Notes:

z The effect of completeness in polynomial model :

If ⎣N ⎦ is not complete in all the lower powewrs of x , then we cannot achieve a
term-by term matching of the Taylor expansion. In other words,
~
∆ − ∆ ≤ M 0 h n +1 does not hold any longer.

z Effect of singularity : Near singular points, Taylor expansion is not adequate

→ The above discussion breaks down.

3.5 Isoparametric Elements

Coordinate system can be transformed by the coordinate transformation (Mapping)

as follows:

η (ξ ,η )
y
6
7 1
5

8
4
ξ
mapping 1
1 3
2
(ξ ,η )
x
1
(x,y)

-1 1
ξ
Called “parent element”

-1
( x, y ) ↔ (ξ ,η )
⎧ 8

⎪⎪ x = ∑ Fi (ξ ,η ) xi
Coordinate transformation
⎨
i =1
n ←
⎪ y = ∑ Fi (ξ ,η ) y i i.e. Mapping
⎪⎩ i =1

© 2005 by T. H. Kwon 136

Meanwhile, the field variable is interpolated in terms of shape functions

8
φ ( x, y ) = φ (ξ ,η ) = ∑ N i (ξ ,η )φ i
i =1

i) if ⎣F ⎦ = ⎣N ⎦ : isoparametric element → most used

ii) if ⎣F ⎦ is of higher order than ⎣N ⎦ : super parametric element

iii) if ⎣F ⎦ is of lower order than ⎣N ⎦ : subparametric element

eq.
sometimes used in stress
analysis of shell structure

mid-point on straight line

Features of isoparametric formulation

a. If the shape functions are inter- & intra-element continuous in local coordinate, then
intraelement continuity & C 0 -interelement continuity is assured in global
coordinate ( beam case : C 0 -interlement continuity )

b. If the shape functions provide a rigid body motion in local (or natural ) coordinate,
then both a rigid body motion and constant first-derivative state (e.g. constant
strain) are guaranteed in the global system.

In 2-D case :

u = ∑ N i u i , v = ∑ N i vi

x = ∑ N i xi , y = ∑ N i yi

© 2005 by T. H. Kwon 137

i) for a rigid body motions to exist

u0 = ∑ N i ui = ∑ N i u0 = u0 ∑ N i
i i

∴ ∑N i =1 : necessary and sufficient condition

ii) if uniform 1st derivatives (constatant strains) exist in global coordinate, then

u = a + bx + cy

at node i : u i = a + bxi + cyi

multiply by N i & sum over i :

∑ N u = ∑ aN + ∑ bN x + ∑ cN y
i
i i
i
i
i
i i
i
i i

u
= a ∑ N i + b∑ N i xi + c ∑ N i y i
i i i

x y

∴ u = a ∑ N i + bx + cy
i

i.e ∑N i
i = 1 thus gives constant 1st derivatives.

( Note that ∑N ξ = ∑N η = 0
i
i, i, ξ ,η being natural 2 local coordinate )

c. For models of higher order than linear, the geometric mapping may be considered
as a transformation between curvilinear local coordinate & global cartesian
coordinates. Hence, we can represent elements with curved sides very conveniently
in terms of isoparametric shape functions

© 2005 by T. H. Kwon 138

3.6 Mesh Generation Technique

Mesh generator needs to do the following jobs :

1) introduce ( input ) geometry

2) generate nodes with a specific nodal numbering
3) determine nodal connectivity for elements
4) take care of boundary condition (recommended)
5) output data file

3.6.1 Mapped mesh generation

3 subregions

1 2

6
4 5

For each subregion : η

6 6
y 7
5
5
7

8 4
4 8

x
1 2 3
ξ
1 2 3

⎧ 8

⎪⎪ x = ∑
i =1
N i (ξ ,η ) xi
Use the same shape functions as
⎨ 8
⎪ y = ∑ N i (ξ ,η ) xi for field variable
⎪⎩ i =1

Ni given earlier.

