Chapter 5

Two-Dimensional and Transient

Heat Conduction
5.1 IMPORTANCE OF TWO-DIMENSIONAL AND TRANSIENT
HEAT CONDUCTION PROBLEMS
In the previous chapter, finite difference method for solving the one-
dimensional steady state heat conduction systems has been presented.
Although it only focuses on the heat conduction systems, boundary condi-
tions coupling the convective and radiative heat transfer are also
discussed.
Nonetheless, in many engineering systems, heat conduction exhibits pre-
dominantly in two- or three-dimensional nature. Engineering systems consist
of multiple components where thermal properties could be significantly dif-
ferent, leading to second- or third-dimension temperature gradient and its
corresponding heat transfer. In some cases, inhomogeneous temperature dis-
tributions could lead to substantial or even significant thermal stress to the
systems—causing deformation, crack, or even fracture. A simple daily exam-
ple of such phenomenon is the cracking of a glass cup when suddenly heat is
imposed inside. When hot water pouring inside the cup, the inner glass sur-
faces are subjected to high temperature. However, due to the low thermal
conductivity, outer glass surfaces are still maintained in low temperature.
The unevenly temperature distribution causes significant thermal stress that
eventually cracks the glass cup.
Solving the two-dimensional heat conduction could obtain useful tem-
perature distribution for engineers and design to optimize their designs.
Since most of the engineering systems are designed for long-term opera-
tion, design considerations are commonly focused on the steady state
temperature distribution. Nevertheless, in some situation (especially for
high-power devices), transient heat generation and conduction could incur
significant temporal thermal load. It is also important to resolve the tran-
sient heat conduction to obtain the time-dependent temperature generation
and dissipation characteristics.

Demystifying Numerical Models. DOI: https://doi.org/10.1016/B978-0-08-100975-8.00005-9

© 2019 Elsevier Ltd. All rights reserved. 87
88 Demystifying Numerical Models

5.2 DIRECT VERSUS ITERATIVE METHODS

Through the finite difference method, heat conduction governing equation
can be discretized into a set of simultaneous algebraic equations. Solutions
of the simultaneous equations can be obtained by simple matrix inversion
function available in spreadsheets. Such matrix inversion procedures are
referred as direct method in literature. As demonstrated in Chapter 4, the
matrix inversion procedures require user manually formulate and input
the coefficient matrices. Referring the problem example in Chapter 4, a one-
dimensional problem with six mesh nodes results in a 6 3 6 coefficient
matrix. For simple one-dimensional steady state heat conduction, the solution
procedures seem to be manageable for most applications. Nevertheless, for
two-dimensional problems, it is common to have 10 or even 100 nodes in
one domain. The large number of mesh results a significant large coefficient
matrix, marking the formulation and manual input procedures unbearable
(i.e., 10 nodes require a 10 3 10 matrix with 100 coefficients).
To eliminate the tedious procedures in direct method, for large number of
mesh, it is common to adopted iterative method where the solutions are
obtained by a “trial-and-error” procedure with an initial-guess. In theory the
direct and iterative methods should yield identical solutions. Obviously, this
assumes that a converged solution could be obtained through the iterative
method. To ensure an accurate result could be obtained, one should specify a
convergence target to terminate the iterative procedures. With the advance-
ment of software, aforementioned iterative procedures could also be imple-
mented easily in the spreadsheet. Implementation of such two-dimensional
heat conduction iterative method is discussed in this chapter.

5.3 TWO-DIMENSIONAL STEADY STATE HEAT

CONDUCTION
In this session, following the one-dimension heat conduction problem, we
will consider extending our formulation to two-dimensional steady state con-
duction problem in rectangular domain. Finite difference method is adopted
to discretize the heat conduction governing equation. Demonstrating the for-
mulation aims in twofold, readers can follow similar formulation procedures
extending the problem to three-dimensional cases; it presents the mathemati-
cal foundation for implementing the iterative procedures.

