FEM Zabaras 2DFElementCalculations

MAE4700/5700
Finite Element Analysis for

Mechanical and Aerospace Design
Cornell University, Fall 2009
Nicholas Zabaras
Materials Process Design and Control Laboratory
Sibley School of Mechanical and Aerospace Engineering
101 Rhodes Hall
Cornell University
Ithaca, NY 14853-3801
MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (10/06/2009)
1
Two-dimensional element calculations
• Similarly to 1D, we here also define
the concept of a master element in
natural coordinates (ξ ,η )
• Performing calculations on
elements with curved sides as
shown is difficult in the ( x, y )
coordinates.
– For example, the limits of
integration will vary from element
to element
Can we find a master element Ω l that is mapped
appropriately to each element Ωe ? Then with
transformation Ω l → Ω we can perform all
e
integrations for all elements at the master element.

N. Zabaras (10/06/2009)
2
• We consider the master element to
be a square with: −1 ≤ ξ ,η ≤ 1
• We define the map Ωl → Ωe as follows
x = x(ξ ,η )
Te :
y = y (ξ ,η )
• Note that the boundary segment ξ = 1
of Ωl is mapped on the curved segment
x = x(1,η )
y = y (1,η ) with η as a parameter.
Similarly for the other 3 boundary
segments.

N. Zabaras (10/06/2009)
3
• The complete finite element mesh is seen
as a sequence of transformations
{T1 , T2 ,..., TE } with each of them
mapping Te : Ωl → Ωe , e = 1, 2,..., E
• We need to figure out how we

can transfer all needed calculations
in each element Ωe to the master
element Ωl .
• The plan is to make all these
calculations defined in terms of
the mappings {T1 , T2 ,..., TE }.
N. Zabaras (10/06/2009)
4
Transformation equations
• We assume that the mappings
x = x(ξ ,η ), y = y (ξ ,η )
are differentiable. We can then
write:
∂x ∂x
dx = dξ + dη ,
∂ξ ∂η
∂y ∂y
dy = dξ + dη
∂ξ ∂η
• We can write this as a system of

⎡ ∂x ∂x ⎤
Eqs: dx ⎢ ∂ξ ∂η ⎥ dξ
⎧ ⎫ ⎧ ⎫
⎨ ⎬= ⎢ ⎥ ⎨ ⎬
⎩dy ⎭ ⎢ ∂y ∂y ⎥ ⎩dη ⎭
⎢⎣ ∂ξ ∂η ⎥⎦

Jacobian matrix [ J ]
T

N. Zabaras (10/06/2009)
5
• The Jacobian maps segments
d ξ , dη from Ω
l to Ωe . We define this as the
transpose of the
⎡ ∂x ∂x ⎤ Jacobian matrix [ J ]
⎢ ∂ξ ∂η ⎥
for notational purposes
⎧ dx ⎫ ⎧dξ ⎫
⎨ ⎬= ⎢ ⎥ ⎨ ⎬
⎩dy ⎭ ⎢ ∂y ∂y ⎥ ⎩dη ⎭
⎢⎣ ∂ξ ∂η ⎥⎦

Jacobian matrix [ J ]
T
• Assume the inverse

transformation exists:
l , e = 1, 2,..., E
Te−1 : Ωe → Ω
⎡ ∂y ∂x ⎤
− Determinant of the Jacobian
⎧d ξ ⎫ 1 ⎢ ∂η ∂η ⎥ ⎧ dx ⎫
⎨ ⎬= ⎢ ⎥⎨ ⎬ matrix
⎩dη ⎭ | J | ⎢ − ∂y ∂x ⎥ ⎩ dy ⎭ | J |= det [ J ] = det [ J ] =
T ∂x ∂y ∂x ∂y
−
⎢⎣ ∂ξ ∂ξ ⎥⎦ ∂ξ ∂η ∂η ∂ξ

N. Zabaras (10/06/2009)
6
ξ = ξ ( x, y ) ⎡ ∂y ∂x ⎤
−
⎧d ξ ⎫ 1 ⎢ ∂η ∂η ⎥ ⎧ dx ⎫
Te−1 : ⎨ ⎬= ⎢ ⎥⎨ ⎬
⎩dη ⎭ | J | ⎢ − ∂y ∂x ⎥ ⎩dy ⎭
η = η ( x, y ) ⎢⎣ ∂ξ ∂ξ ⎥⎦
• From the equations above

we can conclude that:
∂ξ 1 ∂y ∂ξ 1 ∂x
= , =− ,
∂x | J | ∂η ∂y | J | ∂η
∂η 1 ∂y ∂η 1 ∂x
=− , =
∂x | J | ∂ξ ∂y | J | ∂ξ

N. Zabaras (10/06/2009)
7
Constructing the transformation Te
• We are interested for differentiable mappings {T1, T2 ,..., TE }
of the form x = x(ξ ,η ), y = y (ξ ,η ) for each element that
do not create gaps or overlapping between elements
and which are easy to construct from the geometry of
each element Ωe .
• Based on an earlier lecture on construction of element
basis functions, suppose that we have in place element
basis functions Nl j (ξ ,η ) in Ωl such that any function
gl (ξ ,η ) can be approximated as:
lg (ξ ,η ) = M l j (ξ ,η ), g = gl (
∑ gjN ξ j ,η j )
j
j =1 N
coordinates of node j
in t he master element
where M is the number of nodes in the master element.

N. Zabaras (10/06/2009)
8
Constructing the transformation Te
• The finite element basis
functions can be used to
construct the mapping Te as
follows:
M
l j (ξ ,η ),
x = ∑ xj N
j =1 Quadratic shape functions
M
l j (ξ ,η ),
y = ∑ yj N
j =1
where ( x j , y j ) are the

coordinates of node j in the
element Ωe . Biquadratic shape functions
• Can you convince yourselves Note that the common side between
that this mapping does not elements e and e+1 is uniquely
defined in terms of
create gaps or element
the nodal coordinates that are shared
overlapping? by the 2 elements.
N. Zabaras (10/06/2009)
9
Geometric interpretation of the Jacobian
• The value of |J| is the ratio of
areas of elements at points
x = x(ξ ,η ), y = y (ξ ,η ) and (ξ ,η ).
dA =| J | d l
A
• We need to select the element

