TWO-DIMENSIONAL,

STEADY-STATE
CONDUCTION

1
In many situations the one-dimensional approach of
heat conduction is oversimplification.
Multidimensional effects must be accounted for.
4.1 Alternative Approaches
Here the two-dimensional, steady state heat
conduction will be treated. Again the two major
objectives in conduction analysis is determining the
temperature field, T(x,y) and then the heat transfer.
For two-dimensional steady-state conduction with no
generation and constant k, the PDE to be solved is
∂T ∂T
2 2
+ 2 =0
∂x 2
∂y 2
Methods of solving include analytical, graphical, and
numerical (finite difference, finite-element, or
boundary element) approaches.
The analytical gets the exact solution of the PDE for
only a restricted set of simple geometries and
boundary conditions. Often the solutions are in
series form. The method of separation of variables
will be used for this solution.
Graphical and numerical solutions give approximate
results at discrete points.
Graphical solution is restricted to two-dimensional
problems involving adiabatic and isothermal
boundaries. 3
The method is based on the fact that isotherms must be
perpendicular to heat flow lines figchp-
4\[Link]. It requires good construction skills.
Numerical methods may be used to obtain accurate
results for complex, two- or three-dimensional
geometries involving a variety of boundary
conditions.
4.2 The method of Separation of variables
Consider the system shown in figchp-4\[Link].
Assumption made is that heat transfer from end
surfaces is negligible or (∂2T/∂z2)≈0 . This will
make it a two dimensional problem.
4
To simplify the solution, the following transformation
is used
T − T1
θ≡
T2 − T1
Substituting gives
∂θ ∂θ
2 2
+ 2 =0
∂x 2
∂y
For the complete solution of the above PDE two
boundary conditions are required for each of the
coordinates and these are
T(0,y)=T1 and T(x,0)=T1 or θ(0,y) = 0 and θ(x,0)=0
T(L,y)=T1 and T(x,W)=T2 or θ(L,y) = 0 and θ(x,W)=1 5
Because of the transformation, the BC’s have been
restricted to 0’s and 1.
The method of separation of variables assumes a
product solution of the form
θ(x,y) = X(x) . Y(y)
∂θ ∂ 2
θ
= X ( x ) Y ( y) and
'
= X ''
( x ) Y( y)
∂x ∂x 2

Similarly
∂θ ∂ 2
θ
= X ( x ) Y ( y) and
'
= X ( x ) Y ''
( y)
∂y ∂y 2

Substitution in the PDE of θ gives

2 2
1 dX 1 dY
X Y + XY = 0 Or −
'' ''
2
=
X dx Y dy 2 6
Indeed the differential equation is separable. Another
observation is that the left hand side expression is
only a function of x and the right one only a function
of y. The equality can be satisfied only if the
expression is independent of x and y. Only a
constant can satisfy this condition. Let this constant
be λ2. This makes the equation
2 2
1 dX 1 dY
− 2
= 2
=λ2

X dx Y dy

0 or –λ2 could have been the constants. But λ2

satisfies the prescribed boundary conditions.
7
The resulting differential equations are
d2X d2Y
2
+λ =0
2
2
−λ =0
2

dx dy
And the solutions are
X = C1 cos λx + C2 sin λx
Y = C3 eλy + C4 e-λy
This will give θ as
θ = XY = (C1 cos λx + C2 sin λx)(C3 eλy + C4 e-λy)
Using the boundary conditions
(1) θ(0,y) = 0 → C1 = 0
(2) θ(x,0) = 0 → (C2 sin λx)(C3+ C4) = 0
The only possibility is (C3+ C4) = 0. If C2 = 0
8
It will result in θ = 0 (not the required solution)
This will then give C3 = – C4
(3) θ(L,y) = 0 Gives C2C4 sin λL (eλy – e-λy) = 0
As C2 and C4 can not be zeros, the only possibility is
sin λL = 0 → λL = nπ n=0, 1, 2, 3,…
n=0 makes θ(x,y) = 0 which is not acceptable. So n=0
will be precluded or n = 1, 2, 3, …
The solution then becomes
nπx nπy / L − nπy / L
θ = C2C4 sin
L
(e −e )
Combining the two constants, knowing that the new
constant may depend on n will give 9
 nπx nπy
θ ( x, y) = Cn sin sinh
L L
Since (e nπy / L − e − nπy / L ) = 2 sinh (nπy/L)
As the sum of the solutions is also a solution, then the
general solution becomes
∞
nπx nπy
θ ( x, y) = ∑ Cn sin sinh
n =1 L L
To obtain Cn, we will apply the last boundary
condition ∞
nπx nπW
(4) θ(x,W)=1= ∑ C n sin L sinh L
n =1

