You are on page 1of 143

Modelling of Conjugate Heat

Transfer in Near-Wall Turbulence

A dissertation submitted to the University of Manchester


for the degree of MSc in Thermal Power and Fluids Engineering

2006

Amir Keshmiri

School of Mechanical, Aerospace and Civil Engineering


(MACE)
Dedication

Dedicated To My
Lovely Family
List of Contents

List of Contents
LIST OF CONTENTS ...................................................................................................I

LIST OF FIGURES .................................................................................................... IV

LIST OF TABLES...................................................................................................... VI

ABSTRACT ............................................................................................................ VIII

DECLARATION........................................................................................................ IX

COPYRIGHT STATEMENT ...................................................................................... X

ACKNOWLEDGEMENTS ........................................................................................ XI

NOMENCLATURE ..................................................................................................XII

1 INTRODUCTION ................................................................................................ 1

1.1 Computational Fluid Dynamics (CFD).......................................................................................... 1

1.2 Turbulence Modelling.................................................................................................................... 1

1.3 Reynolds Averaged Navier-Stokes (RANS) .................................................................................. 2

1.4 Large Eddy Simulation (LES)........................................................................................................ 2

1.5 Direct Numerical Simulation (DNS).............................................................................................. 2

1.6 Conjugate Heat Transfer ................................................................................................................ 3

1.7 Thermal Stresses ............................................................................................................................ 5

1.8 Objectives ...................................................................................................................................... 5

1.9 Outline of the Dissertation ............................................................................................................. 6

2 LITERATURE SURVEY ..................................................................................... 7

2.1 Conjugate Heat Transfer of a Fully-Developed Turbulent Flow ................................................... 7

2.2 Heat Transfer Modelling.............................................................................................................. 12

3 GOVERNING EQUATIONS OF FLUID FLOW AND HEAT TRANSFER ...... 16

i
List of Contents

3.1 Introduction.................................................................................................................................. 16

3.2 Normalised Equations and Parameters......................................................................................... 17

3.3 Reynolds-Averaged Navier-Stokes (RANS)................................................................................ 22

3.4 Dynamic-Field Modelling............................................................................................................ 23


3.4.1 Zero- and One-Equation Eddy-Viscosity Model..................................................................... 23
3.4.2 Two-Equation Eddy-Viscosity Model..................................................................................... 25
3.4.2.1 Standard Two-Equation k − ε Model............................................................................ 25
3.4.2.2 Near-Wall and LRN Modifications ............................................................................... 28

3.5 Thermal-Field Modelling............................................................................................................. 31


3.5.1 Governing Equations............................................................................................................... 31
3.5.2 Two-Equation θ 2 − ε θ Model ............................................................................................... 32

3.6 The Log-Law ............................................................................................................................... 34

4 PROBLEM DEFINITION AND COMPUTATIONAL DETAILS ...................... 36


4.1 Introduction.................................................................................................................................. 36

4.2 Problem Definition ...................................................................................................................... 37


4.2.1 Preliminary Remarks............................................................................................................... 37
4.2.2 Geometry................................................................................................................................. 38
4.2.3 Grid Type ................................................................................................................................ 38

4.3 Governing Equations ................................................................................................................... 39


4.3.1 Preliminary Remarks............................................................................................................... 39
4.3.2 Dynamic-Field ........................................................................................................................ 40
4.3.3 Thermal-Field.......................................................................................................................... 41
4.3.4 Summary of the Governing Equations .................................................................................... 43

4.4 Discretisation ............................................................................................................................... 44


4.4.1 Preliminary Remarks............................................................................................................... 44
4.4.2 Finite Volume Method ............................................................................................................ 45
4.4.3 Stability of Discretisation Schemes......................................................................................... 50
4.4.4 Cell Storage Method ............................................................................................................... 51
4.4.5 Discretisation of Bulk Velocity and Temperature................................................................... 52

4.5 Boundary Conditions ................................................................................................................... 52

4.6 Modifications of the Governing Equations for Conjugate Heat Transfer .................................... 54
4.6.1 Different Specific Heat Capacities in Fluid and Solid ............................................................ 54
4.6.2 Discontinuity in Dissipation of Heat at Solid-Fluid Interface ................................................. 55
4.6.3 Summary of the Modified Governing Equations .................................................................... 56

4.7 Computational Details ................................................................................................................. 57


4.7.1 Initial Values ........................................................................................................................... 57
4.7.2 Convergence Criteria .............................................................................................................. 58
4.7.3 Under-Relaxation Factors ....................................................................................................... 59
4.7.4 TDMA Solver ......................................................................................................................... 60

5 RESULTS AND DISCUSSIONS........................................................................ 61

ii
List of Contents

5.1 Preliminary Remarks ................................................................................................................... 61

5.2 Non-Conjugate Heat Transfer...................................................................................................... 61


5.2.1 Preliminary Remarks............................................................................................................... 61
5.2.2 Results of Dynamic-Field Properties ...................................................................................... 62
5.2.3 Results of Thermal-Field......................................................................................................... 64

5.3 Conjugate Heat Transfer .............................................................................................................. 72


5.3.1 Preliminary Remarks............................................................................................................... 72
5.3.2 Results and Modifications....................................................................................................... 73
5.3.2.1 Preliminary Remarks ..................................................................................................... 73
5.3.2.2 Accounting for the Difference in Specific Heat Capacities ........................................... 74
5.3.2.3 Solving for the Total Dissipation Rate of Temperature Variance.................................. 78
5.3.2.4 Creating a Discontinuity in the Dissipation Rate at the Interface.................................. 81
5.3.3 Diagnostic Tests...................................................................................................................... 84
5.3.3.1 Changing the Grid Spacing in the Solid ........................................................................ 84
5.3.3.2 Variation of non-dimensional wall thickness ................................................................ 85
5.3.3.3 Variation of the Ratio of Molecular Diffusivities.......................................................... 87

6 CONCLUSIONS AND SUGGESTIONS FOR FURTHER WORK .................... 89


6.1 Concluding Remarks.................................................................................................................... 89

6.2 Suggestions for Further Work...................................................................................................... 91

REFERENCES ........................................................................................................... 92

APPENDIX A ............................................................................................................ 97

APPENDIX B........................................................................................................... 102

APPENDIX C........................................................................................................... 114

APPENDIX D .......................................................................................................... 118

APPENDIX E........................................................................................................... 122

iii
List of Figures

List of Figures
Figure 1-1 : The geometry of a channel flow................................................................. 4

Figure 1-2 : The geometry of a flume............................................................................ 4

Figure 3-1 : Variation of mixing length-scale with distance from the wall ................... 24

Figure 4-1 : The solution domain of a half channel flow problem, indicating three types
of grid generations in both fluid and solid region ........................................................ 39

Figure 4-2 : The control volume of a node .................................................................. 45

Figure 4-3 : The current node P, along with its two neighbouring nodes...................... 47

Figure 5-1 : Profiles of the mean velocity.................................................................... 63

Figure 5-2 : Normalised turbulent kinetic energy for Pr=0.71, 5 and 7........................ 63

Figure 5-3 : Production of turbulent kinetic energy for Pr=0.71, 5 and 7 .................... 64

Figure 5-4 : Normalised total dissipation of turbulent kinetic energy for Pr=0.71, 5 and
7 ................................................................................................................................. 64

Figure 5-5 : Profile of the mean temperature for a)Pr=0.71 b)Pr=5 c)Pr=7................ 68

Figure 5-6 : Profiles of the turbulent wall-normal heat flux for a)Pr=0.71 b)Pr=5
c)Pr=7........................................................................................................................ 69

Figure 5-7 : Profiles of normalised temperature variance (RMS of temperature


fluctuations) for isoflux and isothermal boundary conditions for a)Pr=0.71 b)Pr=5
c)Pr=7........................................................................................................................ 70

Figure 5-8 : Production of temperature variance for Pr=0.71, 5 and 7......................... 71

Figure 5-9 : Profiles of the total dissipation of heat with isoflux boundary condition for
Pr=0.71, 5 and 7......................................................................................................... 71

Figure 5-10 : Profiles of the total dissipation of heat with isothermal boundary condition
for Pr=0.71, 5 and 7 ................................................................................................... 71

Figure 5-11 : Profiles of mean temperature for Pr=0.71, 5 and 7................................. 72

Figure 5-12 : Variation of Nusselt number against Prandtl number.............................. 72

iv
List of Figures

Figure 5-13 : DNS of Tiselj et al. (2001a) for variation of temperature variance a)in the
fluid b)inside the wall ................................................................................................. 74

Figure 5-14 : Effects of taking into account the difference in c p on the variation of

temperature variance for Tests 1-6 a)in the fluid b)inside the wall............................... 76

Figure 5-15 : Effects of taking into account the difference in c p on the variation of

temperature variance for Tests 3-6 a)in the fluid b)inside the wall............................... 77

Figure 5-16 : Normalised total dissipation of heat through the solid-fluid interface ..... 77

Figure 5-17 : Effects of solving for the total dissipation rate on the variation of
temperature variance for Tests 1-6 a)in the fluid b)inside the wall............................... 79

Figure 5-18 : Effects of solving for the total dissipation rate on the variation of
temperature variance for Tests 2-6 a)in the fluid b)inside the wall............................... 80

Figure 5-19 : Normalised total dissipation rate of temperature variance through the
solid-fluid interface..................................................................................................... 80

Figure 5-20 : Effects of creating a discontinuity in the dissipation rate of temperature


variance at the interface on the variation of temperature variance for Tests 1-6 a)in the
fluid b)inside the wall ................................................................................................. 82

Figure 5-21 : Effects of creating a discontinuity in the dissipation rate of temperature


variance at the interface on the variation of temperature variance for Tests 2-6 a)in the
fluid b)inside the wall ................................................................................................. 83

Figure 5-22 : Normalised total dissipation rate of temperature variance through the
solid-fluid interface..................................................................................................... 83

Figure 5-23 : Effects of changing the grid size of the solid region on the variation of
temperature variance a)in the fluid b)inside the wall ................................................... 85

Figure 5-24 : Effects of changing the dimensionless wall thickness on the variation of
temperature variance a)in the fluid b)inside the wall ................................................... 86

Figure 5-25 : Effects of changing the ratio of diffusivities on the variation of


temperature variance a)in the fluid b)inside the wall ................................................... 88

v
List of Tables

List of Tables
Table 3-1 : Coefficients used for the standard k − ε model .......................................... 27

Table 3-2 : Coefficients used for the two-equation heat transfer model of Hanjalic et al.
(1996) ......................................................................................................................... 33

Table 3-3 : Equations of velocity and thermal log-laws ............................................... 34

Table 3-4 : Experimental measurements of the critical normal wall distance for air and
water........................................................................................................................... 35

Table 4-1 : The values of main dimensionless parameters used in the present work to
match with DNS ......................................................................................................... 37

Table 4-2 : Properties of different types of grids used in grid generation ..................... 38

Table 4-3 : The total diffusion coefficients and the source terms of various turbulent
properties for the fluid region...................................................................................... 43

Table 4-4 : The total diffusion coefficients and the source terms of various turbulent
properties inside the wall ............................................................................................ 44

Table 4-5 : Coefficients of the neighbouring nodes and the source terms of various
turbulent properties in the fluid ................................................................................... 49

Table 4-6 : Coefficients of the neighbouring nodes and the source terms of various
turbulent properties inside the wall.............................................................................. 49

Table 4-7 : Types of boundary conditions which are applied to various turbulent
properties.................................................................................................................... 53

Table 4-8 : Modified coefficients of the neighbouring nodes and the source terms of
various turbulent properties in the fluid....................................................................... 57

Table 4-9 : Modified coefficients of the neighbouring nodes and the source terms of
various turbulent properties inside the wall ................................................................. 57

Table 4-10 : Initial inputs used in the code for various turbulent properties ................. 58

Table 4-11 : Under-relaxation factors used in the code for different turbulent properties
................................................................................................................................... 59

vi
List of Tables

Table 5-1 : The Log-Law equations for different Pr numbers...................................... 67

Table 5-2 : The value of different parameters corresponding to each test case ............. 73

Table 5-3 : Different type and number of grids used in the wall .................................. 84

Table 5-4 : Wall thickness in different tests................................................................. 86

Table 5-5 : The value of different parameters corresponding to each test case ............. 87

vii
Abstract

Abstract
The existence of temperature fluctuations both in the fluid region and inside the solid
wall of a channel flow is investigated by means of applying standard LRN k − ε and

θ 2 − ε θ models for the dynamic- and the thermal-fields respectively. For this purpose, a
very simple test case was considered which involved a fully developed flow through a
straight one-dimensional half-channel with a heated wall. A CFD code, written in
FORTRAN was used to solve equations for the mean flow, the turbulent kinetic energy
and the temperature fluctuations as well as their dissipation rates across the channel.
Simulations were performed at a constant friction Reynolds number Reτ = 150 and
Prandtl numbers of Pr=0.71, 5 and 7. The analysis was presented in two sections: 1)
Non-conjugate and 2) Conjugate systems. The predictions were compared to the DNS
data to evaluate the effectiveness of the turbulence models used in the present work.
Two types of thermal wall boundary conditions for the dimensionless temperature
equation were studied: 1) Isothermal and 2) Isoflux wall boundary condition. The
profile of the mean temperature was not affected by the type of boundary condition;
however, it had a profound effect on the statistics of the temperature fluctuations in the
near-wall region. Generally, the predictions were in fairly good agreement with the
DNS, except for the temperature fluctuations at the solid-fluid interface and inside the
wall. Some attempts were then made to improve the predictions which resulted in
obtaining better results.

viii
Declaration

Declaration
No portion of the work referred to in the dissertation has been submitted in support of
an application for another degree or qualification of this or any other university or other
institute of learning;

ix
Copyright Statement

Copyright Statement
i. Copyright in text of this dissertation rests with the author. Copies (by any
process) either in full, or of extracts, may be made only in accordance with
instructions given by the author. Details may be obtained from the appropriate
Graduate Office. This page must form part of any such copies made. Further
copies (by any process) of copies made in accordance with such instructions
may not be made without the permission (in writing) of the author.

ii. The ownership of any intellectual property rights which may be described in this
dissertation is vested in the University of Manchester, subject to any prior
agreement to the contrary, and may not be made available for use by third
parties without the written permission of the University, which will prescribe the
terms and conditions of any such agreement.

iii. Further information on the conditions under which disclosures and exploitation
may take place is available from the Head of the School of Mechanical,
Aerospace and Civil Engineering (MACE).

x
Acknowledgements

Acknowledgements
The author would like to express his sincere gratitude to:

Professor H. Iacovides and Dr. T. Craft for their creative supervision and constant
guidance throughout the project;

All those academic staff and postgraduate students at the thermo-fluids division at the
University of Manchester, who helped me with my questions and problems;

My family for their constant support and encouragement.

xi
Nomenclature

Nomenclature
aN Coefficient of the scalar in the north cell

aP Coefficient of the scalar in the current cell


aS Coefficient of the scalar in the south cell
A Cross-sectional area of the cell face
cp Specific heat capacity at constant pressure

Cµ Constant in Eqn. 3-39

Cε~ 1 Constant in Eqn. 3-44 and 3-56

Cε~ 2 Constant in Eqn. 3-44 and 3-56

Cεθ1 Constant in Eqn. 3-65

Cεθ2 Constant in Eqn. 3-65

Cεθ3 Constant in Eqn. 3-65

Cεθ4 Constant in Eqn. 3-65

d Thickness of the wall


E Constant in velocity log-law equation
Eε~ Molecular effect in Eqn. 3-56

Eεθ Molecular effect in Eqn. 3-65

fµ Damping function in Eqn. 3-56

f εθ Damping function in Eqn. 3-65

h Convection coefficient
H Height of half channel

ui u j
k Turbulence kinetic energy,
2
ρ f c pf λ f
K Thermal activity ratio,
ρ wc pwλw

l Length-scale
lm Mixing-length scale

xii
Nomenclature

m& Mass flow rate


hL
Nu Nusselt number,
λ
P Mean pressure
Pk Production term of k

Pθ 2 Production term of θ 2

qw Wall heat flux

µ cp
Pr Prandtl number,
λ
2H U B
Re Reynolds number,
ν
k2
Ret Turbulent Reynolds number,
νε
H Uτ
Reτ Reynolds number based on frictional velocity,
ν
Sij Strain-rate tensor

S P , SU Source terms in the source model, Eqn. 4-23

Sφ Source terms in transport and momentum equations, Eqn. 4-12

t Time
uε Kolmogorov velocity scale, (νε )1 / 4

ui u j Turbulent stress tensor

u jθ Turbulent heat flux vector

τw
Uτ Friction velocity,
ρ
U
U+ Dimensionless velocity,

UB Bulk velocity
x Cartesian coordinates in stream-wise directions
y Cartesian coordinates in wall-normal directions
Y Yap Term in Eqn. 3-56

xiii
Nomenclature

y Uτ
Y+ Dimensionless distance from the wall,
ν
+
Ycrit . Critical dimensionless distance from the wall

Greek Symbols:

λ
α Molecular diffusivity,
ρ cp
αt Turbulent diffusivity

∆y Grid spacing in the normal direction


δ ij Kronecker delta

ε Dissipation rate of the turbulence kinetic energy


ε~ Homogenous part of ε
ε̂ Wall-value of ε
εν
ε+ Normalised dissipation rate of the turbulence kinetic energy, 2

εθ Dissipation rate of the temperature variance

ε̂ θ Wall-value of ε θ

ε~θ Homogenous part of ε θ

ε θν
εθ + Normalised dissipation rate of the temperature variance, 2 2
U τ Θτ
φ Scalar quantity
Γ Turbulent exchange coefficient, Eqn. 4-12
γ Grid expansion factor
κ Von Karman constant
λ Thermal conductivity
µ Dynamic viscosity
µt Turbulent dynamic viscosity
ν Kinematic viscosity
νt Turbulent kinematic viscosity

xiv
Nomenclature

ΘB Bulk temperature
Θw Wall temperature

Qw
Θτ Friction temperature,
c p .ρ .Uτ

Θ+ Dimensionless temperature

θ2 Temperature variance
ϑ Velocity scale
ρ Density
σk Constant in Eqn. 3-43 and 3-55

σt Turbulent Prandtl number

σε Constant in Eqn. 3-44 and 3-56

σ εθ Constant in Eqn. 3-65

σθ 2 Constant in Eqn. 3-64

τw Wall shear stress

ξ Residual

Subscripts

( − )i , j Scalar node position

( − )in Initial value

( − )f Fluid

( − )n North cell face

( − )N North node position

( − )P Current node position


( − )s South cell face

( − )S South node position

( − )w Wall

xv
Nomenclature

Superscripts

( − )+ Non-dimensionless quantity normalised by Uτ , Θτ and ν

( − )+ + Non-dimensionless length normalized by α f / αw

Acronyms:

BC Boundary Condition
CFD Computational Fluid Dynamics
DNS Direct Numerical Simulation
EVM Eddy-Viscosity Model
FVM Finite Volume Method
GGDH Generalized Gradient Diffusion Hypothesis
LES Large Eddy Simulation
LEVM Linear Eddy-Viscosity Model
LRN Low Reynolds Number
MLH Mixing Length Hypothesis
RANS Reynolds Averaged Navier-Stokes
RMS Root-Mean-Squared
TDMA Tri-Diagonal Matrix Algorithm

xvi
Chapter 1 : Introduction

1 Introduction
1.1 Computational Fluid Dynamics (CFD)
Heat transfer from solid walls to flowing fluids is a topic of extreme scientific interest as
well as of immense practical importance. Experimental methods aimed at enhancing it
may be costly and time-consuming. This is where Computational Fluid Dynamics (CFD)
becomes a useful tool. CFD is the analysis of systems involving fluid flow, heat transfer
and associated phenomena such as chemical reactions by means of computer-based
simulation. This technique is very powerful and spans a wide range of industrial and non-
industrial application areas. On the other hand, with the growing complexity of the
involved geometries and hence the need for finer grids, more than ever, researchers
attempt to improve the existing turbulence models and other simulation techniques, to
eventually make CFD as economical and practical as possible.

Generally, all CFD codes have the following three main elements:

1) Pre-processor: it consists of the input of a flow problem to a CFD program. This is


sometimes done through a user-friendly interface, which transforms the input into a
form which is suitable for the Solver.
2) Solver: it approximates the unknown flow variables by using simple functions,
discretises the flow domain and then solves the appropriate governing equations by
means of various numerical methods available e.g. TDMA.
3) Post-processor: this element would present the results of the CFD code in a way
which is much easier to understand for the user. This is done through displaying
graphs, vectors and contours, etc.

1.2 Turbulence Modelling


A turbulence model is a computational procedure to close the system of mean flow
equations, namely Continuity and Navier-Stokes equations, so that as wide a range of flow
problems as possible can be calculated. For most engineering purposes it is unnecessary to
resolve the details of the turbulent fluctuations. Only the effects of the turbulence on the

1
Chapter 1 : Introduction

mean flow are usually sought. To be useful in a general purpose CFD code, the turbulence
model should have a wide applicability, be accurate, simple and economical to run.

Some of the most common methods of computationally modelling turbulent flows are
briefly described below.

1.3 Reynolds Averaged Navier-Stokes (RANS)


These turbulence models are based on time-averaged Reynolds equations and the analogy
between the molecular gradient-diffusion process and turbulent motion (in the case of
LEVM). The Reynolds stresses and turbulent scalar fluxes in these models are directly
linked to the local gradients of the mean flow field through a turbulent viscosity and
diffusivity, with the turbulent viscosity determined by a characteristic turbulence velocity-
and length-scale. The choices for the turbulent velocity- and length-scale are very
numerous, ranging from prescribed profiles to the popular two-equation models. Special
models are also employed to characterise the flow near walls.

Note that, since the present work is based on two-equation models (or known as k − ε
models), they are going to be discussed in more details in the subsequent chapters.

1.4 Large Eddy Simulation (LES)


These turbulence models involve a three-dimensional, time-dependent computation of the
large-scale turbulent motions responsible for turbulent mixing whilst those with scales
smaller than the computational grid (the sub-grid stresses), are modelled. The main
difference between conventional turbulence modelling approaches and LES is the
“averaging” procedure used to derive the equation of motion. The LES technique does not
involve the use of ensemble averages, but rather it consists of applying a spatial filter to
the Navier-Stokes equations.

1.5 Direct Numerical Simulation (DNS)


DNS of turbulence is, conceptually at least, perhaps the simplest approach to the flow
simulation problems. In DNS, some arbitrary initial dynamic field is set up and the Navier-

2
Chapter 1 : Introduction

Stokes equations are used directly to describe the evolution of this field over time. It is
then possible to carry out measurements on this field.

In terms of application, DNS is a very versatile technique and can be employed to provide
data for comparison with turbulence theories and models. Nowadays, it is considered to be
an acceptable and reliable alternative to experimental methods in many situations. In other
words, DNS can be regarded as a numerical experiment, which may replace its laboratory
counterpart, but its numerical and physical accuracy must first be confirmed.

In fact, as will be seen later, the results of the present work are also going to be compared
with some DNS data, to evaluate their accuracy and validity.

However, the published DNS studies, especially those on turbulent heat transfer, are still
relatively few in the literature and the flows which may be investigated by DNS are
currently limited to moderate Reynolds and Prandtl numbers due to the resolution barrier
and need for very considerable computational resources, which eventually makes DNS an
extremely expensive method.

1.6 Conjugate Heat Transfer


In the context of flow simulations and CFD, Channel flows and Flumes (as shown in
Figure 1-1 and 1-2), are among the most common geometries in literature due to their
simplicity and possibility of performing very accurate numerical simulations.

Most of the studies which involve channel flows however, tend to focus on the fluid region
and only solve the appropriate equations which apply to the fluid, and therefore the heat
conduction effects within the wall are usually ignored. In heat transfer problems though,
the existence of the wall and its properties, could significantly affect the thermal transport
and the boundary conditions as far as the solid-fluid interface is concerned.

3
Chapter 1 : Introduction

Plane of
Upper Heated Wall Symmetry
y,j

x,i
z,k Flow Direction

H, height of
half Channel

d, thickness
of the wall Lower Heated Wall

Figure 1-1 : The geometry of a channel flow

y,j
Free surface
x,i
z,k

Flow Direction
H, height of
Flume

d, thickness
of the wall Heated Wall

Figure 1-2 : The geometry of a flume

Therefore, the concept of “Conjugate Heat Transfer” in channel flows, say, corresponds to
problems in which the governing equations are solved both in the fluid region and inside
the wall. However, clearly the governing equations which apply to these two regions are
not going to be exactly the same as the velocity in the wall is zero, which implies that the

4
Chapter 1 : Introduction

governing equations as well as the transport equations of the turbulent properties require
some modifications.

Thus, if one needs to obtain the details of the heat transfer near the wall with a given wall
thickness and material properties, a coupled problem of turbulent heat transfer and heat
conduction in the solid wall has to be solved.

1.7 Thermal Stresses


As far as engineering is concerned, the study of temperature fluctuations in the wall,
presents a special interest. For instance, when a heated surface, under uniform wall heat
flux condition, is exposed to a turbulent cooling flow, the flow turbulence can cause
temperature fluctuations at the surface, which are then conducted within the solid wall.
Hence, from the practical point of view, it is important for engineers and designers to
evaluate such fluctuations in the wall temperature, as they can give rise to fluctuating
thermal stresses which can lead to failure, due to fatigue of the material. This signifies the
importance of measuring the thermal fluctuations within the wall in channel flows, flumes
and pipes, etc., especially where the temperature difference between the fluid and the
surrounding is significant e.g. the cooling pipes used in reactors.

1.8 Objectives
The main objective of the present work is to investigate the computational modelling of
temperature fluctuations both in the fluid region and inside the solid wall. For this purpose,
a very simple test case will be considered which involves a fully developed flow through a
straight half channel with a heated wall. A CFD code written in FORTRAN will be used to
solve equations for the mean flow, the turbulent kinetic energy, the turbulent dissipation
rate, the temperature variance and its dissipation rate across the channel, while the
calculations for the thermal field are extended inside the wall as well. The resulting
predictions are compared with available DNS data to evaluate the effectiveness of the
turbulence models used in the present work.

The simulations in this work are performed at a constant Reynolds number and Prandtl
numbers between 0.71 and 7, and a range of different “thermal activity ratio” values.

5
Chapter 1 : Introduction

1.9 Outline of the Dissertation


This chapter is followed by a comprehensive literature review on the topic of conjugate
heat transfer and available turbulence and heat transfer models which are going to be used
in the present work.

Chapter 3 will then focus on introducing the general governing equations of the fluid flow
and some dimensionless numbers which are going to be used in the present work.
Available models for dynamic- and thermal-fields will be discussed in depth.