5.3.1 Two-Dimensional Discretization

We firstly consider the steady two-dimensional heat conduction through
a rectangular solid block where temperature variation is significant in both
x- and y-directions. Aiming to determine the temperature distribution in the
two-dimensional space, a numerical mesh is needed dividing the rectangular
 Two-Dimensional and Transient Heat Conduction Chapter | 5 89

FIGURE 5.1 A two-dimensional numerical mesh for the finite difference discretization.

domain in a number of nodal points with the spacings of Δx and Δy along

the x and y direction. A visualization of the numerical mesh is depicted in
Fig. 5.1.
Since the formulation for all internal nodes is identical, it is convenient
to refer the nodal point based on numbers or notations instead of actual phys-
ical location. As depicted in Fig. 5.1, it is common to use double subscript
notation (i, j) to number a two-dimensional numerical mesh. In x-direction,
nodal point is notated as i 5 1, 2, 3 . . ., N, while j 5 1, 2, 3 . . ., M is counted
along the y-direction. For a uniformly spaced mesh the actual coordinates of
the nodal point can be easily calculated based on the nodal number. One
should notice that choosing double subscript for notation is mainly for the
convenience of programming. Most of spreadsheet calculations are based on
cells that are also notated in double subscription using columns (i.e., A, B,
. . ., Z and more) and rows (i.e., 1, 2, 3, . . ., 10 and more). The following for-
mulation could be also easily referred to cells operation working in
spreadsheet.
Similar to the one-dimensional problem, we focus on developing a
generic formulation for all internal nodal points (i.e., point (i, j)).
Considering the nodal point (i, j) with the volume of Δx 3 Δy 3 1, the sur-
faces of the control volume are placed at the middle among all nodal points.
Heat is being conducted into the control volume through the surface from
both vertical and horizontal direction. Meanwhile, heat is also generated
within the control volume (Fig. 5.2).
Considering the heat balance on the control volume, the governing equa-
tion can be expressed as
qx;left 2 qx;right 1 qy;bottom 2 qy;top 1 qb 5 0 ð5:1Þ
90 Demystifying Numerical Models

i–1 Δx i Δx i+1

j+1

Δy
(i, j)
j

Δy

j–1

FIGURE 5.2 Heat balance considerations for the two-dimensional internal nodal points.

Based on the Fourier’s law of heat conduction, the heat balance is given
by
       
@T @T @T @T
kA 2 kA 1 kA 2 kA 1 qb ΔV 5 0
@x left @x right @y bottom @y top
ð5:2Þ
Following the first-order differencing scheme discussed earlier, the tem-
perature gradient at the left and right is discretized as follows:
Ti21; j 2 Ti;j Ti; j 2 Ti11;j Ti; j21 2 Ti;j Ti; j 2 Ti;j11
kΔy 2 kΔy 1 kΔx 2 kΔx 1 qb Δ xΔy 5 0
Δx Δx Δy Δy
ð5:3Þ
Simplifying the equation gives
Ti21; j 2 2Ti;j 1 Ti11;j Ti; j21 2 2Ti;j 1 Ti;j11
kAx 1 kAy 1 qb ΔxΔy 5 0 ð5:4Þ
Δx Δy
where Ax and Ay are the area where heat conduction flux is passing through
in x and y direction, respectively. Based on Eq. (5.4), one can notice that
governing equation is now discretized into algebraic equation which depends
on temperature difference among nodal points (or the temperature stored in
the cells of the spreadsheet).
Similar to the one-dimensional case, we could also rearrange the equation
in form as following:
ai;j Ti;j 5 ai11;j Ti11;j 1 ai21;j Ti21;j 1 ai;j11 Ti;j11 1 ai;j21 Ti;j21 1 S ð5:5Þ
where ai,j, ai21,j, ai11,j, ai,j11, ai,j21, and S are the coefficients of the
equation which could be easily implemented in spreadsheet. Details of the
implementation could be demonstrated in the following problem
examples.
 Two-Dimensional and Transient Heat Conduction Chapter | 5 91