basis functions Nl i and ( xi , yi )
such that |J|>0 (area cannot be
transformed to a line or a point or
turned inside out!).
• This guarantees that the
mapping Te is invertable.
N. Zabaras (10/06/2009)
10
• Let us return to the definition of
|J| and the earlier stated
transformation equations:
∂x ∂y ∂x ∂y
| J |= det J = −
∂ξ ∂η ∂η ∂ξ
∂ξ 1 ∂y ∂ξ 1 ∂x ∂η 1 ∂y ∂η 1 ∂x
= , =− , =− , =
∂x | J | ∂η ∂y | J | ∂η ∂x | J | ∂ξ ∂y | J | ∂ξ
• Applying the mapping:
M M
l j (ξ ,η ), y = ∑ y N
x = ∑ xj N l j (ξ ,η ),
j
j =1 j =1
leads to:
⎛M l j (ξ ,η ) ⎞⎛ M
∂N l j (ξ ,η ) ⎞ ⎛ M
∂N l j (ξ ,η ) ⎞⎛ M
∂N l j (ξ ,η ) ⎞
∂N
| J |= det J = ⎜ ∑ x j ⎟⎜ ∑ y j ⎟ −⎜ ∑ xj ⎟⎜ ∑ y j ⎟
⎝ j =1 ∂ ξ ⎠⎝ j =1 ∂ η ⎠ ⎝ j =1 ∂ η ⎠⎝ j =1 ∂ξ ⎠
∂ξ 1 M l j (ξ ,η ) ∂ξ
∂N 1 M l j (ξ ,η ) ∂η
∂N 1 M l j (ξ ,η ) ∂η
∂N 1 M l j (ξ ,η )
∂N
= ∑ yj , =− ∑ xj , =− ∑ yj , = ∑ xj
∂x | J | j =1 ∂η ∂y | J | j =1 ∂η ∂x | J | j =1 ∂ξ ∂y | J | j =1 ∂η

N. Zabaras (10/06/2009)
11
Example problem
• Consider the following master element.
• The basis functions for this element are:

l1 (ξ ,η ) = 1 (1 − ξ )(1 − η )
N
4
l 2 (ξ ,η ) = 1 (1 + ξ )(1 − η )
N
4
l 3 (ξ ,η ) = 1 (1 + ξ )(1 + η )
N
4
l 4 (ξ ,η ) = 1 (1 − ξ )(1 + η )
N
4

N. Zabaras (10/06/2009)
12
Example problem
• Examine if the following mapping is invertable.
T1
T1−1 ?
M
l j (ξ ,η ) = 3 N
x = ∑ xj N l 2 (ξ ,η ) = 3 (1 − η )
l1 (ξ ,η ) + 3 N
j =1 2
T1 :
M
l j (ξ ,η ) = N
y = ∑ yj N l 3 (ξ ,η ) = 1 (1 + ξ )
l 2 (ξ ,η ) + N
j =1 2
⎡ 3⎤
⎢ 0 −
2⎥ 3
• Note that | J |= det ⎢
1
⎥ = >0⇒
4
the map T1 is invertable.
⎢ 0 ⎥
⎣⎢ 2 ⎦⎥
• |J| is the ratio of the area of Ω1 to area of l

Ω .
N. Zabaras (10/06/2009)
13
Example problem
T2
T2−1 ?
M
l j (ξ ,η ) = 3 N
x = ∑ xj N l 4 (ξ ,η ) = 3 (1 + η )
l 3 (ξ ,η ) + 3 N
j =1 2
T2 :
M
l j (ξ ,η ) = N
y = ∑ yj N l 3 (ξ ,η ) = 1 (1 + ξ )
l 2 (ξ ,η ) + N
j =1 2
⎡ 3⎤
⎢0 2⎥ 3
• Note that | J |= det ⎢ ⎥ =− <0⇒ the map T2 is not invertable.
⎢1 0⎥
4
⎣⎢ 2 ⎦⎥
• The clockwise numbering of nodes should be avoided.

N. Zabaras (10/06/2009)
14
Example problem
T3
T3−1 ?
M
l j (ξ ,η ) = N
x = ∑ xj N l 3 (ξ ,η ) = 1 (3 + 3ξ + η + ξη )
l 2 (ξ ,η ) + 2 N
j =1 4
T3 :
M
l j (ξ ,η ) = 2 N
y = ∑ yj N l 4 (ξ ,η ) = 1 (3 + ξ + 3η + ξη )
l 3 (ξ ,η ) + N
j =1 4
⎡1 1 ⎤
⎢4 (3 + η ) (1 + ξ ) ⎥ 1 1
4 1
| J |= det ⎢ ⎥= + ξ+ η⇒ |J| cannot be zero inside the element so
⎢ 1 (1 + η ) 1 (3 + ξ ) ⎥ 2 8 8
the map T3 is invertable.
⎢⎣ 4 4 ⎥⎦
• |J| is smaller near node 1, largest near node 3.

N. Zabaras (10/06/2009)
15
Example problem
T4
T4−1 ?
M
l j (ξ ,η ) =3 N
x = ∑ xj N l 3 (ξ ,η ) = 1 (4 + 4ξ − 2η − 2ξη )
l 2 (ξ ,η ) + N
j =1 4
T4 :
M
l j (ξ ,η ) = N
y = ∑ yj N l 4 (ξ ,η ) = 1 (3 − ξ + 3η − ξη )
l 3 (ξ ,η ) + 2 N
j =1 4
|J| is not >0 everywhere inside the element
⎡1 1 ⎤ so the map T3 is not invertable. The
⎢4 (4 − 2 η ) − (2 + 2 ξ ) ⎥ 1
4
| J |= det ⎢ ⎥ = (5 − 3ξ − 4η ) ⇒ marked region near node 3 -- above the
⎢ − 1 (1 + η ) 1
(3 − ξ ) ⎥
8
line ξ = 5 3 − 4 3 η -- is mapped outside the
⎢⎣ 4 4 ⎥⎦
element.
• All angles of quadrilateral elements need to be < π.
N. Zabaras (10/06/2009)
16
Finite element equations
• To solve our boundary value problem of interest
examined earlier, we need to compute the following
matrices and vectors ( 1 ≤ e ≤ E , 1 ≤ i, j ≤ N e ):
⎡ ⎛ ∂N e
∂N e
∂N e
∂N e
⎞ ⎤
kij = ∫ ⎢ k ⎜
e i
⎜ ∂x ∂x
j
+
∂
i
∂
j
⎟⎟ + bN i N j ⎥ dxdy
e e
fi e = ∫ fN ie dxdy
⎢ ⎝
Ωe ⎣
y y ⎠ ⎦⎥ Ωe
P =
e
ij ∫ i
e
pN N dse
j
γ ie = ∫ γ N ie ds
∂Ωe2 h
∂Ωe2 h
• k and f are assumed to be functions of (x,y), whereas

p and γ are assumed functions of s defining ∂Ω2 h ( x( s), y ( s))
• It remains to discuss how to compute these using the
master element and the transformation Te.

