Upload
others
View
8
Download
0
Embed Size (px)
Citation preview
A NUMERICAL STUDY OF BUOYANT TURBULENT
FLOWS USING LOW-REYNOLDS NUMBER k-e MODEL
by
BUNG RYEOL SEO. B.E., M.E.
A DISSERTATION
IN
MECHANICAL ENGINEERING
Submitted to the Graduate Faculty of Texas Tech University m
Partial Fulfillment of the Requirements for
the Degree of
DOCTOR OF PHILOSOPHY
Approved
May, 2001
ACKNOWLEDGEMENTS
Most of all, I would like to thank my graduate advisor. Dr. S. Parameswaran for his
guidance, persistent support, priceless advice, patience, and encouragement for this
research and my life. My thanks are extended to my dissertation committee members. Dr.
Timothy T. Maxwell, Dr. J. Walter Oler, Dr. Darryl L. James, and Dr. Zhimin Zhang, for
their constructive criticism and valuable suggestions and comments. 1 would like to-
extend my thanks to Dr. Wijesuriya P. Dayawansa and Dr. Karlene A. Hoo, who gave
valuable recommendations for this dissertation.
Sincere gratitude is also expressed to my friends M the CFDLAB, Sivapalan
Senthooran, Suthaharan Nagendran, Saman Jayantha, Kumar Aravamudhan and K
Prabaharan, for the special time we had.
I especially thank my wife, Soo Young Ahn, for her enduring love and support, and I
would like to dedicate this work to my son and daughter, Andrew and Hannah.
This research was supported by the U.S. Department of energy (DOE), Cooperative
Aggreement No. DE-FC04-5AL85832, conducted through the Amarillo National
Resource Center for Plutonium.
11
TABLE OF CONTENTS
ABSTRACT
LIST OF TABLES
LIST OF FIGURES
NOMENCLATURE
CHAPTER
I. INTRODUCTION
1.1 Problem Statement
VI
Vlll
IX
Xlll
1
1
1.2 Objectives 2
1.3 Contents of Dissertation 3
II. BUOYANT FLOW 4
2.1 Introduction 4
2.2 Literature Review 4
2.3 Buoyant, Laminar Boundary Layer Flow g
2.4 Buoyant, Turbulent Flow \ j
III. MATHE^lATICAL FORMULATION 13
3.1 Governing Equation for Momentum Transport 13
3.2 Standard k-s Model 16
3.3 Wall Functions Ig
3.4 Low-Reynolds Number k-s Model 20
3.5 Governing Equation for Energy Transport 22
3.6 Turbulent Prandtl Number 23
IV. METHOD OF COMPUTATION 27
111
4.1 Governing Equations in a General Coordinate System 27
4.2 Discretization 29
4.3 Pressure Correction 29
4.4 QUICK Scheme 31
4.5 Normal Distance Calculation 33
4.6 Solution Procedure 33
V. TEST CASE 1 : TURBULENT CHANNEL FLOW 37
5.1 Problem Statement and Boundary Conditions 37
5.2 Grid arrangement and Grid Independent Study 38
5.3 Results without Buoyant Effects 39
5.3.1 U Velocity Profile 39
5.3.2 Turbulent Kinetic Energy Profile 40
5.3.3 Turbulent Kinetic Energy Dissipation Rate Profile 41
5.4 Results with Buoyant Effects 42
5.4.1 U Velocity Profile 42
5.4.2 Temperature Profile 43
5.4.3 Turbulent Kinetic Energy Profile 43
5.4.4 Turbulent Kinetic Energy Rate Profile 43
VI. TEST CASE 2 : BACKWARD-FACING STEP 62
6.1 Problem Statement and Boundary Conditions 62
6.2 Grid Arrangement and Grid Independent Study 63
6.3 Influence oflnlet Boundary Conditions 65
6.4 Influence of Discretization Schemes for the Convective Terms 66
IV
6.5 Results without Buoyant Effects 66
6.5.1 U Velocity Profile 67
6.5.2 V Velocity Profile 67
6.5.3 Reattachment Length 68
6.5.4 Skin Friction Coefficient Comparison 68
6.5.5 Turbulent Kinetic Energy Profile 70
6.5.6 Turbulent Kinetic Energy Dissipation Rate Profile 70
6.5.7 Turbulent Prandtl Number Profile 70
6.5.8 Stanton Number Comparison 72
6.6 Results with Buoyant Effects 73
6.6.1 Reattachment Length 73
6.6.2 Turbulent Scales 74
6.6.3 Skin Friction Coefficient Comparison 76
6.6.4 Temperature Contour Comparison 77
6.6.5 Stanton Nimiber Comparison 77
6.6.6 Vortex Shedding 79
VII. CONCLUSIONS AND RECOMMENDATIONS 116
7.1 Conclusions 116
7.2 Recommendations 118
REFERENCES 119
APPENDIX 127
ABSTRACT
Numerical computation has been performed to determine the influence of buoyant
eflects on convective flows with the standard k-e and the low-Reynolds number k-s
models. The present study was motivated by the need to overcome the shortcoming of the
standard k-e model in separating and reattaching flows because of the wall-function
approach employed in the model.
The low-Reynolds number k-e model is considered to be an appropriate model for
recirculating flows because the model does not employ wall functions. Results of the two
different models are compared against the available experimental data and direct
numerical simulation (DNS) data. In this work, Kolmogorov velocity, u^={ve)^'*, is
introduced instead of shear velocity, u^=^T,„/p, to avoid the singularity that appears at
the separating and reattaching point for both thurbulence models. QUICK differencing
scheme is employed for the convective terms. Eddy-diffusivity concept is used in
modeling buoyant term. Momentum equation for the velocity field and the energy
equation for the temperature field are solved altematively because of the strong coupling
that exists between temperature and the velocity fields in a buoyant flow. Turbulent
Prandtl nimiber was allowed to vary in the low-Reynolds number k-e model to mimic the
experimental data.
Buoyant effects have been investigated with various Richardson numbers for the
backward-facing step flow. The reattachment lengths are calculated, and the skin friction
coefficient and the Stanton numbers are examined. Various separation patterns as well as
vortex shedding were observed beyond a critical Richardson number. Strouhal numbers
VI
based on the frequencies of the vortex shedding are also calculated. In addition, the
required grid configration for accurate results for low-Reynolds number k-e model has
been discussed for the backward-facing step flows. Numerical solution obtained with the
standard k-e model showed that the buoyancy-induced vortex shedding enhanced heat
transfer.
vu
LIST OF TABLES
3.1 Constants for the k-e Group of Models 25
3.2 Functions of the k-e Group of Models 26
6.1 Svmimary of the Reattachment Lengths for Various Cases 68
Vlll
LIST OF FIGURES
3.1 Time History of Turbulent Instaneous Velocity Component 14
4.1 Control Volume for Node P and Neighboring Control Volumes 27
4.2 Two possible cases of normal distance calculation 35
4.3 Normal Distance Contour for complex geometries 35
4.4 Flow Chart for Overall Procedure 36
5.1 Computational Domain and Boundary Conditions for Channel Flow 44
5.2 Grid Formation for Channel Flow(52x42 Grids with r^ =1.2) 45
5.3 Grid Independent Solution for U Profile Using Standard k-e Model 46
5.4 Grid Independent Solution for Nondimensional U Profile Using Standard k-e Model 47
5.5 Grid Independent Solution for Nondimensional k Profile Using Standard k-e Model 48
5.6 Grid Independent Solution for Nondimensional e Profile Using Standard k-e Model 49
5.7 Grid Independent Solution for U Profile Using Low-Reynolds Number k-e Model 50
5.8 Grid Independent Solution for Nondimensional U Profile Using Low-Reynolds Number k-e Model 51
5.9 Grid Independent Solution for Nondimensional k Profile Using Low-Reynolds Number k-e Model 52
5.10 Grid Independent Solution for Nondimensional e Profile Using Low-Reynolds Number k-e Model 53
5.11 Comparison of the Fully Developed U Profiles 54
5.12 Comparison of the Fully Developed Nondimensional k Profiles 5 5
IX
5.13 Comparison of the Fully Developed Nondimensional e Profiles 56
5.14 Effect of Buoyancy on U profiles 57
5.15 Effect of Buoyancy on Nondimensional U profiles 58
5.16 Effect of Buoyancy on Dimensionless Temperature 59
5.17 Effect of Buoyancy on Nondimensional k Profiles 60
5.18 Effect of Buoyancy on Nondimensional e Profiles 61
6.1 Problem Description and Boundary Conditions for Turbulent Flow 82
6.2. Grid Configuration for the Turbulent Flow
(LKE Model, 122x62 Grid) 83
6.3. Partial View of the Grid Configuration near Step End 83
6.4 Grid Independent Solution for U velocity at x/H=5.5 Using Low-Reynolds Number k-e Model 84
6.5 Grid Independent Solution for V velocity at x/H=5.5
Using Low-Reynolds Number k-e Model 85
6.6 Profiles for Inlet Boundary Conditions 86
6.7 Influence of the Inlet Velocity Profile on Reattachment Length 87
6.8 Influence of the Schemes for Discretization of Convective Term
on Reattachment Length 88
6.9 U Profiles for Reference Conditions 89
6.10V Profiles for Reference Conditions 90
6.11 Comparison of Model Results for Skin Friction Coefficient to Vogel's Experiment (8 / / / = 1.1) 91
6.12 Mean Velocity Profiles in Recirculating Region(Re=38000, ER=1.67) 92
6.13 y+ variations along the south wall 93
6.14k Profiles for Reference Conditions 94
6.15 e Profiles for Reference Conditions 95
6.16 Contour for Turbulent Prandtl Number Using Eq. (3.28) 96
6.17 Range of Turbulent Prandtl Number Using Eq. (3.28) 96
6.18 Turbulent Prandtl Number Profiles Variation 97
6.19. Comparison of Stanton Number to Experimental
Results with Various Inlet Boundary Layer Thickness 98
6.20. Particle Trace for Mixed Turbulent Convection Using standard k-e Model 99
6.21. Particle Trace for a Mixed Turbulent Convection Using Low-Reynolds number k-e Model 100
6.22. Influence of the Hichardson Number on Reattachment Length 101
6.23. Contour Plot for the Kolmogorov Length Scale (Ri„ =0.1,10 contour between 4.62x 10" and 6.89x 10' ) 102
6.24. Contour Plot for the Kolmogorov Time Scale {Ri„=0.\, 10 contour between 5.97x10"^ and 1.22x10"°) 102
6.25. Influence of the Richardson Number on Skin Friction Coefficient 103