5.3.2 Two-Dimensional Heat Conduction in a

Square Steel Column
We first consider a steady two-dimensional heat conduction through a square
steel column. The steel column with a cross-sectional area of 5 cm 3 5 cm is
used to support the structure a furnace. One side of the steel column is
exposed to high temperature of 160oC due to the heat loss from the combus-
tion chamber. The other side is exposed to a constant temperature of 60 C
which is adjacent to a recovery heat exchanger (see also in Fig. 5.3). The
temperature difference forms a constant temperature gradient cross the hori-
zontal direction. On the other hand, insulations were deployed covering the
other two surfaces of the steel column. Heat loss through the two surfaces is
negligible. Due to the considerably low-temperature change, thermal proper-
ties of the steel column are assumed to be independent to the temperature
change. The thermal conductivity and capacity are of 54 W/m K and
0.465 kJ/kg K, respectively. The density of steel column is 7750 kg/m3.
In the heat transfer viewpoint, due to the adiabatic condition at both top
and bottom boundaries. There is no heat conduction in vertical direction.
The heat transfer within the steel column can simply be treated as one-
dimensional heat transfer. Moreover, as temperature gradient and all thermal
properties are constant, temperature distribution within the column exhibits a
simple linear relationship. In other words the temperature within the column
can be determined by simple interpolation with respected to the temperature
gradient as follows:
x
TðxÞ 5 160 2 3 ð160 2 60Þ ð5:6Þ
0:05
Obviously, solution of above problem is trivial. Solving the problem with
numerical method seems to be over-killed. However, this also set an example
for the reader to verify their results.
Back to the numerical method, for the ease of implementation, we should
further simplify the governing equation to suit the problem. First of all, a
20 3 20 numerical mesh is adopted to discretize the square column. All the

q = 0 W/m2

Th = 160ºC Tc = 60ºC

q = 0 W/m2
FIGURE 5.3 A schematic of the two-dimensional heat condition in a square steel column.
92 Demystifying Numerical Models

meshes are therefore in square shape of 2.5 mm 3 2.5 mm. In other word, Ax
and Ay are identical and Δx is equal to Δy. Moreover, there is no internal
heat generation within the column. The governing equation could be simpli-
fied as follows:
Ti21; j 2 2Ti;j 1 Ti11;j Ti; j21 2 2Ti;j 1 Ti;j11
kA 1 kA 50 ð5:7Þ
Δx Δx
kA
Dividing the equation by Δx and rearrange the equation give
Ti21; j 1 Ti11;j 1 Ti; j21 1 Ti;j11
Ti;j 5 ð5:8Þ
4
which is simply an arithmetic average of the temperature at the four-
neighboring nodal points. One should also notice that this statement is also
valid for three-dimensional case. For three-dimensional problem, with square
mesh and constant thermal properties, the temperature is the arithmetic
average of all 6 four-neighboring nodal temperature. Eq. (5.8) can also be
implemented easily in spreadsheet by calculating the average of the four-
neighboring cells, as shown in Fig. 5.4. For heat transfer problem with
non-uniform meshes, the discretized formulation would be similar to the
equation 5.8 while the area and spacing of the mesh are not constant.
In this problem, two types of boundary conditions are adopted in the
numerical procedure. Unlike the boundary condition introduced in the

FIGURE 5.4 Spreadsheet showing computation of Eq. (5.8).

 Two-Dimensional and Transient Heat Conduction Chapter | 5 93

qx, right qx, right

Δx Δx Δx Δx

0 1 2 0 1 2
qb
Tb Δx Δx

FIGURE 5.5 Mesh arrangement at the constant temperature and heat flux boundary condition
with additional imaginary node.