Chapter 4 will mainly focus on investigating different aspects of the current flow problem
including the modifications of the governing equations, boundary conditions and
discretisation. Important computational details of the FORTRAN code used in this study
are also discussed in this chapter.

Chapter 5 will present and discuss the results obtained from the code. Also this chapter
contains a series of diagnostic tests which are carried out aiming at improving the results.

Finally, Chapter 6 will list all the concluding remarks which are drawn from the present
study and then some suggestions are made for further work on this topic.

6
Chapter 2 : Literature Survey

2 Literature Survey
2.1 Conjugate Heat Transfer of a Fully-Developed
Turbulent Flow
In the last decade, DNS has become an important research tool in turbulent heat transfer
and particular attention has been paid to the DNS of fully developed turbulent channel
flow, as it reveals the basic mechanisms of the convective heat transfer between the fluid
and the solid wall.

The first DNSs of fully developed turbulent channel flow were made by Kim & Moin
(1989), Lyons & Hanratty (1991), Kasagi et al. (1992), Kasagi & Ohtsubo (1993) and Lu
& Hetsroni (1995).

Kim & Moin (1989) made simulations for various Prandtl numbers ( Pr =0.10, 0.71 and
2.0 with Ret =180), since turbulent heat transfer is characterized not only by the Reynolds
number but also by the Prandtl number of the fluid. They assumed a constant volumetric
heating with a uniform wall temperature. Profiles of the mean temperature, temperature
variance and turbulent heat flux were obtained, but detailed budgets of the transport
equations for those quantities were not reported.

Later, Lyons & Hanratty (1991) made the similar DNS for Pr=1 at a lower Reynolds
number of Ret =150. Both walls of their channel problem were kept at different
temperatures.

Kasagi et al. (1992) and Kasagi & Ohtsubo (1993) performed DNS for Pr=0.025 and 0.71
with Ret =150. The averaged heat flux, qw , was uniform over both the heated walls; but
the instantaneous heat flux varied with respect to time and position. They obtained the
budget for the temperature variance and the turbulent heat fluxes. The turbulent Prandtl
number and the time constant ratio for the scalar transport were also examined.

Later, Kawamura et al. (1998) and Na & Hanratty (2000) performed DNS of turbulent
channel at Prandtl numbers up to ten.

7
Chapter 2 : Literature Survey

Kawamura et al. (1999) studied the influence of the Prandtl and Reynolds numbers on the
turbulent heat transfer statistics (mean temperature, RMS-fluctuations, turbulent heat
fluxes) and showed very weak influence of the Reynolds number in the near-wall region.

However, all the simulations mentioned above used isothermal boundary condition for the
dimensionless temperature, which corresponds to a physical configuration where the fluid
with very small density, heat capacity and thermal conductivity is heated by a thick wall
with high density, high heat capacity and high thermal conductivity.

Kong et al. (2000) compared the isoflux and isothermal boundary conditions by
performing DNS of the developing turbulent thermal boundary layer at Re=300 (based on
inlet momentum thickness) and Pr=0.71 and observed the temperature fluctuations close
to the wall.

Tiselj et al. (2001b) performed DNS of a turbulent flume flow with ideal isothermal and
ideal isoflux thermal boundary conditions and analyzed the differences between both types
of boundary condition at Re=171 (based on wall shear velocity and the height of the
flume) and Pr=1 and 5.4. They found that the temperature RMS fluctuations retain a non-
zero value at the wall for isoflux boundary condition, and zero for isothermal one. They
also found that the temperature field under the isothermal wall BC was highly correlated
with the stream-wise velocity. This correlation was weaker for the temperature field under
isoflux wall boundary condition.

Conjugate heat transfer which solves a coupled problem of turbulent heat transfer and heat
conduction in the solid wall with a given thickness and material properties, was initially
studied by Polyakov (1974), who made a mathematical analysis of the coupled heat
transfer through an infinitely thick wall, using a linearized energy equation in the viscous
sub-layer, and found that the value of the temperature fluctuations at the fluid-solid
interface differs for different fluid-solid systems (i.e. as thermal activity ratio, K, increases,
the temperature fluctuations at the interface increases too). This finding was later
confirmed, analytically by Sinai (1987), Kasagi et al. (1989) and Sommer et al. (1994) and
experimentally by Iritani et al. (1985), Khabakhpasheva (1986), Hetsroni & Rozenblit
(1994) and Mosyak et al. (2001).

8
Chapter 2 : Literature Survey

Sinai (1987) suggested a simple technique for approximating the wall function for the
temperature variance when heat is transferred, in the time-mean, between a thick solid and
a well-mixed turbulent fluid. In his paper, he assumed a linear initial temperature
distribution in the solid as part of a surface-renewal model and this in turn resulted in a
straight forward quadrature equation. In his work though, only low-Prandtl-number fluids
were considered. His results however, showed good agreement with the experimental data
of Hochreiter & Sesonske (1969) and Bobkov et al. (1965).

Heat transfer calculations, assuming both zero and non-zero fluctuations of the wall
temperature, were performed by Kasagi et al. (1989) and Sommer et al. (1994). It was
demonstrated that the fluctuations are strongly influenced by the thermal properties and
thickness of the wall. Two types of wall boundary conditions were investigated: (1)
constant wall temperature, and (2) constant wall heat flux. In practical applications and in
experimental studies however, these ideal conditions are not likely to occur and the
calculations should be verified experimentally.

Kasagi et al. (1989) studied the conjugate heat transfer for Pr=0.02~70, by using an
unsteady two-dimensional Stream-wise Pseudo-Vortical Motion (SPVM) model of near-
wall turbulence, which was initially proposed by Kasagi et al. (1986). The unsteady heat
conduction inside the wall associated with the turbulent flow unsteadiness, was also taken
into account.

Kasagi et al. (1989) showed that in the experiments performed with air, the wall
temperature fluctuations are usually negligible. The experiments with water though, can be
performed at almost isothermal boundary condition if a thick copper plate is used as a
heater (Khabakhpasheva, 1986) or at almost isoflux boundary condition, if the heated wall
is very thin.

On the other hand, the main feature of the work in Kasagi et al. (1989) was that, unlike the
typical methodology based on Reynolds decomposition, the algebraic expressions for the
three fluctuating velocities given by the SPVM model, were directly introduced into the
governing energy equation. However, in this work, while the budget of the temperature
variance was predicted reasonably well, the dissipation process could not be reproduced
well for Pr ≥ 1.

9
Chapter 2 : Literature Survey

Sommer et al. (1994) used a near-wall two-equation turbulence model (based on transport
equations for the temperature variance and its dissipation rate) to investigate the validity
and the extent of the zero fluctuating wall temperature boundary condition (K = 0) for heat
transfer modelling and calculations. They concluded that the zero and the non-zero
fluctuating wall temperature boundary conditions yield the same mean temperature
distributions. On the other hand, they found that non-zero fluctuating wall temperature
boundary condition had a limited effect on the turbulence statistics in a small region near
the wall, while beyond this region, the calculated temperature variance and its dissipation
rate using both zero and non-zero fluctuating wall temperature boundary condition were
essentially identical. Thus, the zero fluctuating wall temperature boundary condition was
recommended for adoption in practical turbulent heat transfer calculations, as it appears to
be valid for the mean field as well as the simplest available.

Among the experiments which were carried out with near-isoflux boundary conditions,
Mosyak et al. (2001) provides probably the most accurate measurements of the wall
temperature fluctuations, when compared to the experiments done by Iritani et al. (1985),
Khabakhpasheva (1986) and Hetsroni & Rozenblit (1994).

Mosyak et al. (2001) carried out their experiments in a horizontal turbulent channel heated
from above. A very thin stainless steel foil was used as a heater and thermal images were
taken from the top side of the foil with an infrared camera. The boundary condition in this
experiment was very close to the ideal isoflux. They also used water as a working fluid.
They found that the wall temperature fluctuations depend strongly on the type of wall
thermal boundary conditions. Also, the pictures that emerged from their investigations
confirmed the hypothesis that moderate-Prandtl-number heat transfer at a solid wall is
governed by the large-scale coherent flow structures. Another finding of their experiments
was that the spacing of the thermal streaks on the wall is independent of the thermal
entrance length.

The value of the dimensionless RMS of the wall temperature fluctuations (which was non-
dimensionalised by friction temperature) measured by Mosyak et al. (2001) at Pr=5.4 was
9.3, which is higher than the theoretical results predicted by Kasagi et al. (1989) and Tiselj
et al. (2001a).

10
Chapter 2 : Literature Survey

One of the most comprehensive works on the topic of conjugate heat transfer taking into
account the conduction within the wall, has been reported by Tiselj et al. (2001a). In Tiselj
et al. (2001a), DNS was coupled with unsteady heat conduction in the wall and simulations
were performed at friction Reynolds number Reτ =150 and Pr=0.71, 5.0 and 7.0. The
results were compared with other DNS studies and with the turbulence model and
numerical approach of Kasagi et al. (1989). Moreover, in Tiselj et al. (2001a) the main
deficiencies of the simpler and cheaper methods used for conjugate heat transfer
calculations in the past by Polyakov (1974) and Kasagi et al. (1989) were discussed. Its
main improvement was the finding of the value of the wall temperature fluctuations for the
ideal isoflux boundary condition, which was shown to be higher than that found by Kasagi
et al. (1989).

Tiselj et al. (2001a) is also among a very few works that have investigated the profile of
the temperature fluctuation in the wall in addition to the fluid region, for different Prandtl
numbers and thermal activity ratios. Thus, their results provide a very good reference data
to compare with the results which are going to be obtained by the present work.

On the other hand, Tiselj et al. (2004) for the first time proved analytically the existence of
different temperature profiles for the isothermal and isoflux wall temperature boundary
conditions.

It is worth noting that, Tiselj et al. (2004) used a different terminology to other works in
respect of the boundary conditions applied: non-fluctuating and fluctuating boundary
conditions instead of isothermal and isoflux, respectively. This was to avoid possible
confusion with the conventional meaning of “isothermal” and “isoflux” boundary
conditions, which were studied and compared, for example, by Kong et al. (2000) and
Teitel & Antonia (1993), etc. In these latter works, the applied thermal boundary
conditions were isothermal and isoflux, where each of them could be further divided into
the fluctuating and non-fluctuating wall temperature boundary conditions. The non-
fluctuating temperature boundary conditions were applied in Teitel & Antonia (1993)
whereas both types of wall temperature fluctuations were applied in Kong et al. (2000).

11
Chapter 2 : Literature Survey

In order to estimate the differences between the temperature profiles calculated with the
non-fluctuating and fluctuating wall temperature thermal boundary conditions, Tiselj et al.
(2004) presented the main results of 12 various DNSs, performed for different Reynolds
and Prandtl number flows in the flume and the channel geometry. One finding of the
aforementioned 12 test cases was that heat transfer rate depends on the properties of the
heater and the Nusselt number is slightly higher for fluid heated from a wall that allows
temperature fluctuations to propagate into the solid wall than for fluid heated from a wall
that does not allow temperature fluctuations.

Tiselj et al. (2004) also proved that the results of Kawamura et al. (1998) are also valid for
non-fluctuating boundary condition. In other words, they proved that in both types of
boundary conditions, the Reynolds number has a relatively weak influence on the near-
wall behaviour of the turbulent heat transfer statistics (mean temperature, RMS
fluctuations, and turbulent heat fluxes) in a turbulent channel.

Although the results of Tiselj et al. (2004) are limited to low Reynolds numbers Reτ =150,
170.8 and 424 and Prandtl numbers Pr=0.025, 0.71, 1.0 and 5.4, it can be assumed that the
same small differences in heat transfer, generated in the diffusive sub-layer of the near-
wall turbulent flow, also appear at other Reynolds and Prandtl numbers.

2.2 Heat Transfer Modelling


Heat Transfer effects in incompressible turbulent flows are influenced by the following
forces: Inertia, viscous, shear and possibly buoyancy. If buoyancy is neglected, the
temperature becomes a passive scalar, because the mean velocity equations are decoupled
from the mean temperature equations.

In the past few decades, numerical prediction of turbulent heat transfer phenomena has
attracted great interest. A set of differential equations for the Reynolds stress ( − uiu j ) and

turbulent heat fluxes ( − uiθ ), should be solved simultaneously for the time-mean velocity
and temperature.

12
Chapter 2 : Literature Survey

In predicting the turbulent heat transfer, a very common approach is to model the turbulent
heat flux with the Boussinesq approximation, where the thermal diffusivity is usually
taken to be proportional to the turbulent viscosity, via a constant turbulent Prandtl number
( α t = ν t / σ t ). The disadvantage of this approach is that it is based on the assumption that
there is a direct analogy between eddy-diffusivities of momentum and heat. In fact, the
DNS of Kim & Moin (1989) indicates that there is no universal value for σ t , since its
value varies across the flow field, contrary to the commonly assumed constant value.
Later, Sommer et al. (1992) argued that in order to calculate the turbulent heat transfer in
wall shear flows correctly, the constant turbulent Prandtl number assumption has to be
relaxed. Therefore, it is highly desirable to obtain a turbulent heat transfer model which is
capable of accounting for the variation of σ t in response to the variation of Prandtl
number and flow structure in different types of flows.

One approach to the above problem is to solve transport equations for the heat fluxes
directly. However, this requires the adoption of a Reynolds stress model for the dynamic
field, and the associated modelling of the heat flux equation is rather complex. An
alternative method is to adopt a gradient transport of the heat fluxes. In other words,
making the thermal diffusivity a function of dynamic and thermal time scales. This is
achieved through solving two other transport equation, in addition to k and ε . These
additional equations are typically those governing the transport of temperature variance
( θ 2 ) and its dissipation rate ( ε θ ). This forms the basic idea behind the so-called: “Four-

equation turbulence model” ( k − ε −θ 2 − εθ ). Employing this approach enables one to


prescribe the turbulent Prandtl number, explicitly, which gives perhaps a more general
framework, though it would still be very empirical.

To account for near wall effects, previous researchers have adopted two methods: wall
functions and LRN modifications. The first group, represented by George & Capp (1979)
and Cheesewright (1986), used the modified wall functions concept, originally derived for
high Reynolds number forced convection flows, to avoid the need for a fine mesh to be
allocated in the sub-layer. However, the strong dependence on particular characteristic
parameters to certain flow patterns and the lack of generality has very much limited the
application of this method. (Henkes & Hoogendoorn, 1990)

13
Chapter 2 : Literature Survey

An alternative method is the so-called: Low-Reynolds number (LRN) modified k − ε


model, which integrates the flow equations right up to the wall. With ready access to more
powerful computer resources over the past decades, the method has attracted more
attention for modelling turbulence of LRN flows. A number of such LRN k − ε models
have been proposed by Jones & Launder (1972) and Davidson (1990).

A reasonable amount of effort has been directed to the development and improvement of
LRN four-equation models. Examinations of the near-wall limiting behaviour of
turbulence have resulted in various modifications in modelling the governing equations
and schemes have been proposed by different researchers, including Nagano & Hishida
model (1987) (NH model), Myong & Kasagi (1990) and Nagano & Tagawa (1990) (NT
model). Recently, based on these latter models, various extended models of k − ε type
have also been constructed. For example, Abe et al. (1994) proposed a new two-equation
heat transfer model which incorporates some essential characteristics of second-order
closure models. The main improvement made in this model is that it is extended to a LRN
type by the introduction of the Kolmogorov velocity scale, u ε [= ( νε )1 / 4 ], instead of the

friction velocity, uτ , which makes it possible to apply the model to more complex
turbulent heat transfer problems.

Nagano & Kim (1988) (NK model), Nagano et al. (1991) (NTT model) have also proposed
new versions of two-equation models for the thermal-field calculations. Nagano & Kim
(1988) expressed the heat flux by using eddy-diffusivity for heat and due to the fact that
the there was no need to prescribe σ t , their model had much higher universality than the
conventional zero-equation models. A weakness though, was that the NK model was
developed aiming mainly at the heat transfer analysis under the uniform wall-temperature
condition. In order to analyze heat transfer problems under various wall thermal
conditions, further improvements of the NK model or development of a more sophisticated

θ 2 − ε θ heat transfer model was needed. (Youssef et al., 1992)

Generally, as far as numerical calculations are concerned, those four-equation models


which satisfy the wall-limiting behaviour of turbulence do not necessarily provide a good
convergence. This problem mainly originated from the applied boundary conditions for ε

14
Chapter 2 : Literature Survey

and εθ at the wall, i.e., non-zero values. Thus, the NH and NK models have modified the
dissipation rates in a way that they become zero at the wall, which effectively makes these
models very stable in computations. The NH and NK models, however, do not reproduce
exactly the wall limiting behaviour; thus the models are less accurate in predicting the heat
transfer under arbitrary wall thermal conditions. (Nagano et al., 1997)

Youssef et al. (1992) modified the NK model to try and capture the wall-limiting
behaviour of turbulence quantities under various wall thermal conditions. Their proposed
heat-transfer model discovered novel features of the decaying process of turbulent
thermal-field quantities in a boundary layer and added a reasonable comprehension about
the scalar transport mechanisms in this physical phenomenon.

Using the near-wall behaviour of turbulence quantities, Deng et al. (2001) developed a

new near-wall θ 2 − ε θ model. They modified the modelling of the production term in the

ε θ equation, and proposed a new boundary condition for ε θ under uniform wall heat flux
conditions, which is more convenient for complex heat transfer, compared with that of
Youssef et al. (1992). Both the LRN k − ε turbulence model (Abe et al., 1994) and their
proposed two-equation heat transfer model were applied to study convective heat transfer
for a two-dimensional turbulent channel flow.

Hanjalic et al. (1996) developed a new buoyancy-modified turbulence model based on the
four-equation model, incorporating LRN modifications, which allowed integration up to
the wall and prediction of turbulence transition. Their model has produced very successful
results in many buoyancy-driven flow problems including those reported by Kenjeres &
Hanjalic (1995).

Neglecting buoyant terms, this model can also be applied to non-buoyant flows, including
channel flow, which is the geometry used in the present work. Therefore, as will be seen in
the subsequent chapters, the transport equations of the turbulent properties in this project
will be based on this Hanjalic four-equation model.

15
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

3 Governing Equations of Fluid Flow


and Heat Transfer
3.1 Introduction
The conservation laws of physics are stated as follows:

1) The mass of a fluid is conserved.


2) The rate of change of momentum equals the sum of the forces on a fluid particle
(i.e. Newton’s second law of motion).
3) The rate of change of energy is equal to the sum of the rate of heat addition to
and the rate of work done on a fluid particle (i.e. first law of thermodynamics).

In fact, the governing equations of fluid flow, which are going to be discussed in this
chapter, represent the above laws, mathematically.

In studying the governing equations of the fluid flow, one needs to make a number of
assumptions. One of them is that the fluid is assumed to be continuum. Also, the
behaviour of the flow is described in terms of macroscopic properties such as: velocity,
pressure, density, temperature and their space and time derivatives. For the analysis of
fluid flow at macroscopic length scale, the molecular structure of matter and molecular
motions may be ignored.

In this chapter, first a number of non-dimensional parameters are introduced which as


will be seen later, are very common in the context of turbulence and heat transfer
modelling. Then, the governing equations of the dynamic- and thermal-fields, along
with their models which have been used at the present work, are introduced. Finally, the
velocity and temperature log-law are going to be discussed briefly, which will be used
in Chapter 5 to check validity of the obtained velocity and temperature profiles in the
channel flow.

16
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

3.2 Normalised Equations and Parameters


Probably the most common non-dimensional number in fluids is the Reynolds number
which is the ratio of inertial to the viscous forces. It also often characterises whether or
not the flow is laminar or turbulent. Reynolds number can be expressed in different
ways, depending of the nature of the problem. The two types of Reynolds numbers
which are going to be used in the present work are defined as follows:

1) Reynolds number based on the height of the channel:

2H U B
Re = Eqn. 3-1
ν

2) Reynolds number based on the frictional velocity:

H Uτ
Reτ = Eqn. 3-2
ν

In the above equations, H is the height of the half channel and the frictional velocity,
Uτ and bulk velocity, U B , are defined as:

τw
Uτ = Eqn. 3-3
ρ

UB =

0
U dy
Eqn. 3-4
H

The Prandtl number, Pr , is a dimensionless parameter which signifies the ratio between
molecular conduction of momentum and molecular conduction of thermal energy, or:

cpµ ν
Pr = = Eqn. 3-5
λ α

However, unlike Re number, Pr is a property of the fluid.

17
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

It is worth mentioning that although Pr is conventionally defined for fluids, in


conjugate heat transfer one could define the viscosity for the wall (being the same as the
viscosity of the fluid) and thus, the definition of Pr can be extended to the solid wall as
well (Mosyak et al., 2001). The reason for defining Prw in this manner will be
discussed in the next chapters.

On the other hand, in heat transfer studies, the most frequently used non-dimensional
number is the Nusselt number, defined as:

Hh
Nu = Eqn. 3-6
λ

Alternatively, this can be expressed in terms of the wall heat flux as:

2H qw
Nu = Eqn. 3-7
µ
( ΘW - ΘB ) cp
Pr

where the bulk temperature, Θ B , is defined as:

ΘB =
∫ ρ c U Θ dA
p
Eqn. 3-8
∫ ρ c Θ dA
p

There is an empirical equation for calculating the Nusselt number in fully-developed


pipe flow which was introduced by Dittus & Boelter in 1930:

Nu = 0 .023 Re D0.8 Pr 0.3 Eqn. 3-9

Later, in 1976, Gnielinski suggested another empirical equation for calculating Nusselt
number:

Nu = 0.0214 (Re D0.8 − 100 ) Pr 0.3 Eqn. 3-10

Note that Eqn. 3-9 and 3-10 are used for a fully developed turbulent flow in a geometry
of a pipe and that is why they are a function of ReD , which is a Reynolds number based

18
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

on the diameter of the pipe. However, one can assume that these correlations are
applicable to channel flows too and as will be seen in Chapter 5, this proves to be a
reasonably valid assumption.

In the framework of boundary layers, the dimensionless velocity and near-wall distance
are usually defined as:

U
U+ = Eqn. 3-11

y Uτ
Y+ = Eqn. 3-12
ν

It is worth considering the non-dimensionalised form of the velocity and temperature


equations too.

The momentum equation with, at constant wall temperature is:

DρU i ∂  ∂U i  ∂P
= µ − Eqn. 3-13
Dt ∂x j  ∂x  ∂x
 j  i

and the temperature equation is given by:

Dρ Θ ∂  µ ∂Θ 
=   Eqn. 3-14
Dt ∂x j  Pr ∂x j 

where the total derivative is defined as:

D / Dt = ∂ / ∂t + U j ∂ / ∂x j Eqn. 3-15

If Eqn. 3-13 and 3-14 are non-dimensionalised according to:

x y U V P Θ − Θw
x∗ = y∗ = U∗ = V∗ = P∗ = Θ∗ =
H H U∞ U∞ ρU ∞ 2 Θ∞ − Θ w

they are transformed to:

19
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

DU i* 1 ∂  ∂U i*  ∂P*
= − Eqn. 3-16
Dt Re ∂x*j  ∂x*j  ∂xi*

DΘ* 1 ∂  ∂Θ* 
=   Eqn. 3-17
Dt Re Pr ∂x*j  ∂x*j 

These equations imply that at zero pressure gradient and at Pr = 1 , the solution of these
velocity and temperature equations would be identical with exactly the same boundary
conditions. In other words, the profile of Θ* is identical to that of U * .

As far as the results from the four-equation models are concerned, it is usually desirable
to present the variation of the turbulent kinetic energy, k , and its dissipation, ε , as well

as the temperature variance, θ 2 , and its dissipation, ε θ , in non-dimensional form.


Therefore, they are normalised according to:

k
k+ = 2
Eqn. 3-18

εν
ε+ = 2
Eqn. 3-19

+ θ2
θ2 = 2
Eqn. 3-20
Θτ

ε θν
εθ + = 2 2
Eqn. 3-21
Uτ Θτ

where the frictional temperature, Θτ , is defined as:

Qw
Θτ = Eqn. 3-22
c p . ρ .U τ

Using the frictional temperature defined as above, one can introduce the non-
dimensionalised temperature as:

20
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

Θw − Θ
Θ+ = Eqn. 3-23
Θτ

In studying conjugate heat transfer, there are two dimensionless parameters which are
very common and useful, which will be briefly introduced here:

The first one is known as the “Thermal Activity Ratio, K ”, which is defined as:

ρ f c pf λ f λ f α w
K= = Eqn. 3-24
ρ wc pwλw λw α f

The main purpose of using this parameter in the context of conjugate heat transfer is
that different solid-fluid combinations can be represented by varying this parameter.

The role of this parameter is more significant in studying the variation of θ 2 , both in

the fluid and within the wall. Therefore, provided the results of θ 2 are obtained for a
broad range of K , one can easily relate the results to any solid-fluid combinations.

The other dimensionless parameter which plays a crucial role in the results of θ 2 and
its dissipation rate, ε θ , in particular within the channel walls, is the non-dimensional

wall thickness, d + + , which is defined as:

αf
d ++ = d + Eqn. 3-25
αw

where d + is defined as:

d Uτ
d+ = Eqn. 3-26
ν

The effects of these two non-dimensional parameters on the results of the thermal-field
equations will be discussed in Chapter 5.

21
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

3.3 Reynolds-Averaged Navier-Stokes (RANS)


Development of the Reynolds-averaged equations to describe turbulent flows that have
a statistically steady mean, is achieved by substituting the various velocity and scalar
decompositions into the instantaneous equations and operating on the equations with the
time-averaging operator:

1 ~
t →∞ t ∫
φ ( xi ) = lim φ (xi ,t )dt Eqn. 3-27

where xi indicates the vector location.

~ ~
The decomposition of instantaneous velocity, U i , static pressure, P , and scalar
~
quantities (which here is temperature, Θ ) are done as:

~
U i = U i + ui Eqn. 3-28

~
P = P+ p Eqn. 3-29

~
Θ = Θ +θ Eqn. 3-30

where the over-bar symbol represents the mean quantity while the lowercase letters
correspond to the fluctuating component of that quantity.

From above decomposition, it is clear that the equations for the fluctuating components
can be developed by subtracting the mean equations from their instantaneous forms.
Then the transport equations for various turbulence statistics are obtained by
manipulating the fluctuating component equations and then time-averaging the residual
equations. These become the transport equations of various turbulence quantities such
as: turbulence kinetic energy, dissipation rate, Reynolds stresses and turbulent heat flux
components.