10
To determine Cn, some manipulation is required. This
requires the knowledge of orthogonality.
An infinite set g1(x), g2(x), …, gn(x) is said to be
orthogonal in the domain a ≤ x ≤ b if
b
∫a
g m ( x )g n ( x )dx = 0 m≠n
Most functions exhibit orthogonality including
trigonometric functions
nπx nπx
sin and cos for 0 ≤ x ≤ L
L L

11
If f(x) is expressed in terms of series of orthogonal
functions as
∞
f (x) = ∑ Angn (x)
n =1

Then manipulating the above equation as follows

∞
f ( x )g m ( x )dx = ∫ g m ( x )∑ A n g n ( x )dx
b b
∫a a
n =1

And then using the principles of orthogonality, on the

right side all the expressions will be zero except one
term and the result is given by
b b
∫a
f ( x )g m ( x )dx =A m ∫ g 2m ( x )dx
a
12
Solving for Am and switching the subscript to n will
give
b

An =
∫a
f ( x )g n ( x )dx
b
∫
2
f ( x )g ( x )dx
n
a

For our situation gn(x) = sin (nπx/L), f(x)=1 and

substitution will give
nπx
L
∫0 sin L dx 2 (−1)n+1 + 1
An = L =
2 nπx π
∫0 sin dx n
L 13
Substitution in the series representation of f(x) gives
∞
2 ( −1) ( n +1)
+1 nπx
1= ∑ sin
n =1 π n L
Recalling the equation of the result of the fourth
boundary condition and comparing gives
n +1
2[( −1) + 1]
Cn = n = 1, 2, 3, ...
nπ sinh( nπW / L)
And the resulting final solution will be
nπy
2 ∞ ( −1) n +1 + 1 nπx sinh L
θ ( x, y) = ∑ sin
π n =1 n L sinh nπW
L
Isotherms and heat flow lines are shown in figchp-
4\[Link]. 14
4.3 THE CONDUCTION SHAPE FACTOR AND
DIMENSIONLESS CONDUCTION HEAT
RATE
Conduction shape factors, S are used to report the
existing solutions of heat conduction equations. The
steady-state heat transfer rate is expressed in
*
dimensionless conduction heat rate, ss .q
Heat transfer rate in terms of shape factors may be
expressed as
q = SkΔT1-2
Where ΔT1-2 is the temperature difference between
boundaries.
15
Two dimensional conduction in a square channel of
length L. (a) Symmetry planes. (b) Flux plot. (c)
Typical curvilinear square. 16
If properly constructed, qi will be the same for all
lanes
M
q = ∑ qi = M qi
i =1
M = Number of lanes
Application of Fourier’s law for the approximate
square in a lane
∆T j ∆T j
qi ≈ kAi ≈ k (∆y.l )
∆x ∆x
For equal temperature increments across the isotherms
N
∆T1− 2 = ∑ ∆T j =N∆T j
j =1 17
If the construction is made with Δx ≈Δy then the heat
transfer can be estimated by
Ml
q≈ k∆T1− 2
N

The above equation gives the two dimensional

conduction shape factor, S, as
S = Ml/N

18
Accordingly the two-dimensional conduction
resistance as
1
Rt ,cond ( 2 D ) =
Sk
Shape factors have been obtained analytically for
numerous two- and three-dimensional systems and
the results are summarized in [Link]. Case 9 is
a three dimensional case.
For one dimensional cases, shape factors may also be
defined for the familiar shapes as follows:
r2
Plane wall: A/L Cylinder: 2πL / ln
r1
Sphere: 4πr1r2/(r2-r1) 19
Cases 12 through 15 are for conduction heat rates
from isothermal objects at T1 embedded within an
infinite medium of uniform temperature (T2) and to
facilitate the procedure a characteristic length is
defined as
Lc ≡ (As/4π)1/2
where As is the surface area of the object.
Heat transfer rate to the infinite medium is given in
terms dimensionless conduction heat rate as
q ss ≡ qLc /[kAs (T1 − T2 )]
*