previous chapter, it is more convenient to use additional cells for storing

data as the boundary condition.
For the constant temperature, as depicted in Fig. 5.5, temperature is
stored in an additional nodal point outside the computational domain (i.e.,
steel column). One should notice this additional node is imaginary and does
not practically exist in the problem. However, since the imaginary node has
the same spacing with all internal nodal points, Eq. (5.8) is still valid.
Therefore the first nodal point can be determined from the arithmetic average
of the four-neighboring nodal points. The only problem left is how to specify
a correct temperature in the imaginary node according to the boundary
condition.
For the constant temperature boundary condition, as discussed in
Chapter 4, the heat transfer from the boundary to the first node has only half
grid size. In order to maintain the same temperature gradient, temperature at
the imaginary node can be determined by extrapolation as follows:
T1 2 Tb T1 2 T0
kA 5 kA
Δx=2 Δx ð5:9Þ
T0 5 T1 2 2 3 ðT1 2 Tb Þ
where Tb is the temperature at the boundary (i.e., 160 and 60 C at both sides
in this problem).
For the adiabatic boundary condition the treatment could even more sim-
ple. In physical viewpoint, adiabatic condition has no heat flux passing
through the boundary. To set the heat flux as zero, we could simply specify
the temperature at the imaginary nodal to be identical to the adjacent node.
T1 2 T0
kA 50
Δx ð5:10Þ
T0 5 T1
94 Demystifying Numerical Models

FIGURE 5.6 Constant temperature boundary condition implementation.

Using Eq. (5.10) the boundary condition can be implemented in the addi-
tional node. Figs. 5.6 and 5.7 demonstrate how the formulation be implemen-
ted as the boundary condition in the spreadsheet.
After implementing the formulation, all the cells are correlated to each
other. This also forms a set of simultaneous equations governing all cells
with appropriate physical interpretations. The solution can now be obtained
by iterative procedures. In most of the spreadsheet, iterative option requires
user to enable the iteration functionality. For Microsoftr excel users, the
option can be enabled by following the file menu and select option. Under
the Formulas tab, user can enable the iterative calculation and specify the
maximum iteration number and maximum change as the convergence target
as shown in Fig. 5.8.
After enabling the iterative calculation the spreadsheet will carry out
the iterative process and obtain a converged solution as shown in
Fig. 5.9. Depend on the hardware of the computer, such iterative process
normally takes less than a few seconds to complete. Comparing to solv-
ing the solution by matrix operation, the iterative method appears to be
faster and more convenient. In fact, similar procedure could be also
implemented for one-dimensional heat conduction problem. It depends
totally on user preference if direct or iterative method is adopted in any
given problem.
 Two-Dimensional and Transient Heat Conduction Chapter | 5 95

FIGURE 5.7 Adiabatic boundary condition implementation.

FIGURE 5.8 Enabling iterative calculation in spreadsheet.

96 Demystifying Numerical Models

FIGURE 5.9 Converaged solution for the temperature distribution within the square steel
column.

As shown in Fig. 5.9, the converged solution shows a linear temperature

variation from the high- to the low-temperature boundary (i.e., left to the
right hand side in the spreadsheet). The linear variation follows exactly the
same temperature gradient with the interpolation expression in Eq. (5.6).

5.4 TIME-DEPENDENT HEAT CONDUCTION PROBLEM

Apart from solving the steady state heat conduction, the transient thermal
stress imposes to the system is also a common problem where engineers or
designers constantly require a thorough evaluation of the time-dependent
heat conduction within the system. The discussion earlier so far focuses only
on steady state problem where spatial temperature distribution within the
domain is concerned. In this section, we introduce two numerical methods
for incorporating time-dependent consideration in the iterative process.
In Chapter 4, Steady State Heat Conduction Systems, finite difference
approach is introduced to approximate the spatial temperature gradient and
solving the temperature distribution with a set of discrete numerical mesh.
The solution obtained represents temperature distribution when the heat con-
duction process is fully developed and the temperature at any given spatial
location is independent of time. In the calculation, time is not involved in
the governing equations (as well as the discretized algebraic equations).
Nonetheless, for transient heat conduction problem, temperature change with
 Two-Dimensional and Transient Heat Conduction Chapter | 5 97

t=n+1
T ni–1,j
+1 n+1
T i,j T ni ++ 1,j
1

t=n
n
T i–1, j T i,n j T in+ 1,j

Δt
t=1

Δt Δx Δx

t=0
i–1, j i, j i + 1, j
FIGURE 5.10 Notation for the spatial and temporal temperature change in transient problem.