Substitution of the Reynolds decompositions into the instantaneous continuity and x i -


direction momentum and temperature equations, then time-averaging and finally

22
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

applying the derivative rule, one would arrive at the following equations, which form
the basis of the Reynolds-Averaged Navier-Stoke technique:

∂ρ U i
Continuity: =0 Eqn. 3-31
∂xi

Dρ U i ∂P ∂  ∂U i 
U-Momentum: =− + µ − ρ ui u j  Eqn. 3-32
Dt ∂xi ∂x j  ∂x j 

Dρ Θ ∂  µ ∂Θ 
Temperature: =  − ρ u jθ  Eqn. 3-33
Dt ∂x j  Pr ∂x j 

It is worth noting that density, ρ , does not carry an over-bar symbol, which implies that
the density decomposition is not considered here. Should fluctuations in density need to
be considered however, one would have to mass- or Favre-average rather than time-
average.

Also note that, for the sake of simplicity, the over-bar symbol, representing the mean
quantity, is going to be removed from now on, in the coming sections and chapters.

3.4 Dynamic-Field Modelling


3.4.1 Zero- and One-Equation Eddy-Viscosity Model

As the name of Eddy-Viscosity Model (EVM) suggests, these types of models are based
on the eddy viscosity formulation, which in its general form, is given by:

2  ∂U ∂U j 
− ρ uiu j = − ρ k δ ij + 2 µt  i +  Eqn. 3-34
3  ∂x ∂x 
 j i 

Eqn. 3-34 implies that the LEVMs employ a linear relation between the stresses and
strains.

The turbulent viscosity, which is a scalar quantity is determined by:

23
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

µt = ρ lm 2 ( 2 Sij S ij ) Eqn. 3-35

where the mixing-length, lm is prescribed by algebraic equations and the mean strain

rate, Sij , is defined as:

1  ∂U ∂U j 
Sij =  i + Eqn. 3-36
2  ∂x j ∂xi 

In some flow situations, such as two-dimensional thin shear layers, the changes in the
flow direction are always so slow that the turbulence can adjust itself to local
conditions. Thus, if convection and diffusion of turbulence properties can be neglected,
the concept of mixing-length and its prescribed value would produce a valid assumption
for expressing the influence of turbulence on the mean flow.

It is worth mentioning that the mixing length-scale varies linearly in normal direction
from the wall. However, in the viscous sub-layer it is no longer linear and therefore
should be damped by a function (e.g. damping function proposed by Van Driest in
1956). Figure 3-1 shows the variation of mixing-length scale against the distance from
the wall.

Figure 3-1 : Variation of mixing length-


scale with distance from the wall

Despite the simplicity of zero-equation models, they are not appropriate for more
complex flows, as prescribing the mixing length becomes difficult in these flows, which
makes the application of these models, limited to only simple flow problems.

24
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

In one-equation models, the turbulent kinetic energy, k , is found by solving a closed


form of its transport equation while its dissipation term is found by an algebraic
relation:

k3/ 2
ε = Eqn. 3-37

Therefore, the only unknown variable in the transport equation of k , would be its
diffusion term. Prandtl in 1945, proposed that this diffusion term should be taken as a
gradient transport process:

∂  ν t ∂k 
Diff k =   Eqn. 3-38
∂y  σ t ∂y 

where the turbulent Prandtl number, σ t , for the diffusion of k , is given as unity and the
turbulent viscosity is found through:

µt = ρCµ k 1 / 2lµ Eqn. 3-39

where the algebraic prescription of lµ is analogous to that of the mixing length, lm ,

which makes this model, not reliable for situations where the prescription of length
scale is complex, e.g. re-circulating and separating flows.

The main advantage of one-equation models over zero-equation ones, is that the
turbulent energy and thus the viscosity, do not vanish when the velocity gradient
vanishes. This is because of solving an equation for k .

3.4.2 Two-Equation Eddy-Viscosity Model


3.4.2.1 Standard Two-Equation k − ε Model

As was mentioned earlier, in most of the flow problems which one would encounter in
CFD, the convection and diffusion of turbulence properties cannot be ignored (e.g.
recirculation flows) and thus the prescription of length-scale becomes infeasible. In
other words, the two-equation models (also known as the standard k − ε model)
evolved out of the desire to eliminate the need to prescribe the turbulent length-scale in

25
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

an ad hoc manner. The first successful two-equation model, was the k − ε model of
Launder and Spalding (Launder and Spalding, 1974).

In the standard k − ε model, the Boussinesq eddy-viscosity hypothesis is employed,


yielding the same representation for the Reynolds stress tensor as in Eqn. 3-34, but with
the eddy viscosity defined as:

k2
µt = ρ Cµ = Clϑ Eqn. 3-40
ε

Where Cµ is a constant that is taken to be 0.09 in order to yield results consistent with

the Law of the Wall. Also, l and ϑ correspond to the length- and velocity-scale,
respectively, and are defined as:

k3/ 2
l= Eqn. 3-41
ε

ϑ = k1/ 2 Eqn. 3-42

The standard k − ε model is a semi-empirical model based on model transport


equations for the turbulent kinetic energy (k) and its dissipation rate ( ε ). The model
transport equation for k is derived from the exact equation, while the model transport
equation for ε is obtained using physical reasoning and bears little resemblance to its
mathematically exact counterpart.

In the derivation of the standard k − ε model, it is assumed that the flow is fully
turbulent, and the effects of molecular viscosity are negligible. The standard k − ε
model is therefore valid only for fully turbulent flows.

The modelled transport equations, at high Reynolds numbers are given by (Launder and
Spalding, 1974):

Dρk ∂  µt  ∂k 
=    + ρ (Pk − ε ) Eqn. 3-43
Dt ∂x j  σ k  ∂x j 

26
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

Dρε~ ∂  µt  ∂ε~  ε~ ε~ 2
=    + Cε~ 1 Pk − Cε~ 2 ρ Eqn. 3-44
Dt ∂x j  σ ε~  ∂x j  k k

Where the production of turbulent kinetic energy, Pk , is defined as:

2
∂U i  ∂U 
Pk = −uiu j =νt  i  Eqn. 3-45
∂x j  ∂x 
 j 

A commonly used set of coefficients in the standard k − ε model is given in Table 3-1.

Coefficients Cµ σk σ ε~ C ε~ 1 Cε~ 2
Values 0.09 1.0 1.3 1.44 1.92
Table 3-1 : Coefficients used for the standard k − ε model

These coefficients have been optimised over a range of benchmark turbulent flows. In
fact, it is worth noting that the magnitude of Cε~ 1 and Cε~ 2 have significant effects on the

spreading rates of free shear flows. Therefore, Cε~ 1 is typically chosen from computer
optimisation by calibration against free flows (e.g. plane jet or mixing layer), whereas
Cε~ 2 is independently fixed from considering the decay of grid turbulence.

Despite its short comings, the k − ε model is far superior to the older zero- and one-
equation models that preceded it and still remains popular to this day. It was the first
complete turbulence model that could be solved in general flow configurations without
the need for ad hoc assumptions that depend on the geometry. The major criticism
however, is its reliance on the Boussinesq eddy-viscosity hypothesis. It is now well
accepted that the eddy-viscosity should be anisotropic and is only theoretically justified
for turbulent flows that are close to equilibrium.

However, it should be noted that even two-equation models are not reliable for some
flow problems including: rotating flows, curved boundary layers and swirling flows,
etc.

27
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

3.4.2.2 Near-Wall and LRN Modifications

It is important to remember that the k − ε model described above is appropriate for high
Re flow regions, therefore certain modifications have to be made if one wants to apply
this model to near wall regions, where the Reynolds number is low.

Generally, there are five modifications, which need to be implemented, in order to


account for the LRN effects in k and ε transport equations:

• Homogenous Dissipation:

The first step is to consider the boundary conditions of k and ε , which are applied at
the solid-fluid interface. It is clear that at the wall, k has the simple boundary condition,
k = 0 , while ε is non-zero at the wall.

If the fluctuating quantities are assumed to be analytic near a wall, they can be expanded
in Taylor series in terms of y, the normal direction to the wall. By using such Taylor
series, one can find the wall-value of dissipation, ε̂ w . One can show that ε̂ is given by:

2
 ∂k 1 / 2 
εˆ w = 2ν   Eqn. 3-46
 ∂y  w

However, Eqn. 3-46 could be rather difficult to implement in a numerical scheme, and
so instead, some researchers have used a new parameter, known as “homogenous
dissipation rate, ε~ ”, which is defined as:

ε~ = ε − εˆ Eqn. 3-47

where:
2
 ∂ k1/ 2 
εˆ = 2ν   Eqn. 3-48
 ∂x 
 j 

The advantage of using this homogenous dissipation rate is that outside the viscous sub-
layer, ε~ and ε are identical, which means all the earlier modelling considerations are
still valid. However, at the wall, ε~ = 0 .

28
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

• Viscous Damping Function:

The second modification is done through introducing a viscous damping function, f µ ,

into the turbulent viscosity equation. Therefore, Eqn. 3-40 becomes:

k2
µt = ρCµ f µ Eqn. 3-49
ε

The f µ function should be defined so that it takes a value of unity in the fully-turbulent

part of the flow, and then decreases across the viscous sub-layer.

Different researchers have proposed different techniques to calculate f µ . Some of them

make f µ a function of non-dimensional distance from the wall, y + . However, this is

not desirable in complex flows.

An alternative method is to make f µ a function of turbulent Reynolds number, Ret . For

example, Launder-Sharma (1974) used:

 
 
− 3 .4 
f µ = exp  Eqn. 3-50
 Re t  
2

 1 +  
 50  

where the turbulent Reynolds number is defined as:

k2
Ret = Eqn. 3-51
ν ε~

There is also a damping function embedded in the dissipation term in the ε transport
equation, which is defined as:

(
fε~ = 1 − 0.3 exp − Ret
2
) Eqn. 3-52

• E-Term:

The third modification is to introduce and add a new source term, commonly known as
“E-term” in the ε transport equation, which is defined as:

29
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

2
µ  ∂ 2U i 
Eε~ = 2 µ t  Eqn. 3-53
ρ  ∂x j ∂x j 

This term acts to increase the dissipation rate in the near wall region, where velocity
gradients are changing rapidly.

• Yap-Term:

The fourth modification is to add another new term, called “Yap Term”, which is used
to correct the length-scale in the model and is defined as:

 l  l  ε~ 
2 2
Y = max 0.85 e − 1  e   Eqn. 3-54
 l  l  k 

k3/ 2
where le = and l = 2.5 y .
ε

This new source term was first proposed by Yap (1987) designed to prevent the
Launder-Sharma model from returning excessively large length-scales, especially in
reattaching and impinging flows. Thus, only when the predicted turbulent length-scale
exceeds the equilibrium length-scale, does this term come into effect.

However, in the present flow problem the Yap term was ignored as the length-scale here
is an equilibrium one, which means it does not require any correction.

• Molecular Diffusion:

Finally the last modification is to add the molecular viscosity, µ , to the diffusion term
in both k and ε transport equations.

Therefore, once all the aforementioned modifications are applied, the LRN k − ε
transport equations become:

30
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

Dρk ∂  µt  ∂k    ∂(k )1 / 2  
2

=  µ +   ~ 
 + ρPk − ρ ε + 2ν    Eqn. 3-55
 
Dt ∂x j  σk ∂
 j 
x
  ∂x j  

Dρε~ ∂  µt  ∂ε~  ε~ ε~ 2
=  µ +   + Cε~ 1 ρ Pk − Cε~ 2 fε~ ρ + Eε~ + Y Eqn. 3-56
Dt ∂x j  σ ε~  ∂x j  k k

These equations form the basic governing equations which are used in the present work
to model the dynamic-field.

3.5 Thermal-Field Modelling


3.5.1 Governing Equations
For temperature, one can write a very similar equation to the momentum equation,
considered earlier. However, unlike the momentum equation, there is no source term in
the mean temperature transport equation which is given by:

DρΘ ∂  ∂Θ 
=  ρα  Eqn. 3-57
Dt ∂x j  ∂x j 

Since the thermal diffusivity is given by:

λ ν
α= = Eqn. 3-58
ρc p Pr

Eqn. 3-57 becomes:

DρΘ ∂  λ ∂Θ  ∂  µ ∂Θ 
=  =   Eqn. 3-59
Dt ∂x j  c p ∂x j  ∂x j  Pr ∂x j 

From the transport equation of the instantaneous temperature, one can derive an

equation for the temperature variance, θ 2 and its dissipation, ε~θ .

The exact equations, governing the transport of θ 2 and ε~θ are given by (Launder,
1976):

31
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

∂  ∂θ 2  ∂
Dρθ 2
Dt
=
∂x j 
ρα −
∂x j  ∂x j
( )
ρ u jθ 2 − 2 ρ u jθ
∂Θ
∂x j
− 2 ρα
∂θ ∂θ
∂x j ∂x j
+ 2θSθ Eqn. 3-60

Dρε~θ ∂  ∂ε~θ  ∂ ∂θ ∂u j ∂Θ ∂θ ∂ 2 Θ
= ρα
Dt ∂x j  ∂x j
−
 ∂x
ρ( )
u j ε~θ − 2ρα
∂xk ∂xk ∂x j
− 2ραu j
∂xk ∂xk ∂x j
 j

2
Eqn. 3-61
∂θ ∂θ ∂U j ∂θ ∂u j ∂θ  ∂ 2θ 
− 2ρα − 2ρα − 2ρα  + 2ρα ∂θ ∂Sθ
∂xk ∂x j ∂xk ∂xk ∂xk ∂x j  ∂x ∂x  ∂xk ∂xk
 k j 

In above equations Sθ is a source term involving fluctuating viscous stresses and


fluctuating velocity gradients.

It is worth noting that, by comparing the above equations with the transport equations of
turbulent kinetic energy ( k ) and its dissipation rate ( ε ), one can realize that the relative

importance of different terms in the θ 2 and ε θ budgets is similar to that of the

corresponding terms in the k and ε transport equations. In fact, several experimental


studies including Krishnamoorthy & Antonia (1987a) and Krishnamoorthy & Antonia
(1987b) have proved this similarity as well as the resemblance of the thermal- and
dynamic-fields.

On the other hand, the measurements of ε θ by the latter studies, have enabled the
temperature dissipation time scale to be estimated in the near-wall region and
approximately the same distribution as the velocity dissipation time scale was obtained.
In view of this and the assumption of incompressible flow, the last term in Eqn. 3-60
and 3-61 is small compared to other terms and can be neglected. Therefore, they do not
need to be modelled as far as the present work is concerned.

2
3.5.2 Two-Equation θ − ε θ Model

Using the concept of eddy-diffusivity for heat, α t , (analogous to Eddy-Viscosity


model) the governing equations of the two-equation heat transfer model of Hanjalic et
al. (1996) can be written as (neglecting buoyancy-related terms):

32
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

DρΘ ∂  ∂Θ 
=  ρα − ρ u jθ  Eqn. 3-62
Dt ∂x j  ∂x j 

∂Θ ν t ∂Θ
− u jθ = α t = Eqn. 3-63
∂xi σ t ∂x j

Dρ θ 2 ∂  µ µt  ∂θ 2 
=  + 
 ∂x j 
+ 2 ρPθ 2 − 2 ρε θ Eqn. 3-64
Dt ∂x j  Pr σ tσ θ 2  

Dρε~θ ∂  µ µt  ∂ε~θ  θ ε~θ θ ε~θ θ ε~θ 2 θ θ ε~ε~θ θ


=  +  +Cε1ρPk +Cε 3ρPθ2 2 −Cε 2ρ 2 −Cε 4 fε ρ + Eε Eqn. 3-65
Dt ∂xj  Pr σtσεθ  ∂xj  k θ θ k

where:

∂Θ
Pθ 2 = −u jθ Eqn. 3-66
∂x j

∂U i
Pk = −ui u j Eqn. 3-67
∂x j

( )
2
 ∂ θ 2 1/ 2 
εθ = ε θ − α 
~  Eqn. 3-68
 ∂x j 
 
2
 ∂ 2Θ 
E = 2 ραα t 
θ  Eqn. 3-69
ε  ∂x ∂x 
 j j

fεθ = 1 Eqn. 3-70

Table 3-2 lists the coefficients used in the above two-equation heat transfer model.

Coefficients σt σθ 2 σ εθ Cεθ1 Cεθ2 Cεθ3 Cεθ4


Values 0.9 1.0 1.3 0.72 2.2 1.3 0.8
Table 3-2 : Coefficients used for the two-equation heat transfer model of Hanjalic et al.
(1996)

It is clear from the above equations that this two-equation heat transfer model has been
designed to account for LRN effects, as damping functions, a homogenous dissipation

33
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

rate of temperature variance and an extra source term similar to the aforementioned E-
term have been introduced in Eqn. 3-65.

Also note that, in this model, fεθ was assumed 1 due to absence of reliable data for
model validation, in anticipation that for fluids with Pr numbers close to unity, the
thermal turbulence will follow closely the dynamics of the mechanical turbulence. Any
inadequacy in this assumption may only marginally influence the predictions of the
turbulence field at the very edge of a turbulence region.

Moreover, note that the original two-equation heat transfer model of Hanjalic et al.
(1996) used Algebraic Flux Model (AFM) for modelling the turbulent heat flux and thus
in the present work, the method of AFM for modelling the turbulent heat flux has been
replaced by a simple eddy-diffusivity model.

It is worth mentioning however, that the eddy-diffusivity approach requires explicit


prescription of the value of the turbulent Prandtl number and the commonly adopted
approach is to assume a constant value which was generally taken to be 0.9 (Jones &
Launder, 1973). However, as indicated in the DNS data (Kim & Moin, 1989), the
turbulent Prandtl number is not constant and the thermal diffusivity is not necessarily
related to the eddy-diffusivity (Nagano & Kim , 1988).

3.6 The Log-Law


By means of simple derivation, one can introduce a universal log-law for both velocity
and temperature profiles in a two-dimensional boundary layer. In this section, only the
equations which represent these log-laws are shown (See Table 3-3), however full
derivation of these equations can be found in Appendix A.

Y+ Velocity Log-Law Thermal Log-Law

Y + < Ycrit
+
. U+ =Y+ Θ+ = Pr Y +

1 σt  Y + 
Y + > Ycrit
+
. U+ = ln( Y + ) + E Θ+ = ln +  + Ycrit
+
. Pr
κ κ  Ycrit . 
Table 3-3 : Equations of velocity and thermal log-laws

34
Chapter 3 : Governing Equations of Fluid Flow and Heat Transfer

Note that the constants used in the log-law equations shown in Table 3-3 are as follows:

κ = 0.41

E = 5.0

σ t = 0.9

Moreover, it is clear that unlike the velocity log-law, the thermal log-law depends on
the Pr of the fluid since the critical wall-normal distance decreases as Pr number
increases. Thus, one approach to this problem would be to refer to the experimental data
+
which indicate the value of Ycrit . for some fluids such as air and water (shown in Table 3-

+
4) and then by means of a simple interpolation, the value of Ycrit . for Pr=0.71, 5 and 7,

can be estimated. This is how the equations for the thermal log-law have been derived in
the present work. The validity of this assumption will be examined in Chapter 5.

Pr 0.7 5.9
+
Ycrit . 13.2 7.55
Table 3-4 : Experimental measurements of the critical
normal wall distance for air and water

35
Chapter 4 : Problem Definition and Computational Details

4 Problem Definition and Computational


Details
4.1 Introduction
As mentioned earlier, channel flow is a type of geometry which is very popular in CFD
studies due to its simplicity. In conjugate heat transfer problems, where one needs to
solve the appropriate equations, both in the fluid and in the wall, the choice of “a half-
channel flow” seems ideal. Also, one of the most important advantages that the half
channel flow geometry has, in the context of conjugate heat transfer, is that it could be
only one-dimensional and still be able to perform the necessary analysis and obtain the
required results. This type of geometry could significantly simplify the governing
equations and the numerical methods and consequently would become much cheaper in
terms of computational resources as it needs grid generation only in one direction.

For the present work, a simple FORTRAN code (Developed by Prof. H. Iacovides) is
used to perform the analysis. This code initially had the following main features:

• One-dimensional half-channel flow geometry, which included both the fluid and
the wall region, with separate grid generation.
• Launder-Sharma k − ε model for the dynamic-field.
• Finite-Volume discretisation scheme.
• Standard TDMA (the Tri-Diagonal Matrix Algorithm) solver.

Therefore, in order to apply the four-equation model, it was necessary to introduce the

two-equation heat transfer model ( θ 2 − ε θ ) for the thermal field, in addition to the
existing two-equation k − ε model.

In this chapter, the flow problem considered in the present work is studied and the
governing equations as well as the boundary conditions which apply to this particular
flow problem are discussed in more detail. Also, some of the computational features
used in the FORTRAN code, including the discretisation method, stabilisation
techniques and the type of the solver are discussed.

36
Chapter 4 : Problem Definition and Computational Details

It should be mentioned that a copy of this FORTRAN code can be found in Appendix
B. Also Appendix C contains a guidance list which describes the notations used in the
code.

4.2 Problem Definition


4.2.1 Preliminary Remarks

Before discussing different features of the present flow problem, it is important to


mention the values of the main non-dimensional parameters which are used. The values
of these parameters are listed in Table 4-1. Note that some of these parameters,
including the Reynolds number (based on frictional velocity) and non-dimensional wall
thickness are fixed throughout the project, while some other parameters, including
Prandtl number and thermal activity ratio, are varied within a certain range.

As was mentioned earlier, one of the objectives of the present study is to compare the
results obtained here to the available DNS data. So, clearly, the flow problem defined
here, should exactly match the flow problem studied earlier by the DNS technique,
which the present results are going to be compared with. Since, the DNS data of Tiselj
et al. (2001a) provides this means of comparison, the dimensionless parameters, shown
in Table 4-1 are set according to this DNS data.

Parameters Re τ d ++ α f / αw Pr K

Values 150 20 1 0.71-7.0 0.1-100


Table 4-1 : The values of main dimensionless parameters used in the present work to
match with DNS

The flow problem here is steady-state, which means all ∂ / ∂t terms are ignored. The
flow is also assumed to be fully-developed, which means the turbulent flow properties
of the flow do not vary in the stream-wise direction.

37
Chapter 4 : Problem Definition and Computational Details

4.2.2 Geometry

As mentioned earlier, the geometry of the present study is a one-dimensional half


channel. This geometry, as shown in Figure 1-1, contains a wall with thickness d and
the fluid region with height H .

It is worth mentioning that in order to fix the Reynolds number the height of the half-
channel is taken to be 0.5 m, while as will be seen later, the thickness of the wall may
vary in order to vary the thermal activity ratio, K , or the ratio of the diffusivities,
α f / αw , while the non-dimensional wall thickness, d + + , is kept constant.

4.2.3 Grid Type

Generally, there are two methods of generating grids in geometries such as the one-
dimensional half-channel:

1) Expanding grids
2) Uniform grids

Using the above methods, one can have four different arrangements in the present work,
although only three of these are tested (see Table 4-2).

Grid Type Type of Grid in the Wall Type of Grid in the Fluid
A Expanding Expanding
B Uniform Expanding
C Uniform Uniform
Table 4-2 : Properties of different types of grids used in grid generation

As will be seen later, “grid type B” is found to be the most suitable and efficient
arrangement for this particular problem and therefore all the tests are carried out using
this type of grid, with 160 nodes in the fluid region with an expansion ratio of
γ = 1.025 and 50 nodes inside the wall.

It is clear that one needs to have immensely finer grids near the wall, in the fluid region
as there are steep wall-normal gradients across the viscosity-affected layer of a near

38
Chapter 4 : Problem Definition and Computational Details

wall turbulent flow. However, it would be very expensive from computational point of
view to use fine grids throughout the domain, therefore expanding type of grid are very
desirable.

In terms of coordinates, as shown in Figure 4-1, the geometric configuration considered


at the present work, has a flat solid-fluid interface at y = 0 , with solid material located
in the region y < 0 and the fluid region located at y > 0 .

Symmetry
line

jn

N
+y Fluid

jw
Solid-Fluid
Interface

−y Solid

S j1
Type A Type B Type C
Figure 4-1 : The solution domain of a half channel flow problem, indicating three types of grid
generations in both fluid and solid region

4.3 Governing Equations


4.3.1 Preliminary Remarks

The general forms of the governing equations were discussed in Chapter 3. In this
section, the specific forms of these equations for the present application are presented,
both for the fluid and solid region.

39
Chapter 4 : Problem Definition and Computational Details

4.3.2 Dynamic-Field

Since the problem, is a steady-state, the ∂φ / ∂t term on the left hand side of all the
transport equations is ignored. Also the convection terms become zero as well, since the
flow here is fully-developed and thus all gradients (except of pressure and temperature)
in the stream-wise direction are zero.

• Fluid Region

The governing equations in the fluid region then become:

∂  ∂U  ∂p
U-momentum: 0 =  ( µ + µt )  − Eqn. 4-1
∂y  ∂y  ∂x
{
14 442444 3 source term
diffusion term

∂  µt  ∂k    ∂ k1/ 2  
2

k -equation: 0 =  µ +   + ρPk ~


− ρ ε + 2ν    Eqn. 4-2
∂y  {
σ k  ∂y  production  ∂
 y  
144 42444 3 term
144424443
diffusion term destruction term

2
~ ∂  µt  ∂ε~  ε~ ε~2 µt  ∂2Ui 
ε -equation: 0 =  µ +   + Cε~1ρ Pk − Cε~2 fε~ ρ + 2µ  Eqn. 4-3
∂y  σε~  ∂y  142k4 k ρ  ∂x j∂x j 
14 4
42444 3 production 3 14243
term destructi on term
14 42443
diffusion term extra source term

The coefficients used in above equations, can be found in Table 3-1.

• Solid Region

In the solid, one has the following:

U i , Pk ,k ,ε~ = 0 Eqn. 4-4

Therefore, no equation needs to be solved for the dynamic-field inside the wall.

