Computational Modelling : MEC3014/MEC8043

Chapter 1
Computational modelling of
steady diffusion process
Lecture 1: Solving 2D Steady Heat Conduction Equation

Prof. Nilanjan Chakraborty

E-mail: [email protected]

Governing law of conduction heat transfer

Conductive heat transfer rate in x direction is given by Fouriers law of heat
conduction

Qx = kA
x

T
q x = k
x

or

where T is temperature, k is the thermal conductivity and A is the area

normal to the direction of heat transfer and qx is the heat flux in the x
direction.
Heat flux is a vector. The generalised expression for conductive heat flux is
given by:

q = kT = k grad (T )

where
SI Units:

T = grad (T ) =
k W/m.K

T T T
k
i+
j+
x
y
z
qx and q W/m2
2

.
Qx Watt

General Heat Conduction Equation

Cartesian co-ordinates

Energy balance of the control volume gives us

E st = Q x + Q y + Q z (Q x + dQ x ) (Q y + dQ y ) (Q z + dQ z ) + E g
T
= qxyz

where Est = C ( xyz )

and E
g
t
q = volumetric heat generation
4

Let us consider x direction

Qx = (yz )q x = k (yz )
x
Q x + dQ x = (yz )(q x + dq x )
q x + dq x = q x + x

q x
= qx +
x
x

Similarly in y and z directions

Q y = (xz )q y = k (xz )
y

Qz = (xy )q z = k (xy )
z

Q y + dQ y = (xz )(q y + dq y ) Q z + dQ z = (xy )(q z + dq z )

q z
q y
z
y q z + dq z = q z + z = q z +
q y + dq y = q y + y = q y +
z
y
5

Est = C (xyz )
t
E = qxyz
g

Energy balance of the control volume gives us

E st = Q x + Q y + Q z (Q x + dQ x ) (Q y + dQ y ) (Q z + dQ z ) + E g
E = q yz + q xz + q xy
st

(q x + dq x )yz (q y + dq y )xz (q z + dq z )xy + E g

q y
q x
T
q
C
xyz =
xyz
xyz z xyz + qxyz
t
x
y
z
6

Energy balance of the control volume gives us

q x q y q z
T
C
=

+ q
z
t
x
y
According to Fouriers law of heat conduction

T
q x = k
x

T
q y = k
y

T
q z = k
z

T
T T T
+ k
+ q
C
+ k
= k

t x x y y z z
7

Generalised conduction
equation in Cartesian
co-ordinates.

Generalised conduction equation in Cartesian co-ordinates.

T T T T
+ k
+ q
C
= k
+ k

t x x y y z z
In vector notation the same equation is given by:

T
C
= .(kT ) + q
t

or

T
C
= div(k grad (T )) + q
t

If the thermal conductivity k remains constant

2T 2T 2T
T
= k 2 +
+
+ q
C
2
2
y
z
t
x
The above equation can be rewritten as:

2T 2T 2T q
1 T
+
+
+
=
2
2
2

x
y
z

k t
where

Thermal diffusivity
8

k
C

Generalised conduction equation in Cartesian co-ordinates.

2T 2T 2T q
1 T
+
+ =
2 +
2
2
y
z k t
x
For steady state

T
=0
t

For 2D problem

T
=0
z

Under this condition the governing equation becomes

2T 2T q
+ =0
2 +
2
y k
x

or

q
T + =0
k
2

2 (_) 2 (_) 2 (_)

(_) =
+
+
2
2
x
y
z 2
2

In 2D Laplace operator becomes

2 (_) 2 (_)
(_) =
+
2
y 2
x
2

Laplace operator

2D steady heat conduction equation with internal heat generation

2T
2T
q
+
=

2
2

y
k
x

2T =

or

q
k

The above partial differential equation is known as Poisson Equation

2D steady heat conduction equation without internal heat generation

2T
2T
+
=0

2
2

x
y

2T = 0

or

The above partial differential equation is known as Laplace Equation

The analytical solutions of Poisson and Laplace equations are often
mathematically involved and are beyond the scope of the present module.
Interested readers should consult the following text for analytical solutions.
Conduction of Heat in Solids: H.S. Carslaw & J.A. Jaeger (OUP)
We will opt for Numerical Solution Technique for 2D heat conduction problems.
10