Values which have been determined analytically and

numerically are given as cases (12)-(15) in
20
Table 4.1.
Example 4.1
A metallic electrical wire of diameter d=5mm is to be
coated with insulation of thermal conductivity
k=0.35 W/m.K. It is expected that, for the typical
installation, the coated wire will be exposed to
conditions for which the total coefficient associated
with convection and radiation is h=15W/m2.K. To
minimize the temperature rise of the wire due to
ohmic heating, the insulation thickness is specified
so that the critical insulation radius is achieved.
During the wire coating process, however, the
insulation thickness sometimes varies around the
periphery of the wire, resulting in eccentricity
21
of the wire relative to the coating. Determine the
change in the thermal resistance of the insulation
due to an eccentricity that is 50% of the critical
insulation thickness.

22

Figure for example 4.1 22

Solution
rcr =k/h = 0.35/15= 0.023 m = 23 mm
Critical insulation thickness
tcr = rcr – d/2 = 0.023 – 0.005/2 = 0.021 m = 21 mm
For concentric geometry the thermal resistance is
given by
ln[rcr /(d / 2)] ln[0.023 /(0.005 / 2)]
R '
= = = 1.0 m.K / W
2πk 2π (0.35)
t ,cond

For the eccentric wire, use case 7 where

z=0.5tcr = 0.5 x 0.021 = 0.010 m

23
 −1  D + d − 4z 
2 2 2
cosh  
R '
=
1
=  2 Dd 
2πk
t ,cond
Sk
−1  ( 2 x 0.023) + 0.005 − 4 x 0.010 
2 2 2
cosh  
=  2x ( 2x 0.023) x 0.005 
2πx 0.35

= 0.91 m.K/W
The reduction in the thermal resistance of the
insulation is 0.09 m.K/W or 9%.
24
4.4 FIFNITE DIFFERENCE EQUATIONS
Large number of two-dimensional problems involve
geometries and/or boundary conditions which makes
analytical solutions impossible. The best
alternative, in such cases is to use numerical
technique such as finite difference, finite element or
boundary element method. Due to its ease of
application, finite difference method will be treated.
4.4.1 The Nodal Network
Numerical solution enables determination of
temperature at only discrete points. This requires
subdividing the medium of interest into a number of
small regions with reference points at the centers25
of the regions as shown in figchp-4\[Link]. This
reference point is termed as nodal point or node,
and the aggregate of points is termed a nodal
network, grid, or mesh.
m, n represents the node and the temperature at m, n is
a measure of the average temperature of the region
(shaded area), in this case represented by Tm,n. The
temperature distribution shown in (b) is
discontinuous. Accuracy of the method depends on
how fine the mesh is.

26
4.4.2 Finite Difference Form of the Heat Equation
Considering the second derivative (∂2T/∂x2) of an
interior node, the value of this derivative at the
nodal point m, n may be approximated as
∂ T
2
≈
∂T / ∂x m +1 / 2,n − ∂T / ∂x m −1 / 2,n
∆x
∂x 2
@ m ,n

The above can be determined as

∂T ≈
Tm +1,n − Tm ,n
∆x
∂x m +1 / 2 ,n

∂T ≈
Tm ,n − Tm −1,n
∆x
∂x m −1 / 2 ,n
27
Substitution in the 2nd derivative gives
∂T
2
Tm +1,n + Tm −1,n − 2Tm ,n
=
∂x @ m , n
2
( ∆x ) 2