40
Chapter 4 : Problem Definition and Computational Details

4.3.3 Thermal-Field

Similar to the dynamic-field, as a result of a steady-state one-dimensional problem with


a fully developed flow problem, the rate of change and convection terms in the transport

equations of θ 2 and ε~θ become zero. However, the convection term in the mean
temperature transport equation in the fluid is non-zero and therefore is retained. The
transport equations for the thermal-fields, in both the fluid and the solid region are thus
as follows:

• Fluid Region

In the fluid, the transport equations become:

∂ρΘ ∂  µ µt  ∂Θ 
Temperature equation: U =  +   Eqn. 4-5
12 ∂3x ∂y  Pr σ t  ∂y 
convection term
14442444 3
diffusion term

 1/ 2 2 
∂ µ 
θ 2 -equation: 0 =  + t 
µ ∂θ  2 
 + 2 ρ Pθ 2 − 2ρ ε~θ + α 
∂θ 2 
 ( ) Eqn. 4-6
∂y  Pr σ tσθ 2  ∂y  1 2 3  ∂x j  
14 44 424444 3 production term 14 44   
diffusion term
424444 3
destructio n term

  ~  ~ ~
ε~θ -equation: 0 = ∂  µ + µ t  ∂ε θ  + C εθ1 ρ Pk ε θ + C εθ3 ρPθ ε θ
 
∂ y  Pr σ tσ εθ  ∂ y  2
k
2

14 44 424 4 44 3
14243 1424θ3
production term 1 production term 2
diffusion term
2
ε~ 2 ε~ε~  ∂ 2Θ 
− C ρ θ2 − Cεθ4 fεθ ρ θ + 2ρααt 
θ  Eqn. 4-7
θ
ε2
k  ∂x ∂x 
14243 1 42 43 1442443 
j j
destruc t ion term1 destruction term2
extra source term

• Solid Region

The transport equations for the thermal-field inside the wall are much simpler than in
the fluid region, since there is no production term and the turbulent viscosity is also
zero. Thus, one has:

ε~ , k , µt , Pθ , Pk = 0
2 Eqn. 4-8

41
Chapter 4 : Problem Definition and Computational Details

so, the transport equations become:

∂  λ  ∂Θ 
Temperature equation: 0 =   Eqn. 4-9
∂y  c p  ∂y 
1442443
 w
diffusion term

∂  λ  ∂θ 2 
θ 2 -equation: 0 =   − 2 ρεθ Eqn. 4-10
∂y  c p  ∂y  1 2 3
14  424 w
43 destruction term
diffusion term

∂  λ  ∂ε~θ  θ ε~θ 2


ε θ -equation: 0 =   − Cε 2 ρ 2 Eqn. 4-11
∂y  c p  ∂y  1 θ3
1442443 de struction
 w 424
term
diffusion term

The coefficients used in above equations, can be found in Table 3-2.

Finally, it is worth mentioning that as can be seen from the governing equations
considered above, temperature is assumed to be a passive scalar. This assumption
introduces two approximations:

1) Neglected buoyancy.
2) Neglected temperature dependence of the material properties especially viscosity
and heat conductivity.

Therefore, results of the present study tend to be more accurate for the systems, where
the temperature differences are not too large, while some caution is required for the
systems, where the temperature differences are not negligible. The buoyancy can be for
example neglected in the water flume experiment of Hetsroni & Rozenblit (1994), while
the viscosity and heat conductivity vary for approximately 10% in the range where the
experiment was performed.

42
Chapter 4 : Problem Definition and Computational Details

4.3.4 Summary of the Governing Equations

As mentioned before, the general form of a transport equation of a variable φ , is written


in the form of a differential equation which contains the rate of change of the variable,
convection, diffusion and a source term. This is shown as:

∂  ∂φ 
0= Γ + Sφ
∂y  ∂y  { Eqn. 4-12
14243 source term
diffusion term

where Γ and Sφ represent the diffusion coefficient and the total source term,

respectively.

Therefore, using the above general form, Table 4-3 and 4-4 summarise the governing
equations of the dynamic- and thermal-fields, for both the fluid and the solid regions.

φ Γ Sφ
∂p
U µ + µt −
∂x
µt   ∂ k1/ 2  
2
~
ρPk − ρ ε + 2ν   
k µ+
σk  
  ∂y  
µt ε~ ε~ 2 µ  ∂ 2U i 
ε~ µ+
σ ε~
Cε~ 1 ρ Pk − Cε~ 2 fε~ ρ
k k
+ 2 µ t 
ρ  ∂y∂y 

µ µt ∂Θ
Θ + − ρU
Pr σt ∂x
  ∂ θ 2 1/ 2  
( )
2
µ µt
θ2 + 2 ρ Pθ 2 − 2 ρ ε~θ + α   
Pr σ tσ θ 2
 ∂y  
   
ε~ ε~ ε~ 2
Cεθ1 ρPk θ + Cεθ3 ρPθ 2 θ − Cεθ2 ρ θ
µ µt k θ2 θ2
ε~θ Pr
+
σ t σ εθ 2
ε~ε~θ  ∂ 2Θ 
θ θ
− Cε 4 fε ρ + 2 ραα t  

k  ∂y∂y 
Table 4-3 : The total diffusion coefficients and the source terms of various turbulent
properties for the fluid region

43
Chapter 4 : Problem Definition and Computational Details

φ Γ Sφ
U 0 0
k 0 0
ε~ 0 0
λ
Θ cp
0

  ∂ θ 2 1/ 2  
( )
2
λ  
θ2 − 2 ρ ε~θ + α  
cp  ∂y  
   
λ ε~ 2
ε~θ cp
− Cεθ2 ρ θ
θ2
Table 4-4 : The total diffusion coefficients and the source terms
of various turbulent properties inside the wall

4.4 Discretisation
4.4.1 Preliminary Remarks

The numerical solution of differential equations cannot produce a continuous


distribution of the variables over the solution domain, thus the aim instead becomes to
produce a set of discrete values at a number of nodes that cover the solution domain.
Therefore, a continuous algebraic equation is approximated by a set of discrete values.
This process is called “Discretisation”.

There are however, different ways to approximate the flow variables:

a) Finite Difference: it describes the unknown φ of the flow problem by means of


point samples at the node points of a grid of co-ordinate lines. Truncated Taylor
series expansions are often used to generate finite difference approximations of
derivatives of φ at each grid point and its immediate neighbours.
b) Finite Element: it uses simple piecewise functions (e.g. linear or quadratic)
valid on elements to describe the local variations of unknown flow variables φ .
The governing equations are precisely satisfied by the exact solution of φ . As a

44
Chapter 4 : Problem Definition and Computational Details

result, one can obtain a set of algebraic equations for the unknown coefficients
of the approximating functions.
c) Spectral Method: it approximates the unknowns by means of truncated
Fourier series or series of Chebyshev polynomials. Unlike Finite Difference or
Finite Element, the approximations are not local but valid throughout the entire
computational domain.
d) Finite Volume: it was originally developed as a special finite difference
formulation. This is a method which has been used in the present study and
therefore is going to be discussed in depths, in the next section.

4.4.2 Finite Volume Method

The finite volume method has three main steps:

1) Integrating the governing equations formally over the control volumes of the
entire solution domain.
2) Converting the integral equations into a system of algebraic equations by
substitution of a variety of Finite-Difference-type approximations for different
terms in the governing equations.
3) Solving the algebraic equations by an iterative method.

A very unique feature of the finite volume method is the conservation of relevant
properties for each finite cell which makes this method very common and useful for
unstructured grids.

∆x P

P
∆y P

Figure 4-2 : The control volume of a node

45
Chapter 4 : Problem Definition and Computational Details

As mentioned above, the first step is to integrate the governing equation over the control
volume (see Figure 4-2), which for a three-dimensional problem, results in the
following:

Dρφ ∂  ∂φ 
∫ Dt d ( Vol .) = ∫ ∂x j Γ

d ( Vol .) + ∫ Sφ d ( Vol .) Eqn. 4-13
Vol . Vol  x 
j  Vol

Since here the flow problem is steady-state and the convection term is zero, the integral
of the transport equations become:

∂  ∂φ 
∫ ∂y Γ ∂y dxdydz + ∫ Sφ dxdydz = 0
Vol Vol
Eqn. 4-14

It would then be simplified as:

 ∂φ 
∫ Γ ∂y dxdz + Vol∫ Sφ dxdydz = 0
Vol
Eqn. 4-15

But since the problem is one-dimensional:

dx = dz = 1 Eqn. 4-16

Thus, if the integral of Eqn. 4-15 is evaluated for “cell P”, which is shown in Figure 4-3,
one would get:

 ∂φ   ∂φ 
Γ ∂y  − Γ ∂y  + Sφ AP = 0 Eqn. 4-17
 n  s

where Sφ is the average value of Sφ over the control volume and AP is the cross-

sectional area of the control volume face of cell P (i.e. AP = ∆y P , as shown in Figure 4-
2).

Eqn. 4-17 has a clear physical interpretation. It states that the diffusive flux of φ ,
leaving the east face minus the diffusive flux of φ entering the south face is equal to the
generation of φ within the cell.

46
Chapter 4 : Problem Definition and Computational Details

Cell North Face n

Cell Centre P

Cell South Face s

Figure 4-3 : The current node P, along


with its two neighbouring nodes

In order to derive useful forms of discretised equations, the interface diffusion


coefficient, Γ , and the gradient ∂φ / ∂y at north ( n ) and south ( s ) faces are required.
Since the values of the property φ and the diffusion coefficient are defined and
evaluated at the centre of the each cell, linear approximations seem to be the obvious
and simplest way of calculating interface values and the gradients. Therefore, by using
central differences, the diffusion coefficients and the diffusive flux terms are calculated
as follows:

ΓN + ΓP
Γn = Eqn. 4-18
2

ΓP + ΓS
Γs = Eqn. 4-19
2

 ∂φ  φ N − φP 
Γ ∂y  = Γn  ∆y  Eqn. 4-20
 n  n 

 ∂φ   φ P − φS 
Γ ∂y  = Γs  ∆y  Eqn. 4-21
 s  s 

Therefore, Eqn. 4-17 becomes:

φ − φ P  φ P − φ S 
Γn  N  − Γs   + Sφ AP = 0 Eqn. 4-22
 ∆ y n   ∆ y s 

47
Chapter 4 : Problem Definition and Computational Details

Sometimes, the source term in the transport equation is a function of the dependent
variable. In this case the finite volume method approximates the source term by means
of a linear form (refer to Section 4.4.3 for practical implementation of this method):

Sφ ∆y = SU + S PφP Eqn. 4-23

where S P is strictly positive.

Substituting the above equation into Eqn. 4-22 gives:

φ − φP  φP − φS 
Γn  N  − Γs   + S U + S Pφ P = 0 Eqn. 4-24
 ∆yn   ∆ys 

Re-arranging yields:

 Γn Γ   Γ   Γ 
 + s + S P φ P =  n φ N +  s φ S + S U Eqn. 4-25
1∆4
yn ∆ ys
4
42 4 443 1∆2
yn 
3 1∆2
ys 
3
aP aN aS

Thus, in each cell, the variable φ j has a coefficient, a j . So, for the cell arrangements

shown in Figure 4-3, these coefficients would have a form:

aPφP = aN φN + aSφS − SU Eqn. 4-26

where:

a P = a N + aS − S P Eqn. 4-27

Discretised equations of the form of Eqn. 4-26 must be set up at each of the nodal points
in order to solve a problem. The forms of aN , aS , S P and SU for the transport
equations in the present problem are given in Table 4-5 and 4-6.

48
Chapter 4 : Problem Definition and Computational Details

φ aN aS SP SU
∆P
U [µ + (µ ) ]/(∆y )
t n n [µ + (µ ) ] / (∆y )
t s s 0 AP
∆x P
 (µt )n   (µt )s  εˆ
k µ + / (∆y )n µ + / (∆y )s − ρAP ρAP (Pk − ε~ )
 σ k   σ k  k
 ε~ ε~2 
C P − C
 ε1 k k ε2 ε k 
~ ~ f ~
 (µt )n   (µt )s 
ε~ µ +
 σ ε~ 
/ (∆y )n µ +
 σ ε~ 
/ (∆y )s 0 ρAP 
  ∂2Ui 


+ 2ννt  ∂y∂y  
   
 µ (µt )n   µ (µt )s  ∂Θ
Θ  +  / (∆y)n  +  / (∆y )s 0 − ρAPU
 Pr σ t   Pr σ t  ∂x
 µ (µt )n   µ (µt )s  εˆθ
θ2  +  / (∆y)n  +  / (∆y)s −2ρ AP 2 ρAP (Pθ 2 − ε~θ )
 Pr σtσθ2   Pr σ tσθ 2 
2
θ
 θ ε~θ θ ε~θ θ ε~θ2 
Cε1Pk +Cε3Pθ2 2 −Cε2 2 
 µ (µt )n   µ (µt )s  k θ θ 
ε~θ  +  /(∆y)n  +  / (∆y)s 0 ρAP  2
 Pr σ tσ εθ   Pr σ tσεθ  ~
εε~  ∂2
Θ 
−Cε4 fε
θ θ θ
+2ααt   
 k  ∂y∂y 
Table 4-5 : Coefficients of the neighbouring nodes and the source terms of various turbulent
properties in the fluid

φ aN aS SP SU
U 0 0 0 0

k 0 0 0 0

ε~ 0 0 0 0
λ λ
Θ   / (∆y )n   / (∆y )s 0 0
c  c 
 p  p
λ λ εˆθ
θ2   / (∆y )n   / (∆y )s − 2ρ AP − 2 ρAP ε~θ
c  c  θ2
 p  p
λ λ  θ ε~θ 2 
ε~θ
  / (∆y )n
c 
  / (∆y )s
c  0 − ρ AP Cε 2 2 
 p  p  θ 
Table 4-6 : Coefficients of the neighbouring nodes and the source terms of various
turbulent properties inside the wall

49
Chapter 4 : Problem Definition and Computational Details

Finally, with respect to evaluating the grid spacing ∆y , and the turbulent viscosity, µt ,
at the faces of the cell, which have been used in calculating the north and south
coefficients ( a N and aS ), it is worth noting the following equations:

(∆y )n = yN − yP Eqn. 4-28

(∆y )s = y P − y S Eqn. 4-29

(µt )n = 1 [(µt )N + (µt )P ] Eqn. 4-30


2

(µt )s = 1 [(µt )P + (µt )S ] Eqn. 4-31


2

4.4.3 Stability of Discretisation Schemes

One of the most important and fundamental properties of a discretisation scheme is its
stability. In the previous section, it was indicated that by using the source model (Eqn.
4-23), the averaged source term is split into two parts. In this section, the reason for
such splitting of the source term is discussed.

As can be seen from the transport equations discussed earlier (or by looking at Table 4-
3 and 4-4), the source terms in some of these equations are very significant. This is
however undesirable as far as the stability and convergence of the solution are
concerned. Therefore, the aim is to have a matrix of coefficients which is “diagonally
dominant”. To achieve a diagonally dominant matrix, the net coefficient of the central
node, P (i.e. aP − S P ) needs to be large. This implies that the linearization practice of
source terms should ensure that S P is always negative, which results in a large and
positive value of ( aP − S P ).

Therefore, in the code, this is achieved by initially splitting the source term as in Eqn. 4-
23) and then defining a condition for each variable, φ , which states that if SU is

negative, it should be set to zero and its value has to be added to S P . This is shown as:

50
Chapter 4 : Problem Definition and Computational Details

Min (SU ,0 )
S ′p = S P + Eqn. 4-32
φ

SU′ = Max (SU ,0 ) Eqn. 4-33

where S P′ and SU′ indicate the new values of S P and SU , respectively.

However, it is very important to mention that the Eqn. 4-32 and 4-33 are only applicable

when the variable φ is known to be positive e.g. φ = k or θ 2 . Thus, in situations where


the sign of the variable is unknown such as in the momentum equation, the source term
cannot be represented by using the above equations.

Another essential requirement for boundedness, particularly in three-dimensional


problems is that all coefficients of the discretised equations should have the same sign
(usually positive). Physically, this implies that an increase in the variable φ , at one node
should result in an increase in φ at neighbouring nodes.

4.4.4 Cell Storage Method

When solving for more than one variable, one can either use “Collocated” or
“Staggered” meshes. The difference between these two types of mesh is in the location
of the storage of the variables. On a collocated mesh, all variables are stored at the same
computational nodes, usually at the centre of the cell. However, if both velocities and
pressures are defined at the nodes of an ordinarily control volume, a highly non-uniform
pressure field can act like a uniform field in the discretised momentum equations. The
alternative would be to employ staggered mesh, in which, the pressures are stored at the
centre of each cell, while velocities are stored at the mid-point if the faces of the cell.
This would enhance the numerical stability.

However, in the present study, where the problem is only one dimensional, there is no
need to use a staggered mesh and thus all the variables are stored at the centre of each
cell and as mentioned earlier, linear interpolation is used to evaluate the variables at the
faces of the cell.

51
Chapter 4 : Problem Definition and Computational Details

4.4.5 Discretisation of Bulk Velocity and Temperature

In the present problem, the discretised form of bulk velocity and temperature are given
by:

UB =
∫0
U P ∆y P
Eqn. 4-34
H
H H

ΘB =
∫0
ρ U P Θ P ∆yP
=

0
ρ U P Θ P ∆yP
Eqn. 4-35
H H
∫ 0
ρ U P ∆yP ∫0
m&

4.5 Boundary Conditions


As mentioned earlier, the present flow problem is a steady-state problem, which means
its governing equations are of Elliptic type of equations. Generally, in order to obtain a
unique solution to elliptic problems, one needs to specify conditions on the dependent
variable (e.g. here temperature and heat flux, etc.) on all the boundaries of the solution
domain. These types of problems, which require data over the entire boundary, are
called: “Boundary-value problems”.

A boundary-value problem considered at the present work consists of partial differential


equations and the some initial or boundary values which are required to solve the
equations. The solution to these differential equations will not only satisfy the
differential equations everywhere inside the boundary, but will also satisfy the boundary
conditions themselves.

In the context of conjugate heat transfer, it is a common practice to define the following
two general boundary conditions for temperature:

a) Isothermal: The temperature inside the wall is fixed i.e. Θ = Θ w

∂Θ
b) Isoflux: The wall is imposed to a constant heat flux i.e. λ = −qw (the
∂y
Fourier’s Law)

52
Chapter 4 : Problem Definition and Computational Details

In the present study, as will be seen later, most of the numerical simulations are
performed using isoflux boundary condition.

On the other hand, at the solid-fluid interface, the boundary conditions for temperature
and heat flux are:

Θ + w = Θ+ f Eqn. 4-36

 ∂Θ +   ∂Θ + 
 + +  = K  +  Eqn. 4-37
 ∂y  w  ∂y  f

Also note that, at the wall, the “no-slip condition” states:

Velocity: U wall = Vwall = 0

Turbulent Kinetic Energy: k = 0

Finally, by definition, at the wall:

Homogenous turbulent dissipation rate: ε~ = 0

Also at the symmetry line, the gradients of all variables are zero.

Boundary Conditions U k ε~ Θ θ2 ε~θ


φ ( j1 ) = φ ( j1 + 1) • •
φ ( jw ) = 0 • •
φ ( jn ) = φ ( jn − 1) • • • • • •
aS ( j1 + 1) = 0 • • •
a N ( jw ) = aS ( jw + 1) • • •
aN ( jn − 1) = 0 • • • • • •
Table 4-7 : Types of boundary conditions which are applied to various turbulent properties

For the present flow problem, Table 4-7 indicates different boundary conditions which
are applied to the velocity, temperature and other turbulent properties (Also see Figure
4-1). Note that for different sections of the solution domain including the outer surface
of the wall, the wall-fluid interface and the symmetry line, two different boundary

53
Chapter 4 : Problem Definition and Computational Details

conditions are defined, one in terms of the coefficient of the scalar (i.e. aN or aS ) and
the other one, in terms of the scalar itself, φ .

4.6 Modifications of the Governing Equations


for Conjugate Heat Transfer
4.6.1 Different Specific Heat Capacities in Fluid and Solid

As was mentioned earlier, the Fourier’s Law is given by:

∂Θ
qw = λ Eqn. 4-38
∂y

If both sides of the above equation are divided by the specific heat capacity at a constant
pressure, c p , it becomes:

q w λ ∂Θ µ ∂Θ
= = Eqn. 4-39
c p c p ∂y Pr ∂y

In most CFD codes, it is a very common practice to define q w as one single parameter
cp

for the sake of simplicity. This was an assumption which was made in the original
version of the FORTRAN code which was used in the present study. However, it should
be noted that this assumption is only valid when the flow problem is concerned with
only one material, such as most of the flow problems studied by CFD. Though, in the
present problem, where one is not only dealing with the fluid, but also with the solid as
well which obviously would have different c p from the fluid. Therefore, this difference

in the value of c p would mean that the assumption of treating qw as one single
cp

parameter in the code is not valid for conjugate heat transfer problems. This implies that
the governing equations of the thermal-field which were considered earlier have to be
modified accordingly.

54
Chapter 4 : Problem Definition and Computational Details

The solution to the above problem would be to multiply the transport equations of the
thermal-field by c p . Therefore, the general form of the transport equations would

become:

D ρ c pφ ∂  ∂φ 
= c p Γ  + c p Sφ Eqn. 4-40
Dt ∂ x j  ∂ x j 

Thus for transport equations in the fluid region, the equation would be multiplied by c p

of the fluid and similarly, the transport equations inside the wall are multiplied by c p of

the wall.

4.6.2 Discontinuity in Dissipation of Heat at Solid-Fluid


Interface
Due to the difference in the properties of the fluid and solid materials, another
modification has to be made to the governing equations when conjugate heat transfer

problems are studied. This modifications is associated with the dissipation rates of θ 2
in the fluid and inside the wall and arises from the fact that this dissipation rate cannot
be continuous across the interface. In this section, this modification is discussed and its
results will be shown in Chapter 5.

As mentioned in Section 4.5, the boundary condition for the heat flux through the solid-
fluid interface is given by Eqn. 4-37. Using the definition of K and y + + , this equation
can be re-written as:

 ∂Θ +   ∂Θ +  λ
 +  =  +  f Eqn. 4-41
 ∂y  w  ∂y  f λw

By considering the standard definition of homogenous dissipation of heat, ε~θ and its

wall value, εˆθ , and employing the above substitution, one can arrive at the following
relationships:

ε~θ w = K 2ε~θ f Eqn. 4-42

55
Chapter 4 : Problem Definition and Computational Details

εˆ θ w = K 2εˆ θ f Eqn. 4-43

Thus, the total dissipation rate of θ 2 inside the wall would also be related to that in the
fluid by:

ε θ w = K 2ε θ f Eqn. 4-44

Substituting Eqn. 4-42 to 4-44 into the transport equations of θ 2 and ε~θ inside the wall,
would effectively mean that these transport equations are solving for the same

dissipation rate of θ 2 (which is that of the fluid) throughout the entire solution domain
i.e. allows one to solve for a variable that is continuous across the interface. Therefore,

the transport equations of θ 2 and ε~θ inside the wall become:

Dρ wc p wθ 2 ∂  ∂θ 2 
λ  − 2 ρ w c p w K ε~θ f − 2 ρ w c p wεˆθ w
2
= Eqn. 4-45
Dt ∂y  ∂y 
w

Dρ w c p wε~θ f ~
∂  ∂ε  ~ 2
2 εθ f
= λw θ f θ
 − Cε 2 ρ w c p w K Eqn. 4-46
Dt ∂y  ∂y  θ2

The main effect of this type modification is that, now there is going to be an abrupt

change (i.e. a discontinuity) in the profile of the dissipation of θ 2 , through the solid-
fluid interface, which is more desirable, since a continuous dissipation of heat through
the interface is physically invalid as the properties of the fluid and the solid are clearly
different.

The full derivation of Eqn. 4-42 to 4-46 are shown in Appendix D.

4.6.3 Summary of the Modified Governing Equations


So far two modifications have been discussed which should be applied to the transport
equations of the thermal-field in the present work. In this section, these modifications
have been summarised and shown in Table 4-8 and 4-9.

56
Chapter 4 : Problem Definition and Computational Details

φ aN aS SP SU
µ (µt )n  µ (µt )s  ∂Θ
Θ cp f  +  /(∆y)n cp f  +  /(∆y)s 0 − ρ f c p f APU
Pr σt  Pr σt  ∂x
 µ (µ )   µ (µ )  εˆθ
θ2 cp f  + t n  /(∆y)n cp f  + t s  /(∆y)s −2ρf cp f AP 2 ρ f c p f AP (Pθ 2 − ε~θ )
Pr σtσθ2  Pr σtσθ2 
2
θ
 ε~ ε~ 
Cεθ1Pk θ + Cεθ3 Pθ 2 θ2 
 k θ 

 (µ )   (µ )  ~
ε 2
ε~ε~ 
ε~θ cp f  µ + t n  /(∆y)n cp f  µ + t s  /(∆y)s 0 ρ f cp f AP − Cεθ2 θ2 − Cεθ4 fεθ θ 
Pr σtσεθ  Pr σtσεθ   θ k 
 2 
  ∂ Θ
2

+ 2ααt  ∂y∂y  
   
Table 4-8 : Modified coefficients of the neighbouring nodes and the source terms of various
turbulent properties in the fluid

φ aN aS SP SU

Θ λ / (∆y )n λ / (∆y )s 0 0

εˆθ
θ2 λ / (∆y )n λ / (∆y )s − 2 ρwc p w AP − 2 ρ w c p w K 2 AP ε~θ
θ2
 ε~ 2 
ε~θ λ / (∆y )n λ / (∆y )s 0 − ρwcp wK 2 AP  Cεθ2 θ2 
 θ 
Table 4-9 : Modified coefficients of the neighbouring nodes and the source terms of
various turbulent properties inside the wall

4.7 Computational Details


4.7.1 Initial Values
Before solving the transport equations of the turbulent properties, one needs to define an
initial value for each of them. This value however, should not be chosen arbitrarily,
since it might significantly increase the convergence period and would reduce the order
of convergence. Thus, in iterative methods, it is important to define an initial value
which results in a quick and stable convergence. The initial values which are used in the
present work are listed in Table 4-10.

57
Chapter 4 : Problem Definition and Computational Details

φ φ in
U 5000
2
k 0 .002 U in

ε~ kin
3/ 2
/ 0.34

Θ 300

0.02 (Θ w − Θ B )
2
θ2
ε~θ θ 2 in kin 1 / 2 / H
Table 4-10 : Initial inputs used in the code for various
turbulent properties

4.7.2 Convergence Criteria


Since most CFD codes solve the governing equations by iterative methods, it is
important to know when the solution has converged. One way of making sure that the
solution is converging (rather than oscillating or even diverging) is by monitoring the
“Residuals” denoted by “ ξ ”, which for each variable, φ , are defined as:

ξ ( φ ) = ∑ [aPφP − (aNφN + aSφS + SU )] Eqn. 4-47

The value of ξ is a measure of how close the current solution is to the exact solution.
The process of iteration has to be repeated until ξ or the difference between the
successive iterations becomes smaller than a pre-determined value, known as
“Convergence Criterion”.

However, in some situations, such as the one in the present work, the value of the
convergence criterion is dependent on some parameters, including the properties of the
flow and the solid, etc., thus, the process of iteration has to be repeated until the solution
of different variables, is not changing (or at least the difference between the solutions
after each successive iteration, lies within an acceptable range).

The residual can also be expressed in non-dimensional form as:

ξ(φ )
ξ +(φ ) = Eqn. 4-48
m& φin

58
Chapter 4 : Problem Definition and Computational Details

This is known as “absolute normalised residual” which in the present work is used to
judge the convergence of the solution and as can be seen, it is normalised by using the
mass flow rate and the initial (or inlet) value of the variable, φ .