6.26. Buoyancy-affected Temperature Contour for Recirculating Region
Obtained by Standard k-e Model 104
6.27. Buoyancy-affected Temperature Contour for Recirculating Region
Obtained by Low-Reynolds Number k-e Model 105
6.28. Fig. 6. 28 Temperature Contour Plots at Exit Boundary for Two Turbulent Models: Standard k-e Model(Ri=0); (b) Low-Reynolds Number k-e Model(Ri=0); (c) Standard k-e Model(Ri=0.1); (d) Low-Reynolds Number k-e Model(Ri=0.1) 106
6.29. Influence of the Richardson Number on Stanton Number 107
6.30. Effects of Turbulent Prandtl Number on Stanton Number 108
XI
6.31. Effects of Richardson Number on Wall Normal Distance
for Low-Reynolds Number k-e Model 109
6.32. Vortex Shedding af er the Backward-Facing Step with Ri„ =0.25 using Standard k-e Model 110
6.33. Reattachment Variation of the Flow with Ri„ =0.25 along the time 111
6.34. Vortex Interaction Between Two Bubbles after the Backward-Facing Step with Riff =0.20 using Low-Reynolds Nimiber k-e Model 112
6.35. Effect of Richardson Numberon Strouhal Number of \ onex Shedding Using Standard k-e Model 113
6.36. Stanton Number Change during Vortex Shedding Using Standard k-e Model 114
6.37. Cf Change during Vortex Shedding Using Standard k-e Model 115
xu
NOMENCLATURE
English Letter Symbol
C,, C,, Cy, C , model constants of k-e model
Cf mean skin friction coefficient, T^J{pull2)
f frequency of the vortex shedding,//z
Gjj buoyancy production in k and e equation s, Nl{s-m^)
Gfff Grashof number based on step height, gfiH^{T„ -TQ)/V'
H height of step, m
g^ acceleration due to gravity, m/s^
NUfj local Nusselt number, iS'r/(Re„ Pr)
/» production ofthe kinetic energy, Nl{s-m^)
p mean static pressure, NIrn^
Pr laminar Prandtl number
Pr, turbulent Prandtll number
Re, turbulent Reynolds number, k^ I ve
Re„ Reynolds number based oh step height
Riff Richardson number, Grfj /Re^
Shff Strouhal number,/?//«
St Standton number, q„ / pC,,u„ {T„ - T„)
T temperature, K
xiii
T fluctuation of temperature, K
t time, s
M,v mean velocity in X-and y-directions, respectively, «7/.y
" y fluctuation ofthe velocity in x- and y-direction, respectively, mis
I'r friction velocity, ^ | r„ , | /p , mis
X Cartesian Coordinate in streamwise direction, m
A,. flow reattachment length, m
y dimensionless length in y direction
y cartesian coordinate in normal to streamwise direction, m
y* nondimensional distance from wall surface, u^ylv
Greek symbols
a,a, molecular and eddy diffusivity for heat, m^ Is
Tf, r grid expansion ratio in streamwise and normal to streamwise direction
P volumetric coefficient of thermal expansion, AT''
5 boundary layer thickness, m
K turbulent energy, u^ujl, m^ Is^
e turbulence dissipation rate, Nl{s-m^)
e nondimensional fluid temperature, (7;, - T) l{T^ - T^)
V,V, kinematic and eddy viscosity, respectively, nlp, m' Is
p fluid density, kglrn^
XIV
cr,. ,<Tj model constants in turbulent diffusion terms of k-e model
r„, shear stress evaluated at the wall, NIm'
Subscripts
0 reference values at inlet position
c center of inlet position in y direction
P cell center ofthe control volume
p 'lie point where wall functions are applied to
w wall surface
^ streamwise direction in computational domain
Tj normal direction to s t reamwise direction in computat ional doma in
XV
CHAPTER I
INTRODUCTION
1.1 Problem Statement
The standard k-e model has been widely used for its simplicity and economic
computational cost to account for rotation, buoyancy, chemical reaction, etc. The-
standard k-e model has played an important role in interpreting simple turbulent flows,
such as, duct flow or pipe flow. However, this model shows a weakness for recirculating
flow. It is a well-known problem that the standard k-e model underpredicts the
reattachment length in recirculating flow. In addition, one of the most difficult tasks for
researchers who used the standard k-e model for their studies was to find proper wall
functions for complicated turbulent flow. Conventional wall functions assumed turbulent
flow to be fully developed while separating and reattaching flow does not satisfy the
description of fully developed flow.
In the wall fiinction method, the boimdary conditions for the velocity and
temperature are matched with the logarithmic profile that exists near a wall in a forced
convection boundary layer flow. It has been known for some time that the logarithmic
profile does not hold near a wall for a pure natural convection boundary layer. George
and Capp (1979) have developed a wall function for natural convection turbulent flow
using analytical techniques. Henkes and Hoogendoom (1990), however, found that these
wall functions did not show sufficient agreement with experimental results found by
other researches such as Tsuji and Nagano (1988), while the low-Reynolds number k-s
models by Jones and Launder (1972), Chien (1982) and Lam and Bremhorst (1981)
showed fairly good agreement with other's experimental results. In addition, after
studying wall functions numerically. To and Humphrey (1986) suggest that turbulent
natural convection calculations should be carried out up to the wall, including the viscous
sublayer, with a low-Reynolds number k-e model.
In order to overcome the aforementioned problems of the standard k-s model, the
low-Reynolds number k-e model is investigated for buoyant separating flow in this study.'
It is impossible to predict the temperature field properly without accurate flow field
prediction. Two turbulent models will be used to predict buoyant separating flow and the
results will be compared against available experimental data.
For the standard k-e model, a constant turbulent number is used throughout the
calculation field because of the assumption of a high turbulent Reynolds number.
Re, -k- lv£. The low-Reynolds number k-s model, however, needs varying turbulent
Prandtl number to account for viscous sublayer effects. Finally, a varying turbulent
Prandtl number based on empirical correlations proposed by Kays and Crawford (1993)
will be introduced.
1.2 Objectives
The overall objective of this study is to employ a better turbulent model to improve
the prediction of separating and reattaching buoyant turbulent flows. The specific
objectives of this study are:
1. To develop a computational code that can predict accurate resuhs in the presence
of separating and reattaching flow.
2. To examine buoyant effects on the mixed flow.
3. To compare the low-Reynolds number k-e model with the standard k-e model for
channel flow and backward-facing step flow.
4. To improve temperature field predictions by introducing varying turbulent Prandtl
number based on empirical correlations.
1.3 Contents of Dissertation
The following chapter contains an overview and literature review of buoyant flows
for laminar and turbulence flow fields. Definitions of parameters which are used in this
work to define the characteristics of the buoyant flows are introduced. In Chapter 111,
mathematical formulation for a closure problem and the descriptions for the standard k-e
model and the low-Reynolds number k-e model are covered. Computational methods
such as discretization of the governing equations, pressure correction, and higher-order
convective term discretization techniques are discussed in Chapter IV. Also in Chapter
IV, the normal distjmce calculation algorithm which needs to be found for the damping
functions of the low-Reynolds number k-e model is presented. Chapter V and VI are
devoted to the test cases of turbulent buoyant flows for a channel and a backward-facing
step. For each case, results without and with buoyant effects for two turbulent models are
compared against available experimental data. Chapter VII contains the conclusions and
recommendations for future works.
CHAPTER II
BUOYANT FLOW
2.1 Introduction
Buoyant flows are caused by gravity. In this section, the reader is introduced to the
important features of buoyant flows and the important non-dimensional numbers that
characterize buoyant flow. Buoyancy force is the driving mechanism for natural
convection. Therefore, when the effect of temperature on fluid density is taken into
account, lighter parts of a heated fluid flow upward and the heavier ones flow downward
relative to a reference state in equilibrium. The phenomena are seen in the heating of a
room from convectors or space heaters, in providing the draft in chimneys, in cooling
products in refrigerators or cooling houses, in the cooling of transistors and transformers,
as well as in human beings and animals standing in a quiescent atmosphere. The most
frequent buoyant flows are encountered in many industrial flows such as electronic
cooling, HVAC systems and material processing, large-scale fires, and wild fires.
2.2 Literature Review
Many researchers have conducted studies on buoyant flows for various geometries.
In this literature review, boundary layer flow, flow between two plates, and flow over a
backward-facing step are discussed. In addition, vortex shedding and wall-functions
related literature are discussed.
Tsuji and Nagano (1988) measured velocity and temperature profiles for air along a
vertical plate with a constant wall temperature. They found a curve-fit for the wall-shear
stress measuring velocity using a hot-wire technique. The data is considered to be one of
the best-measured and is widely used by many researchers to study the boundary layer
buoyant flow.