N. Zabaras (10/06/2009)
17
Master element calculations: Step 1
l → Ω , e = 1, 2,..., E
Te : Ω
• We construct the mappings e as follows:
N
l j (ξ ,η )
x = ∑ xj N
e
j =1
N
l j (ξ ,η ),
y = ∑ yj N
e
j =1
• This mapping Te : Ωl → Ωe , is completed defined by the
nodal coordinates ( x j , y j ) of element e.

N. Zabaras (10/06/2009)
18
• Let g(x,y) be any function of (x,y) in Ωe. We can
l as follows:
convert g to a function of (ξ ,η ) defined in Ω
g ( x, y ) = g ( x (ξ ,η ), y (ξ ,η )) ≡ g (ξ ,η )
• The transformation x(ξ ,η ), y (ξ ,η ) is defined in Step 1.

e
• The element shape functions j ( x, y ) are simply
N
obtained from Nl j (ξ ,η ) as follows:
l j ( x(ξ ,η ), y (ξ ,η )), j = 1, 2,..., N

N ej ( x, y ) = N e

N. Zabaras (10/06/2009)
19
• The derivatives of N ej ( x, y ) with respect to x and y are:
∂N ej ( x, y ) l j ∂ξ ∂ N
∂N l j ∂η l k (ξ ,η ) l k (ξ ,η )
= + , ∂x N ∂N ∂x N ∂N
= ∑ xk = ∑ xk
e e
,
∂x ∂ξ ∂x ∂η ∂x ∂ξ k =1 ∂ξ ∂η k =1 ∂η
where: ∂y N l k (ξ ,η )
∂N ∂y N l k (ξ ,η )
∂N
∂N ej ( x, y ) l j ∂ξ ∂ N
∂N l j ∂η = ∑ yk
e
, = ∑ yk
e
= + ∂η k =1 ∂η ∂η k =1 ∂η
∂y ∂ξ ∂y ∂η ∂y
• Finally, with |J| the Jacobian of Te , using
∂ξ 1 M l k (ξ ,η ) ∂ξ
∂N 1 M l k (ξ ,η ) ∂η
∂N 1 M l k (ξ ,η ) ∂η
∂N 1 M l k (ξ ,η )
∂N
= ∑ yk , =− ∑ xk , =− ∑ yk , = ∑ xk
∂x | J | k =1 ∂η ∂y | J | k =1 ∂η ∂x | J | k =1 ∂ξ ∂y | J | k =1 ∂η
we conclude: ∂N ej ( x, y ) 1 ⎧⎪ ∂ N
lj M lj M
l k (ξ ,η ) ∂ N
∂N l k (ξ ,η ) ⎫⎪
∂N
= ⎨ ∑ yk − ∑ yk ⎬,
∂x | J | ⎪ ∂ξ j =1 ∂η ∂η k =1 ∂ξ
⎩ ⎭⎪
∂N ej ( x, y ) 1 ⎧⎪ ∂ N
lj M lj M
l k (ξ ,η ) ∂ N
∂N l k (ξ ,η ) ⎫⎪
∂N
= ⎨− ∑ xk + ∑ xk ⎬
∂y | J | ⎪ ∂ξ k =1 ∂η ∂η jk =1 ∂η ⎪⎭
⎩

N. Zabaras (10/06/2009)
20
Note regarding convergence rates
• We had presented earlier that the asymptotic rates of
convergece are only valid if whe ( x, y) = N∑ w j N ej ( x, y)
e
j =1
contains complete polynomials of degree k for smooth

interpolated functions.
• However, with the transformation
l j ( x(ξ ,η ), y (ξ ,η )), j = 1, 2,..., N
N ej ( x, y ) = N e
e
the shape functions N j ( x, y ) may not even be
polynomials!
• It can be shown that if the master basis functions Nl j (ξ ,η )
contain complete polynomials of degree k and |J|>0, the
error estimates still hold.
N. Zabaras (10/06/2009)
21
• The domain integrals of interest now take the form:
g (ξ ,η ) | J | d ξ dη
∫ g ( x, y )dxdy = ∫ g ( x (ξ ,η ), y (ξ ,η )) | J | d ξ dη ≡ ∫
Ω l l
e Ω Ω l (ξ ,η )
G
l
• Our integrals are now in the domain of Ω
l (ξ ,η )d ξ dη
∫ g ( x, y )dxdy = ∫ g ( x (ξ ,η ), y (ξ ,η )) | J | d ξ dη ≡ ∫ G
Ω e
l
Ω l
Ω
• As in 1D, we use Gauss integration (to be discussed):
N
l l
∫ G (ξ ,η )d ξ dη ≈ ∑ G ( ξN
i ,ηi )
i
wi
l i =1 N
Ω Gaus s Gauss
int egration int egration
po int s weights

N. Zabaras (10/06/2009)
22
• In numerical calculation of e.g. the element stiffness,
⎡ ⎛ ∂N e
∂N e
∂N e
∂N e
⎞ ⎤
kij = ∫ ⎢ k ⎜
e i j
+ i j
⎟⎟ + bN i N j ⎥ dxdy
e e
⎜ ∂x ∂x ∂ ∂
⎢ ⎝
Ωe ⎣
y y ⎠ ⎦⎥
we often need to consider functions k(x,y). It is common

to use the finite element interpolant to approximate this
function, i.e.
N N
k h ( x, y ) = ∑
e
kj N ej ( x, y )
e
l j (ξ ,η )
= ∑ kj N
j =1 N j =1
k(x ,y )
j j
This requires only the nodal values of k. Similar

calculation can be done for integrals involving b(x,y) and
f(x,y).