4.7.3 Under-Relaxation Factors


In situations, where no stable formulation φ = f ( φ ) can be found or when the
convergence is very slow, the use of “under-relaxation factor, ω ”, can be of great
interest, as it can lead to a stable solution sequence. Thus in the present problem,
because of the coupled nature of the discretised equations, under-relaxation factor is
used to improve the convergence of the solution.

In fact, the under-relaxation factor slows down the rate of advance of the solution, by
linearly interpolating between the current iteration value, φ i , and that of the next

iterative level, f ( φ i ) . Thus:

φ i +1 = (1 − ω ) φ i + ω f ( φ i ) Eqn. 4-49

In the code used for this project, the method of under-relaxation is implemented as
follows:

aPφPi = aNφN + aSφS + SU′ Eqn. 4-50

where:

SU′ = SU + (1 − ω ) a′PφPi −1 Eqn. 4-51

aP
a′P = Eqn. 4-52
ω

Table 4-11 lists the values of ω used in the code for each variable.

φ P U k ε~ Θ θ2 ε~θ
ω 0.25 0.425 0.425 0.425 0.425 0.425 0.425
Table 4-11 : Under-relaxation factors used in the code for different turbulent properties

59
Chapter 4 : Problem Definition and Computational Details

4.7.4 TDMA Solver


The iterative method which is used in the present work to solve the discretised
equations, is known as Tri-Diagonal Matrix Algorithm (TDMA) or Thomas algorithm
(Thomas, 1949) which is one of the most common and efficient iterative methods and in
fact, it is computationally inexpensive and has the advantage that it requires a minimum
amount of storage. This method is explained in more depth in Appendix E.

60
Chapter 5 : Results and Discussions

5 Results and Discussions


5.1 Preliminary Remarks

In this chapter, the results obtained from solving the transport equations are presented. The
analysis of the channel flow is presented in two sections:

1) Non-conjugate heat transfer


2) Conjugate heat transfer

As discussed in previous chapters, the main difference between these two cases is the
existence of the wall. In other words, in the first part, the calculations for the dynamic- and
thermal-fields are carried out only for the fluid region and not for the wall, which
physically, is the same as a case where the thickness of the wall is negligible. However, in
the second case, solving the transport equations for the thermal-field is extended to the
wall region as well. This enables one to see the variation of temperature, temperature
variance and its dissipation rate inside the wall too.

As mentioned in Chapter 2, there are many publications on the topic of non-conjugate heat
transfer, where two-equation turbulence models for both dynamic- and thermal-fields,
taking into account the LRN and near-wall effects are employed. However, Tiselj et al.
(2001a) is among a very few publications which can be found on the topic of conjugate
heat transfer, where the thermal-field is modelled using two-equation heat transfer model

of θ 2 − ε θ and therefore it provides a good source for comparing the present results with.
For the dynamic-field however, the present results will be verified against the results of
other DNS data such as those presented by Kasagi et al. (1992).

5.2 Non-Conjugate Heat Transfer


5.2.1 Preliminary Remarks
For the non-conjugate case the simulations are performed at a friction Reynolds number of
150 and Prandtl numbers of 0.71, 5 and 7. At each Prandtl number two different boundary
conditions are applied: Isoflux and Isothermal. In fact, studying these two ideal boundary

61
Chapter 5 : Results and Discussions

conditions is important since the results of the solid-fluid conjugate system generally fall
between these two ideal conditions.

As far as imposing these two boundary conditions are concerned, it is very important to
note that they do not have the same implications in conjugate and non-conjugate cases. In
other words, for non-conjugate case since the assumption is that the thickness of the wall is
negligible, there is no conduction inside the wall and physically, this means that the isoflux
and isothermal boundary conditions are applied at the solid-fluid interface, at y=0, whereas
in the conjugate case these boundary conditions are applied at the outer surface of the wall,
at y=d, where d is the thickness of the wall.

5.2.2 Results of Dynamic-Field Properties


The dynamic-field properties in the present work include the mean velocity, U , the
turbulent kinetic energy, k , and its dissipation rate, ε .

Figure 5-1 shows the predicted profile of the mean velocity along with that obtained from
the log-law (see Section 3.6). As can be seen from this figure, the mean velocity profile is
in fairly good agreement with the universal law of the wall, especially in the near wall
region. The velocity profile obtained from the DNS of Kasagi et al. (1992) is also shown
here.

In the fully turbulent region, the prediction of the present velocity profile is slightly higher
(maximum of 9% difference) than the DNS data, which is within an acceptable range.

Figure 5-2 shows the variation of the non-dimensionalised turbulent kinetic energy against
distance from the wall. As one expects, it agrees well with the results reported by Launder
& Sharma (1974), giving confidence in the model implementation. However, when
compared to DNS data of Kasagi et al. (1992), it is clear that the standard Launder-Sharma
two-equation model under-predicts the turbulent kinetic energy peak.

Figure 5-3 shows the profile of the production term of the turbulent kinetic energy which
as one might expect, has very similar profile to the turbulent kinetic energy itself. Note
however that the distance from the wall at which PK has its maximum value (here at

62
Chapter 5 : Results and Discussions

y + ≈ 12 ) is not necessarily the same as that for the turbulent kinetic energy, where in this

case is at y + ≈ 21 .

Figure 5-4 depicts the dissipation rate of turbulent kinetic energy which again shows that
the present results correctly match those reported by Launder & Sharma (1974). However,
like the turbulent kinetic energy itself, the model used in the present work under-predicts
the dissipation term when compared to the DNS of Kasagi et al. (1992), particularly in the
viscous sub-layer region. In fact the DNS gives a much larger wall-value of dissipation
( ε w = 0.16 ) in comparison to that returned by the Launder & Sharma model ( ε w = 0.01 ).

20
Present
Log-Law Near Wall Region
Log-Law T urbulent Region
15
DNS - Kasagi et al. (1992)

U+ 10

0
0.1 1 10 100 1000
Log y +
Figure 5-1 : Profiles of the mean velocity

4
Present
DNS-Kasagi et al. (1992)
3 Launder & Sharma (1974)

k+ 2

0
1 10
Log y + 100 1000

Figure 5-2 : Normalised turbulent kinetic energy for Pr=0.71, 5 and 7

63
Chapter 5 : Results and Discussions

2.0E+09

1.5E+09

Pk 1.0E+09

5.0E+08

0.0E+00
0.1 1 10 100 1000
+
Log y
Figure 5-3 : Production of turbulent kinetic energy for Pr=0.71, 5 and 7

0.20
Present
DNS-Kasagi et al. (1992)
0.15 Launder & Sharma (1974)

ε+ 0.10

0.05

0.00
0.1 1 10 100 1000
Log y +
Figure 5-4 : Normalised total dissipation of turbulent kinetic energy for Pr=0.71, 5 and 7

5.2.3 Results of Thermal-Field


The thermal-field variables which are going to be discussed in this section include the

mean temperature, Θ , the wall-normal heat flux, vθ , the temperature variance, θ 2 and its
dissipation rate, ε θ . The variation of Nusselt number against Prandtl number will also be
examined.

Figure 5-5 shows the profiles of the dimensionless mean temperature for different Pr
numbers. It is clear that at the same frictional Reynolds number, increasing the Pr number
of the fluid results in an increase in the mean temperature. In Figure 5-5 the present
temperature profiles have been compared with both the Log-law and the DNS data of
Tiselj et al. (2001a). It can be seen that the present prediction is very close to DNS data,
64
Chapter 5 : Results and Discussions

with a slight over-prediction in the turbulent region. However, for higher Pr numbers, the
over-predictions of the results become more significant, particularly in the turbulent
region. However, since the temperature profiles obtained in the present work are close
enough to the ones found by Log-Law, one can argue that the present predictions of
temperature profiles are within an acceptable range of accuracy.

It is worth mentioning that the Log-Laws which have been shown in Figure 5-5, have been
derived based on the assumptions discussed earlier in Section 3.6. However, for deriving
+
the Log-Law for a thermal boundary layer, one needs to find the value of Ycrit . at the

required Pr number. Therefore, in the present work, linear interpolation has been used to
+
estimate the value of Ycrit . at Prandtl numbers 0.71, 5 and 7 and as can be seen from Figure

5-5, this has proved to be a reasonable assumption as it produces results which are in fairly
good agreement with the DNS data. Table 5-1 shows the equations for the temperature
Log-Law which have been used in the present work as a tool to validate the results
obtained from the Code.

Figure 5-6 shows the variation of the wall-normal heat flux for different Pr numbers and
the results are again compared to the DNS of Tiselj et al. (2001a). It is clear that these
profiles are generally in good agreement. However, compared to DNS, the present
predictions appear to produce a slightly thicker viscous layer, followed by a slightly
steeper gradient of vθ across the buffer region. This difference in the buffer region could
well be due to the fact that the mean temperature was over-predicted in the present work. It
also seems that this difference increases with increasing Pr number, similar to the mean
temperature profile, discussed earlier. As one might expect, the value of the wall-normal
heat flux increases with increasing Pr number, though this increase is not very significant
between the results of Pr=5 and Pr=7.

Figure 5-7 shows the temperature fluctuations for different Pr numbers, compared with the
DNS data of Tiselj et al. (2001a). It is important to note that at each Pr number, applying
the two ideal boundary conditions of isoflux and isothermal show the limits for the
conjugate heat transfer calculations, which are going to be discussed in the next section.

Obviously, for isothermal boundary condition, the temperature fluctuations gradually


diminish near the wall and finally become zero at the wall, whereas for the isoflux

65
Chapter 5 : Results and Discussions

boundary condition, the situation is different. In fact, for this latter boundary condition
both the present results and the DNS show that in the region of around y + < 5 for Pr=0.71

and around y + < 3 for Pr=5 and 7, the gradient of the temperature fluctuations becomes
very small. However, at the wall, unlike in the isothermal case, the temperature variance is
non-zero. The wall values of temperature variance found by Tiselj et al. (2001a) are 2.05,
7.0 and 8.8 for Prandtl numbers 0.71, 5 and 7, respectively, while the present calculations
predict these values as 2.2, 6.4 and 9.1, respectively. The turbulence model of Kasagi et al.
(1989) at the same Reynolds number ( Reτ = 150 ), found these values to be 1.8 and 5.8 for
Pr=0.71 and 7, respectively. It is worth mentioning that the wall value of the temperature
variance is important in a sense that it could be verified with the experimental
measurements, including the measurements of Mosyak et al. (2001), where they measured
the temperature fluctuations at the wall and found to be 9.3 for Pr=5.4, which is higher
than the present results or the DNS of Tiselj et al. (2001a).

Comparison of the present results with the study of Kong et al. (2000), who performed
similar tests at Reτ = 300 (but only for Pr=0.71) indicates that the profile of the

temperature variance does not change significantly between Reτ = 150 and Reτ = 300 .

Overall, Figure 5-7(a) shows a fairly good agreement between the present results and the
DNS for both boundary conditions, while Figure 5-7(b) and (c), indicate that the present
work under-predicts the temperature variance in the isothermal situation, while for isoflux
boundary condition, the difference is not that significant. However, for both Pr=5 and 7

the present results fail to produce the peaks in θ 2 which are visible in the DNS data at
around y + < 7 . This can be understood further by comparing the profiles of the production
terms of temperature variance, Pθ 2 , and its dissipation rate, ε θ , which are shown in Figure

5-8 and 5-9, respectively. It can be seen from these figures that, as the Pr number
increases, both the production and the dissipation rates of temperature variance increase
too. Their profiles shapes are also very similar, particularly for Pr=5 and 7.

On the other hand, comparing the profiles of the dissipation rates in Figure 5-9 and 5-10,
indicates that although the profiles in both figures are very similar, the dissipation rate
under the isothermal boundary condition is much higher than that with the isoflux
boundary condition. Moreover, at the wall, dissipation for the isothermal case is

66
Chapter 5 : Results and Discussions

significant, whereas this is not the case for the isoflux situation, where the dissipation rate
at the wall becomes very small.

+
It should be mentioned that the behaviour of ε θ is quite different in different models,
because it is dependent on the boundary conditions imposed by the model. For instance,
+
the two-equation model of Sommer et al. (1992) predicts a zero value for ε θ at the wall,
while the models of Nagano & Kim (1988) or Kasagi et al. (1992) predict a finite value for
ε θ + at the wall.

Figure 5-11 shows the non-dimensionalised mean temperature profiles for a range of Pr
numbers (Pr=0.2 - 50). From this figure, one can closely see that by increasing the Pr
number, the sub-layer within which the heat conduction is significant with respect to the
turbulent transport of thermal energy becomes thinner. Also note that for Pr=1, the
temperature profile is identical to the velocity log-law.

Figure 5-12 shows how the Nusselt number increases with increasing Prandtl number of
the fluid. To validate the results obtained by the present study, the results of the empirical
correlation for pipes given by Eqn. 3-9 are also shown in this figure, as are the results of
another turbulence modelling carried out by Hwang & Lin (1999). It can be seen that the
present results are in accord with the empirical correlation, while Hwang & Lin (1999)
predicted higher values, which could be due to the modifications they made to the two-
equation heat transfer model, including the explicit prescription of σ t , instead of assuming

a constant value ( σ t = 0.9 ) , which was used in the present study.

Pr Near-Wall Region Turbulent Region

0.71 Θ + = Pr Y + Θ + = 2.25 ln( Y + ) + 3.46

5 Θ + = Pr Y + Θ + = 2.25 ln( Y + ) + 37.83

7 Θ + = Pr Y + Θ + = 2.25 ln( Y + ) + 40.30

Table 5-1 : The Log-Law equations for different Pr numbers

67
Chapter 5 : Results and Discussions

20

15

Θ+ 10
Present
5 Log-Law Near Wall Region
Log-Law T urbulent Region
DNS
0
0.1 1 10 100 1000
Log y +
(a)
60

40

Θ+
20
Present
Log-Law Near Wall Region
Log-Law T urbulent Region
DNS
0
0.1 1 10 100 1000
Log y +
(b)
80

60

Θ+ 40
Present

20 Log-Law Near Wall Region


Log-Law T urbulent Region
DNS
0
0.1 1 10 100 1000
+
Log y
(c)
Figure 5-5 : Profile of the mean temperature for a)Pr=0.71 b)Pr=5 c)Pr=7

68
Chapter 5 : Results and Discussions

1.0
DNS
Present
0.8

0.6
+

0.4

0.2

0.0
0.1 1 10 100 1000
+
Log y
(a)
1.0
DNS
Present
0.8

0.6
+

0.4

0.2

0.0
0.1 1 10 100 1000
+
Log y
(b)
1.0
DNS
Present
0.8

0.6
+

0.4 .

0.2

0.0
0.1 1 10 100 1000
+
Log y
(c)
Figure 5-6 : Profiles of the turbulent wall-normal heat flux for a)Pr=0.71 b)Pr=5 c)Pr=7

69
Chapter 5 : Results and Discussions

3
DNS is o flux

DNS is o the rm a l

P re s e nt is o flux
2 P re s e nt is o the rm a l
+
θ2
1

0
0.1 1 10 100 1000
+
Log y
(a)
10
DNS is o flux

DNS is o the rm a l
8
P re s e nt is o flux
P re s e nt is o the rm a l
6
+
θ2 4

0
0.1 1 10 100 1000
Log y +
(b)
12
DNS is o flux

10 DNS is o the rm a l
P re s e nt is o flux
8 P re s e nt is o the rm a l

+
θ2 6

0
0.1 1 10 100 1000
Log y +
(c)
Figure 5-7 : Profiles of normalised temperature variance (RMS of temperature fluctuations)
for isoflux and isothermal boundary conditions for a)Pr=0.71 b)Pr=5 c)Pr=7

70
Chapter 5 : Results and Discussions

180
160 Pr=0.71
Pr=5
140
Pr=7
120

Pθ 2 100
80
60
40

20
0
0.1 1 10 100 1000
+
Log y
Figure 5-8 : Production of temperature variance for Pr=0.71, 5 and 7

1.4
Pr=0.71
1.2
Pr=5
1.0 Pr=7

0.8
+
εθ 0.6

0.4

0.2

0.0
0.1 1 10 100 1000
+
Log y
Figure 5-9 : Profiles of the total dissipation of heat with isoflux boundary condition for
Pr=0.71, 5 and 7

140
Pr=0.71
120
Pr=5
100 Pr=7

80
+
εθ 60

40

20

0
0.1 1 10 100 1000
+
Log y
Figure 5-10 : Profiles of the total dissipation of heat with isothermal boundary condition
for Pr=0.71, 5 and 7

71
Chapter 5 : Results and Discussions

120
Pr=0.2
100 Pr=1.0
Pr=10
80 Pr=50

Θ+60
40

20

0
0.1 1 10 100
Log y +
Figure 5-11 : Profiles of mean temperature for Pr=0.71, 5 and 7

1000
Hwang & Lin (1999)

Empirical

100 Present

Nu
10

1
0.01 0.1 1 10 100 1000 10000
Pr
Figure 5-12 : Variation of Nusselt number against Prandtl number

5.3 Conjugate Heat Transfer


5.3.1 Preliminary Remarks
As mentioned earlier, the modelling of conjugate heat transfer problems has rarely been
studied in the past, and therefore finding a technique to model the thermal-field is the main
aim of the present work.

This section focuses on the results obtained by applying the four-equation model

( k − ε − θ 2 − εθ ) to the thermal-field in a conjugate system as well as the assessment of


how successful this technique would be. Further refinements in order to improve the
predictions are also presented.

72
Chapter 5 : Results and Discussions

5.3.2 Results and Modifications


5.3.2.1 Preliminary Remarks

After implementing the governing equations as described earlier, the code was tested and
the results are shown in this section.

To be consistent with the DNS data of Tiselj et al. (2001a), the code was tested by
computing 6 test cases for the Prandtl number of 7. The difference between these test cases
is in the value of the thermal activity ratio, K. Values of the parameters used are shown in
Table 4-1. These values were chosen to ensure the non-dimensional parameters Reτ , Pr ,

K and d + + matched those in the DNS of Tiselj et al. (2001a)

Figure 5-13 shows the DNS profiles of θ 2 across the fluid and solid obtained by Tiselj et
al. (2001a) for a conjugate channel flow problem at Pr=7 and Reτ = 150 , while the value
of thermal activity ratio, K, was varied. In fact, as can be seen from Table 5-2, the test
cases which have been defined for the present analysis, cover the same range of K as
indicated by Tiselj et al. (2001a) i.e. K = 0 − 100 . Moreover note that by definition, the
isothermal and the isoflux boundary conditions shown in Figure 5-13 correspond to K = 0
and K = ∞ , respectively. However, in the present study, these two limiting boundary
conditions are assumed to correspond to K = 0.0001 and K = 100 .

Parameter Test 1 Test 2 Test 3 Test 4 Test 5 Test 6


Pr 7 7 7 7 7 7
qw 10 10 10 10 10 10
ρf 1 1 1 1 1 1
cp f 0.01 1 1 1 1 10
λf 0.00143 0.143 0.143 0.143 0.143 1.43
µ 1 1 1 1 1 1
ρw 1 1 1 1 1 1
cp w 100 10 2 1 0.5 0.1
λw 14.3 1.43 0.286 0.143 0.0286 0.0143
d 0.066 0.066 0.066 0.066 0.066 0.066
d ++ 20.06 20.06 20.06 20.06 20.06 20.06
α f / αw 1 1 1 1 1 1
K 0.0001 0.1 0.5 1 5 100
Table 5-2 : The value of different parameters corresponding to each test case

73
Chapter 5 : Results and Discussions

12
Isothermal
10
K=0.5

8 K=1.0
+ K=5.0
θ2 6
Isoflux
4

0
0.1 1 10 100 1000
+
Log y
(a)
10
K=0.1

8 K=0.5

K=1.0
6
K=5.0
+
2
θ 4
Isoflux

0
0 5 10 15 20
++
d
(b)
Figure 5-13 : DNS of Tiselj et al. (2001a) for variation of temperature variance a)in the
fluid b)inside the wall

5.3.2.2 Accounting for the Difference in Specific Heat Capacities

As discussed in Section 4.6.1, the original governing equations have to be modified for a
conjugate system, not only by setting the production and turbulent terms to zero, but also
by taking into account the differences between the fluid and the solid materials. This can
be taken care of, by multiplying the transport equations of the thermal-field by either the
specific heat capacity of the fluid or that of the wall, depending on where the equations are
to be applied at.

Figure 5-14 shows how this modification affects the behaviour of the temperature variance
in the fluid region and inside the wall. In the fluid (Figure 5-14a), Tests 1 and 2 would give
74
Chapter 5 : Results and Discussions

nearly the same results which are very similar to the isothermal condition in the non-
conjugate system, while on the other hand, Tests 3-6, tend to give very close results to the
isoflux condition in the non-conjugate case, shown in Figure 5-7(c).

Inside the wall however, one can see from Figure 5-14(b) that, for Tests 1 and 2, the
temperature fluctuations are negligible, while for Tests 3-6, this is not the case.

In Figure 5-15, a different scale has been used to signify the effects of variation of K on the
temperature variance in the fluid and solid region for Tests 3-6. It seems to suggest that

reducing K would lead to a decrease in θ 2 at the solid-fluid interface which is theoretically


in agreement with the DNS data, though, comparing these results with those found by the

DNS shows that by decreasing K, the increase in the dissipation rate of θ 2 predicted by the
present work, both in the fluid and inside the wall is much lower, which is why the present
results for Test 3-6 are very close to each other and the rate of decay in the solid is also
very slow.

In order to further understand the reason for such behaviour of θ 2 , it is worth plotting the
total dissipation rate of temperature variance, ε θ ( = ε~θ + εˆθ ), through the solid-fluid
interface for all the test cases. This is shown in Figure 5-16. It can be seen that in the
+
viscous sub-layer, the behaviour of εθ becomes different for different tests. It shows that
at the interface, Tests 1 and 2 have much higher dissipation rates than Tests 3-6, which

consequently results in predicting nearly zero θ 2 at the interface, as shown in Figure 5-


+
14(a). This high prediction of εθ at the interface for Test 1 and 2 is due to the high

contribution of ε̂ θ to the total dissipation rate, in the viscous sub-layer. As a result of this

high contribution of ε̂ θ in the fluid region for Tests 1 and 2, one can see that the
dissipation rate drops dramatically through the interface, which is obviously not desirable
and thus should be avoided.

It is worth mentioning that without making the aforementioned modification to the

governing equations, the profiles of θ 2 would show no sensitivity to the value of K, which
is clearly incorrect.

75
Chapter 5 : Results and Discussions

10
T est1
T est2
8
T est3
T est4
6 T est5
+ T est6
2
θ 4

0
0.1 1 10 100 1000
+
Log y
(a)
10

8
T est1
T est2
6
+ T est3
2
θ 4
T est4
T est5
T est6
2

0
0 5 10 15 20

d ++
(b)
Figure 5-14 : Effects of taking into account the difference in c p on the variation of
temperature variance for Tests 1-6 a)in the fluid b)inside the wall

9.15
T est3
T est4
T est5
9.05 T est6

+
θ2
8.95

8.85
0.1 1 10
Log y +
(a)

76
Chapter 5 : Results and Discussions

9.15
T est3
T est4
T est5
9.05 T est6
+
θ2
8.95

8.85
0 5 10 15 20
++
d
(b)
Figure 5-15 : Effects of taking into account the difference in c p on the variation of
temperature variance for Tests 3-6 a)in the fluid b)inside the wall

1.0E+00

1.0E-03

εθ + Solid Fluid
1.0E-06

1.0E-09 Test1
Test2
Test3-6

1.0E-12
-20 -15 -10 -5 0 5 10 15 20
+ +
−y +y
Figure 5-16 : Normalised total dissipation of heat through the solid-fluid interface

77
Chapter 5 : Results and Discussions

5.3.2.3 Solving for the Total Dissipation Rate of Temperature Variance

As discussed in Chapter 3, one of the modifications made to the governing equations both
in the dynamic- and thermal-field to account for the near-wall effects, was to define the
homogenous dissipation rate (i.e. ε~ and ε~ ), which had the role of returning a non-zero
θ

dissipation rate at the wall. This could be a useful modification as far as a non-conjugate
system is concerned. However, in conjugate heat transfer problems, this modification,
particularly the homogenous dissipation rate of temperature variance, ε~ , needs to be
θ

applied with care.

As far as the governing equations of a conjugate system is concerned, it might be useful to


solve for the total dissipation rate of temperature variance instead of the homogenous one.

Figure 5-17 shows how this modification affects the results. It can be seen that although

the results of Tests 3-6 are not affected at all, the profiles of θ 2 for Tests 1 and 2 no longer
become negligible at the interface. The difference between these test cases can be seen
more clearly by changing the scales as shown in Figure 5-18.

It is also important now to see how the total dissipation rate varies through the interface.
Figure 5-19 shows that in the fluid region (including the viscous sub-layer), the dissipation
rate for all the tests are very close to each other. However, at the wall, the dissipation rate
of Test 1 is slightly lower than those of Tests 2-6 and it also undergoes a sudden drop in
the solid region, which is because its sink term becomes very large inside the wall (since

this sink term is divided by θ 2 which is very small at the interface).

Therefore, from the above analysis one can realize that in the conjugate heat transfer
problems, solving for the total dissipation rate for the thermal-field, tends to improve the

prediction of θ 2 both in the fluid and solid. This would also mean that to account for the
near-wall effects in conjugate heat transfer problems, instead of introducing homogenous

dissipation rate of θ 2 , an alternative method (or function) is required.