Yin et al. (1990) developed a model which was originally based on the low-
Reynolds number k-e model to predict turbulent natural convection boundary layers. In
the model they divided the velocity into two components, i.e., a forced convection
component and a buoyancy-influenced component. A two-equation model for energy
equation by Nagano and Kim (1988) has been used for the calculation of temperature
field. They foimd that the combination of the modified low-Reynolds number k-e and the
two-equation model for energy equation is the best way to predict natural convection.
In the mean time, the study to develop the wall functions for the buoyant flow had
been done by several researchers (George & Capp, 1979; Tsuji & Nagano, 1989; Henkes,
1990; Henkes & Hoogendoom, 1990; Yuan et al., 1993). Henkes and Hoogendoom
(1990) suggested that the calculation for natural convection should been done up to the
wall with the low-Reynolds number k-e model for turbulence. They listed two differences
between the wall functions for the forced and natural convection boundary layer. First,
the inertial sublayer is fully turbulent, whereas the turbulent viscosity is very small in the
buoyant sublayer. Second, the log-law in the inertial sublayer gives a good approximaion
until close to the outer edge, >;"' >1000, while the extension ofthe buoyant sublayer is
restricted to the velocity maximum.
Yuan et al. (1993) tried to find the wall function for natural convection to avoid a
large number of grid points in the near-wall to save the computing costs, especially for 3-
D calculations. They introduced heat flux temperature and heat flux velocity as a
temperature and velocity scale. However, they found that their wall functions still
required a very fine grid system for accuracy.
Blackwell and Armaly (1993) conducted laminar mixed convection flow over the
backward-facing step as a benchmark test for the codes used for this study. The results of
many contributors (Acharya et al., 1993; Blackwell & Pepper, 1992; Chopin, 1993;
Cochran et al.. 1993; Hong et al., 1993; Iglesias et al., 1993; Sanchez & Vradis, 1993)
indicated that Uie reattachment length had been reduced as the Grashof number was
increased. With a fixed Reynolds number of 100 and an expansion ratio of 2, the
reattachment length, after the step, was found to be 5 and 3 for Grashof numbers of 0 and
1000, resj)ectively. They found that the Nusselt number and friction coefficient had been
increased due to buoyancy effects when Gr=1000 is used. Iwai et.al. (1999) conducted
the numerical simulations for mixed convective laminar flow, Re=125, over a backward-
facing step in the duct. It was found that the locations ofthe reattachment and the peak of
Nusselt number moved upstream as the Richardson number was increased, which is
similar to the study by Blackwell and Armaly (1993).
Kishinami et al. (1998) studied the laminar mixed convection over the backward-
facing step, which can be adjusted to change its height. They found that as the step height
is increased, the position of peak Nusselt number moves downward. Further, unstable
self-excited oscillation was noticed near the reattachment point within the limiting range
ofthe Richardson number, 0.3<Ri<2.5. In the oscillating region, the unsteady fluctuating
behaviors ofthe streamline and temperature contours were observed.
The problem of laminar flow over a backward-facing step geometry in natural,
forced, and mixed convection has been investigated rather extensively in the past, both
numerically and experimentally. On the other hand, studies of turbulent flow over a
backward-facing step have dealt mainly with forced convection only. Inagaki (1995),
Abu-Mulaweh et al. (1996), and Abu-Mulaweh (1999) are the only researchers who have
reported some results on free convection over a backward-facing step.
Abu-Mulaweh et al. (1999) conducted an experiment of turbulent natural
convection over a backward-facing step whose step height can be adjusted. Temperature
difference between heated wall and ambient air was maintained at 30 °C. Using three
different step heights, the authors found that both the reattachment length and the heat'
transfer rate from the downstream heated wall increase with increasing step height.
Kwak and Song (1998) conducted an experimental and numerical study of buoyant
flow over a vertical plate with horizontal rectangular grooves. They found that if the
Rayleigh number is not large enough the heat transfer rate of the plate with grooves
would be less than that of a smooth plate. It was found that if the Rayleigh number is not
sufficiently large enough, small recirculation will block the main stream from flowing
into the bottom surface ofthe grooves.
Similar study to augment heat transfer for laminar mixed convection for flat plate
with rectangular grooves has been done numerically by Wu and Pemg (1998). They
found that the installation of an inclined plate, which creates upstream vortex shedding in
the vertical block-heated channel, could effectively augment the blocks' heat transfer
performance in the channel.
Singh et al. (1998) investigated the flow field and the temperature distribution
around a heated/cooled circular cylinder placed in an insulated vertical channel using a
finite volume method. With a fixed Reynolds number of 100, the frequencies of vortex
shedding were examined with various Richardson numbers. For Ri up to 0.1, the
shedding frequency steadily increased, but for Ri>0.15, the shedding frequency suddenly
became zero, an indication that the shedding was stopped.
Heindel et al. (1994) compared two different low-Reynolds number k-e models for
natural convection in an enclosure. They concluded that the predictions of variable
coefficients model, which vary with the value of local turbulence Reynolds number, is
better than that ofthe model with fixed coefficients.
2.3 Buoyant, Laminar Boundary Layer Flow
For laminar boundary layer flow in which gravity acts in the negative x direction,
the body force per unit volume is pg^. With gravitational body force, the steady state
momentum equation for boundary layer can be written as follows.
.dU .,dU dP d ( dU^ pU — +pV — = -— + -^—-pg^. • (2.1)
ox dy ox oy\ ay J
The pressure gradient term can be defined by the pressure outside of the boundary
layer where the hydrostatic pressure can be applied to, dPldx = -pQg^ , and the
momentum equation becomes
-EAP-PO)- (2.2)
If density is a function of temperature and pressure, density can be expanded using
pU + pV = —H p — dx dy dy\ dy ^
the Tayk r series as follows.
/ ' p ^ ^ P = Po +
dp (T-To)^^] iP-Po) + - (2.3)
In a most natural pressure correction term in right hand side is neglected compared
to temperature difference term,
Po-p = P(T-To). (2.4)
where p = -{dpldT),, denotes the coefficient of volumetric thermal expansion, p^ and
TQ are the density and temperature where convection would stop for uniform temperature.
Finally, the momentum equation for natural convection can be written as follows.
d.\ dy dy ;'|^]-^,Ar-r„). (2.5)
In natural convection, some dimensionless parameters are introduced to
characterize buoyant flow. One of the most significant nondimensional parameter is the
Grashof number which is often interpreted as the parameter describing the ratio of
buoyancy to viscous forces,
Gr^MP^lZ^. (2.6)
where / is the characteristic length. The Grashof number is independent of the velocity
and is sometimes written in the following modified form to reveal constant heat flux at
the wall.
Gr =Gr,Nu^=f^ql\ (2.7) kv
Another parameter, the Rayleigh number, can be used to account for the inherent
coupling between energy and momentum balances.
Ra = i ^ ^ = GrPr. (2.8) va
This Rayleigh number is the counterpart of natural convection to the Reynolds
number of forced convection. For a natural convection boundary layer, Incropera and
DeWitt (1990) found that transition from laminar to turbulence occurs at the position
where Ra ~ 10^, regardless ofthe value ofthe Prandtl number. This universal transition
criterion can be supported by many experimental evidences (Fujii, 1959; Humphreys &
Welty, 1975; Godaux & Gebhart, 1974).
For mixed convection, the Richardson number can be defined as accounting for the
ratio of buoyant force to the change in momentum.
_. gP6Tl Gr
U' Re^ ^ ^ ^
where Re = —2-, denotes Reynolds number and U^ , the free stream velocity. It is
known for external flow that if the Richardson number is much less than 1, buoyancy
effects can be neglected. However, the role of the Richardson number in separating flow
has not been studied by many researchers.
Combined forced and free convection heat transfer coupled with thermal
conduction and separated recirculation flow becan-;c important in recent years, both in
academic and practical fields (Kishinami et al., 1995).
In natural convection, velocity profile has peak values close to the wall while in
forced convection the velocity profile is parabolic in which the free stream velocity is the
maximum velocity. For natural convection, Eckert (1950) assumed the following shapes
for velocity and temperature profiles, with the assumption that the momentum and
thermal boundary layers are of equal thickness,
(2.10) ^.=zfi_zV U^ SI s 'J
^-^" =fi_zf (2.11)
In buoyant flow, the amount of heat transferred can be affected by the Grashof
numbers and Prandti number, Pr, of the fluid. Ede (1967) found the correlation of the
local Nusselt number as a function of Pr and Gr^,
Nu=-"" 4
2Pr
5(l + 2Pr"^ + 2Pr)
1/4 ,1/4 (Gr^Pr)"\ (2.12)
A correlating equation for laminar constant-heat-flux Nusselt number was found in
10
a similar way by Fujii and Fujii (1976) using a modified Grashof number to contain the
heat flux, q, explicitly,
Pr
4 + 9Pr"^ + 10Pr)
Kays and Crawford (1993) compared the two cases above and revealed that the
Nu,= ,1/5 (Gr Pr)"^ (2.13)
constant-heat-flux Nusselt number is about 15 percent higher in laminar free convection. '
2.4 Buoyant, Turbulent Flow
The Reynolds-averaged Navier-Stokes equation and energy equation contain two
new terms, u^Uj and u^T' which represent turbulent momentum transport and turbulent
heat transport, respectively. These two terms need to be modeled in order to have a closed
set of equations.