N. Zabaras (10/06/2009)
23
• We need to consider a number of integrals that involve
boundary integration:
Pije = ∫ pN ie N ej ds γ ie = ∫ γ N ie ds
∂Ωe2 h
∂Ωe2 h
• Lets introduce the restriction of the basis functions Nl j (ξ ,η )

on the side ξ = 1 . We denote them as:θ (η ) ≡ Nl j (1,η ), j = 1, 2,.., Ne .
These functions are only non-zero for nodes on the
boundary ξ = 1 . The 2 integrals above can be
| j (η ) | is the Jacobian of the
computed as: transformation of η to s
1
lp (η )θ i (η )θ j (η ) | j (η ) | dη ds =| j (η ) | dη dy ds
P = ∫ pN N ds = ∫
e e e 1/2
ij i j ds = ⎡⎣ dx 2 + dy 2 ⎤⎦
−1 dx
∂Ωe2 h
1/2
⎡⎛ ∂x (1,η ) ⎞2 ⎛ ∂y (1,η ) ⎞ 2 ⎤
γ N ie ds = ∫ γ (η )θ i (η ) | j (η ) | dη
1
γ ie = ∫ −1
| j (η ) |= ⎢⎜
⎢⎣⎝ ∂ η ⎟ + ⎜ ∂η ⎟ ⎥
⎠ ⎝ ⎠ ⎥⎦
∂Ωe2 h
Notes: (a) Use FE interpolants of lp(η ), γ (η ) (b) Use 1D Gauss integration.
N. Zabaras (10/06/2009)
24
Example problem using 3-node triangular elements
• Consider 2D heat conduction using the general
problem discussed in an earlier lecture.
• We are computing T(x,y) such that:
−∇ • ( D∇T ) = f ( x, y )
T = T on ΓT
q = − D∇T • n = q on Γ q
• Recall that the final weak form is: ⎛ ⎞

⎜ ⎟
T ⎜ ⎟
nel nel
{w} ∑ ⎣⎡ Le ⎦⎤ ∫ ⎣⎡ B e ⎦⎤ [ De ] ⎣⎡ B e ⎦⎤ d Ω ⎣⎡ Le ⎦⎤ {d } = {w} ∑ ⎣⎡ Le ⎦⎤ ⎜ ∫ ⎣⎡ N e ⎦⎤ f d Ω − ∫ ⎣⎡ N e ⎦⎤ qd Γ ⎟, ∀{wF }
T T T T T T
e =1

Ωe e =1
⎜ Ω
e

Γeq

⎟
Ke ⎜ { fΩe } { fΓe} ⎟
⎝ ⎠
⎡ ∂N1
e
∂N 2
e
∂N nen ⎤
e
⎢ ∂x ...
∂x ⎥⎥
⎡⎣ B ⎤⎦ = ⎢ e
e ∂x ⎡ KE K EF ⎤ ⎡ d E ⎤ ⎡ f E + rE ⎤
⎢ ⎥=⎢
⎡⎣ N e ⎤⎦ = ⎡⎣ N1e N 2e ... N nen
e
⎤⎦
⎢ ∂N1 ∂N 2e ∂N nen ⎥
e
⎢K K F ⎦ ⎣ d F ⎦ ⎣ f F ⎥⎦
⎥
⎢ ∂y
⎣ ∂y
...
∂y ⎥⎦ ⎣ FE

N. Zabaras (10/06/2009)
25
Example problem using 3-node triangular elements
• Consider the heat conduction problem in the figure.
We assume isotropic
conductivity k=5 WoC-1 and
a source term f=6 Wm-2.
The BCs are as shown.
• We want to compute the temperature field using 2
linear triangular elements.

N. Zabaras (10/06/2009)
26
Example problem : Stiffness, element 1
• Using expressions derived before for triangular
elements, we can write: 3
⎡ ∂N1e ∂N 2e ∂N 3e ⎤
⎢ ∂x ∂x ∂x ⎥⎥ 1 ⎡ y2e − y3e y3e − y11 y1e − y2e ⎤
⎡⎣ B ⎤⎦ = ⎢ e
e
= ⎢ e ⎥ 2
N ⎢ ∂N1 ∂N 2e ∂N 3e ⎥ 2 Ae ⎣ x3 − x2
e
x1e − x3e x2e − x1e ⎦
cons tan t
⎢ ∂y ∂y ∂y ⎥⎦
matrix
⎣
1
2 Ae = ( x2e y3e − x3e y2e ) − ( x1e y3e − x3e y1e ) + ( x1e y2e − x2e y1e )
∫e ⎡⎣ B ⎤⎦ [ D ] ⎡⎣ B ⎤⎦ d Ω = ⎡⎣ B ⎤⎦ k ⎡⎣ B ⎤⎦ A
T T
⎡⎣ K e ⎤⎦ = e e e e e e
1 ⎡ −0.5 1 −0.5⎤
• For e=1: ⎡⎣ B1 ⎤⎦ = ⎢
2 ⎣ −2 0 2 ⎥⎦
1 2 3
⎡ 5.3125 −0.625 −4.6875⎤ 1
K 1 = B1 B1kA1 = ⎢ −0.625 −0.625 ⎥ 2
T
1.25
⎢ ⎥
⎢⎣ −4.6875 −0.625 5.3125 ⎥⎦ 3

N. Zabaras (10/06/2009)
27
Example problem : Stiffness, element 2
• Using expressions derived before for triangular
elements, we can write: 3 2
⎡ ∂N1e ∂N 2e ∂N 3e ⎤
⎢ ∂x ∂x ∂x ⎥⎥ 1 ⎡ y2e − y3e y3e − y11 y1e − y2e ⎤ 1
B =
N
e
⎢ = ⎢ e ⎥
cons tan t ⎢ ∂N1e ∂N 2e ∂N 3e ⎥ 2 Ae ⎣ x3 − x2
e
x1e − x3e x2e − x1e ⎦
matrix ⎢ ∂y ∂y ∂y ⎥⎦
⎣
2 Ae = ( x2e y3e − x3e y2e ) − ( x1e y3e − x3e y1e ) + ( x1e y2e − x2e y1e )
Ke = ∫ B e [ D e ]B e d Ω = B e BkAe
T T
Ωe
⎡ 0 0.5 −0.5⎤
• For e=2: B2 = ⎢
⎣ −2 2 0 ⎥⎦
2 4 3
⎡ 10 −10 0 ⎤ 2
⎡⎣ K 2 ⎤⎦ = ⎡⎣ B ⎤⎦ ⎡⎣ B 2 ⎤⎦ kA2 = ⎢ −10 10.625 −0.625⎥
2 T
4
⎢ ⎥
⎢⎣ 0 −0.625 0.625 ⎥⎦ 3
N. Zabaras (10/06/2009)
28
Example problem : Assembled stiffness
• Assembling the 2 stiffness matrices gives:
⎡ 5.3125 −0.625 −4.6875 0 ⎤ 1

⎢ −0.625 11.25 −0.625 −10 ⎥ 2
[ K ] = ⎡⎣ L ⎤⎦ ⎡⎣ K ⎤⎦ ⎡⎣ L ⎤⎦ + ⎡⎣ L ⎤⎦ ⎡⎣ K ⎤⎦ ⎡⎣ L ⎤⎦ = ⎢
1 T 1 1 2 T 2 2 ⎥
⎢ −4.6875 −0.625 5.9375 −0.625⎥ 3
⎢ ⎥
⎣ 0 −10 −0.625 10.625 ⎦ 4