Finite Difference Technique

S
Tw

Tp

TE

Tp

TN

TS

11

Examples of meshing of engineering structures are shown below. The figures

shown below demonstrate Cartesian adaptive grid for a full scale motorcycle
and an IC engine. It can be seen from the figures shown below that the
geometrical complexities can be handled satisfactorily if the cells are made
sufficiently small (i.e. If the mesh is sufficiently refined).

Courtesy: Profs. W.N. Dawes and R.S. Cant, Cambridge University

12

According to Taylor series expansion

T
1 2T
1 3T
2
3
4
x +

T ( x + x ) = T ( x ) +
x
x
O
x
(
)
(
)
(
)
2
3
x x
6 x x
2 x x
T ( x x ) = T ( x )

(1)

T
1 T
1T
2
3
4
x +

x
x
O
x
(
)
(
)
(
)
2
3
x x
2 x x
6 x x
2

(2)

P
x

T ( x + x ) = TE
T ( x x ) = TW
T ( x ) = TP

From eqs. (1) and (2) one gets:

2T
x 2

=
x

T ( x + x ) + T ( x x ) 2T ( x )
2
+

O
(
x
)
2
( x )

2T
x 2

=
P

TE + TW 2TP
2
+
O

x
(
)
2
( x )
13

According to Taylor series expansion

T
T ( y + y ) = T ( y ) +
y

1 2T
y +
2
2
y

1 3T
( y ) +
3
6
y

T
T ( y y ) = T ( y )
y

1 2T
y +
2
2

y
y

1 3T
( y )
3
6

y
y

( y ) 3 + O ( y 4 )

T ( y + y ) = TN
T ( y y ) = TS
T ( y ) = TP

From eqs. (3) and (4) one gets:

2T
y 2

=
y

T ( y + y ) + T ( y y ) 2T ( y )
2
+
O

y
(
)
2
( y )

2T
y 2

=
P

(3)

(4)

N
y

( y ) 3 + O ( y 4 )

TN + TS 2TP
2
+

O
y
(
)
2
( y )
14

 2T
x 2

T + TW 2TP
2
= E
+
O
(

x
)
2
( x )
Truncation error

2T
y 2

T + TS 2TP
2
= N
+

O
y
(
)
2
( y )

Discretised form of

2T
2T
q
+
=

2
y 2
k
x
Truncation error

TE + TW 2TP TN + TS 2TP
q
2
2
+
+ O ( x ) + O ( y ) =
2
2
( x )
( y )
k
TE + TW 2TP TN + TS 2TP
q
=

+
For small values of x and y:
k
( x ) 2
( y ) 2
For x = y = :

q 2
TE + TW + TN + TS 4TP + = 0
k
15

Discretised form of Poisson equation for equal

grid spacing:

q 2
TE + TW + TN + TS 4TP + = 0
k

P
y

Which can be rewritten as:

4TP = TE + TW + TN + TS +

q 2

The general form of the discretised equation can be written as:

a PT p = anbTnb + b

a PTP = a E TE + aW TW + a N TN + aS TS + b

Where subscript nb refers to neighbouring points

q
a P = 4, a E = 1, aW = 1, a N = 1, aS = 1 and b = 2
k

16

Discretised form of

2T
2T
+
=0

2
2
y
x
W

For small values of x and y:

TE + TW 2TP TN + TS 2TP
+
=0
2
2
( x )
( y )

P
y

For x = y = :
S

TE + TW + TN + TS 4TP = 0
The above discretised equation can be written as:

a PT p = anbTnb + b
where

a P = 4, a E = 1, aW = 1, a N = 1, aS = 1 and

17

b=0

Golden rules of the discretised equation:

a PT p = anbTnb + b
W

Rules
1. The discretisation coefficients aP, aE, aW ,
aN, aS have to be positive or zero.

