ME 478 FINITE ELEMENT METHOD

Chapter 4.5 Field Problems

Thermal problems (1dof – Temperature, T)
Electrostatic problems (1dof – Potential, V)
Antiplane problems (1dof – Out of plane displacement, w)

1-D Field Problems

(1-D heat conduction)
Consider a wall of thickness L

T1 kt T2

Q1 Q2

dT dT T2 − T1
q = − kt A q = Q1 = −Q2 =
dx dx L
which we model as a single element

T1 kt T2

Q1 Q2
We have two equations with two unknowns T1 and T2
kt A
Q1 = (T1 − T2 )
L
k A
Q2 = t (T2 − T1 )
L

4.1
We write this in matrix form as:
 kt A kt A 
 Q1   L −  T1 
=
   kA L
k A  T 
 2  −
Q t t
 2 
 L L 
Flux vector Temperature vector
Conductance
matrix
This is the same thing that we had for the 1-D truss element
Consider a wall made up of two layers

(1) ( 2)
T1 kt T2 kt T3

Q1 Q2

We model this as two elements

(1) ( 2)
T1 kt T2 kt T3

For element 1 we have For element 2 we have

 kt A(1) kt A 
(1)
 kt A(2) kt A 
( 2)

Q1(1)   L − T (1)  Q

( 2)
  − T (2) 
L  1 L L  2
 (1)  =  (1)  (1)   ( 2)  =  ( 2)  ( 2) 
2
(1) ( 2)
 2  − t
Q  k A kt A  2 
T  3  − t
Q  k A kt A T3 
 L L   L L 
Now join the two elements
(1) (2)
equilibrium requires that Q2 + Q2 = Q2
(1)
compatibility requires that T2 = T2(2) = T2

4.2
We really have the following 3 equations with 3 unknowns
In matrix from:
 k t 1 A1 k A 
 − t1 1 0 
 Q1   L1 L1   T1 
   k t 1 A1 k A  
− t 2 2  T2 
k t 1 A1 k t 2 A2
Q 2  =  − +
Q   L1 L1 L2 L2   
 3
 0 k t 2 A2 k t 2 A2  T3 
−
 L2 L2 
Or q = Kt
Nt. We don’t have to rotate this –T has no direction

2-D Field Problems

(General case)
The field variable is φ
The steady state problem is given by
∂  ∂T  ∂  ∂T 
 k xx  +  k yy  + Q = 0
∂x  ∂x  ∂x  ∂y 
Which, when k = constant, reduces to the well known Poisson’s
equation as:
Q
∇ 2φ =
k
Which, when Q = 0, reduces to the well known Laplace equation as:
∇ 2φ = 0
For now let’s consider the more general case
∂  ∂T  ∂  ∂T 
 k xx  +  k yy  + Q = 0
∂x  ∂x  ∂x  ∂y 

4.3
2-D heat transfer finite element derivation
Again we are going to go through the same steps as before

Step 1 Select element type

Consider the three nodded triangular element
m ( xm , y m )
y
Nt. Node
numbering is
clockwise

i ( xi , yi ) j( x j , y j )
x
Here each node has only 1 dof (Temperature)

Step 2 Select a temperature function

Since we have 3 nodes and 1 dof per node, let
T ( x, y ) = a1 + a2 x + a3 y
which upon solving for the a’s can be written as:
 Ti 
 
T ( x, y ) = [N1 N2 N 3 ]T j  = Nt
T 
 m
where the shape functions are solved as follows:
We really have 3 equations here
In matrix form we have for ux
 Ti  1 xi yi   a1 
     
T j  = 1 x j y j  a2 
T  1 x ym  a3 
 m  m

4.4
Which can be solved as:
a = x −1t
To get the inverse of x we use the method of cofactors
1
x −1 = adj[x]
2A
where
1 xi yi 
2 A = det 1 x j y j 
1 xm ym 
A is the area of the triangle and
adj[x] = (cof [x]) T

Step 3 Define the temperature gradient/temperature and heat

flux/temperature gradient relations

 ∂T 
 ∂x 
g =  ∂T  = Bt
 
 ∂y 
where g is the temperature gradient matrix (analogous to strain)
and
1  βi βj βm 
B= 
2 A γ i γj γ m 
The heat flux/temperature gradient relation is:

4.5
  ∂T 
qx  k xx 0   ∂x 
q =   = Dg =   
q y  0 k yy   ∂T 
 ∂y 
Step 4 Derive the element conduction matrix and equations
Here we are going to use the variational method (analogous to
the principle of minimum potential energy seen earlier)
Given the functional
π h = U + ΩQ + Ω q + Ωh
Internal Energy

1   ∂T 
2
 ∂T 
2

U = ∫∫∫  k xx   + k yy    dV
∂x   ∂y 
V 2  
Potential energy of the external forces

Ω Q = − ∫∫∫ QTdV
V

Ω q = − ∫∫ q *T ds q
Sq

1
Ω h = ∫∫ h(T − T∞ ) 2 ds h
Sh 2
Where Q is internal heat generation, q is heat flow and h is
convection loss
(Nt, we can’t specify both q and h on the same surface)
We can rewrite in matrix form as:

4.6
 1 T
π h = ∫∫∫ g Dg dv − ∫∫∫ t T N T QdV
V 2 V

1
− ∫∫ t T N T q * ds q + ∫∫ h ( t T N T − T∞ ) 2 ds h
Sq Sh 2
t is not a function of x and y so lets pull it out if the integrals to get
1 T 1
πh = t ∫∫∫ B T DB dv t − t T ∫∫∫ N T QdV
2 V 2 V

− t T ∫∫ N T q * ds q
Sq

1
+ ∫∫ h ( t T N T Nt − t T N T T∞ + T∞ ) ds h
2

Sh 2

We want to find the minimum energy (differentiate w.r.t. t and = 0)

∂π h
= ∫∫∫ B T DB dv t − ∫∫∫ N T QdV
∂t V V

− ∫∫ N T q * ds q
Sq

1
+ ∫∫ h( N Nt ) ds h + ∫∫ h( N T T∞ ) ds h
T

Sh Sh 2

4.7
And rewriting

 
 ∫∫∫ B DB dv + ∫∫ h(N N )dsh  t =
T T

V Sh 
1
+ ∫∫∫ N QdV + ∫∫ N q dsq − ∫∫ h(N T T∞ )dsh
T T *

V Sq Sh 2

And Voila!! K t −f = 0
We can simplify the conduction matrix since all terms in the B
matrix are constant as

 
K =  ∫∫∫ B DB dv + ∫∫ h(N N)dsh 
T T

V Sh 
= tAB T DB + h ∫∫ (N T N) dsh
Sh

Step 5 Assembly
Use the same approach as shown before by applying compatibility
and equilibrium

Step 6 Solve for the nodal temperatures…

Step 7 Postprocess for the temperature gradients and heat flux

4.8