time as well as spatial locations. The discretization procedure should there-

fore involve temperature derivatives with respect to time and space. In other
word the time variation become an additional dimension on top of the spatial
discretization as shown in Fig. 5.10. As depicted, temperature distribution is
stored in spatial grid which is two-dimensional in our case. To consider time
variation, temperature at every nodal point change with each time step (i.e.,
t 5 Δt). At the initial condition (i.e., t 5 0), temperature at each node equal
to the initial temperature. As the heat conduction take place, temperature
would rise after a certain time internal or time step. This process will con-
tinue with the temperature at each node changing with respect to time until
the steady state condition achieved.
For example, instead of steady state heat conduction, the steel column in
our previous example has an initial temperature of 60oC. High temperature
from the furnace starts heating up the steel column when time equal to zero
(i.e., t 5 0). The additional heat coming through heats up the steel column
gradually. After 2 s (i.e., t 5 2 s), temperature nearby the furnace rise start
transferring heat to the other side connected to the heat exchanger.
Considering that we measured the temperature at each node for every 2 s
time-internal, we would record the temperature development from initial for
every 2 s until there is no temperature rise in any nodal point (which is cor-
responding to the steady state condition).
This example elucidates the discretization nature for transient problem.
Instead of solving a continue time-dependent temperature rise, numerical
method only calculates the “snapshot” or instantaneous temperature value at
a given time for every nodal point. Obviously, adopting a small time-step
will reduce the error in general and provide more comprehensive informa-
tion. Unfortunately, more computational step and resource would require for
small time step. Moreover, with time as additional dimension, the notation
Ti;jn refers to the temperature at nodal point (i, j) at the time (t 5 n 3 Δt).
98 Demystifying Numerical Models

5.4.1 Implicit and Explicit Time Discretization

Different from the steady state heat conduction, for transient heat conduc-
tion, the governing equation should consider the change of energy content at
a given finite element. As discussed in Chapter 4, Steady State Heat
Conduction Systems, the generic governing equation for transient heat con-
duction is given as
       
@ @T @ @T @ @T @ðρCTÞ
k 1 k 1 k 1 qb 5 ð5:11Þ
@x @x @y @y @z @z @t
This generic heat conduction governing equation is applicable for the
transient three-dimensional heat conduction problem with internal volumetric
heat source. Instead of zero the spatial heat conduction is now equal to the
energy change per unit volume of the finite element with respect to time.
Using the finite difference method the time-dependent energy change can be
approximated as

@ðρCTÞ Ti;jn11 2 Ti;jn

5 ρC ð5:12Þ
@t Δt
where Ti;jn and Ti;jn11 refer to the temperature at nodal point (i, j) at the times
of t 5 n 3 Δt and t 5 (n 1 1) 3 Δt, respectively. The physical interpretation
of Eq. (5.12) is the temperature change between the time step Δt at the nodal
point (i, j) caused by the heat transfer and internal heat source. To complete
the discretization, we should now consider what is the dominant factor con-
tributing to the temperature rise during time step.
There are generally two main methods in approaching this matter. One
method assumes that the temperature rise is entirely caused by energy input
due to the heat transfer and internal heat source in the previous time step
(t 5 n 3 Δt). In other word, temperature of the nodal point (i, j) at time t 5
(n 1 1) 3 Δt is determined by its temperature, internal heat generation, and
the heat conduction from the neighboring points (i.e., (i 1 1, j), (i 2 1, j)) at
previous time step (t 5 n 3 Δt), as shown in Fig. 5.11. This is commonly

FIGURE 5.11 Graphrical presentation of the explicit scheme.