In direct analogy to the turbulent momentum transport, the turbulent heat transport
is assumed to be related to the gradient ofthe transport quantity, temperature,
dT -UjT = a , ^ , (2.14)
dxj
where a, is the turbulent heat diffusity. This turbulent heat diffusivity can be modeled by
the analogy between heat transport and momentum transport,
a =— (2.15) ' Pr , '
where v and Pr, are eddy diffusivity and turbulent Prandtl number, respectively. It can
be noted that these two properties are not fluid properties but instead depend on the state
of the turbulence. The turbulent kinetic energy and dissipation transport equations for
buoyant flows contain the buoyant production/destruction term, G^ , which can be written
as follows,
11
G,=-gJuJ\ (2.16)
Using Equations 2.14 and 2.16 can be arranged as follows,
G,=-g^P^^. (2.17) Pr, c\
The buoyancy production/destruction term represents the exchange between
potential energy and turbulent kinetic energy. Heindel et al. (1994) defined G^ as a sink
term if the term is negative, and defined it as a turbulent kinetic energy source for a
positive case. This distmction/production term is added to the k and s equations to
a^^count for turbulent buoyant flow. A set of turbulent transport equations and the Navier-
Stokes equation will be discussed in the following Chapter.
Turbulent buoyant velocity and temperature profiles assumed by Eckert and
Jackson (1950) are, in a similar way, used for laminar free-convection flow.
JLJy. u^ [s
^ ^- = l - f - | (2.19) Pw T^_
12
CHAPTER III
MATHEMATICAL FORMULATION
In analyzing separating flow, more complex governing equations must be solved
compared with tliose for non-separating flows. Separating flow can occur by pressure
difference without a sudden geometry change but it primarily occurs by flow geometry
change, such as, backward-facing step, forward-facing step, cavity flow, disk or cylinder
in flow stream, etc. One of main differences between separating flow and non-separating
flow is the velocity profile. Non-separating flow has a monotonic velocity profile and
provides researchers with rules which can be applied to numerical simulation while
separating flows do not. Especially, when the flow field and temperature field affect each
other, flow profile can be much more complicated than other flows.
Backward-facing step has been attractive to many researchers, and it became the
representative geometry to study separating flow. Backward-facing step can be seen in
many engineering applications, such as sudden expansion in a pipe or a channel. This
sudden expansion can make the flow separate from the wall. Backward-facing step can be
classified into three types: two-dimensional symmetric step, two-dimensional single step,
and axisymmetric aimular step. In this study, a numerical method using the standard k-e
model and the low-Reynolds number k-s model were used to study the flow over the
backward step with buoyancy effects.
3.1 Governing Equation for Momentum Transport
The characteristics of turbulence are random, diffusive, dissipative, and three-
13
dimensional. Figure 3.1 shows the time history of turbulent velocity.
Figure 3.1 Time History of Turbulent Instaneous Velocity Component
In Figure 3.1, instantaneous velocity at any pairticular time can be a summation of
the mean velocity and the fluctuating velocity:
« ' = ( / + «, (3.1)
where U is the mean velocity,
u is the fluctuating velocity.
For constant p and p, incompressible Navier-Stokes (1827 & 1845) equations can
be expressed in the following manner.
Continuity Equation,
5M
CbC; -i- = 0. (3.2)
Momentum Equations
du , du 1 dp Q —'- + U —'- = + — dt ' dx. p dxi dXf
du' du • \ • + •
dx , dx, \ + gJ{T-T,) (3.3)
where /?* = P + /? is the instantaneous pressure.
In engineering application, however, a more practical approach to describing
14
turbulent flows would be to model the averaged turbulent transport quantities. Using
short-term time Reynolds averaging over At, which is short but still much greater than
any significant period of fluctuations, dt, in velocity u, governing equations for
turbulence can be arranged as follows.
Continuity Equation,
dU. - = 0.
dx,
Momentum Equations
dU^^^dU.
(3.4)
dl ' dx, \_dP_ j _ p dXj dx:
( dU, dU,
dx, dx. \ "".I
-u,Uj\ + gJ{T-T,) (3.5)
In Equation (3.5), the double velocity correlation term, -u^u • has been added as a
result of Reynolds time averaging and is known as the Reynolds stress term or the
turbulent shear stress term. The Reynolds stress term makes the set of equations not
closed. Therefore the turbulent shear stress term has to be modeled in terms of known
quantities in order to close this equation set for turbulence.
Many models have been introduced to model the unknown Reynolds stress term
known as zero-equation model or the first-order closure model.
Boussinesq eddy viscosity model (1977) dU
Prandtl mixing length model
Von Karman mixing length model
- MV = V,
- O T = / ^
-l^ = k^
dy
dU dy
dU
dy
dU .d^U •I
dU
dy
(3.6)
(3.7)
(3.8) dy dy^
where / is a length scale characteristic of the size of momentum-transfer eddies. The
Boussinesq eddy viscosity model (EVM) is used in a generalized form.
15
u^Uj = r . - ' -S..k. ^dxj dx, ^ 3 "
(3.9)
Here, k denotes the turbulent kinetic energy and S^j denotes the Kronecker delta function.
For this EVM the eddy viscosity, v,, should be modeled. Substituting Equation (3.9), into
Equation (3.5) gives
dl ' cbc, p dx, dXj \y + v,) —^ + —-1.
I, 5jf, dx, J -^5„^^gJ{T-T,). (3.10)
Prandtl (1925) and Kolmogorov (1941) setup the relation among eddy viscosity,
turbulent kinetic energy, k, and its dissipation rate, s, by dimensional analysis as
^,=C^ (3.11)
Here, C^ is considered as constant for high Reynolds number. For a simple flow, such as
pipe flow or channel flow, the Reynolds stress term can be modeled by a simple Prandtl
mixing length model called zero-equation model. The zero-equation model can be
modeled directly without any additional differential equations. However, more general
and complex flow needs to have a more sophisticated model that includes k and s. This is
referred to as the second-order closure problem. One of the most popular turbulence
models with various combinations of the variables proposed is the k-s model by Launder
and Spalding (1974).
3.2 Standard k-s Model
As previously mentioned, kinetic energy and its dissipation rate are introduced to
calculate eddy viscosity. The standard k-s model has been applied to many different flows,
such as plane jets, mixing layer, boundary layer flows, and so forth. Despite the existence
16
of more advanced turbulent models, such as Direct Numerical Simulation (DNS), Large
Eddy Simulation (LES) or Reynolds Stress Model (RSM), the standard k-s model has
been studied by many researchers because of its easy adaptivity and smaller
computational cost. This model was developed on the assumption that the flow is fully
turbulent and there is a local equilibrium where rate of production of turbulent stress, /"^,
equals rate of turbulent dissipation rate, e , near the wall. The assumption of fially
turbulent flow requires the local turbulent Reynolds number. Re, =k^ Ive, to be high.
For this reason, the standard k-s model is sometimes called the high-Reynolds numbtx k-
e model, compared to the low-Reynolds number k-s model in which calculation is carried
out up to the wall including the viscous sublayer. It is necessary for standard k-s model to
have the first calculation point far away from the wall where the local Reynolds number
is large enough to satisfy the above assumption. Empirical wall functions are used to
bridge the gap between the wall and the first grid node adjacent to the wall. Those wall
functions will be discussed later in this chapter.
The turbulent model, including buoyancy term, originally developed by Launder
and Spalding (1974), is described by the following equations.
The k-s equation
^A,U^^ = M[v.^]f-\.P,-e.U,. (3.12) dt 'dXf dxA\ cyjdxj p
The s-equation
d£ ,, de d — + U
( v,\ds\ ^ £ ^ ^ E^ 1
dt ' dx, dx. v + -
£ ^ C,-P,-C,— + -C,jG,. (3.13) ax, I K K p k
where.
17
A=v, V ^ 7 ^^. J ^j dx,
y^dT^
Pr, dy
The empirical constants are assigned their usual values, these are
G,=-g,P-^—. Pr, dy
( C „ . C , , , C^„ a , , 0-,) = (0.09,1.44,1.92,1.0,1.3).
One more exti-a constant, C,, is necessary. Rodi(1984) suggested that Cj should
be close to 1 in the boundary layer which is in the direction of gravity and close to 0 in
the boundary layer which is in the perpendicular direction of gravity. Following Henkes
et al. (1991), an approximation that satisfied this condition is given by
Cy = tanh (3.14)
3.3 Wall Functions
The fluxes of momentum and heat to the wall are supposed to obey the wall
functions by Launder and Spalding (1974), as follows
(3.15)
(3.16)
"' C"'k';' - ^ in Eyp
= Pr,
(c;'
iln K
kr)'"' V
Eyp V
where
P = 9.0[(Pr/Pr, )°^^ -1][1 + 0.28exp(-0.007 Pr/Pr,)] .
The P in the right-hand side of Equation (3.16) has its origin in an analysis of
experimental data conducted by JayatiUaka (1969). Equations (3.15) and (3.16) are
18
modified forms of u* and T* to avoid a shear velocity, u^ = ^[rj~p , which changes its
sign in a recirculating region. However, k in Equations (3.15) and (3.16) still become
zero at the separating and reattaching points. Therefore, the Kolmogorov velocity,
Mj = (ve)"*, is introduced to avoid this shortcoming using the following local equilibrium
approximation for high turbulent Reynolds number. According to experimental data, e
has a finite non-zero value, even at the wall, while k becomes zero,
e,=[{C]rkr)lKy,\ (3.17)
Here subscript P in Equation (3.17) denotes the first calculation point where wall
functions are applied. Equations (3.15) and (3.16) can be rearranged using Equation
(3.17) and Kolmogorov velocity as follows:
"p 1/3 n *"^ 1 1 ^ -MAf Re =—In
^ K {rip),.
