88 3-5 Numerical Method of Analysis

Note that this relation differs from the one for an insulated surface by the minus sign in the
denominator. Inserting r = 0.15 m and D = 1.0 m we obtain
4π(0.15) 4π(0.15)
S= = = 1.26 m
[π/2 − tan−1 (0.15/2)] [π/2 − 0.07486]

For buried objects the shape factor is based on 1T = Tobject − Tfar field . The far-field temperature
is taken as the isothermal surface temperature, and the heat lost by the disk is therefore
q = kS1T = (2.1)(1.26)(95 − 20) = 198.45 W

EXAMPLE 3-4 Buried Parallel Disks

Two parallel 50-cm-diameter disks are separated by a distance of 1.5 m in an infinite medium
having k = 2.3 W/m · ◦ C. One disk is maintained at 80◦ C and the other at 20◦ C. Calculate the
heat transfer between the disks.
Solution
This is a shape-factor problem and the heat transfer may be calculated from
q = kS1T

where S is obtained from Table 3-1 as

4πr
S= for D > 5r
[π/2 − tan−1 (r/D)]

With r = 0.25 m and D = 1.5 m we obtain

4π(0.25) 4π(0.25)
S= = = 2.235
[π/2 − tan−1 (0.25/1.5)] [π/2 − 0.1651]

and
q = kS 1T = (2.3)(2.235)(80 − 20) = 308.4 W

3-5 NUMERICAL METHOD OF ANALYSIS

Figure 3-5 Sketch An immense number of analytical solutions for conduction heat-transfer problems have
illustrating nomenclature used been accumulated in the literature over the past 150 years. Even so, in many practical
in two-dimensional numerical situations the geometry or boundary conditions are such that an analytical solution has not
analysis of heat conduction.
been obtained at all, or if the solution has been developed, it involves such a complex series
solution that numerical evaluation becomes exceedingly difficult. For such situations the
m, n + 1 most fruitful approach to the problem is one based on finite-difference techniques, the basic
principles of which we shall outline in this section.
∆y Consider a two-dimensional body that is to be divided into equal increments in both
m − 1, n m, n m + 1, n the x and y directions, as shown in Figure 3-5. The nodal points are designated as shown,
the m locations indicating the x increment and the n locations indicating the y increment.
∆y
We wish to establish the temperatures at any of these nodal points within the body, using
m, n − 1
Equation (3-1) as a governing condition. Finite differences are used to approximate differ-
∆x ∆x ential increments in the temperature and space coordinates; and the smaller we choose these
finite increments, the more closely the true temperature distribution will be approximated.
 CHAPTER 3 Steady-State Conduction—Multiple Dimensions 89

The temperature gradients may be written as follows:


∂T Tm+1,n − Tm,n
≈
∂x m+1/2,n 1x

∂T Tm,n − Tm−1,n
≈
∂x m−1/2,n 1x

∂T Tm,n+1 − Tm,n
≈
∂y m,n+1/2 1y

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

≈ =
∂x2 m,n 1x (1x)2
 
∂T ∂T
2
−
∂y m,n+1/2 ∂y m,n−1/2 Tm,n+1 + Tm,n−1 − 2Tm,n

∂ T
2
≈ =
∂y m,n 1y (1y)2
Thus the finite-difference approximation for Equation (3-1) becomes
Tm+1,n + Tm−1,n − 2Tm,n Tm,n+1 + Tm,n−1 − 2Tm,n
+ =0
(1x)2 (1y)2

If 1x = 1y, then
Tm+1,n + Tm−1,n + Tm,n+1 + Tm,n−1 − 4Tm,n = 0 [3-24]

Since we are considering the case of constant thermal conductivity, the heat flows may all
be expressed in terms of temperature differentials. Equation (3-24) states very simply that
the net heat flow into any node is zero at steady-state conditions. In effect, the numerical
finite-difference approach replaces the continuous temperature distribution by fictitious
heat-conducting rods connected between small nodal points that do not generate heat.
We can also devise a finite-difference scheme to take heat generation into account. We
merely add the term q̇/k into the general equation and obtain
Tm+1,n + Tm−1,n − 2Tm,n Tm,n+1 + Tm,n−1 − 2Tm,n q̇
+ + =0
(1x)2 (1y)2 k