 Two-Dimensional and Transient Heat Conduction Chapter | 5 99

referred as explicit method where the discretization equation for two-

dimensional heat conduction problems becomes
n
Ti21;j 2 2Ti;jn 1 Ti11;j
n n
Ti;j21 2 2Ti;jn 1 Ti;j11
n
Ti;jn11 2 Ti;jn
kAx 1 kAy 1 qbn ΔV 5 ρCΔV
Δx Δy Δt
ð5:13Þ
From the formulation, one could notice that the explicit method is a time
marching scheme where temperature change in next time step is totally relied
on temperature in present. Neighboring nodal points in next time step has no
influence at all. The main advantage of such a time marching scheme is the
relatively low computational requirement. Based on the given guess or initial
condition, temperature at next time step could be determined in a process
somewhat analogy to extrapolation from the present values. Nonetheless, It
will become more transparent in next section; the explicit method suffers
from the limitation of relatively small time step. Nonrealistic large time step
could pose stability issue in the time-marching or even causing diverge
situation.
The other method adopts an opposite approach by assuming the tempera-
ture rise is governed by the temperature distribution at the next time step and
temperature value at present time. As shown in Fig. 5.12, temperature at the
nodal point (i, j) at next time step depends on the heat transfer from neigh-
boring points (i.e., (i 1 1, j), (i 2 1, j)) at the next time step t 5 (n 1 1) 3 Δt
and the present temperature values. This is commonly referred as implicit
method where the discretization equation for two-dimensional heat conduc-
tion problems becomes
n11
Ti21;j 2 2Ti;jn11 1 Ti11;j
n11 n11
Ti;j21 2 2Ti;jn11 1 Ti;j11
n11
kAx 1 kAy
Δx Δy
ð5:14Þ
Ti;j 2 Ti;j
n11 n
1 qbn ΔV 5 ρCΔV
Δt
Based on the formulation, one could notice that majority of temperature
variables are the next time step where the exact temperature values are
unknown. This is the disadvantage of the implicit method. Unlike explicit

t=n+1
T ni–1,j
+1 T ni,j+ 1 T ni ++ 1,j
1

t=n
n
Ti–1, j Ti,n j T in+ 1,j

i–1, j i, j i + 1, j
FIGURE 5.12 Graphrical presentation of the implicit scheme.
100 Demystifying Numerical Models

method where the “future” temperature is extrapolated from the present

values, implicit method requires information from the future value which are
unknowns in the equation. To obtain these unknown values, it requires to
formulate a set of simultaneous equation for all future values at the nodal
points. Iterative processes are then adopted to obtain all future value. In other
word, implicit method requires iteration process for each time step which
obviously is time consuming and computationally heavy and expensive.
Nonetheless, as the value is obtained from iteration process at each time
step, a larger time step could be used without suffering instability issue as
experienced in explicit method.

5.4.2 Stability and Choice of Time Step for Explicit Method

To understand the stability issue, we should firstly understand the formula-
tion in both methods. Considering a transient one-dimensional heat conduc-
tion problem (similar to the problem discussed in Chapter 4, Steady State
Heat Conduction Systems), the governing equation could then be reduced as
follows:
 
@ðρCTÞ @ @T
5 k 1 qb ð5:15Þ
@t @x @x
Adopting the finite difference method, the governing equation for any
element with a volume of ΔV could be discretized as follows:
Tin11 2 Tin Ti21 2 Ti Ti 2 Ti11
ρCΔV 5 kA 2 kA 1 qb ΔV ð5:16Þ
Δt Δx Δx
Notice that the ΔV 5 ΔxA, where A is the cross-sectional area of any ele-
ment. Canceling the area A and rearrange Eq. (5.16) gives
ρC Δx2
Δx2 ðTin11 2 Tin Þ 5 ðTi21 2 2Ti 1 Ti11 Þ 1 qb ð5:17Þ
kΔt k
Introducing a dimensionless number—mesh Fourier number as
kΔt
τ5 ð5:18Þ
ρCΔx2
One should notice that Eq. (5.17) is valid for both implicit and explicit
method. Depends on the method, present or future temperature values are
adopted in the right-hand side. For the explicit method, the formulation is
expressed as
ðTin11 2 Tin Þ Δx2
5 ðTi21
n
2 2Tin 1 Ti11
n
Þ 1 qb ð5:19Þ
τ k
Arranging the equation gives
 Two-Dimensional and Transient Heat Conduction Chapter | 5 101