EK'" Re »4/3
\n {Tp-TJCppK"'Rc .1/3
= Pr, In K
EK'" Re-,4/.1
+ P
(3.18)
(3.19)
where
Re = ypu,
Using wall functions related parameters, the Nusselt number and friction
coefficient can be found
where
Nu.= hH qlH {T^-T,.)u,H
" k k{T^-TJ T;{T„-TJa
• \l2pU^ ii2u;ul
.1/3
(3.20)
(3.21)
19
H is the characteristic length,
T^ is the reference stream temperature,
( / , is the reference velocity,
u*p denotes the dimensionless velocity for the first calculation point next to the
wall,
Tp denotes the dimensionless temperature for the first calculation point next
to the wall.
3.4 Low-Reynolds Number k-e Model
The first low-Reynolds-number k-e model was proposed by Jones and Launder
(1972). Low-Reynolds-number k-e model was made for the modification of turbulence
closure models that would enable their use at low Reynolds numbers and to calculate the
flow close to the wall. This model includes damping functions based on the local
turbulent Reynolds mmiber. The attempt has been made to predict the appropriate values
ofthe eddy viscosity and the dissipation near the wall where turbulent Reynolds number
is small. Low-Reynolds-number k-e by Lam and Bremhorst (1981), Jones and Launder
(1972) and its modification by Launder and Sharma (1974) are still considered adequate
for modeling turbulent natural convection. Chien (1982) was one ofthe researchers who
improved the low-Reynolds number k-e model by applying the Taylor series expansion
technique to investigate the proper behavior of the turbulent shear stress and the kinetic
energy and its dissipation near a solid wall. Craft et al. (1993) proposed a non-linear low-
Reynolds-number model that has a nonlinear relation between strain-stress and vorticity
that includes quadratic and cubic terms. Nagano and Tagawa (1990) developed the low-
Reynolds number k-e model to satisfy the physical requirements of wall and free
20
turbulence. Abe et al. (1994) introduced Kolmogorov velocity scale, u^ = (v / / f ) ' ^^
instead of the friction velocity, u^, to account for the near-wall and low-Reynolds-
number effects in both attached and detached flows. For a flow over backward-facing
step, experimental data by Kim et al. (1987), Driver and Seegmiller (1985), and Vogel
and Eaton (1985) are frequently used to verify the models. Unlike experimental data,
DNS data are often used to develop new turbulence models because of its detailed
information of flow quantities even very close to the wall.
Chen (1998) summarized k-e group of models as shown in Tables 3.1 and 3.2. A
variety of low-Reynolds number k-e models can be written in general form
dk ,, dk d — + U, = — dt ' dx, dx.
v + -\
'kj
dk dx,
•dU, •u,u i
' ' dx. £-D, (3.22)
d£ ,, d£ d — + U dt ' dXi dXf
^
v + V, d£
\ ^e)
•dU,
STf-'^'^'i-'-^^-^^^^T^^- ("^'
where.
r uv = v.
dU, dUj +
^ dXj dx. -f' '
y, =cj„--.
Rcy — — , ve
Re^ = 4ky
u, =
y ' = ^
21
y =::—£. V
,1/4 «, = {vsY
Here, D in Table 3.1 has been introduced to compensate for the boundary condition of
turbulent kinetic energy dissipation rate, e, because physically in the Table 3.1, £•„, is not
zero.
Chien's model (hereafter referred as the CH model) was developed basically for
simple flows, such as, channel flow or pipe flow, or in other words nonseparating flow.
Wall dissipation, 2vkly^, in the k-equation is equal to the molecular diffusion at y=0. In a
similar way, another wall dissipation term has been introduced in the e-equation.
Turbulent eddy viscosity has been modified to include the damping effect due to the
presence of the wall. This damping function usually has an exponential form so that
damping effects can be faded out away from the wall.
The major difference between the Abe-Kondoh-Nagano model (hereafter referred
to as the AKN model) and CH model is the velocity scale. In AKN model, Kolmogorov
velocity scale, u^ =(v£-)"*, is used while CH model used friction velocity, «_., to account
for the near-wall and low-Reynolds effects in both attached and detached flows. Wall
dissipation term is given as a boundary condition for turbulent kinetic dissipation rate. In
this study the AKN model with the same buoyancy terms defined in the standard k-s case
has been selected to solve the problem of recirculating flow.
3.5 Governing Equation for Energy Transport
Analogous to the momentum equation, after Reynolds time averaging, the energy
equation has an extra term due to turbulence, known as the Reynolds flux,
22
-^^"'-S^'a^l''-^-"''^]- (3.24)
Reynolds heat fluxes in Equation (3.24) also need to be modeled. In Chapter II, The
modeling of Reynolds fluxes were discussed. Using Equations (2.14) and (2.15), the
energy equation can be rearranged as follows
This energy equation is strongly coupled with the momentum equation so that all
the numerical calculations for the governing equations can be performed simultaneously.
3.6 Turbulent Prandtl Number
The turbulent Prandtl number cannot be considered as a constant if calculations for
the turbulent model are carried up to the wall. According to the experimental data for
fully developed flow by Blackwell et al. (1972) and Hollingsworth et al. (1989), the
turbulent Prandtl number is sharply increased in the viscous sublayer region. In addition,
if the flow is not fully developed a more precise turbulent Prandtl number calculation will
be needed. From Equations (2.15) and (2.14), and the definition ofthe eddy viscosity.
Equation (3.6), the turbulent Prandtl number can be expressed as follows
v^^-^dTldy
a, -vT dUldy
To define the turbulent Prandtl number, four quantities in Eq, (3.26) should be
defined. Kays and Crawford (1993) showed that for fluids with 0.7 and 5.9 of laminar
Prandtl number, Pr , such as air and water, the turbulent Prandtl number in the
logarithmic region is close to 1. The experimental data by many researchers (Reynolds et
23
al., 1958; Blackwell, 1972; Gibson & Verriopoulous, 1984; Hollingsworth, 1989) show
that the turbulent Prandtl number has a range from 0.7 to 0.9, with a preponderance of
data around 0.85. For this reason, 0.85 is used as the turbulent Prandtl number for
standard k-s model. Kays and Crawford (1993) fitted the experimental data for boundary
layer into the following equation
1 Pr,=
2Pr. ' +CPe,]^-{CPe,y
700 fPr, 7oO
1-exp 1
CPe,^, /CO /
(3.27)
where Pe, = {\; /v)Pr, Pr,„ is the value of Pr, far from the wall that is usually assigned a
value of 0.85. C is an experimental constant with a value of 0.3. It .should be noted that
Equation (3.27) and all the experimental data mentioned above have been performed for
boundary layer flow, which is attached flow.
24
Table 3.1 Constants for the k-e Group of Models
Model
Standard
Launder-
Sharma
Hassid-
Poreh
Chien
Lam-
Bremhorst
Lam-
Bremhorst
Nagano-
Hishida
Nagano-
Tagawa
Abe-
Kondoh-
Nagano
Code
STD
LS 2\'
HP
CH
LB
LBl
NH 2v
NT
AKN
D
0
'd^] < 'y J
2A \'"
2A
0
0
0
0
f..,-B.C.
Wall
0
0
0
d-k
'V ^ = 0 dy
i ' 2 v —
y
2v — y
C.
0.09
0.09
0.09
0.09
0.09
0.09
0.09
0.09
0.09
f,
1.44
1.44
1.45
1.35
1.44
1.44
1.45
1.40
1.40
^2
1.92
1.92
2.00
1.80
1.92
1.92
1.90
1.30
1.40
o-A
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.45
1.50
^e
1.30
1.30
1.30
•
1.30
1.30
1.30
1.30
1.90
1.90
25
Table 3.2 Functions ofthe k-e Group of Models
Code f J n
STD l.O
LS
exp
LB
- 3 . 4
HP l-exp(-0.0015Re,)
CH l-exp(-0.0115.v')
[l-exp(-0.0165Re,,)f
L B l [l-exp(-0.0165Re,)f
20.5' 1 + -Re,
N H [l-exp{-Re,/26.5)f
NT ^ 26
AKN
4'^^] l -exp(-^)
^ 14 ( / „ N2>'\
. 5 f fRe Y
/ . fl
1.0
1.0
1.0
I - 0.3exp(-ReJ)
1.0 l-0.3exp(-Rc;)
1.0 l-0.22exp(-(Re,/6)^)
l + (0.05//^,) l-exp(Re.J.)
l+(0.05//^,) l-exp(Re^.)
1.0
1.0
1.0
l-exp(-^^) ^ 6.0
1 - 0.3exp 6.5
0.0
2vv, ^d'-U^"
dy' )
-2v I f> J
-2v{£ly') X exp(-0.5_v)
0
l-0.3exp(-Re7) '» , ( • - / „ )
l - e x p ( - ^ 3.1
l-0.3exp r / „ ^2^^
Re, ~6l
0.0
0.0
26
are
CHAPTER IV
METHODS OF COMPUTATION
A typical control volume for a node P and its neighbor node N,S,E and W
shown in Figure 4.1. Governing equations are integrated over the control volume for
nodes. A nonstaggered grid system has been used. Therefore, velocity, pressure, and'
temperature are stored at the cell center P. For the pressure correction, a modified version
of the SIMPLE algorithm by Patankar (1980) is used. The momentum equation and
energy equation are solved altematively because ofthe strong coupling between velocity
and temperature. Stone's (1968) sti-ongly implicit scheme is used to solve the resulting
discretized equations.