2. The discretistaion coefficients will be such

that it will satisfy :

a P a E + aW + a N + aS

or a P anb

Explanation:
1. If temperature increases in the neighbouring nodes, temperature will
increase in node P.
2. The second criterion satisfies isothermal condition. The inequality holds
when there is a source term which depends on temperature.
18

Flux Boundary Condition Specification in Finite Difference Method

T
1 2T
2
3
x +

T ( x + x) = T ( x) +
x
O
x
(
)
(
)
2
x x
2 x x

T
x

T
x

T ( x + x) T ( x)
+ O ( x)
x

TE TP
+ O ( x)
x

Boundary condition for node P is given by:

T
q = k
x

T TP
q = k E
+ O ( x)
x

For small x the boundary condition is given by:

19

T TE
q = k P

Convective Boundary Condition Specification in Finite Difference Method

h
y

T
P

T
T ( y y ) = T ( y )
y

T
y

T
y

T ( y ) T ( y y )
+ O ( y )
y

TP TS
+ O ( y )
y

1 2T
2
3
y +

y
O
y
(
)
(
)
2
2 y x
y

Boundary condition for node P is given by:

T
k
y

= h(TP T )
P

Discretised boundary condition is given by:

+ O ( y )

P
TS TP

(
)
h
T
T
k

=
For small y the boundary condition is given by:

P
y
h(TP T ) = k

T
y

T TS
= k P
y

20

Solving 1D Heat Conduction problem using Finite Difference Method

L= 0.25m
50 oC

-20 oC

Discretised form
T1

T2

T3

T4

T5

T6

x= 0.05m

T1 = 50
T1 2T2 + T3 = 0
T2 2T3 + T4 = 0
T3 2T4 + T5 = 0
T4 2T5 + T6 = 0
T6 = 20

1 0 0
1 -2 1
0 1 -2
0 0 1
0 0 0
0 0 0

0 0 0
0 0 0
1 0 0
-2 1 0
1 -2 1
0 01

Tri-diagonal matrix
21

T1
T2
T3
T4
T5
T6

50
0
= 0
0
0
-20

Solving Tri-diagonal system of equations by Tri-Diagonal Matrix Algorithm (TDMA)

liTi 1 + aiTi + uiTi +1 = bi

0
0

d iTi + eiTi +1 = f i

(i)

(ii)

From eq. (ii) one gets:

d i 1Ti 1 + ei 1Ti = f i 1

Ti 1 = ( f i ei 1Ti ) / d i 1

(iii)

Substituting eq. (iii) in eq. (i) one gets:

f i 1 ei 1Ti
li
d i 1

e
f
ai li i 1 Ti + uiTi +1 = bi li i 1
d i 1
d i 1

+ aiTi + uiTi +1 = bi

ei 1

;
d i = ai li
d i 1

ei = ui

f i 1

f i = bi li
d i 1

22

Solving 1D Heat Conduction problem using Finite Difference Method

L= 0.25m
50 oC

-20 oC

Discretised form
T1

T2

T3

T4

T5

T6

Analytical solution

T ( x) = 280 x + 50
x= 0.05m
T1 = 50

T1

T1 2T2 + T3 = 0

T2

T2 2T3 + T4 = 0

T3

T3 2T4 + T5 = 0

T4

T4 2T5 + T6 = 0

T5

T6 = 20

T6

50
36
= 22
8
-6
-20

T1
T2
T3
T4
T5
T6

50
36
22
= 8
-6
-20

The numerical scheme is second order accurate and the actual profile is linear so the
numerical scheme here gives exact solution because there is no error due to truncation of
higher order terms.
23