Δx2
Tin11 5 τðTi21
n
1 Ti11
n
Þ 1 ð1 2 2τÞTin 1 qb τ ð5:20Þ
k
For all internal points, Eq. (5.20) solves the temperature at next time step
directly from the present temperature value. This could also consider as an
extrapolation of new temperature from the present value. On the other hand,
for the implicit method, the formulation becomes
ðTin11 2 Tin Þ Δx2
5 ðTi21
n11
2 2Tin11 1 Ti11
n11
Þ 1 qb ð5:21Þ
τ k
Arranging the equation gives
Δx2
ð1 1 2τÞTin11 5 τðTi21
n11
1 Ti11
n11
Þ 1 qb τ 1 Tin ð5:22Þ
k
As discussed, the formulation reveals that implicit method requires itera-
tive process to obtain the new temperature on the right-hand side of the
equation. Comparing both methods, we can understand that explicit method
is relatively more straight-forward and easier to be implemented in
spreadsheet.
Nevertheless, the choice of time step for explicit method must be careful.
Otherwise, no realistic result or even divergence situation could occur. As
shown in Eq. (5.20), the coefficient for present temperature value (i.e., Tin ) is
12τ. From the physical viewpoint the present temperature should have a
positive effect on the future temperature. For example, if the heat conduction
has resulted a temperature rise in the element, the temperature at next time
step is equal to the temperature rise plus previous temperature. Thus the
coefficient must be positive to avoid unrealistic calculation. In other word, to
maintain stability or realistic explicit calculation, the criteria could be
expressed as
kΔt 1
ð1 2 2τÞ $ 0 or τ5 # ð5:23Þ
ρCΔx 2 2
With the given thermal property and mesh spacing, one should carefully
choose the time step size to avoid instability occur. On the other hand, unre-
alistic calculation will not occur for the implicit method (see also
Eq. (5.22)).

5.4.3 Transient Heat Conduction in a One-Dimensional

Steel Bar
To demonstrate the difference between explicit and implicit methods, a one-
dimensional transient heat conduction problem is discussed here. In this
problem, similar to the two-dimensional problem, we are considering the
heat conduction through a steel bar. The steel bar with a 5 cm length is used
102 Demystifying Numerical Models

to support the structure a furnace. One side of the steel bar is exposed to
high temperature of 160 C due to the heat loss from the combustion cham-
ber. The other side is exposed to a constant temperature of 60 C which is
adjacent to a recovery heat exchanger. All other surfaces of the steel bar are
covered by insulation where heat loss is negligible. The thermal conductivity
and capacity are of 54 W/m K and 0.465 kJ/kg K, respectively. The density
of steel column is 7750 kg/m3. Initially the steel bar has a uniform tempera-
ture of 60oC. We aim to determine the time for the steel bar to achieve stea-
dy state temperature distribution using both methods.
Again, although the problem is one-dimensional, readers could notice
that the temperature distribution within the steel bar at the steady state condi-
tion exhibits as a linear function as follows:
x
TðxÞ 5 160 2 3 ð160 2 60Þ ð5:24Þ
0:05
To simplify the calculation, we adopt a total of 10 mesh points to discre-
tize the problem. The mesh spacing equals to 5 cm/10 5 0.5 cm or 0.005 m.
For the boundary condition, identical boundary treatment can be adopted
directly in this case (see Eq. (5.9)).
With the boundary condition, all internal nodal points can now be evalu-
ated based on the explicit method. Using the explicit method, the temperature
at the next time step is given by
Tin11 5 τðTi21
n
1 Ti11
n
Þ 1 ð1 2 2τÞTin ð5:25Þ
For the time step, we should first evaluate the maximum allowable time
step based on the stability criteria. Using the given thermal properties and
the mesh spacing, the stability criteria is evaluated as
ρCΔx2
Δt # 5 0:8 s ð5:26Þ
2k
For simplicity, we could choose the time step as 0.5 s. Afterward, we can
now implement Eq. (5.26) for the first node at the first-time step. Details of
the implementation in the spreadsheet are shown in Fig. 5.13.
Applying the same formulation for the rest of the mesh nodes, we can
obtain the temperature distribution within the steel bar at the first-time step
(i.e., t 5 0.5 s) as shown in Fig. 5.14. One could notice that temperature rise
only occurs in the first grid node. This is because only the first grid node is
subject to the heat transfer from the boundary condition. Other nodal points
have no energy input from neighboring points since all temperatures in pre-
vious time step are 60oC.
To obtain the temperature in other time step, one could simply apply for-
mulation to other rows in the spreadsheet. This can be done by simply
highlighting the first row of our time step (i.e., t 5 0.5 s) then drag the
AutoFill Handle down to the bottom of the spreadsheet (Fig. 5.15).
FIGURE 5.13 Implementation of explicit method for the first grid node.