Using similar procedures for the y-coordinate will

give
∂ T 2
∂T / ∂y m ,n +1 / 2 − ∂T / ∂y m ,n −1 / 2
≈
∆y
∂y 2
@ m ,n

Tm ,n +1 + Tm ,n −1 − 2Tm ,n
≈
( ∆y ) 2

28
Substitution in the differential equation gives
∂T ∂T
2 2
Tm +1,n + Tm −1,n − 2Tm ,n
+ 2 =0 ≈
∂x 2
∂y ( ∆x ) 2

Tm ,n +1 + Tm ,n −1 − 2Tm ,n
+
( ∆y ) 2

Using a network for which Δx = Δy will give

Tm,n+1 + Tm,n-1 + Tm+1,n + Tm-1,n - 4Tm,n = 0

Thus the PDE is reduced to an algebraic equation
which gives the temperature at a node as the average
of the temperatures of the four neighboring nodes.
29
4.4.3 The Energy Balance Method
This is an alternative method to develop the finite
difference equations for nodes. The approach is
more versatile than the one seen before. It uses
energy conservation principle and assumes that all
heat transfer is into the node as shown in figchp-
4\[Link] analysis will consider heat
generation too. For a unit depth
4
E in + E g = 0 ∑ q( i )→m,n + q [∆x ∆y(1)] = 0

i =1
 '''

30
The conduction heat rate from each node can be
written as
Tm −1,n − Tm ,n
q ( m −1,n )→( m ,n ) = k ( ∆y .1)
∆x
Tm+1,n − Tm ,n
q ( m+1,n )→( m ,n ) = k ( ∆y .1)
∆x
Tm ,n +1 − Tm ,n
q ( m ,n +1)→( m ,n ) = k ( ∆x .1)
∆y
Tm ,n −1 − Tm ,n
q ( m ,n −1)→( m ,n ) = k ( ∆x .1)
∆y 31
Using Δx =Δy and substituting in the summation
equation will give
Tm,n+1 + Tm,n-1 + Tm+1,n + Tm-1,n + 
q ( ∆x ) 2
- 4Tm,n = 0
k
With no heat generation it gives the same equation.
Equations must also be developed for nodes lying on
the boundary surface of the object. The surface may
enjoy convection heat transfer as shown in figchp-
4\[Link] which is an internal corner.
The equations can be developed as

32
 Tm −1,n Tm ,n
q ( m −1,n )→( m ,n ) = k ( ∆y .1)
∆x
1 Tm +1,n − Tm ,n
q ( m +1,n )→( m ,n ) = k ( ∆y .1)
2 ∆x
Tm ,n +1 − Tm ,n
q ( m ,n +1)→( m ,n ) = k ( ∆x .1)
∆y
1 Tm ,n −1 − Tm ,n
q ( m ,n −1)→( m ,n ) = k ( ∆x .1)
2 ∆y
Convection:
 ∆x   ∆y 
q ( ∞ )→( m ,n ) =  h .1(T∞ − Tm ,n ) +  h .1(T∞ − Tm ,n )
 2   2  33
With no energy generation and steady-state and
network with Δx = Δy
E in = 0
1 h∆x  h∆x 
Tm −1,n + Tm ,n +1 + (Tm +1,n + Tm ,n −1 ) + T∞ −  3 + Tm ,n = 0
2 k  k 

Nodal energy balance equations for several common

geometries is given in Table 4.2.
Example 4.2
Using the energy balance method, derive the finite
difference equation for the m, n nodal point located
on a plane, insulated surface of a medium with
34
uniform heat generation.

35
Solution
Using the energy conservation for node m,n
 ∆x 
q1 + q 2 + q 3 + q 4 + q 
'''
.∆y.1 = 0
 2 
where
Tm −1,n − Tm ,n
q1 = k( ∆y.1)
∆x
∆x Tm ,n −1 − Tm ,n
q 2 = k( .1)
2 ∆y
q3 = 0
∆x Tm ,n +1 − Tm ,n
q 4 = k( .1)
2 ∆y 36
Substituting in the energy balance equation and
simplifying gives
q ''' ( ∆x.∆y)
2Tm −1,n + Tm ,n −1 + Tm ,n +1 + −4Tm ,n + =0
k

37
The finite difference equations may also be formulated
using the thermal resistance concept. For the
example of the corner node (internal)
Tm −1,n − Tm ,n Tm −1,n − Tm ,n
q( m −1,n )→( m ,n ) = =
Rt ,cond ∆x /(∆y.1.k )
T∞ − Tm ,n T∞ − Tm ,n
q( ∞ )→( m ,n ) = = −1
Rt ,cond   ∆x   ∆y  
h  .1 +  .1 
  2   2  
The utility of the resistance can come when
considering two nodes of dissimilar materials with a
contact resistance between them figchp-
4\[Link]. T − T
m ,n −1
q ( m ,n )→( m ,n −1) = m ,n