Solving 2D Heat Conduction problem using Finite Difference Method

100
100

0
1

0
0

For each nodes with

unknown temperature we
can write the discretised
equation in the form :
TE + TW + TN + TS 4TP = 0

100

100

100

100

100

For point 1:
T2 + 100 + 0 + T4 4T1 = 0

0
For point 4:

For point 2:

For point 3:

T3 + T1 + 0 + T5 4T2 = 0

0 + T2 + 0 + T6 4T3 = 0 T5 + 100 + T1 + 100 4T4 = 0

For point 5:

For point 6:

T6 + T4 + T2 + 100 4T5 = 0

0 + T5 + T3 + 100 4T6 = 0

We have a closed set of 6 equations with 6 unknowns

24

Solving 2D Heat Conduction problem using Finite Difference Method

Set of equations:
100
100

100

100

0
1

0
0

T2 + 100 + 0 + T4 4T1 = 0
T3 + T1 + 0 + T5 4T2 = 0

0 + T2 + 0 + T6 4T3 = 0

100

100

T5 + 100 + T1 + 100 4T4 = 0

T6 + T4 + T2 + 100 4T5 = 0

100

0 + T5 + T3 + 100 4T6 = 0

One can solve the above set of equations by calculator. However we have 6
equations but in the actual problem we might have 1000 equations. As direct
Matrix inversion is a computationally expensive process the system of
equation is solved by iterative numerical methods. One such method is
Gauss Seidel Iteration scheme.

25

Gauss Seidel Iteration

1. Equations should be constructed in the following manner

a PT p = anbTnb + b
It is necessary to make sure that coefficients aP, aE, aW , aN, aS follow
the Golden rules of discretisation.
2. Make an initial guess about the unknowns.
3. Recast the equation in the following manner:

Tp =
k

a
a
a
b
a
(k )
(k )
( k 1)
( k 1)
+ W TW
+ N TN + E TE
+ S TS
aP
aP
aP
aP
aP

where k and k -1 are iteration counts and are not exponents. For initial
guess k = 0.
4. Check whether a prescribed convergence criterion is satisfied:
TP

(k )

TP

( k 1)

<

is a small number

5. If not carry out steps 3 and 4 until convergence.

26

Solving 2D Heat Conduction problem using Finite Difference Method

Set of equations:
100

0
1

100

0
0

4T1 = T2 + 100 + T4
4T2 = T3 + T1 + T5

4T3 = T2 + T6

100

100

100

100

4T4 = T5 + T1 + 200
4T5 = T6 + T4 + T2 + 100

100

4T6 = T5 + T3 + 100

Initial guess based on the boundary condition & common sense

T1

(0)

T1

(1)

T2

(1)

= 60, T2

= (T2

(0)

(0)

= (T3

(0)

= 50, T3

+ 100 + T4
+ T1

(1)

= 50, T4

(0)

(0)

+ T5

(0)

= 80, T5

(0)

= 70 and T6

(0)

= 50

) / 4 = (50 + 80 + 100) / 4 = 57.5

(0)

) / 4 = (50 + 57.5 + 70) / 4 = 44.4

The process is repeated till T6 using most recent values. When improved values of T1 to
T6 are obtained the whole process is repeated until convergence
27

Solving 2D Heat Conduction problem using Finite Difference Method

Initial guess

1st iteration

2nd iteration

3rd iteration

4th iteration

T1 = 60

57.5

56.6

54.7

54.0

T2 = 50

44.4

37.3

35.7

35.1

T3 = 50

23.6

21.4

20.7

20.4

T4 = 80
T5 = 70
T6 = 50

81.9

81.4

80.4

80.0

69.0

66.7

65.8

65.4

48.2

47.0

46.6

46.5

It is worth doing a couple of iterations before the convergence

Actual solution of discretised equation:
T1 = 53.6, T2 = 34.8, T3 = 20.3, T4 = 79.7, T5 = 65.2 and T6 = 46.4
The actual solution of discretised equation may not be the exact solution of the physical
problem. In general it is appropriate to solve the problem with finer mesh. If the solution
does not change appreciably with finer mesh then grid independence is said to be
achieved. At that point the solution can be taken as the exact solution of the problem.
28