Then for a square grid in which 1x = 1y,

q̇(1x)2
Tm+1,n + Tm−1,n + Tm,n+1 + Tm,n−1 + − 4Tm,n−1 = 0 [3-24a]
k
To utilize the numerical method, Equation (3-24) must be written for each node within the
material and the resultant system of equations solved for the temperatures at the various
nodes. A very simple example is shown in Figure 3-6, and the four equations for nodes 1,
2, 3, and 4 would be
100 + 500 + T2 + T3 − 4T1 = 0
T1 + 500 + 100 + T4 − 4T2 = 0
90 3-5 Numerical Method of Analysis

Figure 3-6 Four-node problem.

T = 500˚C

T = 100˚C
T = 100˚C
1 2

3 4

T = 100˚C

100 + T1 + T4 + 100 − 4T3 = 0

T3 + T2 + 100 + 100 − 4T4 = 0

These equations have the solution

T1 = T2 = 250◦ C T3 = T4 = 150◦ C

Of course, we could recognize from symmetry that T1 = T2 and T3 = T4 and would then
only need two nodal equations,
100 + 500 + T3 − 3T1 = 0
100 + T1 + 100 − 3T3 = 0

Once the temperatures are determined, the heat flow may be calculated from
X 1T
q= k 1x
1y

where the 1T is taken at the boundaries. In the example the heat flow may be calculated
at either the 500◦ C face or the three 100◦ C faces. If a sufficiently fine grid is used, the two
values should be very nearly the same. As a matter of general practice, it is usually best to
take the arithmetic average of the two values for use in the calculations. In the example, the
two calculations yield:
500◦ C face:
1x
q = −k [(250 − 500) + (250 − 500)] = 500k
1y
100◦ C face:
1y
q = −k [(250 − 100) + (150 − 100) + (150 − 100) + (150 − 100)
1x
+ (150 − 100) + (250 − 100)] = −500k

and the two values agree in this case. The calculation of the heat flow in cases in which
curved boundaries or complicated shapes are involved is treated in References 2, 3, and 15.
 CHAPTER 3 Steady-State Conduction—Multiple Dimensions 91

When the solid is exposed to some convection boundary condition, the temperatures
at the surface must be computed differently from the method given above. Consider the
boundary shown in Figure 3-7. The energy balance on node (m, n) is
Tm,n − Tm−1,n 1x Tm,n − Tm,n+1 1x Tm,n − Tm,n−1
−k 1y −k −k
1x 2 1y 2 1y
= h 1y(Tm,n − T∞ )

If 1x = 1y, the boundary temperature is expressed in the equation

 
h 1x h 1x 1
Tm,n +2 − T∞ − (2Tm−1,n + Tm,n+1 + Tm,n−1 ) = 0 [3-25]
k k 2

An equation of this type must be written for each node along the surface shown in
Figure 3-7. So when a convection boundary condition is present, an equation like (3-25) is
used at the boundary and an equation like (3-24) is used for the interior points.
Equation (3-25) applies to a plane surface exposed to a convection boundary condition.
It will not apply for other situations, such as an insulated wall or a corner exposed to
a convection boundary condition. Consider the corner section shown in Figure 3-8. The
energy balance for the corner section is
1y Tm,n − Tm−1,n 1x Tm,n − Tm,n−1 1x 1y
−k −k =h (Tm,n − T∞ ) + h (Tm,n − T∞ )
2 1x 2 1y 2 2
If 1x = 1y,
 
h 1x h 1x
2Tm,n +1 −2 T∞ − (Tm−1,n + Tm,n−1 ) = 0 [3-26]
k k

Other boundary conditions may be treated in a similar fashion, and a convenient sum-
mary of nodal equations is given in Table 3-2 for different geometrical and boundary
situations. Situations f and g are of particular interest since they provide the calcula-
tion equations that may be employed with curved boundaries, while still using uniform
increments in 1x and 1y.