R tot 38
Where
∆y / 2 ∆y / 2
''
R
R tot = + t ,c
+
k A ( ∆x.1) ∆x.1 k B ( ∆x.1)

4.5 SOLVING THE FINITE DIFFERENCE

EQUATIONS
These equations are the result of the nodal equations.
For N-nodal points there will be N equations. The
result is N linear algebraic with N unknown
temperatures.
Two types of solutions will be dealt with
• Direct and iterative 39
For small equations, direct is suitable while for larger
ones, iterative is better.
4.5.1 The Matrix Inversion Method
This is a direct solution. The nodal equations are
a11T1 + a12T2 + a13T3 + ... + a1N TN = C1
a 21T1 + a 22T2 + a 23T3 + ... + a 2 N TN = C2
. . . . .
. . . . .
. . . . .
a N1T1 + a N 2T2 + a N 3T3 + ... + a NN TN = C N
40
Using matrix notation, the above can be written as
[A][T] = [C]
where
a11 a12 ... a1N  T1  C1 
a a ... a  T  C 
 21 22 2N   2  2
. . . , .  . 
A≡ , T ≡  , C≡ 
. . .  .
  .
 
. . .  .  . 
     
a N1 a N 2 ... a NN  TN  C N 

41
Since
[A]-1[A] = I , then [A]-1[A][T]= [A]-1[C] or
[T] = [A]-1[C]
If the inverse of A is defined as

 b11 b12 ... b1N 

 b b ... b 
 21 22 2N 

. . . 
[A] ≡ 
−1

. . . 
. . . 
 
 b N1 b N 2 ... b NN 
42
The solution then becomes
T1 = b11C1 + b12C2 + … + b1NCN
T1 = b11C1 + b12C2 + … + b1NCN
. . . … .
. . . … .
. . . … .
T1 = b11C1 + b12C2 + … + b1NCN

Matrix inversion can easily be done by calculators and

computers. For large equation, this procedure is
time consuming.
43
4.5.2 Gauss-Seidel Iteration
This is an iteration procedure and is good for large
equations. The following procedure is followed.
1. As much as possible, the equations should be
reordered to have the largest coefficient at the
diagonal for quick convergence.
2. After reordering, each of the N equations should be
written in explicit form as
i −1 N
Ci a ij a ij ( k −1)
T i
(k)
= − ∑ Tj − ∑ Tj
(k)

a ii j=1 a ii j= i +1 a ii

44
where i = 1, 2, ….., N. The superscript k refers to the
level of iteration.
3. An initial (k=0) value is assumed for each
temperature Ti.
4. Setting k=1 in the equation, values of Ti(1) are then
calculated by substituting assumed values in the
second summation or new values in the first
summation. This step is the first (k=1) iteration.
New values of Ti(k)are determined from Tj(k) values
of the current iteration, where 1 ≤ j≤ i-1, and the
Tj(k-1) values of the previous iteration, where
i+1 ≤ j ≤N
45
5. The iteration is terminated when a prescribed
criterion is satisfied.
Ti( k ) − Ti( k −1) ≤ ε
where ε represents an error margin.
Example 4.3
A large industrial furnace is supported on a long
column fire clay brick, which is 1m by 1m on a
side. During steady-state operation, installation is
such that three surfaces of the column are
maintained at 500 K while the remaining surface is
exposed to an airstream for which T∞ = 300 K and
h = 10 W/m2.K. Using a grid of Δx = Δy = 0.25 m,
determine the two dimensional
46
temperature distribution in the column and the heat
rate to the airstream per unit length of column.

47
Solution
There are 12 nodes. By symmetry the unknowns will
be reduced to 8.
For interior nodes 1, 2, 3, 4, 5, and 6, the finite
difference equations are
Node 1: T2 +T3 + 1000 - 4T1 = 0
Node 2: 2T1 +T4 + 500 - 4T2 = 0
Node 3: T1 +T4 + T5+ 500 - 4T3 = 0
Node 4: T2 +2T3 + T6 – 4T4 = 0
Node 5: T3 +T6 +T7 + 500 - 4T5 = 0
Node 6: T4 +2T5 + T8 - 4T6 = 0
48
For the surface nodes 7 and 8, use equation 4.42 using
h Δx/k=2.5
Node 7: 2T5 + T8 + 2000 - 9T7 = 0
Node 8: 2T6 + 2T7 + 1500 - 9T8 = 0