Energy Balance Method

x / 2

Q
N

y / 2
W

P
q

Q
W

According to Fouriers law of heat conduction

= k TE TP ( y.1)
= k TN TP ( x.1) Q
= k TS TP ( x.1) Q
Q
N
; S
; E
x
y
y
= k TW TP ( y.1)
Q
W
x
From energy balance: Q N + Q S + Q E + QW + q (xy.1) = 0

TE + TW + TN + TS 4TP +
29

q 2
=0
k

Q
E

Q
S

For x = y = :

Application of Energy balance method in constructing boundary conditions

Case 1: Boundary with convection
h

Q
W

Q
N

Q
E

= h( x.1)(T T )
Q
N

y / 2

= k TS TP ( x.1)
Q
S
y

Q
S

= k TE TP y .1
Q
E
x 2

S
x / 2

= k TW TP y .1
Q
W
x 2

From energy balance method:

+Q
+Q
+Q
+ q xy .1 = 0
Q
N
S
E
W

For x = y = :
(TE + TW + 2TS ) + 2

h
h
q 2

T 4 + 2
TP + = 0
k
k
k

30

Application of Energy balance method in constructing boundary conditions

Case 2: External corner

Q
W

Q
N

Q
E
h
Q
S

= h x .1(T T )
Q
N

P
2
= k TS TP x .1
Q
S
y 2
= h y .1(T T )
Q
E

P
2

y / 2
y

= k TW TP y .1
Q
W
x 2

x / 2
From energy balance method:
+Q
+Q
+Q
+ q xy .1 = 0
Q
N
S
E
W

For x = y = :
h
h
q 2

=0
T +
(TW + TS ) 2 + 2
TP + 2
k
k
2k

31

Application of Energy balance method in constructing boundary conditions

Case 3: Internal corner
x

N
Q
N

W QW

T h

Q
E

Q
S

= h x .1(T T ) + k x .1 TN TP
Q

N
P

2 y
2
= h y .1(T T ) + k y .1 TE TP
Q
E
P

2 x
2

= k TW TP ( y.1)
Q
W
x
= k TS TP ( x.1)
Q
S
y

y / 2

S
x / 2
From energy balance method:
+Q
+Q
+Q
+ q 3xy .1 = 0
Q
N
S
E
W

For x = y = :
3q 2
h
h

(TN + TE + 2TS + 2TW ) 2 3 +

=0
T +
TP + 2
2
k
k
k

32

EXAMPLE
The cross-section of a water-cooled component is illustrated below. The wall is
symmetrical about AB. The faces AF and EF are maintained at 200o C whilst face ED is
well insulated. For faces BC and CD, which are exposed to water at 20o C, the convective
heat transfer coefficient is 500 W/m2K. As shown in the figure a 10 mm square mesh of
nodal points is employed in a finite difference analysis of the two-dimensional, steadystate, temperature distribution. The thermal conductivity of the material is 0.25 W/mK.
(a) Derive discretised equations for nodes 1 to 7 based on energy balance method.
(b) Determine the temperatures at nodes 1 to 7.
(c) Calculate the rate of convective heat transfer from the component per metre depth into
the page.
200o C

F
1
200o C

4
6

2
5

3
C

20mm

h = 500 W/m2K
T= 20oC

10mm

7
30mm

30mm

D
33

Epilogue
You will need to understand this part to understand the rest of the
module. Please devote time on this material.
And

Most of the practical heat transfer problems in engineering analysis

are solved using the numerical techniques described in this
lecture.
Moreover

The numerical techniques described above are also valid for 3D

problems
The next lecture will be on solving unsteady (transient) heat conduction
problems.
34