78
Chapter 5 : Results and Discussions

10

8
T est1
T est2
6 T est3
+ T est4

θ2 4
T est5
T est6

0
0.1 1 10 100 1000
Log y +
(a)

10

T est1
6 T est2
+ T est3
θ2 4
T est4
T est5
T est6
2

0
0 5 10 15 20
d ++
(b)
Figure 5-17 : Effects of solving for the total dissipation rate on the variation of temperature
variance for Tests 1-6 a)in the fluid b)inside the wall

9.4

9.2

9.0
+ T est2
2
θ 8.8
T est3
T est4
T est5
T est6
8.6

8.4
0.1 1 10
Log y +
(a)

79
Chapter 5 : Results and Discussions

9.4

9.2

T est2
9.0 T est3
+ T est4
θ2 8.8
T est5
T est6

8.6

8.4
0 5 10 15 20
++
d
(b)
Figure 5-18 : Effects of solving for the total dissipation rate on the variation of temperature
variance for Tests 2-6 a)in the fluid b)inside the wall

1.0E+01

1.0E-01

εθ + 1.0E-03

Solid Fluid
1.0E-05
Test1
Test2-6

1.0E-07
-20 -15 -10 -5 0 5 10 15 20
+ +
−y +y
Figure 5-19 : Normalised total dissipation rate of temperature variance through the solid-
fluid interface

80
Chapter 5 : Results and Discussions

5.3.2.4 Creating a Discontinuity in the Dissipation Rate at the Interface


The results presented in this section, are based on the modifications discussed in Section
4.6.2, in which the aim was to form a discontinuity in the profile of the normalised total

dissipation rate of θ 2 , by replacing εθ of the wall by that of the fluid in the transport

equations of θ 2 and εθ inside the wall. As a result of this, the source terms in the

transport equations of both θ 2 and εθ would be multiplied by K 2 . (see Appendix D)

Figure 5-20 shows how this modification affects the behaviour of the temperature
fluctuations in the fluid and inside the wall. Figure 5-20(b) suggests that increasing the

thermal activity ratio results in an increase in the decay rate of θ 2 inside the wall. By
changing the scale, as shown in Figure 5-21, the difference between Tests 2 to 6 become
more visible. Figure 5-21(a) indicates that, unlike previous results (and even the DNS
data), increasing K results in a decrease in the temperature variance in the fluid and at the

interface, which is obviously incorrect. Such behaviour of θ 2 can be understood better by


again plotting the variation of εθ across the fluid and the solid region which is shown in
Figure 5-22. This figure shows that as before, in the fluid the dissipation rates are very
close to each other, however, inside the wall for K > 1 (Tests 5 and 6), since the sink terms

in both θ 2 and εθ equations are multiplied by a large number, the total dissipation rate
undergoes a sudden jump at the interface, which is what the initial purpose of this
modification was. However, for K ≤ 1 (Tests 1-4), this abrupt increase in the dissipation
rate at the interface, either does not occur at all or acts inversely i.e. results in an abrupt
decrease in the dissipation rate, which is obviously undesirable.

Furthermore, from Figure 5-21(b) one can see that inside the wall, by decreasing the value
of the thermal activity ratio, K, the decay rate of εθ decreases as well (note however that
for K < 1 , this decay rate becomes zero). It is worth mentioning that direct proportionality
of thermal activity ratio with the decay rate of εθ in the wall is considered to be a desirable
feature as far as a heat transfer model for the solid is concerned, however, the above
modification cannot provide this feature properly which in overall, makes it an unsuitable
modification.

81
Chapter 5 : Results and Discussions

10
Test1
Test2
8 Test3
Test4
Test5
6 Test6
+
2
θ
4

0
0.1 1 10 100 1000
+
Log y
(a)
10

6 Test1
Test2
+ Test3
θ2 4 Test4
Test5
Test6
2

0
0 5 10 15 20
++
d
(b)
Figure 5-20 : Effects of creating a discontinuity in the dissipation rate of temperature
variance at the interface on the variation of temperature variance for Tests 1-6 a)in the
fluid b)inside the wall

9.4
Test2
Test3
9.3 Test4
Test5
Test6
9.2
+
θ2
9.1

9.0

8.9
0.1 1 10
+
Log y
(a)

82
Chapter 5 : Results and Discussions

9.4

9.0

Test2
+ 8.6 Test3
2
θ Test4
Test5
Test6
8.2

7.8
0 5 10 15 20
++
d
(b)
Figure 5-21 : Effects of creating a discontinuity in the dissipation rate of temperature
variance at the interface on the variation of temperature variance for Tests 2-6 a)in the
fluid b)inside the wall

1.0E+03

1.0E+00

1.0E-03

εθ +
1.0E-06

Test1
Test2
Solid Fluid
1.0E-09 Test3
Test4
Test5
Test6
1.0E-12
-20 -15 -10 -5 0 5 10 15 20
+ +
−y +y
Figure 5-22 : Normalised total dissipation rate of temperature variance through the solid-
fluid interface

83
Chapter 5 : Results and Discussions

5.3.3 Diagnostic Tests


5.3.3.1 Changing the Grid Spacing in the Solid

One of the diagnostic tests which had to be carried out to check the validity of the results
obtained by the code was to test the effects of changing the mesh size, particularly, inside
the wall.

As mentioned earlier, all aforementioned tests of the code were carried out using 50
uniform nodes for the wall region and 160 expanding nodes with the expansion factor of
1.025 for the fluid region (for more details see Section 4.2.3). In order to check the effects
of changing the mesh size in the wall, three more tests were carried out which are shown in
Table 5-3. These tests were computed at the same thermal activity ratio, K = 1 . Also note
that + ∆y ( jw ) and − ∆y( jw ) correspond to the size of the first cell in the fluid and in the
wall region, respectively (located on both sides of the interface).

Figure 5-23 shows the effects of changing the type and the size of the grid in the wall
region. As can be seen from this figure, while the fluid region remains unaffected, there is

only a very small change in θ 2 inside the wall, which can be assumed to be negligible.
This implies that the present problem is grid independent. However, it is worth mentioning
that as indicated by Tiselj et al. (2001a), an appropriate number of grids for the wall, in the
case of d + + = 20 , is about 40, and perhaps that is why in Figure 5-23(b), the temperature
variance is predicted slightly higher when using 20 nodes as it is rather a coarse mesh for
this case.

Type and number of grids in the solid + ∆y ( jw ) − ∆y ( j w )


20 nodes – uniform spacing 0.0745 1.0554
50 nodes – uniform spacing 0.0745 0.4092
100 nodes – uniform spacing 0.0745 0.2025
100 nodes – expanding with ratio of 1.025 0.0745 0.0476
Table 5-3 : Different type and number of grids used in the wall

84
Chapter 5 : Results and Discussions

10
20 uniform nodes

8 50 uniform nodes

100 uniform nodes


6
+ 100 expanding nodes -
θ2 4
ratio=1.025

0
0.1 1 10 100 1000

(a)
9.12
20 uniform nodes
9.10
50 uniform nodes

9.08 100 uniform nodes


+
θ2 9.06 100 expanding nodes -
ratio=1.025

9.04

9.02

9.00
0 5 10 15 20
++
d
(b)
Figure 5-23 : Effects of changing the grid size of the solid region on the variation of
temperature variance a)in the fluid b)inside the wall

5.3.3.2 Variation of non-dimensional wall thickness


As part of the investigation on the variation of the temperature variance inside the wall,
which has so far shown to be decaying with a much slower rate in comparison to the DNS
data, it is worth to test the effects of changing the non-dimensional wall thickness, d + + , on

the decay rate of θ 2 inside the wall. Therefore, five different tests were carried out at the
same thermal activity ratio of K = 1 , but with different wall thickness, as indicated in
Table 5-4.

The results are shown in Figure 5-24. It can be seen than in the fluid region, increasing the
wall thickness results in a decrease in the temperature variance at the interface. The trend
of this change is very similar to the change caused by decreasing the thermal activity ratio

85
Chapter 5 : Results and Discussions

which was discussed earlier. However, as indicated in Figure 5-24(b), the situation inside
the wall is different. As one might expect, increasing the wall thickness results in a

dramatic increase in the decay rate of θ 2 .

Note that Tiselj et al. (2001a) have also carried out tests with different wall thicknesses

( d + + = 0.5 − 20 ) and have shown similar effects on the decay rate of θ 2 inside the wall.
However, changing the wall thickness in their analysis has much greater effects on the

variation of θ 2 .

d ++ 10 20 40 80 160
d 0.033 0.066 0.132 0.264 0.528
Table 5-4 : Wall thickness in different tests

9.5

9.3

9.1
+
θ2 8.9
d++=10
d++=20
8.7 d++=40
d++=80
d++=160
8.5
0.1 1 10
Log y +
(a)
10

8
+
θ2 7
d++=10
d++=20
6 d++=40
d++=80
d++=160
5
0 20 40 60 80 100 120 140 160 180
++
d
(b)
Figure 5-24 : Effects of changing the dimensionless wall thickness on the variation of
temperature variance a)in the fluid b)inside the wall

86
Chapter 5 : Results and Discussions

5.3.3.3 Variation of the Ratio of Molecular Diffusivities


It is important to mention that in all the existing conjugate heat transfer articles including
Tiselj et al. (2001a) (DNS), Kasagi et al. (1986) (turbulence modelling) and Mosyak et al.
(2001) (experimental), the value of α f / α w (ratio of molecular diffusivities) used in the

analysis, has not been indicated explicitly. Though, by studying the governing equations in
these latter articles, it can be realized that the Pr number appears in the temperature
transport equation in both fluid and solid. This implies that in these papers, the definition
of the Pr number is not only limited to the fluid region, but it has also been extended to the
wall region (by assuming that the wall has a viscosity equal to that of the fluid), which
eventually leads to a conclusion that all the existing data on the topic of conjugate heat
transfer, have assumed α f / α w to be equal to one.

However, it is important to find out whether or not, the value of α f / α w would really

affect the results. Therefore, the code was tested with five different tests as indicated in
Table 5-5 and the results are shown in Figure 5-25. This figure indeed suggests that the
ratio of diffusivities, despite previous assumptions, has no effects on the results. In other
words, there is no need to specify this ratio explicitly as a non-dimensional parameter
when studying conjugate heat transfer problems. Thus, the current dimensionless numbers
including Pr , Reτ , d + + and K are sufficient for defining a particular test case.

Parameter Test 1 Test 2 Test 3 Test 4 Test 5


Pr 7 7 7 7 7
qw 10 10 10 10 10
ρf 1 1 1 1 1
cp f 0. 1 0.316 1 3.16 10
λf 0.0143 0.0452 0.143 0.452 1.43
µ 1 1 1 1 1
ρw 1 1 1 1 1
cp w 0.01 0.1 1 10 100
λw 0.143 0.143 0.143 0.143 0.143
d 0.66 0.208 0.066 0.0208 0.0066
d ++ 20.06 20.06 20.06 20.06 20.06
K 1 1 1 1 1
α f / αw 0.01 0.1 1.0 10 100
Table 5-5 : The value of different parameters corresponding to each test case

87
Chapter 5 : Results and Discussions

10
T est1
T est2
8
T est3
T est4
6 T est5
+
2
θ
4

0
0.1 1 10 100 1000
+
Log y
(a)
9.30
T est1
T est2
9.25 T est3
T est4
T est5
+
θ2 9.20

9.15

9.10
0 5 10 15 20
++
d
(b)
Figure 5-25 : Effects of changing the ratio of diffusivities on the variation of temperature
variance a)in the fluid b)inside the wall

88
Chapter 6 : Conclusions and Suggestions for Further Work

6 Conclusions and Suggestions for


Further Work
6.1 Concluding Remarks

The steady-state turbulent heat transfer in a one-dimensional fully-developed channel

flow with a heated wall was analysed by applying standard LRN k − ε and θ 2 − ε θ
models for the dynamic- and the thermal-fields respectively. Simulations were
performed at a constant friction Reynolds number Reτ = 150 and Prandtl numbers of
Pr=0.71, 5 and 7. Whenever available, the results were compared with the DNS data.

To solve all the required equations in the present study, a simple FORTRAN code was
used which was exclusively modified for this study by the author and a copy of it can be
found in Appendix B.

The following conclusions can be drawn from the present work:

• In conjugate heat transfer problems, the thermal-field variables including the


mean and fluctuating temperature, turbulent heat flux, temperature variance and
the dissipation rate of temperature variance are affected by the thermal wall
conditions i.e. isothermal or isoflux boundary conditions. In the near wall regions,
these variables are affected more severely by these boundary conditions.

• For low Pr numbers, the profiles of the mean temperature and the turbulent wall-
normal heat flux are in good agreement with the DNS and the Log-Law.
Increasing the Pr number however, results in over-predictions of the mean
temperature compared to DNS.

• In non-conjugate systems, the profiles of the mean temperature and the turbulent
wall-normal heat flux are not affected by the type of wall temperature boundary
condition. However, these boundary conditions have profound effects on statistics
of the temperature fluctuations in the near-wall region, y + < 10 .

89
Chapter 6 : Conclusions and Suggestions for Further Work

• The temperature variance profile in both conjugate and non-conjugate systems


falls between the two ideal cases of isothermal and isoflux walls.

• In non-conjugate heat transfer system, the temperature variance is predicted


generally with an acceptable level of agreement with DNS, for both isothermal
and isoflux boundary conditions.

• For conjugate heat transfer, due to the difference in the properties of the fluid and
solid materials, certain modifications need to be made to the governing equations
including multiplying the transport equations of the thermal-field by a c p and

creating a discontinuity in the profile of εθ across the solid-fluid interface.

• Although replacing εθ w by ε θ f in the transport equations of θ 2 and εθ inside

the wall might be desirable for limiting cases (for K >1), it does not generally

produce the required change to the profiles of θ 2 or εθ in either the fluid or the
solid region.

• Replacing the homogenous dissipation rate of θ 2 by the total dissipation rate

improves the predictions of θ 2 , both in the fluid and solid.

• The temperature variance profile is dependent on the solid-fluid material


combinations and the wall thickness as well as the Prandtl and Reynolds number
of the fluid.

• The present results do not show any dependency on the ratio of the diffusivity,
α f / α w , or the grid size in the solid region.

• The turbulent kinetic energy and its dissipation rate are under-predicated by the
k − ε model when compared to DNS data, especially in the near-wall regions.
These under-predications feed into the transport equations of the thermal-field
which could thus be a source of the discrepancies between the thermal-field of the
present results and those of the DNS.

• The calculated Nu numbers for a wide range of Pr numbers are in close


agreement with the empirical results.

90
Chapter 6 : Conclusions and Suggestions for Further Work

6.2 Suggestions for Further Work

Despite obtaining fairly encouraging results during the course of the present work on
the topic of conjugate heat transfer, further work on this topic can be done in order to
obtain better results which are in more accord with the available DNS data. Here are
some suggestions as to what could be done to improve results in the future:

• More attempts should be made to introduce other modifications to the governing


equations of the thermal-field which make the temperature variance and its
dissipation rate more sensitive to the variation of K.

• Other models for dynamic- or thermal-field should be tested as they might


produce better results.

• The only coefficient in the ε θ transport equation inside the wall ( C εθ 2 ), needs to

be tuned as it does not necessarily need to have the same value as in the εθ
transport equation in the fluid region.

• A fixed value of turbulent Prandtl number ( σ t = 0.9 ) was used in the present
work, but studies have shown that this value is not constant across the channel.
This could introduce inaccuracy in the present results and an alternative method
should be found to model this.

• Instead of using a simple eddy-diffusivity to model the turbulent heat fluxes,


other methods could be used, which might improve the present results.

91
References

References
Abe, K., Kondoh, T. and Nagano, Y., 1994, “A new turbulence model for predicting
fluid flow and heat transfer in separating and reattaching flows”, I. Flow field
calculations, Int. J. Heat Mass Transfer, 37, pp. 139-151.

Bobkov, P. V., Gribanov, Yu., I., Hragimov, M. Kh., Nomofilov, E. V. and Subbotin,
V. I., 1965, “Measurement of Intensity of Temperature Fluctuations in Turbulent Flow
of Mercury in a Tube”, Teplofiz. Vys. Temp., 3, pp. 708–716.

Cheesewright, R., 1986, “The scaling of turbulent natural convection boundary layer in
the asymptotic limit of infinite Grashof number”, Euromech, Colloquium, 207, (The
Netherlands)

Davidson, L., 1990, “Second-order corrections of the k − ε model to account for non-
isotropic effects due to buoyancy”, Int. J. Heat Mass Transfer, 33, pp. 2599-2608

Deng, B., Wu, W. and Xi, S., 2001, “A near-wall two-equation heat transfer model for
wall Turbulent flows” , Int. J. Heat and Mass Transfer, 44, pp. 691-698.

Donohue, G. L., Tiederman, W. G. and Reischman, M. M., 1972, ‘‘Flow Visualization


of the Near-Wall Region in a Drag-Reducing Channel Flow’’ J. Fluid Mech., 56, pp.
559–575.

Gavrilakis S., Tsai, H. M., Voke, P. R. and Leslie, D. C., 1986, ‘‘Direct and Large Eddy
Simulation of Turbulence’’ Notes on Numerical Fluid Mechanics, 15, U. Schumann, R.
Friedrich, Vieweg, and D. B. R. Braunschweig, eds., pp. 105.

George, W.K. and Capp, S.P., 1979, “A theory for natural convections turbulent
boundary layers next to heated surfaces”, Int. J. Heat Mass Transfer, 22, pp. 813-826

Hanjalic, K., 1996, ”Some resolved and unresolved issues in modelling non-equilibrium
and unsteady turbulent flows”, Proc. 3rd Int. Symp. on Engineering turbulence
modelling and measurements, Crete, Greece.

Hanjalic, K., Kenjeres, S. and Durst, F., 1996, “Natural convection in partitioned two-
dimensional enclosures at higher Rayleigh numbers”, lnt. J. Heat Mass Transfer, 39,
pp. 1407-1427.

Hartnett, J. P., 1955, ‘‘Experimental Determination of the Thermal Entrance Length for
the Flow of Water and Oil in Circular Pipes’’, ASME J. Heat Transfer, 77, pp. 1211–
1220.

Henkes, R.A.W.M. and Hoogendoorn, C.J., 1990, “Numerical determination of wall


functions for the turbulence natural convection boundary layer”, Int. J. Heat Mass
Transfer, 33, pp. 1087-1097.

92
References

Hetsroni, G. and Rozenblit, R., 1994, ‘‘Heat Transfer to a Liquid-Solid Mixture in a


Flume’’ Int. J. Multiphase Flow, 20, pp. 671–689.

Hetsroni, G., Zakin, J. L., and Mosyak, A., 1997, ‘‘Low-Speed Streaks in Drag-Reduced
Turbulent Flow’’ Phys. Fluids, 9, pp. 2397–2404.

Hochreiter, L. E. and Sesonske, A., 1969, “Thermal Turbulence Characteristics in


Sodium-Potassium”, Int. J. Heat Mass Transfer, 12, pp. 114-118.

Hwang, C.B. and Lin, C.A., 1999, “A low Reynolds number two-equation kθ − ε θ
model to predict thermal fields”, Int. J. Heat Mass Transfer, 42, pp. 3217-3230.

Iritani, Y., Kasagi, N. and Hizata, M., 1985, ‘‘Heat Transfer Mechanism and Associated
Turbulence Structure in the Near-Wall Region of a Turbulent Boundary Layer’’, Proc.
of the 4th Symp. on Turbulent Shear Flows, Karlsruhe, Germany.

Johnson, R.W. (edited by), 1998, “The Handbook of fluid dynamics”, Published by
Springer-Verlag GmbH and CRC press LLC.

Jones, W.P. and Launder, B.E., 1972, “The prediction of laminarization with a two-
equation model of turbulence”, Int. J. Heat Mass Transfer, 15, pp. 301-314.

Jones, W.P. and Launder, B.E., 1973, “The calculation of low Reynolds number
phenomena with a two-equation model of turbulence”, Int. J. Heat Mass Transfer, 16,
pp. 1119-1130.

Kasagi, N., Hirata, M. and Nishino, K., 1986, ‘‘Streamwise Pseudo-Vortical Structures
and Associated Vorticity in the Near-Wall Region of a Wall-Bounded Shear Flow’’ Exp.
Fluids, 4, pp. 309–318.

Kasagi, N., Kuroda, A. and Hirata, M., 1989, ‘‘Numerical Investigation of Near-Wall
Turbulent Heat Transfer Taking Into Account the Unsteady Heat Conduction in the
Solid Wall’’ J. Heat Transfer, 111, pp. 385–392.

Kasagi, N., Tomita, Y. and Kuroda, A., 1992, “Direct Numerical Simulation of Passive
Scalar Field in a Turbulent Channel Flow” J. Heat Transfer, 114, pp. 598–606.

Kasagi, N. and Ohtsubo, Y., 1993, “Direct Numerical Simulation of Low Prandtl
Number Thermal Field in a Turbulent Channel Flow”, in Turbulent Shear Flows, 8,
Springer, Berlin, pp. 97-119.

Kasagi, N. and Iida, O., 1999, ‘‘Progress in Direct Numerical Simulation of Turbulent
Heat Transfer’’, Proc. of 5th ASME/JSME Joint Thermal Engineering Conference, San
Diego, CA.

Kawamura, H., Ohsaka, K., Abe, H. and Yamamoto, K., 1998, ‘‘DNS of Turbulent Heat
Transfer in Channel Flow with low to medium-high Prandtl number fluid’’, Int. J. Heat
Fluid, 19, pp. 482–491.

93
References

Kawamura, H., Abe, H. and Matsuo, Y., 1999, ‘‘DNS of Turbulent Heat Transfer in
Channel Flow with Respect to Reynolds and Prandtl Number Effects’’, Int. J. Heat
Fluid, 20, pp. 196–207.

Kays, W.M. and Crawford, M.E., “Convective Heat and Mass Transfer”, McGraw-Hill
series in Mech. Eng., 3rd Edition.

Kenjeres, S. and Hanjalic, K., 1995, “On the prediction of thermal convection in
horizontal slender enclosures”, Proc. 10th Symp. Turbulent Shear Flows, Pennsylvania
State Univ., U.S.A

Khabakhpasheva, Y. M., 1986, ‘‘Experimental Investigation of Turbulent Momentum


and Heat Transfer in the Proximity of the Wall’’, Proc. of 8th Int. Heat Transfer
Conference, San Francisco, CA, USA.

Kim, J.,1989, “On the structure of pressure fluctuations in simulated turbulent channel
flow”, J. Fluid Mech., 205, pp. 421-451.

Kim, J. and Moin, P., 1989, ‘‘Transport of Passive Scalars in a Turbulent Channel
Flow’’, in Turbulent Shear Flows, 6, Springer-Verlag, Berlin, pp. 85.

Kong, H., Choi, H. and Lee, J. S., 2000, ‘‘Direct Numerical Simulation of Turbulent
Thermal Boundary Layer’’ Phys. Fluids, 12, No. 10, pp. 2555–2568.

Krishnamoorthy L. V. and Antonia, R. A., 1987a, “Temperature dissipation


measurements in a turbulent boundary layer”, J. Fluid Mech., 176, pp. 265-281.

Krishnamoorthy L. V. and Antonia, R. A., 1987b, “Turbulent kinetic energy budget in


the near-wall region”, AIAA J., 26, pp. 300-302.

Lam, K. L. and Banerjee, S., 1988, ‘‘Investigation of Turbulent Flow Bounded by a


Wall and a Free Surface’’, Fundamentals of Gas-Liquid Flows, Sharma Michaelides,
ed., 72, ASME, Washington DC, pp. 29–38.

Lam, K. L., 1989, ‘‘Numerical Investigation of Turbulent Flow Bounded by a Wall and
a Free-Slip Surface’’, Ph.D. Thesis, Univ. Calif. Santa Barbara.

Launder, B. E., 1976, “Heat and Mass Transport”, Turbulence Topics in Applied
Physics (Edited by P. Bradshaw), pp. 232-287. Springer, Berlin.

Launder, B. E. and Sharma, B.I., 1974, “Application of the energy dissipation model of
turbulence to the calculation of flow near a spinning disc”, Lett. Heat Mass Transfer, 1,
pp. 131-138.

Launder, B.E. and Spalding, D.B, 1974, “The numerical computation of turbulent
flows”, Comput. Methods Appl. Mech. Eng., 3, pp. 269-289

94
References

Lu, D. M. and Hetsroni, G., 1995, ‘‘Direct Numerical Simulation of a Turbulent Open
Channel Flow with Passive Heat Transfer’’ Int. J. Heat Mass Transfer, 38, pp. 3241–
3251.

Lyons, S. L. and Hanratty, T. J., 1991, ‘‘Direct Numerical Simulation of Passive Heat
Transfer in a Turbulent Channel Flow’’ Int. J. Heat Mass Transfer, 34, pp. 1149–1161.

Moin, P. and Mahesh, K., 1999, ‘‘Direct Numerical Simulation: A Tool in Turbulence
Research’’, Annu. Rev. Fluid Mech., 30, pp. 539–578.

Mosyak, A., Pogrebnyak, E. and Hetsroni, G., 2001, ‘‘Effect of Constant Heat Flux
Boundary Condition on Wall Temperature Fluctuations’’, ASME J. Heat Transfer , 123,
pp. 213–218.

Myong, H.K. and Kasagi, N., 1990, “A new approach to the improvement of
k − ε turbulence model for wall-bounded shear flows”, JSME Int. J. II, 33, pp. 63-72.

Na, Y. and Hanratty, T. J., 2000, ‘‘Limiting Behavior of Turbulent Scalar Transport
Close to a Wall’’, Int. J. Heat Mass Transf., 43, pp. 1749–1758.

Nagano, Y. and Hishida, M., 1987, “Improved form of the k − ε model for wall
turbulent shear flows”, Trans. ASME, J. Fluids Eng. 109, pp.156-160.

Nagano, Y., and Kim, C., 1988, “A two-equation model for heat transport in wall
turbulent shear flows”, Trans. ASME, J. Heat Transfer, 110, pp. 583-589.

Nagano, Y. and Tagawa, M., 1990, “An improved k − ε model for boundary layer
flows”, Trans. ASME, J. Fluids Eng., 112, pp.33-39.

Nagano, Y., Tagawa, M. and Tsuji, T., 1991, “An improved two-equation heat transfer
model for wall turbulent shear flows”, in: Proc. ASME-JSME Thermal Eng. Joint
Conf., eds. J. Lloyd and Y. Kurosaki, 3, pp. 233-240.

Nagano, Y., Hattoria, H., Abe, K., 1997, “Modelling the turbulent heat and momentum
transfer in flows under different thermal conditions” Fluid Dynamics Research, 20, pp.
127-142.

Polyakov, A. F., 1974, ‘‘Wall Effect of Temperature Fluctuations in the Viscous


Sublayer’’, Teplofiz. Vys. Temp., 12, pp. 328–337.

Sinai, Y. L., 1987, ‘‘A Wall Function for the Temperature Variance in Turbulent Flow
Adjacent to a Diabatic Wall’’, J. Heat Transfer, 109, pp. 861–865.

Sommer, T.P., So, R.M.C. and Lai, Y.G., 1992, “A near-wall two-equation model for
turbulent heat fluxes”, Int. J. of Heat and Mass Transfer, 35, pp. 3375-3387.

95
References

Sommer, T. P., So, R. M. C. and Zhang, H. S., 1994, ‘‘Heat Transfer Modelling and the
Assumption of Zero Wall Temperature Fluctuations’’, J. Heat Transfer, 116, pp. 855–
863.

Teitel, M. and Antonia, R. A., 1993, “Heat Transfer in Fully-Developed Turbulent


Channel Flow-Comparison between Experiment and Direct Numerical Simulations”,
Int. J. Heat and Mass Transfer, 36, pp. 1701–1706.

Tennekes, H. and Lumley, J. L., 1972, “A First Course in Turbulence”, MIT Press,
Cambridge, MA.

Thomas, L.H., 1949, “Elliptic problems is linear difference equations over a network”,
Watson Sci. Comput. Lab. Report, Columbia University, New York.