49
-4T1 + T2 + T3 + 0 + 0 + 0 + 0 + 0 = -1000
2T1 - 4T2 + 0 + T4 + 0 + 0 + 0 + 0 = -500
T1 + 0 - 4T3 + T4 + T5 + 0 + 0 + 0 = -500
0 + T2 +2T3 - 4T4+ 0 + T6+ 0 + 0 = 0
0 + 0 + T3 + 0 -4T5 +T6 +T7 + 0 = -500
0 + 0 + 0 + T4 +2T5-4T6 + 0 + T8 = 0
0 + 0 + 0 + 0 +2T5+ 0 -9T7+ T8 = -2000
0 + 0 + 0 + 0 + 0 +2T6+2T7-9T8= -1500

50
In matrix form [A][T]=[C]

− 4 1 1 0 0 0 0 0   − 1000 
 2 − 4 0 1 0 0 0 0  − 500 
   
 1 0 −4 1 1 0 0 0   − 500 
   
 0 1 2− 4 0 1 0 0   0
A= [C] =
 0 0 1 0− 4 1 1 0   − 500 
   
 0 0 0 1 2− 4 0 1   0
 0 0 0 0 2 0− 9 1  − 2000
   
 0 0 0 0 0 2 2 − 9  − 1500 

51
The solution given by
[T]=[A]-1[C] is the column vector

T1  489.30
T  485.15
 2  
T3  472.07
   
 T4  462.01
[T ] = =
T5  436.95
   
T6  418.74
T  356.99
 7  
T8  339.05

52
The heat transfer from the column to the airstream per
unit length will be
 q  ∆x
  = 2h (Ts − T∞ ) + 2h∆x (T7 − T∞ ) + h∆x (T8 − T∞ )
 L 2
 ∆x  ∆x 
= 2 h  (Ts − T∞ ) + ∆x (T7 − T∞ ) + (T8 − T∞ )
 2  2 
= 2x10[0.125x 200 + 0.25x56.99 + 0.125x 39.05 = 883 W / m

To check the solution, the equality of the conduction

and convection heat transfers can be checked.

53
The temperature distribution of the above example can
also be determine using the Gauss-Seidel iteration
method.
As the equations give a diagonal dominant coefficient
matrix, it does not need any reordering. This is a
typical behavior of finite difference solutions to
conduction problems.
The explicit form of the equation according to step 2 is

54
T1( k ) = 0.25T2( k−1) + 0.25T3( k−1) + 250
T2( k ) = 0.50T1( k ) + 0.25T4( k −1) + 125
T3( k ) = 0.25T1( k ) + 0.25T4( k −1) + 0.25T5( k −1) + 250
T4( k ) = 0.25T2( k ) + 0.50T3( k ) + 0.25T6( k −1)
T5( k ) = 0.25T3( k ) + 0.25T6( k −1) + 0.25T7( k −1) + 125
T6( k ) = 0.25T4( k ) + 0.50T5( k ) + 0.25T8( k −1)
T7( k ) = 0.2222T5( k ) + 0.1111T8( k −1)
T8( k ) = 0.2222T6( k ) + 0.2222T7( k ) + 166.67

With initial assumption at k=0 and convergence

criterion of ε=0.2 K, the following table is
generated. 55
 k T1 T2 T3 T4 T5 T6 T7 T8
0 480.0 470.0 440.5 430.0 400.0 390.0 370.0 350.0
1 477.5 471.3 451.9 441.3 428.0 411.8 356.2 337.3
2 480.8 475.7 462.5 453.1 432.6 413.9 355.8 337.7
3 484.6 480.6 467.6 457.4 434.3 415.9 356.2 338.3
4 487.0 482.9 469.7 459.6 435.5 417.2 356.6 338.6
5 488.4 484.0 470.8 460.7 436.1 417.9 356.7 338.8
6 488.7 484.5 471.4 461.3 436.5 418.3 356.9 338.9
7 489.0 484.8 471.7 461.6 436.7 418.5 356.9 339.0
8 489.1 485.0 471.9 461.8 436.8 418.6 356.9 339.0

The results given in row 8 are in excellent agreement

with those obtained by matrix inversion

56