© 2005 by T. H. Kwon 139

- given (ξ ,η ) → calculate nodal coordinates ( x, y ) with a constant ∆ξ , ∆η

- Numbering nodes & elements with nodal connectivity.

with this scheme,one can generate various elements such as

- Easy to extend to 3-D

3.6.2 Free mesh generation (I-DEAS)

Put nodes along the boundary and form elements from boundary to inside with the
following basic operations:

Trimming

Notching

etc.

© 2005 by T. H. Kwon 140

3.6.3 Triangulation with charged particle method

- generate boundary nodes with a certain spacing between them

- generate internal nodes by electrostatics (placing many electrically charged
particles and allowing them to move under a certain electric field and settle
down at nodes at an equilibrium state)
- generate elements by triangulation as depicted by the following figure:

convex polygon
(Voronoi polygon)

- Pi : generating points for Dirichlet tessellation (or Voronoi tessellation)

- Connecting generating points sharing the common boundary.
(Watson’s algorithm for Delaunay Triangulation)

[C.D.Musicant, N.A.Langrana, T.H.Kwon, ASME computers in Eng.Vol.3, 163-169(1988)]

[S.W.Sloam, G.T.Houlsby, Adv.Eng.Software Vol6, Bo.4, 192-197(1984)]
[P.S.Rusnock, R.J.Cipra, ASME J.Mechanism Transmission &Automatics in Design Vol.107,
326-333(1985)]

© 2005 by T. H. Kwon 141

3.6.4 Solution of potential equation
[Elliptic Grid Generator: J.L.Steger, R.L.Sorenson, J.Comput.physics 23,405(1979)]

Solve ∇ 2φ = 0 and make use of the solution field.

φ =0

φ =1
∇φ ⋅ ∇ ψ = 0

φ = C1

ψ = C2

Elliptic Grid Generator:

[Elliptic Grid Generator: J.L.Steger, R.L.Sorenson, J.Comput.physics 23,405(1979)]

axξξ − 2bxξη + cX ηη = 0

ayξξ − 2byξη + cyηη = 0

where a = xη2 + y η2 , b = xξ xη + y ξ y η , c = xξ2 + y ξ2

Schwarz-Christoffel method :
[K.P.Sridhar, R.T.Davis, ASME J. of Fluid Eng., vol.107, 330-337, 1985 ]

I
dz
= M ∏ (t − ai ) α i / π
dy i =1

ty
y

zi
αi
x tx
a7 a8 a1 a2 a6

© 2005 by T. H. Kwon 142

 NL
π αm

dz ζ
∏ [sinh 2h
(ζ − bm )] π
= κ exp[(θ + δ) ] mNU=1
dζ 2h π αn
∏n =1
[cosh
2h
( ζ − bn )] π

η
y
αn βn

αm
x ξ
βm

z- plane ζ - plane

3.6.5 Blending Technique [Zienkiewicz, p.207]

A geometry enclosed by four curves can be easily discretized by a blending

technique as long as the equations of the four sides are defined analytically as follows:
(This blending technique can also be used as a method of defining a surface enclosed by
four curves, i.e. as a surface generation technique.)

© 2005 by T. H. Kwon 143

 φ = pζ φ + pξ φ − pξ pζ φ

where pη φ = N 1 (η )φ (ξ ,1) + N 2 (η )φ (ξ ,−1)

pξ φ = N 1 (ξ )φ (1,η ) + N 2 (ξ )φ (−1,η )

pξ pη φ = N 1 (ξ ) N 1 (η )φ (1,1) + N 1 (ξ ) N 2 (η )φ (1,−1)

+ N 2 (ξ ) N 1 (η )φ (−1.1) + N 2 (ξ ) N 2 (η )φ (−1,−1)

N2 N1
where N 1 (ξ) = (1 + ξ) , N 2 (ξ) = (1 − ξ)
2 2

-1 ξ 1

φ(ξ,1) φ(1,η)
φ(-1,η)
φ(1,1)
φ(-1,1)

(-1,1) (1,1) = + -
η φ(-1,-1)
ξ φ(1,-1)

(-1,-1) (1, -1) φ(ξ,-1)

φ = Pη φ + Pξ φ - Pξ Pη φ

© 2005 by T. H. Kwon 144