FIGURE 5.14 Applying the explicit method for the whole domain at the first time step.

FIGURE 5.15 Temperature distribution for the first 10 s based on explicit method.
104 Demystifying Numerical Models

FIGURE 5.16 The steady state temperature distribution predicted by the explicit method.

As depicted in the picture, one could notice that all internal points
record a temperature rise during the first 10 s of heat conduction. The rate
of such temperature rise depends on the thermal diffusivity of the steel bar.
Comparing to our steady state result the temperature distribution is yet to
achieve steady state. Further applying the formulation at more time step,
we could obtained a steady state temperature distribution at 120 s
(Fig. 5.16).
As demonstrated, the explicit method calculation is a simple time march-
ing. With the time step of 0.5 s, it takes 240-time steps to obtain steady state
solution. To accelerate the calculation, one could change the time step to be
0.8 s. The same time-distribution would be obtained. However, using a time
step higher than 0.8 s (e.g., 1.0 s) would cause a diverged solution as shown
in Fig. 5.17.
Aiming to compare both methods, another spreadsheet is used to solve
the transient problem with implicit method. Using the implicit method, the
temperature at the next time step is given by
τ 1
Tin11 5 ðT n11 1 Ti11
n11
Þ1 Tn ð5:27Þ
ð1 1 2τÞ i21 ð1 1 2τÞ i
For the time step, for the sack of comparison, we choose the time step as
0.5 s (Fig. 5.18).
Similar to the explicit method, we firstly implement the formulation in
the first node at the first-time step. As the implicit method requires future
values in the calculation, one should be reminded to the enable the iterative
process option in the spreadsheet. Applying the formulation to the first row,
we could obtain the temperature distribution for the first-time step.
 Two-Dimensional and Transient Heat Conduction Chapter | 5 105

FIGURE 5.17 A divergence problem occurred when 1.0 s time step is employed.

FIGURE 5.18 Implementation of implicit method for the first grid node.

Comparing the result with the explicit method, one can notice that tem-
perature rise occurs not only at the first node but also at three other nodes.
The temperature change in other nodes is caused by the heat conduction
from the first node which sequentially transfers to other nodes. Such heat
conduction effect is neglected in explicit method which imposes some
numerical error in each time step (Fig. 5.19).
106 Demystifying Numerical Models

FIGURE 5.19 Applying the implicit method for the whole domain at the first time step.

FIGURE 5.20 The steady state temperature distribution predicted by the implicit method.

As shown in Fig. 5.20, using the implicit method, temperature change at

all nodal points achieve the steady state distribution around 122 s which
slightly longer than the 120 s predicted by explicit method. In most of the
cases, for the same time step, implicit method tends to yield more accurate
results compared to the explicit method. This is partly due to the additional
 Two-Dimensional and Transient Heat Conduction Chapter | 5 107

FIGURE 5.21 A divergence problem occurred when 1.0 s time step is employed.

considerations of temperature change in future time-step. Obviously, implicit

method requires more computational steps and resources throughout the iter-
ative process.
On the other hand the implicit method is numerically more
stable compared to explicit method. The iterative process in the calculation
ensures each nodal point is having realistic results. Therefore a larger time
step could be employed using implicit method. Fig. 5.21 shows the steady
state temperature distribution predicted by the implicit method using a time-
step of 1.0 s. There is no divergence problem occurred in this case. The time
for achieving steady state is predicted as around 126 s. The different in pre-
dictions could be caused by the numerical error in the time discretization.
One should realize that we have adopted a first-order differencing scheme
for the time discretization for both methods. Thus a larger time step would
impose higher discretization and numerical errors in the calculation.
Reducing the time step could minimize such error, however, it is at the cost
of more computational steps and time.