Figure 3-7 Nomenclature for nodal Figure 3-8 Nomenclature for nodal
equation with convective equation with convection at
boundary condition. a corner section.

m < 1, n m, n
6y
m, n + 1 2
T' 6y T'
m < 1, n m, n 6y
m, n < 1
6y m < 1, n < 1
6x
q 2
6x
m, n < 1
6x
2 Surface
6x
92 3-5 Numerical Method of Analysis

Table 3-2 Summary of nodal formulas for finite-difference calculations. (Dashed lines indicate element volume.)†
Nodal equation for equal increments in x and y
Physical situation (second equation in situation is in form for Gauss-Seidel iteration)

(a) Interior node 0 = Tm+1,n + Tm,n+1 + Tm−1,n + Tm,n−1 − 4Tm,n

m, n + 1 Tm,n = (Tm+1,n + Tm,n+1 + Tm−1,n + Tm,n−1 )/4

∆y
m − 1, n m, n m + 1, n

∆y
m, n − 1

∆x ∆x

 
0 = h1x 1 h1x + 2 T
(b) Convection boundary node k T∞ + 2 (2Tm−1,n + Tm,n+1 + Tm,n−1 ) − k m,n

m, n + 1 Tm−1,n +(Tm,n+1 +Tm,n−1 )/2+Bi T∞

Tm,n = 2+Bi
∆y
m − 1, n h, T∞ Bi = h1x
k
m, n
∆y
m, n − 1

∆x

 
(c) Exterior corner with convection boundary 0 = 2 h1x
k T∞ + (Tm−1,n + Tm,n−1 ) − 2
h1x + 1 T
k m,n
h, T∞
m − 1, n (Tm−1,n + Tm,n−1 )/2 + Bi T∞
m, n Tm,n = 1+Bi
∆y
Bi = h1x
k
m, n − 1

∆x

 
(d) Interior corner with convection boundary 0 = 2 h1x h1x T
k T∞ + 2Tm−1,n + Tm,n+1 + Tm+1,n + Tm,n−1 − 2 3 + k m,n

m, n + 1 Bi T∞ +Tm,n+1 +Tm−1,n +(Tm+1,n + Tm,n−1 )/2

Tm,n = 3+Bi

Bi = h1x
k
m − 1, n m, n m + 1, n

∆y h, T∞
m, n − 1

∆x
 CHAPTER 3 Steady-State Conduction—Multiple Dimensions 93

Table 3-2 (Continued).

Nodal equation for equal increments in x and y
Physical situation (second equation in situation is in form for Gauss-Seidel iteration)

(e) Insulated boundary 0 = Tm,n+1 + Tm,n−1 + 2Tm−1,n − 4Tm,n

Tm,n = (Tm,n+1 + Tm,n−1 + 2Tm−1,n )/4

m, n + 1
Insulated

m < 1, n m, n

6y
m, n < 1

6x
 
(f ) Interior node near curved boundary‡ 2 T
0 = b(b 2+ 1) T2 + a + 2 2 1 1
1 m+1,n + b + 1 Tm,n−1 + a(a + 1) T1 − 2 a + b Tm,n
m, n + 1
h,T' 3
2 c 6 x 6y
1 b 6x
m < 1, n m, n m + 1, n
a 6x
6y
m, n < 1

6x 6x
p p
(g) Boundary node with convection along 0= √ b T1 + √ b T3 + a+1 h1x 2 2 2
b Tm,n + k ( c + 1 + a + b )T∞
a2 + b2 c2 + 1
curved boundary—node 2 for (f ) above§
h p p i
− √ b +√b + a+1 2 2 2 h1x
b + ( c + 1 + a + b ) k T2
a2 + b2 c2 +1

† Convection boundary may be converted to insulated surface by setting h = 0 (Bi = 0).

‡ This equation is obtained by multiplying the resistance by 4/(a + 1)(b + 1)
§ This relation is obtained by dividing the resistance formulation by 2.