N. Zabaras (10/06/2009)
29
Example problem : Load calculations – e=1,2
• The shape functions for 3-node triangular elements are
3 2
⎡⎣ N e ⎤⎦ = ⎡⎣ N1e N 2e N 3e ⎤⎦ 3
N1e =
1
( x2e y3e − x3e y2e + ( y2e − y3e ) x + ( x3e − x2e ) y )
1
e
2A 2
1
N 2e = e ( x3e y1e − x1e y3e + ( y3e − y1e ) x + ( x1e − x3e ) y )
2A
1
N 3e = e ( x1e y2e − x2e y1e + ( y1e − y2e ) x + ( x2e − x1e ) y ) 1
2A
⎡1⎤ This expression
2 Ae = ( x2e y3e − x3e y2e ) − ( x1e y3e − x3e y1e ) + ( x1e y2e − x2e y1e ) { f } = f ∫ {N }
e e T
dΩ =
fAe ⎢ ⎥
1 can be proved for
3 ⎢⎥
Ω
Ωe
⎢⎣1⎥⎦ constant f!
• For e=1: ⎡1⎤ ⎡1⎤ ⎡ 2 ⎤ 1
1
×
{ f 1} = 3 ⎢⎢1⎥⎥ = 3 ⎢⎢1⎥⎥ = ⎢⎢2⎥⎥
fA 6 1
2
⎡ 2 ⎤ ⎡2⎤ 1
⎢⎣1⎥⎦ ⎢⎣1⎥⎦ ⎢⎣ 2 ⎥⎦ 3 ⎢ 2 + 1⎥ ⎢ 3 ⎥
• For e=2: ⎡1⎤ ⎡1⎤ ⎡1⎤ 2 { f } = ⎡⎣ L ⎤⎦ { f } + ⎡⎣ L ⎤⎦ { f } = ⎢ ⎥ = ⎢ ⎥
1 T 1 2 T 2
⎢ 2 + 1⎥ ⎢ 3 ⎥
2
3
2
×
{ f 2 } = 3 ⎢⎢1⎥⎥ = 3 ⎢⎢1⎥⎥ = ⎢⎢1⎥⎥
fA 6 0.5
4
⎢ ⎥ ⎢ ⎥
⎣ 1 ⎦ ⎣1 ⎦ 4
⎢⎣1⎥⎦ ⎢⎣1⎥⎦ ⎢⎣1⎥⎦ 3

N. Zabaras (10/06/2009)
30
Example problem: Applying natural BCs
• For elements with segments on 2
3
Γq, we need to compute
1
{ f } = − ∫ ⎡⎣ N
Γ
e e T
⎤⎦ qd Γ
Γeq
• We only need to worry about e=2:

⎡ 1 ⎤
⎢ 2 A2 ( x2 y3 − x3 y2 + ( y2 − y3 ) x + ( x3 − x2 ) y ) ⎥
2 2 2 2 2 2 2 2
⎢ ⎥ ⎡ 0 ⎤
⎡⎣ N 2 ⎤⎦ | y =1 = ⎢ 2 ( x32 y12 − x12 y32 + ( y32 − y12 ) x + ( x12 − x32 ) y ) ⎥ = ⎢ 0.5 x ⎥
T 1
⎢2A ⎥ ⎢ ⎥
⎢ 1 ⎥ ⎢⎣ −0.5 x + 1⎥⎦
⎢ 2 ( x1 y2 − x2 y1 + ( y1 − y2 ) x + ( x2 − x1 ) y ) ⎥
2 2 2 2 2 2 2 2
⎣2A ⎦ y =1
⎡ 0 ⎤ ⎡ 0 ⎤ 2
• Thus: { fΓe } = −∫0 ⎢⎢ 0.5 x ⎥⎥ 20dx = ⎢⎢−20⎥⎥
2
4
⎢⎣ −0.5 x + 1⎥⎦ ⎢⎣ −20 ⎥⎦
3

N. Zabaras (10/06/2009)
31
Example problem: Final system of equations
• Assembling the load vector gives the following:
⎡ 2 ⎤ ⎡ r1 ⎤ ⎡ r1 + 2 ⎤ r1 , r2 , r3 , are unknown heat fluxes
⎢ 3 ⎥ ⎢r ⎥ ⎢ r + 3 ⎥
{ f } = { f Γ } + { f Ω } + {r} = ⎢ ⎥ + ⎢ 2 ⎥ = ⎢ 2 ⎥ at the nodes with prescribed
⎢ −17 ⎥ ⎢ r3 ⎥ ⎢ r3 − 17 ⎥ temperature (short of reaction
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ −19 ⎦ ⎣ 0 ⎦ ⎣ −19 ⎦ forces!)
• The final system of equations is:

⎡ 5.3125 −0.625 −4.6875 0 ⎤ ⎡ 0 ⎤ ⎡ r1 + 2 ⎤
⎢ −0.625 11.25 −0.625 −10 ⎥ ⎢ 0 ⎥ ⎢ r2 + 3 ⎥
⎢ ⎥⎢ ⎥ = ⎢ ⎥
⎢ −4.6875 −0.625 5.9375 −0.625⎥ ⎢ 0 ⎥ ⎢ r3 − 17 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣ 0 −10 −0.625 10.625 ⎦ ⎣T4 ⎦ ⎣ −19 ⎦
19
• Partitioning this system gives: T 4 =−
10.625
= −1.788
⎡ 0 ⎤
• `Reaction fluxes’:⎡⎢ 5.3125 −0.625 −4.6875 0 ⎤⎢ ⎥ ⎡ r1 + 2 ⎤ ⎡ r1 ⎤ ⎡ −2 ⎤
⎢ r ⎥ = ⎢ 14.88 ⎥
−10 ⎥ ⎢ ⎥ = ⎢ r +3 ⎥ ⇒
0
−0.625 11.25 −0.625
⎢ ⎥⎢ 0 ⎥ ⎢ 2 ⎥ ⎢ 2⎥ ⎢ ⎥
⎣⎢ −4.6875 −0.625 5.9375 −0.625⎦⎥ ⎢ ⎥ ⎣3⎢ r − 17 ⎥⎦ ⎢⎣ r3 ⎥⎦ ⎢⎣18.1175⎥⎦
⎣ −1.788 ⎦