Tiselj, I., Bergant, R., Mavko, B., Bajsic, I. and Hetsroni, G., 2001a “DNS of Turbulent
Heat Transfer in Channel Flow with Heat Conduction in the Solid Wall”, J. Heat
Transfer, 123, pp. 849–857.

Tiselj, I., Pogrebnyak, E., Changfeng, Li., Mosyak, A. and Hetsroni, G., 2001b, ‘‘Effect
of Wall Boundary Condition on Scalar Transfer in a Fully Developed Turbulent Flume
Flow’’ Phys. Fluids, 13, No. 4, pp. 1028–1039.

Tiselj, I., Horvat, A., Mavko, B., Pogrebnyak, E., Mosyak, A. and Hetsroni, G., 2004,
“Wall Properties and Heat Transfer in Near-Wall Turbulent”, Numerical Heat Transfer,
Part A, 46, pp. 717-729.

Versteeg, H. K. and Malalasekera, W., 1998, “An introduction to computational fluid


dynamics, the finite volume method”, Addison Wesley Longman Ltd., England.

Yap, C.R., 1987, “Turbulent heat and momentum transfer in re-circulating and
impinging flows”, Ph.D. Thesis, Faculty of Technology, University of Manchester.

Youssef, M.S., Nagano, Y. and Tagawa, M., 1992, “A two-equation heat transfer model
for predicting turbulent thermal fields under arbitrary wall thermal conditions”, Int. J.
Heat Mass Transfer, 35, pp. 3095-3104.

96
Appendix A

Appendix A
The Log-Law:
• Velocity Log-Law
The total shear stress across the inner region as well as the buffer and the viscous sub-
layers is given by:

∂U
τ =τw = µ − ρ uv Eqn. A- 1
∂y

In the viscous sub-layer, ρ uv = 0 , thus:

∂U
µ = τw Eqn. A- 2
∂y

τw
U= y+c Eqn. A- 3
µ

Since at y = 0 , U = 0 and c = 0 . Thus:

y (τ w / ρ )
U= Eqn. A-4
ν

Or U + = Y + Eqn. A-5

However, in the inner region, τ w = 0 , thus Eqn. A-1 becomes:

∂U
µt = ρ uv Eqn. A-6
∂y

As mentioned earlier, according to Mixing-Length Hypothesis:

∂U
µt = ρlm Eqn. A-7
∂y

97
Appendix A

where:

lm = κy Eqn. A-8

κ = 0.41 Eqn. A-9

By using the definition of U + and Y + , one would get:

∂U + 1
+
= Eqn. A-10
∂Y κY +

Then, by integrating from the outer edge of the viscous sub-layer:

U+ Y+
1 ∂Y +
∫ ∂U = κ 10∫.8 Y +
+
Eqn. A-11
10.8

+
Where 10.8 is typical value of the critical dimensionless distance from the wall, Ycrit .,

for the fundamental case of zero pressure gradient boundary layer.

By re-arranging Eqn. A-11, the following logarithmic equation is obtained:

1
U+ = ln( Y + ) + 5.0 Eqn. A-12
0.41

The above equation is generally called: “Law of the Wall” or “Velocity Log-Law”

The implication of this log-law is that, the velocity profile in a channel flow, say, should
follow Eqn. A-5 and A-12 in the viscous sub-layer and in the turbulent region,
respectively.

• Thermal Log-Law

The total heat flux across the inner region as well as the buffer and the viscous sub-
layers is given by:

µ ∂Θ
− qy = − ρ uθ Eqn. A-13
Pr ∂y

By introducing the effective-viscosity approximation, the heat flux is approximated as:

98
Appendix A

µ t ∂Θ
ρ uθ = − Eqn. A-14
σ t ∂y

Combining Eqn. A-13 and A-14, would result in:

 µ µt  ∂Θ
− q y =  +  Eqn. A-15
 Pr σ t  ∂y

In boundary layers where the wall temperature, Θ w , is constant, the convective

transport term can be neglected i.e. q y = qw across the near-wall regions. Thus, Eqn. A-

15 becomes:

 µ µt  ∂Θ
− qw =  +  Eqn. A-16
 Pr σ t  ∂y

And by defining dimensionless temperature as:

(Θ w − Θ )ρc pUτ
Θ+ = Eqn. A-17
qw

Eqn. A-16 becomes:

 µ µt  ∂Θ +
 +  + = 1 Eqn. A-18
 Pr σ t  ∂Y

Integrating the above equation, would result in the following expression for the
dimensionless temperature across a turbulent boundary layer:

dY +
Θ+ = ∫ Eqn. A-19
[(1 / Pr ) + (α t / ν )]
Following very similar procedures as were used for the momentum boundary layer, one
could introduce a general law for the temperature profile in a boundary layer. However,
+
unlike the momentum boundary layer, Ycrit . in the thermal boundary layer is not fixed

and it varies with the value of Pr . Therefore, only for a fluid with Pr = 1 , might one
+
expect to find Ycrit . = 10.8 (in fact, even this would not be the case, due to high and

varying σ t in the viscous sub-layer).

99
Appendix A

If the integral of Eqn. A-19 is split into two parts corresponding to a viscous sub-layer
and a fully developed region, then for the region Y + > Ycrit
+
. , one would have:

+
Ycrit Y+
.
dY + dY +
∫ +∫
+
Θ = Eqn. A-20
0
[(1 / Pr ) + (α t / ν )] Ycrit+ . [(1 / Pr ) + (α t / ν )]

If the same procedure as for velocity log-law is to be used, α t / ν in the first integral
and 1 / Pr in the second integral have to be neglected. But the magnitude of 1 / Pr can
vary tremendously and therefore can have a large influence on the importance of all the
terms in the denominators.

In reality although the eddy-diffusivity in the sub-layer is very small, it is not zero. But
if Pr is large, a small diffusivity in the sub-layer can be significant, therefore,
significant error would be introduced if α t / ν is neglected. One solution to this problem
+
would be to use a Ycrit . , which is a function of Pr . But this means that this approach

would be restricted to only two Pr numbers, 0.7 and 5.9, as experimental data are only
available for air and water. (see Table 3-4 in Chapter 3)

However, note that at very low Pr , the 1 / Pr term is so dominates the first integral of
Eqn. A-20 that α t / ν can be neglected in any case, which implies that in the viscous
sub-layer, the profile should follow the following equation:

Θ + = Pr Y + Eqn. A-21

However, in the fully turbulent region (i.e. in the second integral in Eqn. A-20), the
situation is reversed, which means, the 1 / Pr term can become large for low- Pr fluids,
and can become significantly larger than α t / ν . The relative magnitude of these terms
can be investigated by means of Reynolds analogy along with the Mixing-Length
Hypothesis. Thus:

αt / ν = κ Y + Eqn. A-22

where κ = 0.41 , as before.

100
Appendix A

Therefore, by applying all the aforementioned assumptions, one would arrive at the
following equation (Kays & Crawford):

+
Ycrit Y+
.
σ t dY +
∫ Pr dY + ∫
+ +
Θ = Eqn. A-23
0 +
Ycrit .
κ Y+

The above equation is called: “Thermal Law of the Wall” which means the temperature
profile in the turbulent region of a boundary layer, should follow this equation.

101
Appendix B

Appendix B
The FORTRAN Code:
PROGRAM CHANNEL
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
DIMENSION REL(JP)
C----OPENING FILES-------------
OPEN(10,FILE="C:\FILE10.TXT")
OPEN(11,FILE="C:\FILE11.TXT")
OPEN(12,FILE="C:\FILE12.TXT")
OPEN(13,FILE="C:\FILE13.TXT")
OPEN(14,FILE="C:\FILE14.TXT")
OPEN(20,FILE='C:\FILE20.TXT')
OPEN(30,FILE='C:\FILE30.TXT')
OPEN(40,FILE='C:\FILE40.TXT')
OPEN(50,FILE='C:\FILE50.TXT')
OPEN(60,FILE='C:\FILE60.TXT')
OPEN(70,FILE='C:\FILE70.TXT')
OPEN(80,FILE='C:\FILE80.TXT')
OPEN(90,FILE='C:\FILE90.TXT')
OPEN(100,FILE='C:\FILE100.TXT')
OPEN(110,FILE='C:\FILE110.TXT')
OPEN(120,FILE='C:\FILE120.TXT')
OPEN(130,FILE='C:\FILE130.TXT')
OPEN(140,FILE='C:\FILE140.TXT')
OPEN(150,FILE='C:\FILE150.TXT')
C --- Total number of grid nodes
NJ=210
C --- Number of grid nodes within solid wall
NW=50
C------ SPECIFY CONSTANTS -------------------------
SGMTE=1.
SGMED=1.3
CMU=0.09
TINY=10.**(-5)
GREAT=10.**(15)
RZER=0.
CE1=1.44
CE2=1.92
SGMT2=1.
SGMET=1.3
CET1=0.72
CET2=2.2
CET3=1.3

102
Appendix B

CET4=0.8
C--- Dynamic Viscosity
VISCOS=1.
C --- Molecular Prandtl number
PR=7.
C --- Turbulent Prandtl number
PRT=0.9
C --- Wall heat flux rate divided by Cp
QW=10
C --- Density of fluid
DENSIT=0.5
C --- Density of solid
DENSITS=5.
C--- Thermal Conductivity of fluid
C RKF=1.
C --- Thermal Conductivity of solid
RKS=1.
C --- Specific Heat Capacity of solid
CPS=5.
C --- Specific Heat Capacity of fluid
CPF=1.
C----------UNDER-RELAXATIONS ------------------
URFU=0.425
URFT=0.425
URFTE=0.425
URFED=0.425
URFET=0.425
URFT2=0.425
URFP=0.25
URFPED=0.425
NSW=30
MAXIT=300000
C --- CALCULATING SOME PARAMETERS---------------------
RKF=CPF*VISCOS/PR
ALPHF=RKF/(CPF*DENSIT)
ALPHS=RKS/(CPS*DENSITS)
ALPHR=ALPHS/ALPHF
RKR=RKF/RKS
C------------------GENERATE GRID --------------------
C --- Channel Half-Height
YLARGE=0.5
C ---- Wall Thickness
HWALL=0.1
C --- DY=YLARGE/FLOAT(NJ-1)
EXY=1.025
DY=YLARGE*(EXY-1.)/((EXY**(NJ-NW))-1.)
C --- Y(J) NODAL POINTS FOR U VELOCITY AND ALL SCALARS
Y(NW)=0.
DO J=NW+1,NJ
Y(J)=Y(J-1)+DY
DY=EXY*DY
ENDDO
C EXYW=1.025
C DYW=HWALL*(EXYW-1.)/((EXYW**(NW-1))-1.)
DYW=HWALL/FLOAT(NW-1)
DO J=NW-1,1, -1
Y(J)=Y(J+1)-DYW
C DYW=EXYW*DYW
ENDDO
C --- YC(J) SOUTH FACE Y LOCATION OF CONTROL VOLUME FOR Y(J)
DO J=2,NJ
YC(J)=0.5*(Y(J-1)+Y(J))
ENDDO
C----NODE BETWEEN J+1 AND J NODE

103
Appendix B

DO J = 1,NJ-1
DYN(J)=Y(J+1)-Y(J)
ENDDO
C----CROSS-SECTIONAL AREA OF CONTROL VOLUME, NORMAL TO THE FLOW ALSO AREA X 1 X 1 =
VOLUME OF C.V.
DO J=2,NJ-1
AREA(J)=YC(J+1)-YC(J)
ENDDO
C--------------- STARTING CONDITIONS -------------------
UIN=5000.
TEIN=0.002*UIN*UIN
EDIN=(TEIN**1.5)/(0.34)
VISIN=DENSIT*CMU*TEIN*TEIN/EDIN
T2IN=0.02*(2.3**2.) !((T(NW)-TB)**2.)
ETIN=T2IN*(TEIN**0.5)/YLARGE
DO J=2,NJ
T(J)=300.
U(J)=UIN
TE(J)=TEIN
ED(J)=EDIN
ET(J)=ETIN
T2(J)=T2IN
TRE(J)=DENSIT*TEIN*TEIN/(VISCOS*EDIN)
ENDDO
ED(1)=ED(2)
DO J=1,NJ
VIS(J)=VISCOS+VISIN
ENDDO
C--------------CALCULATE FLOW RATE -------------------
FLOW1=0
DO J=NW+1,NJ-1
FLOW1=FLOW1+DENSIT*U(J)*AREA(J)
ENDDO
SNORU=FLOW1*UIN
SNORTE=FLOW1*TEIN
SNORED=FLOW1*EDIN
SNORET=FLOW1*ETIN
SNORT2=FLOW1*T2IN
SNORT=FLOW1*10.
C------------ ITERATION LOOP -------------------------
DO NITER = 1,MAXIT
CALL CALCU
CALL CALCT
CALL CALCDP
CALL CALCTE
CALL CALCED
CALL CALCT2
CALL CALCET
CALL PROPS
RESORU=RESORU/SNORU
RESORT=RESORT/SNORT
RESORTE=RESORTE/SNORTE
RESORED=RESORED/SNORED
RESORET=RESORET/SNORET
RESORT2=RESORT2/SNORT2
C --- MONITORING OF EQUATION RESIDUALS
IF(MOD(NITER,1000).EQ.0)
1 WRITE(*,*)NITER,RESORET,RESORT2!,RESORU!,RESORT!,RESORTE,RESORED
ENDDO
C---- OUTPUTS FOR GRAPHICAL PROCESSING ---------------------------------------------
TAW=U(NW+1)*0.5*(VIS(NW)+VIS(NW+1))/DYN(NW)
UTAW=SQRT(TAW/DENSIT)
TTAW=QW/(CPF*UTAW*DENSIT)
RE=(UB*YLARGE*2)/(VISCOS/DENSIT)

104
Appendix B

RETAW=(UTAW*YLARGE)/(VISCOS/DENSIT)
NUA=(QW*YLARGE*2)/((CPF*VISCOS/PR)*(T(NW)-TB))
NUB=(T(NW)-T(NW+1))/(Y(NW+1)-Y(NW))
1 *((VISCOS/PR)+(0.5*((VIS(NW+1)-VISCOS)/PRT)))
2 *(YLARGE*2)/((VISCOS/PR)*(T(NW)-TB))
NU1=0.023*(RE**0.8)*(PR**0.3)
NU2=0.0214*((RE**0.8)-100)*(PR**0.3)
ARK=SQRT((RKF*DENSIT*CPF)/(RKS*DENSITS*CPS))
HPLS=DENSIT*HWALL*UTAW/VISCOS
H2PLS=HPLS*SQRT(ALPHF/ALPHS)
DO J=1,NJ
REL(J)=(TE(J)**1.5)/(ED(J)+TINY)
YPLS(J)=DENSIT*(Y(J)-Y(NW))*UTAW/VISCOS
Y2PLS(J)=YPLS(J)*SQRT(ALPHF/ALPHS)
UPLS(J)=U(J)/UTAW
TPLS(J)=(T(NW)-T(J))/TTAW
ENDDO
DO J=1,NW
TPLS(J)=(T(J)-T(NW))/TTAW
END DO
WRITE(10,*) UIN,T(NW),TB,VISCOS
WRITE(11,*) DENSIT,PR,PRT,RE
WRITE(12,*) NUA,NUB,NU1,NU2
WRITE(13,*) RE,RETAW
DO J=1,NJ
WRITE(20,*) YPLS(J),UPLS(J),Y(J),U(J)
WRITE(30,*) YPLS(J),TPLS(J),Y(J),T(J)
WRITE(40,*) YPLS(J),YC(J),AREA(J)
ENDDO
DO J=NW,NJ
WRITE(50,*)YPLS(J),(ED(J)/UTAW**4),(EDNW(J)+ED(J))/UTAW**4
ENDDO
DO J=NW+1,NJ-1
DKDY=(TE(J+1)-TE(J-1))/(Y(J+1)-Y(J-1))
DEDY=(ED(J+1)+0.*EDNW(J+1)-ED(J-1)-0.*EDNW(J-1))/(Y(J+1)-Y(J-1))
VEFF=1.+.5*(VIS(J)-1.)
SU1=2.*VEFF*(ED(J)+EDNW(J))*DKDY*DKDY/(TE(J)*TE(J)+TINY)
SU1=SU1/(UTAW**6)
SU2=-2.*VEFF*DKDY*DEDY/(TE(J)+TINY)
SU2=SU2/(UTAW**6)
WRITE(60,*)YPLS(J),TAW,UTAW
ENDDO
DO J=NW+1,NJ-1
OMG=(ED(J)+EDNW(J))/(TE(J)+TINY)
OMG=OMG/(UTAW**2)
WRITE(70,*)ALPHF,ALPHS,ALPHR,RKR
ENDDO
DO J=NW+1,NJ-1
DPEDY = ((PK(J+1)*DENSIT/(ED(J+1)+EDNW(J+1)+TINY))-
& (PK(J-1)*DENSIT/(ED(J-1)+EDNW(J-1)+TINY)))/(YPLS(j+1)-YPLS(j-1))
D2UDY2=( ((U(J+1)-U(J))/DYN(J) ) - ((U(J)-U(J-1))/DYN(J-1)))
1 /AREA(J)
E1=SQRT(TE(J))*((TE(J)/(ED(J)+TINY))**2)*0.5*ABS(D2UDY2)
WRITE(80,*)YPLS(J),DPEDY,E1,TRE(J)
ENDDO
DO J=1,NJ
VT(J)=-((T(J+1)-T(J-1))/(Y(J+1)-Y(J-1)))
1 *((VIS(J)-VISCOS)/(DENSIT*PRT))
UV(J)=((U(J+1)-U(J-1))/(Y(J+1)-Y(J-1)))*(VIS(J)-VISCOS)/DENSIT
WRITE(90,*) YPLS(J),TE(J),TE(J)/UTAW**2
WRITE(100,*) YPLS(J),ED(J),EDNW(J),(EDNW(J)+ED(J))/UTAW**4
WRITE(110,*) YPLS(J),Y2PLS(J),T2(J),SQRT(T2(J))/TTAW
WRITE(120,*) YPLS(J),ET(J),ETNW(J)
1 ,(ET(J)+ETNW(J))*(VISCOS/DENSIT)/(TTAW**2*UTAW**2)

105
Appendix B

WRITE(130,*) YPLS(J),VT(J)/(TTAW*UTAW),UV(J)/UTAW
WRITE(140,*) HWALL,H2PLS,ARK
WRITE(150,*) Y(J),YPLS(J),PK(J),PT2(J)
ENDDO
STOP
END
C---------------------------------------------------------------------------------------------------------------
SUBROUTINE CALCU
C===========VELOCITY SUBROUTINE================U=================
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
RESORU=0.
C --- ASSEMBLY OF COEFFIENTS AND SOURCE TERMS
C --- SOLID REGION
DO J=1,NW-1
U(J)=0.
ENDDO
C --- FLUID REGION
DO J=NW+1,NJ-1
VISN=0.5*(VIS(J+1)+VIS(J))
VISS=0.5*(VIS(J-1)+VIS(J))
AN(J)=VISN/DYN(J)
AS(J)=VISS/DYN(J-1)
SU(J)=PPD*AREA(J)
SP(J)=0.
ENDDO
C --- BOUNDARY CONDITIONS
U(NW)=0.
AN(NJ-1)=0.
C --- DIAGONAL COEFFICIENT AND UNDER-RELAXATION
DO J=NW+1,NJ-1
AP(J)=AN(J)+AS(J)-SP(J)
RESOR=AP(J)*U(J)-AN(J)*U(J+1)-AS(J)*U(J-1)-SU(J)
RESORU=RESORU+ABS(RESOR)
AP(J)=AP(J)/URFU
SU(J)=SU(J)+(1.-URFU)*AP(J)*U(J)
DU(J)=AREA(J)/AP(J)
ENDDO
C --- SOLUTION OF TRIADIAGONAL SYSTEM OF DISCRETISED EQUATIONS
CALL SOLVER(U,NW+1,NJ-1,NJ)
U(NJ)=U(NJ-1)
RETURN
END
C------------------------------------------------------------------------------------------------------------------
SUBROUTINE CALCT
C ==============TEMPERATURE SUBROUTINE===========T=================
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),

106
Appendix B

3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
RESORT=0.
C --- ENERGY BALANCE
DTDX=QW/(CPF*FLOW1)
C --- ASSEMBLY OF COEFFIENTS AND SOURCE TERMS
C --- SOLID REGION
DO J=2,NW
VISN=RKS/CPS
VISS=VISN
AN(J)=CPS*VISN/DYN(J)
AS(J)=CPS*VISS/DYN(J-1)
SU(J)=0.
SP(J)=0.
ENDDO
C --- FLUID REGION
DO J=NW+1,NJ-1
VISN=0.5*(VIS(J+1)+VIS(J))
VISS=0.5*(VIS(J-1)+VIS(J))
AN(J)=(((VISN-VISCOS)/PRT)+VISCOS/PR)*CPF/DYN(J)
AS(J)=(((VISS-VISCOS)/PRT)+VISCOS/PR)*CPF/DYN(J-1)
SU(J)=-CPF*DENSIT*U(J)*DTDX*AREA(J)
SP(J)=0.
ENDDO
C --- BOUNDARY CONDITIONS
T(1)= T(2)+DYN(1)*QW/RKS
AS(2)=0.
SU(2)=SU(2)+QW
AN(NJ-1)=0.
C --- INTERFACE CONDITION
AN(NW)=AS(NW+1)
C --- DIAGONAL COEFFICIENT AND UNDER-RELAXATION
DO J=2,NJ-1
AP(J)=AN(J)+AS(J)-SP(J)
RESOR=AP(J)*T(J)-AN(J)*T(J+1)-AS(J)*T(J-1)-SU(J)
RESORT=RESORT+ABS(RESOR)
AP(J)=AP(J)/URFT
SU(J)=SU(J)+(1.-URFT)*AP(J)*T(J)
ENDDO
C --- SOLUTION OF TRIADIAGONAL SYSTEM OF DISCRETISED EQUATIONS
CALL SOLVER(T,2,NJ-1,NJ)
C --- SETTING OF REFERENCE TEMPERATURE AT 300.
T(NJ)=T(NJ-1)
DELT=300.-T(NJ-1)
DO J=1,NJ
T(J)=T(J)+DELT
ENDDO
RETURN
END
C---------------------------------------------------------------------------------------------------------------------------
SUBROUTINE CALCDP
C======================PRESSURE GRADIENT SUBROUTINE=====================
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),

107
Appendix B

2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
C --- CALCULATION OF PRESSURE GRADIENT, PPD, FROM GLOBAL CONTINUITY
FLOW=0.
ARDU=0.
TB=0.
UB=0.
DO J=NW+1,NJ-1
FLOW=FLOW+DENSIT*U(J)*AREA(J)
ARDU=ARDU+DENSIT*DU(J)*AREA(J)
TB=TB+DENSIT*U(J)*T(J)*AREA(J)
UB=UB+U(J)*AREA(J)
ENDDO
TB=TB/FLOW
UB=UB/YLARGE
RATIO=(FLOW1-FLOW)/FLOW1
IF(MOD(NITER,100000).eq.0)
1 WRITE(*,*)RATIO
DP=(FLOW1-FLOW)/ARDU
PPD=PPD+URFU*DP
DO J=NW+1,NJ-1
U(J)=U(J)+DU(J)*DP
ENDDO
U(NJ)=U(NJ-1)
RETURN
END
C-----------------------------------------------------------------------------------------------------------------------
SUBROUTINE CALCTE
C==========TURBULENT KINETIC ENERGY SUBROUTINE====TE=================
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
RESORTE=0.
C --- NEAR WALL DISSIPATION RATE
DO J=NW+1,NJ-1
EDNW(J)=2.*(VISCOS/DENSIT)*((SQRT(TE(J+1))-SQRT(TE(J-1)))
1 /(Y(J+1)-Y(J-1)))**2
ENDDO
EDNW(NW)=EDNW(NW+1)
C --- ASSEMBLY OF COEFFIENTS AND SOURCE TERMS
C --- SOLID REGION
DO J=1,NW-1
TE(J)=0.