Figure 4.1 Control Volimie for Node P and Neighboring Control Volumes
4.1 Governing Equations in a General Coordinate System
The governing equations. Equations (3.10), (3.12), (3.13), (3.22), (3.23) and (3.25)
in the physical domain (x,y), which has curvilinear coordinates, have to be transformed to
the equations in the transformed domain (^,r|) which allows for all the computations to be
27
done on fixed square grids. The transformed grid frees the computational simulation from
resti-iction to certain boundary shapes and allows general codes to be written in which the
boundary shape is specified.
If ^denotes a general scalar variable, the general form ofthe differential equations
can be expressed in the transformed coordinate system {t„r]) as follows:
d d{p^)^]_
e, jik^^^H^'^^^^ = l A f ^ a ^
_i__a_ ^ J dTj , J ^''drjj Jd^[ J ^''drjfjdni J ^ " ^ J
where G^= pUy^-pVx^,
(4.1)
+ 5 .
G,f = pVx^ - pUy^,
11 ^^l+y].
gn =-{Xf^x^+y^y^).
g22=yl +xl.
J = x^y^-x^y^.
Here, J and gjj are the Jacobian of the transformation and the covariant metric
components, respectively, p^^ and Sy are the associated viscosity and source function
for variable , respectively. The detailed derivation of Equation (4.1), and the source
terms and the associated viscosities for each variable are given in the appendix. This
work originated from Parameswaran et al. (1992), and the transformed equations for the
k-e equation and the energy equation are added.
28
4.2 Discretization
Equation (4.1) is integrated over the control volume for node P to yield the
following finite volume equation for <;>,
pp.o,'"" '/"'+[{Frr'). - (F.-r')..,+(F,V-' )„ - ( F , > - ' ). ] Ar
Av
+ '''«'3 <«'"'-«'"'-k -22 \
vo/ ( < ^ P -<PS ) (4.2)
• ^ | / ^ e #
>"e#
-12 ^
vol
-21
vo/
(C-O--12 >
/"^ir vo/ (C-O
Ke-fnJ -2n
/".# vo/ ( C - O + Vvo/
where F^ and F^ are the convective mass fluxes across the east and the north cell faces
G"=(A>.)J+(Ax)J
G'2 = G^' = -[(Ax)^(Ax),, + {t.y)^{^y)rf
G'•'•={^y)\^{^x)\
vol = volume of a general cell.
In Equation (4.2) ^;*' , ^,7 ' , (Z>;*' , and (Z>;*' are calculated according to the
Quadratic Upstream Interpolation for Convective Kinematics (QUICK) scheme by
Leonard (1979). The QUICK scheme is discussed in section 4.4. The superscripts n and
n+1 denote the old and new time levels.
4.3 Pressure Correction
For incompressible flow, there is no separate equation for pressure. The pressure
29
has to be obtained from the continuity and momentum equations. In this study, a modified
version ofthe SIMPLE algorithm by Patankar (1980) is used to obtain the pressure field.
The main aim of the pressure correction is to find a way of improving the guessed
pressure, p', to satisfy the continuity equation. The corrected pressure is called the
pressure correction, p', which can be used to find the corresponding velocity corrections.
The convective fluxes, F, and F^, at the cell faces are computed from the velocity
projections along Uie coordinate direction at the cell faces. All velocity projections are
calculated from the velocities stored at the center of each cell. The velocity projection in
the procedure is described in detail in Parameswaran (1985). Here, only the final form of
the pressure correction equation is given as:
PP P^e , PQ». , pq» , PQ.V . pa ; . pal . pal • pa:
^e -^11. -^.1 A ,
where A^, A„, A„ and A^ are the averaged momentum coefficients at the cell faces
derived from the momentimi coefficients at the cell center, and s,, is the continuity error
associated with the cell P given by:
^p = F\,y, - Fi + Kv - Fi, (4.4)
where F^^, F,',,, F^, and Fl,, are convective fluxes based on velocity projections.
The pressure correction procedure is similar to that of Issa (1986). The following
steps are carried out at each time step,
1. Equation (4.2) is solved to yield the intermediate velocity field «* and v* with the
existing pressure field.
2. The new fluxes F^ and Fj are computed from the velocity projections at the cell
faces.
30
3. The continuity error, e,., is calculated for each cell from the fluxed in step 2 and the
pressure correction equation Equation (4.3) is assembled and solved to yield a new
pressure field p'.
4. New velocity fields for M" and v" are obtained from Equation (4.2) using the new
pressure field in step. 3.
5. A new set of fluxes, F^ and Fj, are calculated using the updated velocity field and
pressure field.
6. The pressure correction equation. Equation (4.3) is solved for an improved pressure
field, p".
7. Steps (4)-(6) are repeated until the momentum and the continuity equations are
satisfied within the preset tolerances.
4.4 QUICK Scheme
It is well known that high accuracy can be achieved either by use of higher-order
discretization or by use of a finer mesh. The higher-order convective term discretization
techniques are necessary to gain accurate, stable, and fast numerical simulation. CFD
code, HEAD2D, which is used for this study, originally employed first-order upwind
scheme to avoid the stability problem of central differencing in discretizing the
convective terms in the governing equations. However, this lower-order upwind scheme
creates artificial dissipation and requires more grid points to secure the accuracy. In order
to overcome these problems, higher order techniques have been proposed by researchers.
Raithby (1976) proposed Skew Upstream Differencing Scheme (SUDS) to improve the
upwind scheme, and Leonard (1979) has developed Quadratic Upstream Interpolation for
Convective Kinematics (QUICK) for steady problems, and for unsteady problems
31
QUICK with estimated Streaming Terms (QUICKEST) scheme to discretize convective
terms. Leonard et al. (1993) developed Uniformly Third-order, Polynomial Interpolation
Algorithm (UTOPIA) and concluded that the UTOPIA scheme is the best scheme in a 2D
problem. However, according to the work by Kiris (1994) there is no significant
difference between the QUICK scheme and the UTOPIA scheme. In this study, the
QUICK scheme has been employed and compared with the results ofthe upwind scheme.'
Using the upwind scheme and QUICK scheme the convective terms, "" , ,"* ,
^"'^', and "" ' in Equation (4.2) can be expressed as follows.
Upwind Scheme
.„.._|«t.r''/^.">o, ' Ur''/ F;' < 0,
i i i + i
K*=-ifw if F;'>O,
<i.r' // F: < 0,
„.,_j«t>r''/^2">o. ^ " I . «+i
„^,_j<t,r ' / /F,">0, (4 5)
^.j,;^' // Fl' < 0
QUICK Scheme
„ , _ u^'+i/2((t.r' -<t>r')-9/3(f;^' - 2 C +<t.r)«/ F - >O, *"''[,!,-'+i/2((t.r' -<i)r')-9/3((t.^' -i^'ir +v;') if ," < o,
, _j(t,r+i/2«i,r' - c ) -9 /3« i ) ' ^ -2<t.r ^r;') if F,- >O, * ' 'Ur'+i/2(<t.r -<t'r')-9/3((t>r' -2^7' +C)?/^ p" < o
32
„„ (<t.r'+i/2((|,r -<t>r)-9/3(,t,r' -ir;' ^r;') if F,- >o, l C + i/2«t)r'-<t)r)-g/3((|,;;;-2C+<t'r')'/ 2"<o,
^„., j<i.rM/2(C-<t)r')-9/3«t.;:;'-2,t)r'+<t>r)'/ K > O , Ur' + i/2«t)r -<t,7')-9/3(f;' -2,^;;' +(|,r),/ F" < o ^ ' ^
where q is assigned as 3/8 and the QUICK scheme has two extra terms compared to the
upwind scheme to secure better accuracy.
4.5 Normal Distance Calculation
Unlike the standard k-e model, the normal distance from the calculation point to the
shortest wall is needed to calculate the damping functions in Table 3.2. For simple
geometry, such as pipe or channel, it is not a difficult task to find the distance from the
walls or obstacles. However, for complex geometry, it is necessary to develop a method
to calculate the normal distance. In this study, the new algorithm to find the distance was
introduced. Two possible cases for the normal distance from the arbitrary point C to
piecewise wall AB are shown in Figure 4.2. For case (a) in Figure 4.2, the shorter
distance between CA and CB is assigned as the normal distance for the piecewise wall
^B. For case (b) in Figure 4.2, the perpendicular distance CJ is assigned as the normal
distance for the piecewise wall AB. After considering all the distance to all piecewise
wall, the minimum value ofthe distance is assigned as the normal distance for point C.
The same approach has been used for the obstacles. Figure 4.3 shows two test geometries
for the normal distance calculation mentioned above.
4.6 Solution Procedure
In buoyancy flow the momentum field is strongly coupled with the temperature
33
field so that all the governing equations should be solved at every time step. The overall
solution procedure implemented in this study is shown in Figure 4.4. The do-loop, which
contains the subroutines from 'store' to 'output', is represented as a box.
34
C{x,.y,) C{x..y,)
J(Xj.y^) A{Xi.y^) B{x,.y,) ^(x,.y,) J{x^.yj)
(a) (b)
Figure 4.2. Two possible cases of normal distance calculation
B{x,.y,)
L.