N. Zabaras (10/06/2009)
32
Example problem: Postprocessing
• The DOF in each element (nodal temperatures) are
now given as:
⎡0 ⎤ ⎡ 0 ⎤
{d 1} = ⎢⎢0⎥⎥ , {d 2 } = ⎢⎢−1.788⎥⎥
⎢⎣0 ⎥⎦ ⎢⎣ 0 ⎥⎦
• The flux vectors can then be computed for each

element as:
⎡0⎤
⎡ 1
⎤ ⎡ − − ⎤ ⎢0⎥ = ⎡0 ⎤
{ } ⎢q1 ⎥ ⎣ ⎦ { } 2 ⎢ −2 0 2 ⎥ ⎢ ⎥ ⎢0⎥
q 1 0.5 1 0.5
q1
≡ x
= − k ⎡ B1
⎤ d 1
= −5
⎣ y⎦ ⎣ ⎦⎢ ⎥ ⎣ ⎦
⎣0⎦
⎡ 0 ⎤
⎡ ⎤ −
{q 2 } ≡ ⎢ qx2 ⎥ = −k ⎣⎡ B 2 ⎦⎤ {d 2 } = −5 ⎡⎢−2 2 2 ⎤⎥ ⎢⎢ −1.788⎥⎥ = ⎡⎢17.88⎤⎥
2
q 0 0.5 0.5 4.47
⎣ y⎦ ⎣ ⎦⎢ ⎣ ⎦
⎣ 0 ⎥⎦

N. Zabaras (10/06/2009)
33
Example problem using 4-node quadrilateral elements
• We repeat the calculations of the earlier example
using one 4-node quadrilateral element. We use again
`matrix notation’.
• The element coordinates are: ⎡ x1e y1e ⎤ ⎡ 0 1 ⎤

⎢ e ⎥
x y2e ⎥ ⎢⎢ 0 0 ⎥⎥
⎡⎣ x e y ⎤⎦ = ⎢ 2e
e
=
⎢ x3 y3e ⎥ ⎢ 2 0.5⎥
⎢ e ⎥ ⎢ ⎥
⎣ x4 y4e ⎦ ⎣ 2 1 ⎦

N. Zabaras (10/06/2009)
34
l1 (ξ ,η ) = 1 (1 − ξ )(1 − η )
N
• The basis functions in 4
l 2 (ξ ,η ) = 1 (1 + ξ )(1 − η )
N
4
natural coordinates are: l 3 (ξ ,η ) = 1 (1 + ξ )(1 + η )
N
4
l 4 (ξ ,η ) = 1 (1 − ξ )(1 + η )
N
4
• We need to compute the Be matrix relating derivatives

wrt x and y and nodal values.
l e4 ⎤ ⎡T1 ⎤
e
⎡ ∂T e ⎤ ⎡ ∂ N
l1e ∂ N
l e2 ∂ N l 3e ∂ N
⎢ ⎥ ⎢ ⎥⎢ e⎥
⎢ ∂x ⎥ = ⎢ ∂x ∂x ∂x ∂x ⎥ ⎢T2 ⎥
⎢ ∂T e ⎥ ⎢ ∂ N ⎢ ⎥
l1e ∂ N
l e2 ∂ N l e4 ⎥ ⎢T3e ⎥
l 3e ∂ N
⎢ ⎥ ⎢ ⎥
⎣ ∂y ⎦ ⎢⎣ ∂y ∂y ∂y ∂y ⎥⎦ ⎢⎣T4e ⎥⎦

[B ]e
⎡∂N l e4 ∂η ⎤ ⎡T1 ⎤
e
⎡ ∂T e ⎤ l1e ∂ξ l1e
∂N∂η l e2
∂N ∂ξ ∂η l e2
∂N ∂Nl 3e ∂ξ ∂ N
l 3e ∂η ∂Nl e4 ∂ξ ∂ N
⎢ ⎥ ⎢ ∂ξ + + + + ⎥⎢ e⎥
∂x ∂η ∂x ∂ξ ∂x ∂η ∂x ∂ξ ∂x ∂η ∂x ∂ξ ∂x ∂η ∂x ⎥ ⎢T2 ⎥
⎢ ∂x ⎥ = ⎢
⎢ ∂T e ⎥ ⎢ l e ⎥⎢ e⎥
l1e ∂η
∂ξ ∂ N l e2 ∂ξ ∂ N
l e2 ∂η l 3e ∂ξ ∂ N
l 3e ∂η l e
l e
∂ N 4 ∂ξ ∂ N 4 ∂η ⎥ ⎢T3 ⎥
⎢ ⎥ ⎢ ∂ N1 +
∂N
+
∂N
+ +
⎣ ∂y ⎦ ⎢⎣ ∂ξ ∂y ∂η ∂y ∂ξ ∂y ∂η ∂y ∂ξ ∂y ∂η ∂y ∂ξ ∂y ∂η ∂y ⎥⎦ ⎢⎣T4e ⎥⎦

N. Zabaras (10/06/2009)
35
⎡∂N l1e l e
∂ξ ∂ N 1 ∂η l e
l e
∂ N 2 ∂ξ ∂ N 2 ∂η l e
l e
∂ N 3 ∂ξ ∂ N 3 ∂η ∂ N 4 ∂ξ ∂ N 4 ∂η ⎤
l e
l e
⎢ + + + + ⎥
∂ξ ∂x ∂η ∂x ∂ξ ∂x ∂η ∂x ∂ξ ∂x ∂η ∂x ∂ξ ∂x ∂η ∂x ⎥
⎡ Be ⎤ = ⎢
⎣ ⎦ ⎢ le l1e ∂η l e2 ∂ξ ∂ N
l e2 ∂η l 3e ∂ξ ∂ N
l 3e ∂η l e
l e ⎥
⎢ ∂ N1 ∂ξ ∂ N ∂N ∂N ∂ N 4 ∂ξ ∂ N 4 ∂η ⎥
⎢ ∂ξ + + + +
⎣ ∂y ∂η ∂y ∂ξ ∂y ∂η ∂y ∂ξ ∂y ∂η ∂y ∂ξ ∂y ∂η ∂y ⎥⎦
• We can write this matrix as:

⎡∂N l e
l
∂N 2
e
l
∂N3
e
l
∂N 4
e⎤
⎡ ∂y e ∂y e⎤⎡ l
∂ N1
e
l
∂N 2
e
l
∂N3
e
l
∂N 4
e⎤
⎡ ∂ξ ∂η ⎤ ⎢ 1 ⎥
⎢ ⎢ − ⎥⎢ ⎥
⎡ B e ⎤ = ⎢ ∂x ∂x ⎥ ⎢ ∂ξ ∂ξ ∂ξ ∂ξ ⎥ 1 ⎢ ∂η ∂ξ ⎥ ⎢ ∂ξ ∂ξ ∂ξ ∂ξ ⎥
⎥ =
⎣ ⎦
⎢ ∂ξ ∂η ⎥ ⎢ l e e
l2 e
l3 e ⎥ | Je |⎢
l4 ⎥ ∂x e ∂x e ⎥ ⎢ le e
l2 e
l3 l e⎥
⎢ ∂ N1 ∂N ∂N ∂N ⎢− ⎥ ⎢ ∂ N1 ∂N ∂N ∂N 4 ⎥
⎢⎣ ∂y ⎥
∂y ⎦ ⎢ ⎥
⎣ ∂η ∂η ∂η ∂η ⎦ ⎣ ∂η ∂ξ ⎦ ⎢⎣ ∂η ∂η ∂η ∂η ⎥⎦
⎡ ∂y e ∂x e ⎤
⎢ − ⎥
This comes from an ⎧dξ ⎫ 1 ⎢ ∂η ∂η ⎥ ⎧ dx ⎫
⎨ ⎬= e ⎨ ⎬
earlier derivation: ⎩dη ⎭ | J | ⎢ ∂y e e ⎥ dy
∂x ⎩ ⎭
⎢− ⎥
⎣ ∂ξ ∂ξ ⎦
N. Zabaras (10/06/2009)
36
∂y e ⎤ ⎡ ∂ N
l1e l e4 ⎤
e e
⎡ ∂y e l2
∂N ∂Nl3 ∂N
⎢ − ⎥⎢ ⎥
∂ η ∂ξ ⎥ ⎢ ∂ξ ∂ξ ∂ξ ∂ξ ⎥
⎡ Be ⎤ = 1 ⎢
⎣ ⎦ | Je |⎢ ⎢ e e⎥
∂x e ∂x e ⎥ ⎢ ∂ N l1 ∂N
e
l2 ∂N
e
l3 l
∂N 4 ⎥
⎢− ⎥
⎣ ∂η ∂ξ ⎢ ∂η ⎥⎦

⎦ ⎣ ∂η ∂η ∂η
4
Inverse of l j (ξ ,η ),
x e = ∑ x ej N
the Jacobian [ J e
]
j =1
• The Jacobian matrix can be computed from:y e = 4
l j (ξ ,η ),
∑ yj N
e
j =1
⎡ e e e e ⎤ ⎡ xe y1e ⎤
⎡ ∂x e
∂y e⎤ l
∂ N1 l
∂N 2 l
∂N3 l
∂N 4 ⎢ 1
⎢ ⎥ ⎢ ⎥ e ⎥
∂ξ ∂ξ ⎥ ⎢ ∂ξ ∂ξ ∂ξ ∂ξ ⎥ ⎢ x2 y2 ⎥
e
⎡J e ⎤ = ⎢ =
⎣ ⎦
N ⎢ ∂x e e⎥ ⎢ e ⎥⎢ e ⎥
∂y l e
l e
l e
l
∂ N 4 ⎥ ⎢ x3 y3e ⎥
Jacobian ⎢ ⎥ ⎢ ∂ N1 ∂N 2 ∂N3
matrix of
element e
⎣ ∂η ∂η ⎦ ⎢⎣ ∂η ∂η ∂η ∂η ⎥⎦ ⎢⎣ x4e ⎥
y4e ⎦

N. Zabaras (10/06/2009)
37
• For element 1:
⎡0 1 ⎤
η −1 1−η 1+η −η − 1⎤ ⎢ 0 0 ⎥ ⎡0 0.125η − 0.375 ⎤
⎡ J 1 ⎤ = 1 ⎡⎢ ⎢ ⎥=
⎣ ⎦ 4 ⎣ξ − 1 −ξ − 1 1 + ξ 1 − ξ ⎦ ⎢ 2 0.5⎥ ⎢⎣1 0.125ξ + 0.125⎥⎦
⎥
⎢ ⎥
⎣2 1 ⎦
⎡1 + ξ ⎤
⎢ 1⎥
−1 3 −η
⎡J1⎤ = ⎢ ⎥ , det J 1 = −0.125η + 0.375
⎣ ⎦ ⎢ 8 0⎥
⎢⎣η − 3 ⎥⎦
⎡1 + ξ ⎤
⎢3 −η 1 ⎥ 1 ⎡η − 1 1 − η 1 + η −η − 1⎤
⎡B ⎤ = ⎢
1
⎥ ⎢
⎣ ⎦ ⎢ 8 4 ξ − 1 −ξ − 1 1 + ξ 1 − ξ ⎥⎦
0⎥ ⎣
⎢⎣η − 3 ⎥⎦
T 1 1 T
⎡ 1⎤ ⎡ 1⎤ ⎡ 1⎤ ⎡ 1⎤ ⎡ 1 ⎤ ⎡ D1 ⎤ ⎡ B1 ⎤ | J 1 | d ξ dη ⇒
• The element stiffness is: ⎣ K ⎦ = Ω∫ ⎣ B ⎦ ⎣ D ⎦ ⎣ B ⎦ d Ω = −∫1 −∫1 ⎣ B ⎦ N
1 ⎣ ⎦⎣ ⎦
k[ I ]
N =2 N =2
⎡ K1 ⎤ = ∑
int int
⎛ ⎡ 1 ⎤T ⎡ 1 ⎤ 1 ⎞
ξ η = ±
1
±
1
),(Wi ,W j ) = (1,1)
∑ k⎜ B B | J | ⎟ W W , ( , ) ( ,
⎣ ⎦ i =1 j =1 ⎝
⎣ ⎦ ⎣ ⎦ ⎠ξ =ξ ,η =η
i j i j
3 3
i j

N. Zabaras (10/06/2009)
38
• The element stiffness is
finally given as:
⎡ 4.76 −3.51 −2.98 1.73 ⎤
⎢ −3.51 4.13 1.73 −2.36 ⎥
⎡K ⎤ = ⎢
1 ⎥
⎣ ⎦ ⎢ −2.98 1.73 6.54 −5.29 ⎥
⎢ 1.73 −2.36 −5.29 5.91 ⎥
⎣ ⎦
• The nodal force is (Gauss integration not shown):
⎡Nl1e ⎤
⎢ ⎥ ⎡ 2.5⎤
⎢l ⎥e ⎢ 2.5⎥
{ fΩe }
1 1 1 1 N
= ∫ f ⎡⎣ N ⎤⎦ d Ω = ∫ ∫ f ⎡⎣ N ⎤⎦ | J | dξ dη = ∫ ∫ 6 ⎢ ⎥ (−0.125η + 0.375)dξ dη = ⎢ ⎥
e T e T 1 2
Ω e
−1 −1 −1 −1 ⎢ N
l 3e ⎥ ⎢ 2 ⎥
⎢ ⎥ ⎢ 2 ⎥
⎢le ⎥ ⎣ ⎦
⎣N 4 ⎦