108
Appendix B

ENDDO
C --- FLUID REGION
DO J=NW+1,NJ-1
VISN=0.5*(VIS(J+1)+VIS(J))
VISS=0.5*(VIS(J-1)+VIS(J))
AN(J)=(VISCOS+(VISN-VISCOS)/SGMTE)/DYN(J)
AS(J)=(VISCOS+(VISS-VISCOS)/SGMTE)/DYN(J-1)
DUDY2=((U(J+1)-U(J-1))/(Y(J+1)-Y(J-1)))**2
PK(J)=(VIS(J)-VISCOS)*DUDY2/DENSIT
SU(J)=(PK(J)-ED(J))*DENSIT*AREA(J)
SP(J)=-DENSIT*EDNW(J)*AREA(J)/(TE(J)+TINY)
SP(J)=SP(J)+MIN(SU(J),RZER)/(TE(J)+TINY)
SU(J)=MAX(SU(J),RZER)
ENDDO
C --- BOUNDARY CONDITIONS
TE(NW)=0.
AN(NJ-1)=0.
C --- DIAGONAL COEFFICIENT AND UNDER-RELAXATION
DO J=NW+1,NJ-1
AP(J)=AN(J)+AS(J)-SP(J)
RESOR=AP(J)*TE(J)-AN(J)*TE(J+1)-AS(J)*TE(J-1)-SU(J)
RESORTE=RESORTE+ABS(RESOR)
AP(J)=AP(J)/URFTE
SU(J)=SU(J)+(1.-URFTE)*AP(J)*TE(J)
ENDDO
C --- SOLUTION OF TRIADIAGONAL SYSTEM OF DISCRETISED EQUATIONS
CALL SOLVER(TE,NW+1,NJ-1,NJ)
TE(NJ)=TE(NJ-1)
RETURN
END
C-----------------------------------------------------------------------------------------------------------------------
SUBROUTINE CALCED
C==================DISSIPATION OF KE SUBROUTINE============ED==========
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
RESORED=0.
C --- ASSEMBLY OF COEFFIENTS AND SOURCE TERMS
DO J=1,NW-1
ED(J)=0.0
END DO
DO J=NW+1,NJ-1
VISN=0.5*(VIS(J+1)+VIS(J))
VISS=0.5*(VIS(J-1)+VIS(J))
AN(J)=(VISCOS+(VISN-VISCOS)/SGMED)/DYN(J)
AS(J)=(VISCOS+(VISS-VISCOS)/SGMED)/DYN(J-1)
D2UDY2=( ((U(J+1)-U(J))/DYN(J) ) - ((U(J)-U(J-1))/DYN(J-1)))
1 /AREA(J)
ETRM(J)=2*(VISCOS/DENSIT)*((VIS(J)-VISCOS)/DENSIT)* (D2UDY2**2)
F2=1.-0.3*EXP(-TRE(J)*TRE(J))
PED(J)=CE1*ED(J)*PK(J)*DENSIT/(TE(J)+TINY)
SU(J)=(CE1*PK(J)-F2*CE2*ED(J))*DENSIT*ED(J)/(TE(J)+TINY)

109
Appendix B

SU(J)=(SU(J)+DENSIT*ETRM(J))*AREA(J)
SP(J)=MIN(SU(J),RZER)/(ED(J)+TINY)
SU(J)=MAX(SU(J),RZER)
ENDDO
C --- BOUNDARY CONDITIONS
ED(NW)=0.
AN(NJ-1)=0.
C --- DIAGONAL COEFFICIENT AND UNDER-RELAXATION
DO J=NW+1,NJ-1
AP(J)=AN(J)+AS(J)-SP(J)
RESOR=AP(J)*ED(J)-AN(J)*ED(J+1)-AS(J)*ED(J-1)-SU(J)
RESORED=RESORED+ABS(RESOR)
AP(J)=AP(J)/URFED
SU(J)=SU(J)+(1.-URFED)*AP(J)*ED(J)
ENDDO
C --- SOLUTION OF TRIADIAGONAL SYSTEM OF DISCRETISED EQUATIONS
CALL SOLVER(ED,NW+1,NJ-1,NJ)
ED(NJ)=ED(NJ-1)
RETURN
END
C-----------------------------------------------------------------------------------------------------------------------
SUBROUTINE CALCT2
C===============TEMPERATURE VARIANCE SUBROUTINE=====T2==============
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
RESORT2=0.
C --- NEAR WALL RATE DISSIPATION OF HEAT
C DO J=NW+1,NJ-1 !<<<<<<<<<<<<<<<<
C ETNW(J)=ALPHF*((SQRT(T2(J+1))-SQRT(T2(J-1)))
C 1 /(Y(J+1)-Y(J-1)))**2
C ENDDO
C ETNW(NW)=ETNW(NW+1)
C DO J=2,NW-1 !<<<<<<<<<<<<<<<<
C ETNW(J)=ALPHS*((SQRT(T2(J+1))-SQRT(T2(J-1)))
C 1 /(Y(J+1)-Y(J-1)))**2
C ENDDO
C ETNW(1)=ETNW(2)
DO J=1,NJ
ETNW(J)=0.0
END DO
C --- ASSEMBLY OF COEFFIENTS AND SOURCE TERMS
C --- SOLID REGION
DO J=2,NW
VISN=RKS/CPS
VISS=VISN
AN(J)=CPS*VISN/DYN(J)
AS(J)=CPS*VISS/DYN(J-1)
SU(J)=-2.*CPS*ET(J)*DENSITS*AREA(J)
SP(J)=-2.*CPS*DENSITS*ETNW(J)*AREA(J)/(T2(J)+TINY)
SP(J)=SP(J)+MIN(SU(J),RZER)/(T2(J)+TINY)
SU(J)=MAX(SU(J),RZER)

110
Appendix B

ENDDO
C --- FLUID REGION
DO J=NW+1,NJ-1
VISN=0.5*(VIS(J+1)+VIS(J))
VISS=0.5*(VIS(J-1)+VIS(J))
AN(J)=((VISCOS/PR)+(VISN-VISCOS)/(PRT*SGMT2))*CPF/DYN(J)
AS(J)=((VISCOS/PR)+(VISS-VISCOS)/(PRT*SGMT2))*CPF/DYN(J-1)
DTDY2=((T(J+1)-T(J-1))/(Y(J+1)-Y(J-1)))**2
PT2(J)=1.*(VIS(J)-VISCOS)*DTDY2/(DENSIT*PRT)
SU(J)=2.*CPF*(PT2(J)-ET(J))*DENSIT*AREA(J)
SP(J)=-2.*CPF*DENSIT*ETNW(J)*AREA(J)/(T2(J)+TINY)
SP(J)=SP(J)+MIN(SU(J),RZER)/(T2(J)+TINY)
SU(J)=MAX(SU(J),RZER)
ENDDO
C --- BOUNDARY CONDITIONS
C T2(NW)=0.
C AS(NW+1)=0.
C T2(NW)=T2(NW+1)
AN(NW)=AS(NW+1)
AS(2)=0.
AN(NJ-1)=0.
C --- DIAGONAL COEFFICIENT AND UNDER-RELAXATION
DO J=2,NJ-1
AP(J)=AN(J)+AS(J)-SP(J)
RESOR=AP(J)*T2(J)-AN(J)*T2(J+1)-AS(J)*T2(J-1)-SU(J)
RESORT2=RESORT2+ABS(RESOR)
AP(J)=AP(J)/URFT2
SU(J)=SU(J)+(1.-URFT2)*AP(J)*T2(J)
ENDDO
C --- SOLUTION OF TRIADIAGONAL SYSTEM OF DISCRETISED EQUATIONS
CALL SOLVER(T2,2,NJ-1,NJ)
T2(1)=T2(2)
T2(NJ)=T2(NJ-1)
RETURN
END
C----------------------------------------------------------------------------------------------------------------------------
SUBROUTINE CALCET
C==========DISSIPATION RATE OF TURBULENT KINETIC ENERGY=== ET ============
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
RESORET=0.
C --- ASSEMBLY OF COEFFIENTS AND SOURCE TERMS
C --- SOLID REGION
DO J=2,NW
VISN=RKS/CPS
VISS=VISN
AN(J)=CPS*VISN/DYN(J)
AS(J)=CPS*VISS/DYN(J-1)
SP(J)=-CET2*CPS*DENSITS*AREA(J)*ET(J)/(T2(J)+TINY)
SU(J)=RZER
ENDDO

111
Appendix B

C --- FLUID REGION


DO J=NW+1,NJ-1
VISN=0.5*(VIS(J+1)+VIS(J))
VISS=0.5*(VIS(J-1)+VIS(J))
AN(J)=((VISCOS/PR)+(VISN-VISCOS)/(PRT*SGMET))*CPF/DYN(J)
AS(J)=((VISCOS/PR)+(VISS-VISCOS)/(PRT*SGMET))*CPF/DYN(J-1)
DTDY2=((T(J+1)-T(J-1))/(Y(J+1)-Y(J-1)))**2
D2TDY2=(((T(J+1)-T(J))/DYN(J))-((T(J)-T(J-1))/DYN(J-1)))
1 /AREA(J)
DUDY2=((U(J+1)-U(J-1))/(Y(J+1)-Y(J-1)))**2
PK(J)=(VIS(J)-VISCOS)*DUDY2/DENSIT
PT2(J)=1.*(VIS(J)-VISCOS)*DTDY2/(DENSIT*PRT)
FET=1.
EETTRM(J)=(2*ALPHF*(D2TDY2**2))
1 *((VIS(J)-VISCOS)/(DENSIT*PRT))
SU(J)= ((CET1*PK(J)*ET(J))/(TE(J)+TINY))
1 +(CET3*PT2(J)*ET(J)/(T2(J)+TINY))
2 -(CET2*(ET(J)**2)/(T2(J)+TINY))
3 -(CET4*FET*ET(J)*ED(J)/(TE(J)+TINY))
SU(J)=(SU(J)+EETTRM(J))*CPF*DENSIT*AREA(J)
SP(J)=MIN(SU(J),RZER)/(ET(J)+TINY)
SU(J)=MAX(SU(J),RZER)
ENDDO
C --- BOUNDARY CONDITIONS
C ET(NW)=0.
C ET(NW)=ET(NW+1)
C AS(NW+1)=0.
AN(NW)=AS(NW+1)
AS(2)=0.
AN(NJ-1)=0.
C --- DIAGONAL COEFFICIENT AND UNDER-RELAXATION
DO J=2,NJ-1
AP(J)=AN(J)+AS(J)-SP(J)
RESOR=AP(J)*ET(J)-AN(J)*ET(J+1)-AS(J)*ET(J-1)-SU(J)
RESORET=RESORET+ABS(RESOR)
AP(J)=AP(J)/URFET
SU(J)=SU(J)+(1.-URFET)*AP(J)*ET(J)
ENDDO
C --- SOLUTION OF TRIADIAGONAL SYSTEM OF DISCRETISED EQUATIONS
CALL SOLVER(ET,2,NJ-1,NJ)
ET(1)=ET(2)
ET(NJ)=ET(NJ-1)
RETURN
END
C====================OTHER PROPERTIES SUBROUTINE=====================
SUBROUTINE PROPS
C -- CALCULATION OF EFFECTIVE DYNAMIC VISCOSITY, VIS ----------------------
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
TINY=10.**(-10)
DO J=NW+1,NJ-1

112
Appendix B

DUDY=(U(J+1)-U(J-1))/(Y(J+1)-Y(J-1))
STR(J)=TE(J)*SQRT(DUDY*DUDY)/MAX((ED(J)-EDNW(J)),TINY)
ENDDO
DO J=NW+1,NJ-1
TRE(J)=DENSIT*TE(J)*TE(J)/(VISCOS*ED(J)+TINY)
CMU=0.09
FMU=EXP(-3.4/(1.+0.02*TRE(J))**2)
VIS(J)=VISCOS+DENSIT*CMU*FMU*TE(J)*TE(J)/(ED(J)+TINY)
ENDDO
VIS(NW)=VISCOS
VIS(NJ)=VIS(NJ-1)
RETURN
END
C-------========================TDMA SOLVER SUBROUTINE=================
SUBROUTINE SOLVER(PHI,IST,IFN,IT)
C--- TRI-DIAGONAL MATRIX SOLVER -----------------------------------------
PARAMETER (JP=460)
IMPLICIT DOUBLE PRECISION (A-H, O-Z)
COMMON U(JP),TE(JP),ED(JP),PED(JP),VIS(JP),Y(JP),YC(JP),
1 YPLS(JP),UPLS(JP),AP(JP),AN(JP),AS(JP),SU(JP),SP(JP),
2 DU(JP),ETRM(JP),EDNW(JP),TRE(JP),AREA(JP),T(JP),TPLS(JP),
3 DYN(JP),PK(JP),STR(JP),ET(JP),T2(JP),EETTRM(JP),ETNW(JP),
4 PT2(JP),VT(JP),UV(JP),Y2PLS(JP)
COMMON URFU,URFP,URFTE,URFED,URFPED,SGMTE,SGMED,SGMPED,CE2,
1 CPE1,CPE2,CE1,CMU,RESORU,RESORT,RESORED,RESORTE,RZER,
2 FLOW1,SNORU,SNORTE,SNORED,SNORT,PPD,YLARGE,GREAT,TINY,
3 DENSIT,DENSITS,VISCOS,PR,RKS,RKF,CPF,CPS,QW,PRT,URFT,
4 FET,CET1,CET2,CET3,CET4,URFT2,URFET,SGMET,SGMT2,ARK,
5 SNORET,SNORT2,RESORET,RESORT2,DTDY2,DUDY2,HWALL,ALPHF,
6 ALPHS,ALPHR,HPLS,H2PLS,RKR
COMMON NSW,NITER,MAXIT,NJ,NW,RE1,RE2,UAVE,USUM,TB,UB,TSUM
DIMENSION PHI(IT),RP(IT),RQ(IT)
RP(IST)=AN(IST)/AP(IST)
RQ(IST)=(AS(IST)*PHI(IST-1)+SU(IST))/AP(IST)
DO J=IST+1,IFN
RP(J)= (AN(J)/AP(J))/(1.-AS(J)*RP(J-1)/AP(J))
RQ(J)=((AS(J)*RQ(J-1)/AP(J))+SU(J)/AP(J))/(1.-AS(J)*RP(J-1)/AP(J))
ENDDO
DO J=IST,IFN
JJ=IFN+IST-J
PHI(JJ)=PHI(JJ+1)*RP(JJ)+RQ(JJ)
ENDDO
RETURN
END
C==========================THE END===============================

113
Appendix C

Appendix C
Nomenclature of the Code:
ALPHF molecular diffusivity of fluid, α f
ALPHS molecular diffusivity of the wall, α w
ALPHR ratio of molecular diffusivities, α f / α w
AN coefficient of Cell N
AP coefficient of Cell P
AREA Area of the cell
ARK Thermal Activity Ratio, K
AS Coefficient of Cell S
CE1 A coefficient, Cε 1
CE2 A coefficient, Cε 2
CET1 A coefficient, Cεθ1
CET2 A coefficient, Cεθ2
CET3 A coefficient, Cεθ3
CET4 A coefficient, Cεθ4
CMU A coefficient, C µ
CPF Specific heat capacity of fluid, C P f
CPS Specific heat capacity of wall, C p w
d 2U
D2UDY2
dy 2
dε~
DEDY
dy

DELT Temperature difference, ∆T


DENSIT Density of fluid, ρ
dK
DKDY
dy

DTDX
dx
DU ∆U
2
 dU 
DUDY2  
 dy 
DYN Mesh size in the fluid region, ∆y f
DYW Mesh size in the wall region, ∆y w
ED Dissipation of kinetic energy, ε

114
Appendix C

EDIN Initial dissipation of kinetic energy, ε in


EDNW Wall-value of dissipation of turbulent kinetic energy, ε̂
EETTRM E-term in ε θ -equation, Eεθ
ETNW Wall-value of dissipation of temperature variance, ε̂ θ
ET Dissipation of temperature variance, ε θ
ETRM E-term in ε -equation, Eε~
EXY Mesh expansion factor in the fluid region, γ f
EXYW Mesh expansion factor in the wall region, γ w
FET Damping function in ε θ -equation, f εθ
FLOW1 Mass flow rate, m&
GREAT Largest possible number
H2PLS dimensionless wall thickness by ratio of molecular diffusivities, d + +
HPLS dimensionless wall thickness, d +
HWALL wall thickness
MAXIT Maximum number of iterations
NITER Number of Iterations
NJ Number of nodes in the Fluid section
NSW Number of Sweeps (in TDMA solver)
NU Nusselt number, Nu
NW Number of nodes in the Wall
PK Production of kinetic energy, Pk
PPD Pressure correction term
PR Prandtl Number, Pr
PRT Turbulent Prandtl Number, σ τ
PT2 Production of temperature variance, Pθ 2
QW Wall heat flux rate, q w
RE Reynolds number
REL Turbulent Length-Scale, l m
RESORU Residual of velocity
RESORT Residual of temperature
RESORED Residual of velocity
RESORTE Residual of turbulent kinetic energy
RESORT2 Residual of temperature variance
RESORET Residual of dissipation of temperature variance
RETAW Frictional Reynolds number, Reτ
RKF Conductivity of fluid, λ f
RKR Ratio of conductivities, λ f / λw
RKS Conductivity of Wall, λw
RZER Zero
SGMTE A coefficient, σ K
SGMED A coefficient, σ ε

115
Appendix C

SGMPED A coefficient, σ Pε
SGMT2 A coefficient, σ θ 2
SGMET A coefficient, σ εθ
SNORU Normalised residual of velocity
SNORT Normalised residual of temperature
SNORTE Normalised residual of turbulent kinetic energy
SNORED Normalised residual of dissipation of kinetic energy
SNORT2 Normalised residual of temperature variance
SNORET Normalised residual of turbulent dissipation of kinetic energy
SP Source Term
SU Source Term
T Temperature, Θ
TAW Wall shear stress, τ w
T2 Temperature variance, θ 2
TB Bulk temperature, Θ B
TE Turbulent kinetic energy, K
TEIN Initial turbulent kinetic energy, K in
TINY Smallest possible number
TRE Turbulent Reynolds number, Reτ
TTAW Friction temperature, Θτ
U Velocity
UB Bulk velocity, U B
UIN Initial velocity, U in
UPLS Dimensionless velocity, U +
URFU Under-Relaxation Factor for velocity
URFT Under-Relaxation Factor for temperature
URFP Under-Relaxation Factor for pressure
URFTE Under-Relaxation Factor for turbulent Kinetic Energy
URFED Under-Relaxation Factor for Dissipation
URFT2 Under-Relaxation Factor for temperature variance
URFET Under-Relaxation Factor for dissipation of temperature variance
UTAW Friction velocity, U τ
UV Shear stress, uv
VEFF Effective dynamic viscosity
VIS Turbulent viscosity, µτ
VISCOS Dynamic viscosity, µ
VISN Dynamic viscosity of the north face, µ n
VISS Dynamic viscosity of the south face, µ s
VT Wall normal heat flux, vθ
Y Normal distance from the wall
Y2PLS Dimensionless normal distance by ratio of molecular diffusivities, Y ++
YLARGE Height of the half channel, H

116
Appendix C

YPLS Dimensionless normal distance, Y +

117
Appendix D

Appendix D
Derivation of Equations used in Section 4.6.2:
From boundary conditions at the solid-fluid interface:
 ∂Θ +   ∂Θ + 
 + +  = K  +  Eqn. D-1
 ∂y w  ∂y f

By using the definition of y + + and K , Eqn. D-1 becomes:

 ∂Θ +  1 λ f α w  ∂Θ + 
 +  =   Eqn. D-2
 ∂y w αf λw α f  ∂y +  f
αw

 ∂Θ +   ∂Θ +  λ
  +  =  +  f Eqn. D-3
 ∂y  w  ∂y  f λw

 ∂Θ   ∂Θ  λ f
   =   Eqn. D-4
 ∂y  w  ∂y  f λw

 ∂θ   ∂θ  λ
   =   f Eqn. D-5
 ∂y  w  ∂y  f λw
2
 ∂θ ∂θ   ∂θ ∂θ   λ f 
   =
 
  
 Eqn. D-6
 ∂y ∂y  w  ∂y ∂y  f  λw 
On the other had, by definition, the wall-value of dissipation rate of temperature
variance in the fluid region is given by:

( )
2
 ∂ θ 2 1/ 2 
εˆ θ f =αf   Eqn. D-7
 ∂y 
 f
while for the solid region, it is given as:

( )
2
 ∂ θ 2 1/ 2 
εˆ θ w = αw  Eqn. D-8
 ∂y 
 w
2
 ∂θ 
= αw
1
 εˆ θ w (2θ ) Eqn. D-9

2θ
2
1
( )
/ 2
∂y 
w

118
Appendix D

by substituting Eqn. D-5 into Eqn. D-9, one gets:


2
 
εˆ θ w = αw
1
(2θ ) λ f ∂θ   Eqn. D-10

2θ ( )
2
1/ 2
 λ w ∂y  f  w
2
 λf 
2
 ∂θ 
 εˆ θ w = α w    1
(2θ ) Eqn. D-11
 λw  
2θ( )
2
1 / 2
∂y 
f
2
α λf 
2
 ∂θ 
 α f 
1
 εˆ θ w = w 

(2θ ) Eqn. D-12
αf  λw  2
( )
1/ 2
∂y 
2θ f
2
α  λf 
 εˆ θ w = w   εˆ Eqn. D-13
α f  λ w  θ f

 εˆ θ w = K 2 εˆ θ f Eqn. D-14

On the other hand, the total dissipation rate of temperature variance is found through:
 ∂θ ∂θ 
ε θ = α  
 Eqn. D-15
 ∂y ∂y 
 ∂θ ∂θ 
 ε θ w = α w  
 Eqn. D-16
 ∂y ∂y  w
The transport equation of temperature variance inside the wall is given by:
 Dρθ 2   λ  ∂θ 2 
  = ∂    − 2 ρε θ w Eqn. D-17
 Dt 
  w ∂x j  c p  ∂x j  w
Substituting Eqn. D-16 into the above equations yields:
 Dρθ 2    2   
  = ∂  λ  ∂θ  − 2 ρα w  ∂θ ∂θ  Eqn. D-18
 Dt     ∂y ∂y 
  w ∂y  c p  ∂y  w  w

 Dρθ 2    ∂θ 2   λf 
2
 ∂θ ∂θ 
  = ∂  λ   − 2 ρα w     Eqn. D-19
 Dt    ∂y  ∂y ∂y 
  w ∂y  c p   w  λw   f

 Dρθ 2    ∂θ 2  α  λf 
2

  = ∂  λ  − 2 ρ w   ε θ
 ∂y 
Eqn. D-20
 Dt   α F  λw 
f
  w ∂y  c p   w

 Dρθ 2    ∂θ 2 
  = ∂  λ 
 ∂y 
− 2 ρK 2 ε θ Eqn. D-21
 Dt   f
  w ∂y  c p  w

119
Appendix D

 Dρθ 2    ∂θ 2 
  = ∂  λ  − 2 ρK 2 ( ε~θ + εˆ θ ) f
 ∂y 
Eqn. D-22
 Dt  
  w ∂y  c p  w
finally since:
1
εˆ θ f = εˆ θ w Eqn. D-23
K2
the transport equation of temperature variance becomes:
 Dρθ 2    ∂θ 2 
  = ∂  λ  − 2 ρK 2ε~θ f − 2 ρεˆ θ w
 ∂y 
Eqn. D-24
 Dt  
  w ∂y  c p  w
On the other hand, the homogenous dissipation rate of temperature variance was defined
as:
ε~θ = ε θ − εˆ θ Eqn. D-25

Inside the wall, this becomes:


ε~ = ε − εˆ
θw θw θw Eqn. D-26

From earlier definitions, one can show:

( )
2
 ∂θ ∂θ   ∂ θ 2 1/ 2 
ε~θ w = α w   −αw  Eqn. D-27
∂ y ∂ y   ∂y 
 w  w
Also since:

( ) ( )
2 2
 ∂ θ 2 1/ 2  2
 1/ 2

 =  α w   ∂ θ
2
  Eqn. D-28
 ∂y   α   ∂y 
 f  f   w
Eqn. D-27 becomes:
 
( )
2
2
 1/ 2

 λf   ∂θ ∂θ  −  ∂ θ
2
ε~ = α w     Eqn. D-29
θw
 λw   ∂y ∂ y    
   f  ∂y f 
 
( )
2
2
 1/ 2

α  λf  α  ∂θ ∂θ  − α  ∂ θ
2
 ε~ = w     Eqn. D-30
θw
α f  λw   f  ∂ y ∂y  f 
∂y  
  f  f 
2
αw  λ f 
 ε~ θw =   (ε θ f − εˆ θ f ) Eqn. D-31
α f  λw 

 ε~θ w = K 2 ε~θ f Eqn. D-32

120
Appendix D

The transport equation of dissipation of temperature variance is given by:

Dρε~θ w ∂  λ  ∂ε~θ w  ε~ 2
=    − Cεθ2 ρ θ w Eqn. D-33
Dt ∂y  c p  ∂y 
w θ2
 
By substituting Eqn. D-32 into the above equation, one can finally arrive at the
following equation:

Dρε~θ f ∂  λ  ∂ε θ f 
~ ε~ 2
2 θ f
=   − Cε 2 ρK
θ
Eqn. D-34
Dt ∂y  c p  ∂y  θ2
 w

121
Appendix E

Appendix E
TDMA Solver:

Generally, the discretisation of fluid flow equations, such as the governing equations of
the present flow problem, often leads to systems that are either tri-diagonal or block tri-
diagonal. A tri-diagonal system is defined as one in which the only non-zero terms in its
coefficient matrix are those along the three leading diagonal lines. This is shown as
(Versteeg & Malalasekera, 1998):

 + B1φ1 − C1φ2   D1 
− A φ + B φ − C φ  D 
 21 2 2 2 3   2
 − A3φ2 + B3φ3 − C3φ4   D3 
   
  =  D4 
 . . . .   . 
   
 − An−1φn−2 + Bn−1φn−1 − Cn−1φn  Dn−1 
 − Anφn + Bnφn+1   Dn 

Alternatively, it can be shown as:

 B1 − C1 0 0 0 0 0   φ1   D1 
− A B2 − C2 0 0 0 0  φ   D 
 2   2   2 
 0 − A3 B3 − C3 0 0 0   φ3   D3 
     
 0 0 − A4 B4 − C4 0 0  ×  φ4  =  D4 
 . . . . . . .   .   . 
     
 0 0 0 0 − An −1 B n −1 − C n −1  φ n −1   D n −1 
 0 0 0 0 0 − An B n   φ n   D n 

In the above set of equations, A j , B j and C j correspond to aS , aP and a N respectively

and φ j is the variable which is to be solved. Thus, φ1 and φn are known from boundary

conditions.

the above matrix can be re-written as:

122
Appendix E

C2 A D
φ2 = φ3 + 2 φ1 + 2 Eqn. E-1a
B2 B2 B2

C3 A D
φ3 = φ4 + 3 φ2 + 3 Eqn. E-1b
B3 B3 B3

C4 A D
φ4 = φ 5 + 4 φ3 + 4 Eqn. E-1c
B4 B4 B4

.........................................

Cn A D
φn = φn+1 + n φn −1 + n Eqn. E-1n
Bn Bn Bn

the above equations are then solved in two steps:

a) forward elimination
b) backward substitution

The forward elimination process begins by eliminating φ2 from Eqn. E-1b by


substitution from Eqn. E-1a, which would give:

    A2 D  
   A3  φ1 + 2  + D3 
 B2 
φ3 = 
C3 φ +   B2 
Eqn. E-2
 C2  4  C2 
B
 3 − A  B − A 
B2   
3 3 3
 B2
 

By adopting the following notations:

C2
β2 = Eqn. E-3
B2

A2 D
γ2 = φ1 + 2 Eqn. E-4
B2 B2

Eqn. E-2 becomes:

123
Appendix E

 C3   A γ + D3 
φ3 =  φ4 +  3 2  Eqn. E-5
 B3 − A3 β 2   B3 − A3 β 2 
142
4 43
4 142
4 43
4
β3 γ3

the above equation becomes:

φ3 = β 3φ4 + γ 3 Eqn. E-6

where:

 C 
β 3 =  3
 Eqn. E-7
B
 3 − A 3β2 

 A γ + D3 
γ 3 =  3 2  Eqn. E-8
 B3 − A3 β 2 

Now, Eqn. E-6 can be used to eliminate φ3 from Eqn. E-1c.

The same procedure can be used up to the last equation of the set. This forms the basis
of forward elimination.

For backward substitution, the general form of Eqn. E-6 is used:

φ j = β jφ j + 1 + γ j Eqn. E-9

where:

 C 
β j =  j


Eqn. E-10
B
 j − A β
j j −1 

 A γ + Dj 
γ j =  j j − 1 

Eqn. E-11
 B j − Aj β j −1 

The above equations can be used to apply at the boundary points, i.e. j=1 and j=n, by
assuming:

β1 = 0 and γ 1 = φ1

124
Appendix E

β n = 0 and γ n = φn

Therefore, in order to solve a system of equations, first they need to be arranged in the
general form of:

− Aiφi −1 + Biφi − C iφi +1 = Di Eqn. E-12

Once the values A j , B j , C j and D j were identified from the governing equations, the

values of β j and γ j are calculated, by starting from j=2 and going up to j=n, using Eqn.

E-10 and E-11. Then, since the values at boundaries are known, the values of φ j can be

obtained in reverse order by means of backward substitution.

125

You might also like