(a) The Geometry with Bluff Body Placed in Channel
i -p- \ >.'•}—
yj
l ^ ^ i
(b) The Geometry with Backward-facing Step and Forward-facing Step
Figure 4.3 Normal Distance Contour for complex geometries
35
RADFLO
uaii input
uaii inietu
uaii caicdist
i uaii store
uaii caicu
uaii caicv
uaii outbc
uaii caicp
uaii caice
uaii caicte
uaii caicea
uaii moapro
uaii output
E N D
Reads input data
Reads input boundary conditions
Calculates the normal distance
Stores the variables for next level
Calculates U velocity
Calculates V velocity
does the block correction for boundaries
carries the block correction for boundaries
Calculates pressure corrections
Calculates turbulent kinetic energy
Calculates dissipation rate
Calculates eddy viscosity
Prints output data
Figure 4.4 Flow Chart for Overall Procedure
36
CHAPTER V
TEST CASE 1 : TURBULENT CHANNEL FLOW
In this chapter, channel flow has been selected as a test case to examine the
buoyancy effects on the flow mechanism. The results by two models, standard k-e model
and low-Reynolds number k-e, have been compared with the budget of DNS. The Two-
dimensional computational fluid code, HEAD2D originally developed by Parameswaran
(1993), is equipped with the standard k-e model. The HEAD2D code has been modified
to predict buoyant turbulent flow. The low-Reynolds number k-e model also has been
added to carry out the calculation up to the wall because buoyancy affected velocity
profile changes close to the wall.
5.1 Problem Statement and Boundary Conditions
The height ofthe channel is selected as 2H and the length ofthe channel is set to be
20 times the channel half-height, H. The Reynolds number based on the mean velocity
and the channel height is 5600, Re,, = lUHIv .
Inlet profiles for the velocity, the turbulent kinetic energy and its dissipation rate
are considered as fully developed which have been acquired numerically by updating
inlet profiles using exit profiles. This correction has been done until the values remain
unchanged within a certain limit. The inlet temperature, however, is given as a constant,
T,„ , which is the same for the temperature of the north wall, Tf, while high wall
temperature, T^ , has been imposed on the south wall. For the exit, zero-gradient
boundary condition is used for the velocities and temperature. Gravitational force acts in
37
the opposite direction ofthe flow to have opposing flow for the heated south wall. Figure
5.1 shows the computational domain and the boundary conditions for the channel flow.
5.2 Grid Arrangement and Grid Independent Study
For the numerical study, a non-uniform grid has been used in the direction normal
to the sti-eam, y, while a uniform grid has been assigned in the streamwise direction, x.'
The grid distribution in y direction is controlled by 5y^^,l5yf =r^. Here /-„ is the
expansion ratio of the grid in the streamwise direction. Values less than 1 have- ''icn .
assigned to the expansion ratio so that more grids can be clustered near both walls. One
ofthe grid configuration used in this study is shown in Figure 5.2.
For the standard k-e model three different sets of meshes, 52x12, 52x22 and 52x32
in the x and y directions are used for grid independent study. The r^ is adjusted for each
case so as not to have a grid point in the viscous sublayer where y* is less than 13 for air.
For the standard k-e model, if the first calculation point close to the wall is in the viscous
sublayer, the numerical prediction tended to overpredict in the viscous sublayer region
while it imderpredicts in the logarithm region. To avoid this problem, r^ is assigned as
1.2, 1.05, 0.95 for 52x12, 52x22 and 52x32 grids, respectively. The predicted velocity
profiles by the standard k-e model using three different grid sets are shown in Figure 5.3
and Figure 5.4. The resuh of 52x12 shows deviation from the results ofthe other two. In
Figure 5.3, U„, denotes the mean velocity. Figure 5.4 shows the same result in semi-
logarithmic coordinates using the nondimensional velocity and the nondimensional
distance from the wall. Figures 5.5 and 5.6 show the nondimensional turbulent kinetic
energy and its dissipation rate, which are defined as follows,
38
k*=klu] (5.1)
£*=£\'lu'. (5.2)
Thus, 52x22 has been selected as a grid independent solution for further study of
standard k-e model.
For the low-Reynolds number k-e model three different sets of meshes, 52x22,,
52x42 and 52x52 in the x and y directions, respectively, are used for grid independent
study. The Low-Reynolds number k-e model requires more grids than the standard k-s
model to have enough grid points within the viscous sublayer region to carry out the
turbulent calculation up to the wall. The grid expansion ratio in the y direction, r^, was
set to be 1.2 for all three sets. The number of grid points in the y direction is decided to
have at least one or two grid points in the viscous sbulayer region. According to Figure
5.7, all three curves of U profiles from three different number of grids are pretty close.
However, when the same results are plotted in a non-dimensional semi-logarithmic
coordinates as shown in Figure 5.8, the U profile with 52x22 grid was clearly deviated
from the other two curves. Figures. 5.9 and 5.10 for non-dimensional turbulent energy
and its dissipation rate show the same grid independent solution for three different
meshes. Therefore, a grid independent solution is obtained with 52x42 mesh for low-
Reynolds number k-s model.
5.3 Results without Buoyant Effects
5.3.1 U Velocity Profile
The mean U profiles for the standard k-e model and the low-Reynolds number k-s
model against DNS data in semi-logarithmic coordinates for channel flow is shown in
39
Figure 5.11. The semi-logarithmic coordinates provide a convenient way to examine the
most important region, the inner region, where most of the velocity change takes place.
Two graphs using two different turbulence models were plotted against DNS data by Kim
(1990) and the law ofthe wall. The law ofthe wall used in this comparison is defined as
following Kays and Crawford (1993)
M* =2.44 1nv'+5.0. (5.3)'
The standard k-s model tends to overpredict velocity while the low-Reynolds
number k-e model (AKN model) tends to underpredict. The shear velocity, u^, for DNS
data is 395 while the shear velocity for current channel flow with Re=5600 is 345. As the
Reynolds number is increased, the outer edge ofthe boundary layer goes to higher.
5.3.2 Turbulent Kinetic Energy Profile
Turbulent kinetic energy has a finite value at the center of the line of the channel
flow. Although there is no production of turbulence at the centerline owing to the zero
velocity gradient, turbulence is continuously diffused towards the center of the channel
from the high turbulence region which is near the wall where a high velocity gradient
exists. Abe et al. (1994) evaluated the near-wall limiting behavior ofthe velocities as
follows,
U = A^y + A^y'^ + Ayy^ +....
V= B2y^ + Byy^ +....
W^C^y + Cjy^ +Cyy^ +....
u = a^y + ajy^ + Oyy +....
40
v= by-+ byy'+ ....
w = co' + c,y^ +Cyy^ +.... (54)
According to Equation (5.4), near-wall limiting behavior of the turbulent kinetic
energy is proportional to ,- In Equation (5.5), only the first term has been written, while
neglecting other higher terms,
,^!r^^J_al^pyl^ (5.5)
This near-wall behavior is valid for the region between the wall and the location of
peak value. In Figure 5.12 DNS data and the result using a low-Reynolds number k-e
model show the near-wall limiting behavior for the turbulent kinetic energy while the
standard k-e model excludes the near-wall behavior assuming that the first calculation
point is located in the logarithmic region. According to Figure 5.12, the standard k-e
model has good agreement with the DNS data if y^ is greater than 100. However, it will
be a hard task to satisfy the condition before one performs the calculation, especially
when the flow is a separating flow. The results using the low-Reynolds number k-s model
show good agreement with DNS data overall including near-wall region and the peak
position at about y*=20.
5.3.3 Turbulent Kinetic Energy Dissipation Rate Profile
In Figure 5.13, a comparison ofthe s profiles calculated by the two models against
the DNS data is shown. In the logarithmic region both models have good agreement with
DNS data. As mentioned earlier, the standard k-s model excludes near-wall behavior for
its calculation and has monotonic decay in the logarithmic region and the outer region.
However, the DNS data and the results using low-Reynolds number k-s model have near
41
wall behaviors and non-zero e at the wall. The dissipation rate, e, near the wall surface
and Equation (5.4) lead to non-zero e at the wall.
£ = v< du V
+ 1 — dy) [dy)
^dw^' • = v(f l , -+cf) + ... (5.6)
\^y)
It is highly debated about where e has its peak and what the value should be at the
wall, though most modelers believe that the peak value of e should be where k peaks.
5.4 Results with Buoyant Effects
Three Richardson numbers, 0, 0.1 and 0.2, are introduced to study the effects of
buoyancy on the flow field in the channel flow. For the case with Re=0, the gravitational
force has been set to 0 to eliminate buoyancy effects.
5.4.1 U Velocity Profile
For the south wall where there is a high temperature, the buoyancy force aids the
convective motion near-wall region resulting in acceleration of the velocity. The
buoyancy effects on the velocity profiles for half of the channel are shown in Figure 5.14.
Velocity profiles using both models show clear peak velocities near the wall due to the
buoyancy force. As the magnitude ofthe buoyancy force increased the velocity gradients,
in the near-wall also increased. The low-Reynolds number k-e model tends to have
velocity peak values close to the wall as the Richardson number is increased. In Figure
5.15, it is found that the value of w at the outer region became smaller as the Richardson
number increased due to the increase of shear velocity at the wall as well as the decrease
ofthe core velocity at the center ofthe channel.
42
5.4.2 Temperature Profile
In Figure 5.16, the effect of buoyancy forces on dimensionless temperature is
shown. It is shown that the trends of the temperature profiles are highly affected by the
velocity profiles in Figures 5.14 and 5.15. Temperature profiles cross each other where
the velocity profiles do in both models.
5.4.3 Turbulent Kinetic Energy Profile
Figure 5.17 shows that turbulent kinetic energy, k, decreased as the Richardson
number becomes high. According to Equation (5.4), k is the function of the velocity
fluctuations. Garg et al. (2000) performed direct numerical simulation (DNS) and found
that in buoyancy affected flows, the changes in the fluctuation component of U, u, are
primarily due to the decreased correlation between the streamwise and vertical velocity
fluctuations, which reduces the Reynolds shear stress and, thus the production. Since u is
the source of the fluctuation velocities v and w components, the behavior of v and w
tends to decrease in buoyancy affected flow. Nakajima et al. (1980) also mentioned the
decrease of turbulent fluctuation with increasing Richardson number. For both models, a
reduction of k was observed as the Richardson number increased. The peak values for the
low-Reynolds mmiber k-e model decreased with the increase of the Richardson number,
while the peak values ofthe standard k-e model are fixed by the wall function.