N. Zabaras (10/06/2009)
39
• We now need to compute the
contribution to element load
from the natural boundary
conditions on the element e=1
at ξ=−1.
⎡1 ⎤
⎢2 (1 − η ) ⎥ ⎡ −20 ⎤
⎢ ⎥ ⎢ 0 ⎥
{ } ⎛ ⎞ −
T 1 T 1
⎢ 0 ⎥ 2 0
f q = − ∫ q ⎡⎣ N ⎤⎦ d Γ = − ∫ q ⎜ ⎡⎣ N ⎤⎦ | J | ⎟
e e e 1
dη = − ∫ 20 dη = ⎢ ⎥
Ω e
−1 ⎝ ⎠ξ =−1 −1 ⎢ 0 ⎥ N 2 ⎢ 0 ⎥
⎢ ⎥ |J |
1
⎢ −20 ⎥
⎢ (1 + η ) ⎥
1 ⎣ ⎦
⎢⎣ 2 ⎥⎦
⎡ r1 − 17.5⎤
• Assembly of all the nodal ⎢ r + 2.5 ⎥
{ f } = { f Ω } + { f Γ } + {r } = ⎢ 2 ⎥
⎢ r3 + 2 ⎥
contributions gives: ⎢ −18 ⎥
⎣ ⎦
N. Zabaras (10/06/2009)
40
• We finally obtain:
⎡ 4.76 −3.51 −2.98 1.73 ⎤ ⎡ 0 ⎤ ⎡ r1 − 17.5⎤

⎢ −3.51 4.13 1.73 −2.36 ⎥ ⎢ 0 ⎥ ⎢ r + 2.5 ⎥
⎢ ⎥⎢ ⎥ = ⎢ 2 ⎥
⎢ −2.98 1.73 6.54 −5.29 ⎥ ⎢ 0 ⎥ ⎢ r3 + 2 ⎥
⎢ 1.73 −2.36 −5.29 5.91 ⎥ ⎢T ⎥ ⎢ −18 ⎥
⎣ ⎦⎣ 4⎦ ⎣ ⎦
which gives T4 = −3.04.

⎡ 0 ⎤
⎢ 0 ⎥
• The nodal DOFs are: {d1} =⎢
⎢ 0 ⎥
⎥
⎢ −3.04 ⎥
⎣ ⎦

N. Zabaras (10/06/2009)
41
• The flux components at each
Gauss point are:
⎡ ∂T e ⎤ ⎡T1e ⎤
⎢ e⎥ ⎡1 + ξ ⎤ ⎡ 0 ⎤
⎢ −k ⎥ ⎢3 −η 1 ⎥ 1 ⎡η − 1 1 − η 1 + η −η − 1⎤ ⎢ 0 ⎥
e ⎤ ⎢T2 ⎥
{ 1
}
q (ξ ,η ) = ⎢ ∂x ⎥
⎢ ∂T e ⎥
⎡
= −k ⎣ B ⎦ ⎢ ⎥ = −k ⎢
e
⎢ 8
⎥ ⎢
⎥ 4 ⎣ξ − 1 −ξ − 1 1 + ξ
⎢
1 − ξ ⎦⎥ ⎢ 0 ⎥
⎥⇒
⎢ − ⎥ ⎢T3 ⎥ 0
k
⎢ e⎥ ⎢⎣η − 3 ⎥⎦ ⎢ −3.04 ⎥
⎣ ∂y ⎦ ⎣ ⎦
⎣ 4⎦
T
⎧ 1 1 1 ⎫ ⎡0.8979 ⎤
⎨ q ( , − )⎬ =
⎩ 3 3 ⎭ ⎢⎣3.5916 ⎥⎦
⎧ 1 1 1 ⎫ ⎡ −2.2965⎤
⎨q ( , ) ⎬ = ⎢
⎩ 3 3 ⎭ ⎣ 19.793 ⎥⎦
⎧ 1 1 1 ⎫ ⎡ 4.9482 ⎤
⎨q (− , ) ⎬ = ⎢ ⎥
⎩ 3 3 ⎭ ⎣19.793 ⎦
⎢ 1 1 1 ⎥ ⎡5.8042 ⎤
q ( − , − ) =
⎢⎣ 3 3 ⎥⎦ ⎢⎣3.5916 ⎥⎦

N. Zabaras (10/06/2009)
42

FEM Zabaras 2DFElementCalculations

Uploaded by

Copyright:

Available Formats

You might also like

FEM Zabaras 2DFElementCalculations

Uploaded by

Document Information

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

FEM Zabaras 2DFElementCalculations

Uploaded by

Copyright:

Available Formats

MAE4700/5700

Finite Element Analysis for

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• We need to figure out how we

• We can write this as a system of

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• Assume the inverse

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• From the equations above

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

where M is the number of nodes in the master element.

where ( x j , y j ) are the

• We need to select the element

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• The basis functions for this element are:

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• |J| is the ratio of the area of Ω1 to area of l

• The clockwise numbering of nodes should be avoided.

• |J| is smaller near node 1, largest near node 3.

• k and f are assumed to be functions of (x,y), whereas

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• The transformation x(ξ ,η ), y (ξ ,η ) is defined in Step 1.

l j ( x(ξ ,η ), y (ξ ,η )), j = 1, 2,..., N

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• Finally, with |J| the Jacobian of Te , using

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

contains complete polynomials of degree k for smooth

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

we often need to consider functions k(x,y). It is common

This requires only the nodal values of k. Similar

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• Lets introduce the restriction of the basis functions Nl j (ξ ,η )

• Recall that the final weak form is: ⎛ ⎞

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

⎡ 5.3125 −0.625 −4.6875 0 ⎤ 1

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

⎢⎣1⎥⎦ ⎢⎣1⎥⎦ ⎢⎣1⎥⎦ 3

• We only need to worry about e=2:

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• The final system of equations is:

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• The flux vectors can then be computed for each

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• The element coordinates are: ⎡ x1e y1e ⎤ ⎡ 0 1 ⎤

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• We need to compute the Be matrix relating derivatives

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

• We can write this matrix as:

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

⎡ 4.76 −3.51 −2.98 1.73 ⎤ ⎡ 0 ⎤ ⎡ r1 − 17.5⎤

which gives T4 = −3.04.

MAE 4700 – FE Analysis for Mechanical & Aerospace Design