.
Because of the variety of strategies as described above, no fixed formulae for
Hermitian interpolation are available. Instead, just use generalized parameters
approach as described below.

N i = a1 + a 2 x + a3 x 2 + L + a nm x nm −1

which can be determined by characteristics of shape functions.


[Huebner & Thornton Sect. 5.6.2]

1
Note that the shape function corresponding to a degree of freedom should provide itself at the node
when the corresponding derivative is taken for the expression of φ. Other order of derivatives at the node
should be zero and any order of derivatives are zero at other nodes. As a result, total nm requirements

© 2005 by T. H. Kwon 104


Example : Beam bending analysis (m=2, C1 –continuity)

θ1 θ2
x ξ
d 1 d
y1 y2 =
dx l e dξ
ξ=0 ξ=1

φ1 = y1 , φ 2 = y2
dy dy
φ1′ = , φ 2′ =
dx 1 dx 2
= θ1 = θ2

y (ξ ) = N 1φ1 + N 2φ1′ + N 3φ 2 + N 4φ 2′

N i (ξ ) = a1(i ) + a 2( i )ξ + a3(i )ξ 2 + a 4(i )ξ 3

Requirements:

1 dN 1 (0) 1 dN 1 (1)
for i=1 : N 1 (0) = 1, = 0, N 1 (1) = 0, =0
l e dξ l e dξ
1 dN 2 (0) 1 dN 2 (1)
for i=2 : N 2 (0) = 0, = 1, N 2 (1) = 0, =0
l e dξ l e dξ
1 dN 3 (0) 1 dN 3 (1)
for i=3 : N 3 (0) = 0, = 0, N 3 (1) = 1, =0
l e dξ l e dξ
1 dN 4 (0) 1 dN 4 (1)
for i=4 : N 4 (0) = 0, = 0, N 4 (1) = 0, =1
l e dξ l e dξ

N1 N3
⎛ 1⎞
N 1 = 2(ξ − 1) ⎜ ξ + ⎟
2

⎝ 2⎠
N 2 = l e (ξ − 1) ξ
2 N2

N 3 = ξ 2 (−2ξ + 3)
N 4 = l eξ 2 (ξ − 1)
N4

* Check if this interpolation function can describe the rigid body rotation.

© 2005 by T. H. Kwon 105

Das könnte Ihnen auch gefallen