Nine-Node Problem EXAMPLE 3-5

Consider the square of Figure Example 3-5. The left face is maintained at 100◦ C and the top face
at 500◦ C, while the other two faces are exposed to an environment at 100◦ C:
h = 10 W/m2 · ◦ C and k = 10 W/m · ◦ C

The block is 1 m square. Compute the temperature of the various nodes as indicated in
Figure Example 3-5 and the heat flows at the boundaries.
Solution
The nodal equation for nodes 1, 2, 4, and 5 is
Tm+1,n + Tm−1,n + Tm,n+1 + Tm,n−1 − 4Tm,n = 0

The equation for nodes 3, 6, 7, and 8 is given by Equation (3-25), and the equation for 9 is given
by Equation (3-26):
h1x (10)(1) 1
= =
k (3)(10) 3
94 3-5 Numerical Method of Analysis

Figure Example 3-5 Nomenclature for Example 3-5.

T = 500˚C

1 2 3

T = 100˚C
1m
4 5 6
T∞ =
100˚C

7 8 9

1m

The equations for nodes 3 and 6 are thus written

2T2 + T6 + 567 − 4.67T3 = 0
2T5 + T3 + T9 + 67 − 4.67T6 = 0

The equations for nodes 7 and 8 are given by

2T4 + T8 + 167 − 4.67T7 = 0
2T5 + T7 + T9 + 67 − 4.67T8 = 0

and the equation for node 9 is

T6 + T8 + 67 − 2.67T9 = 0
We thus have nine equations and nine unknown nodal temperatures. We shall discuss solution
techniques shortly, but for now we just list the answers:

Node Temperature, ◦ C

1 280.67
2 330.30
3 309.38
4 192.38
5 231.15
6 217.19
7 157.70
8 184.71
9 175.62
The heat flows at the boundaries are computed in two ways: as conduction flows for the 100 and
500◦ C faces and as convection flows for the other two faces. For the 500◦ C face, the heat flow
into the face is
X 1T h  i
q= k1x = (10) 500 − 280.67 + 500 − 330.30 + (500 − 309.38) 21
1y
= 4843.4 W/m

The heat flow out of the 100◦ C face is

X 1T h  i
q= k1y = (10) 280.67 − 100 + 192.38 − 100 + (157.70 − 100) 12
1x
= 3019 W/m
 CHAPTER 3 Steady-State Conduction—Multiple Dimensions 95

The convection heat flow out the right face is given by the convection relation
X
q= h1y(T − T∞ )
 h  i
= (10) 13 309.38 − 100 + 217.19 − 100 + (175.62 − 100) 21

= 1214.6 W/m

Finally, the convection heat flow out the bottom face is

X
q= h1x(T − T∞ )
 h    i
= (10) 13 (100 − 100) 12 + 157.70 − 100 + 184.71 − 100 + (175.62 − 100) 12

= 600.7 W/m

The total heat flow out is

qout = 3019 + 1214.6 + 600.7 = 4834.3 W/m

This compares favorably with the 4843.4 W/m conducted into the top face. A solution of this
example using the Excel spreadsheet format is given in Appendix D.

Solution Techniques
From the foregoing discussion we have seen that the numerical method is simply a means
of approximating a continuous temperature distribution with the finite nodal elements. The
more nodes taken, the closer the approximation; but, of course, more equations mean more
cumbersome solutions. Fortunately, computers and even programmable calculators have
the capability to obtain these solutions very quickly.
In practical problems the selection of a large number of nodes may be unnecessary
because of uncertainties in boundary conditions. For example, it is not uncommon to have
uncertainties in h, the convection coefficient, of ±15 to 20 percent.
The nodal equations may be written as
a11 T1 + a12 T2 + · · · + a1n Tn = C1
a21 T1 + a22 T2 + · · · = C2
a31 T1 + · · · = C3 [3-27]
....................................
an1 T1 + an2 T2 + · · · + ann Tn = Cn

where T1 , T2 , . . . , Tn are the unknown nodal temperatures. By using the matrix notation
C  T 
1 1
a
11 a 12 · · · a 1n

 C2   T2 
 a21 a22 · · ·
 ·   · 
    
[A] =  a31 ··· [C] =   [T ] =  
 
 ·   · 

 .................. 
 ·   · 
an1 an2 · · · ann
Cn Tn

Equation (3-27) can be expressed as

[A][T ] = [C] [3-28]