5.4.4 Turbulent Kinetic Energy Dissipation Rate Profile
The decrease of production also suppresses the dissipation rate of turbulent energy,
e, as the Richardson number increases. Profiles of s shown in Figure 5.18 are closely
related with the turbulent kinetic energy behavior affected by buoyancy.
43
T=Tc North W&ll
T=Tm V=0 IHXy)
2H
gc 4
dU_dV_dT
dx dx dx
^ O l i h ^ l L=20H T=TH
X
Figure 5.1 Computational Domain and Boundary Conditions for Channel Flow
44
Figure 5.2 Grid Formation for Channel Flow
(52x42 Grids with r^ =1.2)
45
0.0 0.2 0.4 0.6 y/H
0.8 1.0
Figvire 5.3 Grid Independent Solution for U Profile Using Standard k-s Model
46
10.0 100.0 y+
1000.0
Figure 5.4 Grid Independent Solution for Nondimensional U Profile
Using Standard k-s Model
47
4.0
3.0
i 2.0
1.0
0.0 0.0 100.0 200.0
y+ 300.0
Figure 5.5 Grid Independent Solution for Nondimensional k Profile
Using Standard k-s Model
48
0.25 0.20
^0.15 «0.10
0.05 0.00
0.0 100.0 200.0 300.0
y+
Figure 5.6 Grid Independent Solution for Nondimensional e Profile
Using Standard k-e Model
49
0.0 0.3 0.5 y/H
0.8
Figure 5.7 Grid Independent Solution for U Profile
Using Low-Reynolds Number k-e Model
1.0
50
% 10.0
1.0 10.0 100.0 1000.0
y+
Figure 5.8 Grid Independent Solution for Nondimensional U Profile
Using Low-Reynolds Number k-s Model
51
5.0
4.0
3.0
'2.0
1.0
0.0 0.0 100.0 200.0
y+ 300.0
Figure 5.9 Grid Independent Solution for Nondimensional k Profile
Using Low-Reynolds Number k-e Model
52
0.20
0.15
+ 0.10
0.05
0.00
0.0 100.0 200.0 300.0
y+
Figure 5.10 Grid Independent Solution for Nondimensional e Profile
Using Low-Reynolds Number k-e Model
53
2Q0
1.0 IQO KBO
y^
100QO
Figure 5.11 Comparison ofthe Fully Developed U Profiles
54
5.0
4.0
30 + •^2.0
1.0
00 i
%
•_
0.0
- ^U*
— - -
100.0
y ^ ^ -
—
y+
^ L K E HD-STKE
• DNS
^*"~"~^*'0.~.i i i .^____^ ^
200.0 m
Figure 5.12 Comparison ofthe Fully Developed Nondimensional k Profiles
55
0.0 100.0 200.0 300.0
y+
Figure 5.13 Comparison ofthe Fully Developed Nondimensional e Profiles
56
0.0 0.3 0.5
y/H
0.8 1.0
Figure 5.14 Effect of Buoyancy on U profiles
57
25 .
20 :.
15 .
10
0
,LKE(Ri=0) "
.LKE(Ri=0.1)
.LKE(Ri=0.2)
.STKE(Ri=0)
.STKE(Ri=0.1)
.STKE(Ri=0.2)
10 100 10001
y+
Figure 5.15 Effect of Buoyancy on Nondimensional U profiles
58
1.2
0.9
(0
0 0.6
0.3
0.0 0.0 0.3 0.5
y/H
LKE(Ri=0) LKE(Ri=ai) LKE(Ri=0.2) STKE(Ri=0) STKE(Ri=0.2)i STKE(Ri=0.2):
- t>
0.8 1.0
Figure 5.16 Effect of Buoyancy on Dimensionless Temperature
59
5.0
4.0
3.0
2.0
1.0
0.0 0
—-0_LKE(Ri=0) J_n-LKE(Ri=0.1)
-,A_LKE(Ri=0.2) ^^STKE(Ri=0) ^K_STKE(Ri=0.1)
.STKE(Ri=0.2)
100 200 300'
y+
Figure 5.17 Effect of Buoyancy on Nondimensional k Profiles
60
+ 0>
0.25
0.20
0.15
0 100 200 300
y+
Figure 5.18 Effect of Buoyancy on Nondimensional e Profiles
61
CHAPTER VI
TEST CASE 2: BACKWARD-FACING STEP
Quite often, the boundary layer separates in many flow processes of practical
interest. Model testing is often performed on a backward-facing step because of its
simpler geometry and the availability ofthe experimental data. In this chapter, backward-
facing step has been selected to study buoyant turbulent flow for separating flow.
6.1 Problem Statement and Boimdary Conditions
Numerical computation has been performed for turbulent flqw over a backward-
facing step studied by Vogel and Eaton (1985) which is frequently used in benchmarking
the performance of turbulence models for separating and reattaching flows. The
computational domain consists of two walls and an adiabatic step. The step height is
given as H and the inlet height and the height after the step are assigned as 4H and 5H,
respectively, which make an expansion ratio (ER) of 1.25. The computational domain
was set to 30H in the streamwise direction. A diagram showing the computational
domain is shown in Figure 6.1. For the measurements, the average inlet velocity was 11.3
m/s and the height ofthe step was 3.8 cm. The Reynolds number based on the step height
and a laminar Prandtl number for the experiment are given as 28,000 and 0.71,
respectively. At the inlet (x=-4H), a fully developed velocity profile was prescribed from
the channel flow test case described in Chapter V. The coordinate x is measured from the
step end. Constant heat flux through the south wall was set to 270 W / m \ corresponding
to a maximum temperature of 15 °C above ambient temperature, and the north wall and
62
the step are considered as adiabatic. Constant inlet temperature was imposed, and at the
exit, zero-gradient boundary conditions are assigned for temperature, velocities, and
turbulent quantities. Gravitational force is set to act in the opposite direction of the flow
stream. Numerical predictions with a low-Reynolds number k-e model are compared to
the work by Seo and Parameswaran (2000) which was performed with the standard k-e
model.
6.2 Grid Arrangement and Grid Independent Study
For this study, non-uniform grids are employed which have clustered near the step
end and in the recirculation region. One of the grid arrangements and the partial view of
the grids used in this study are shown in Figures. 6.2 and 6.3, respectively. The grid
distribution is controlled by the parameters r^ and r^. Here r^ and r^ are expansion
ratios of the grids in the streamwise and normal directions, respectively. The
computational domain in the streamwise direction was divided into two regions; the r^^
was assigned to a value of less than one for the region before the step (-4H < x < 0), and
the r„ was assigned to a value of greater than one for the region after the step (x>0). In
the same manner, r , and r^j' which control the grid expansion ratios in the normal
direction of grids for the region between 0 < y < H and H < y < 5H, respectively. Grids
are clustered near the walls in order to place some of grid points inside the viscous
sublayer.
In any numerical calculation, it is necessary to show that the results are
independent of the grid size. In order to achieve this, calculations have been performed
with three grid sizes.
63
An earlier work of Seo and Parameswaran (2000) employed a mesh size of 102x62
with the following expansion factors for the grid distribution:
(r„,r., ,rj,) = (1.02, 0.8, 1.04). (6.1)
In the spanwise direction, only one expansion factor, r,,, was used for 0 < y < 5H
while two expansion factors, r,,, and r,,,. were employed for the current study with the
low-Reynolds number k-e model to have a clustered grid near the step end.
Three mesh sizes employed for the grid independent study for the low-Reynolds
number k-e model are 85x43, 122x62, and 122x93. The grid sizes used for the low-
Reynolds number k-e model are close to those employed for the standard k-e model.
However, more attractions in a crosstream direction have been imposed to place the grids
within the viscous sublayer. The values for the expansion factors are as follows:
('•n.' V ' r^i '•42) = (1 -28, 113, 0.65, 1.025) . (6.2)
These values were adjusted in order to maintain continuity of the grid distribution
at the junction of the grid blocks. A sudden expansion or contraction in the grid size can
cause numerical instabilities. The y* values for the cells adjacent to the wall for 85x43,
122x62, and 122x93 meshes are adjusted to be 18.5, 10.7, and 2.9 at the exit,
respectively.
The low-Reynolds number k-e model with 85x43 mesh predicts the reattachment
length at x/H=5.72, while the predictions of reattachment length with 122x62 and 122x93
meshes are x/H=5.85 and x/H=5.87, respectively. The average reattachment length for
the three test cases is x/H=5.81 which has a 12% increase compared to the value of
x/H=5.24 as predicted by the standard k-e model.
64
Numerical results from three sets of grids are compared at the vicinity of the
reattaching point. Accurate prediction of the reattachment location is one of the key
problems in evaluation of a turbulent model. Location ofthe reattachment point has been
calculated using a skin friction coefficient which changes its sign at the reattachment
point.
Figures 6.4 and 6.5 show the plots of U and V velocities for the three meshes at x/H-
= 5.0. The deviation is not significant because of close predictions of the reattachment
length using the three meshes. However, the U velocity profile of 62x38 mesh shows a
small deviation from the other two profiles in Figure 6.4. Considering velocity profile
deviation and the reattachment length predicted, the 122x62 mesh is selected for further
study.
6.3 Influence oflnlet Boundary Conditions
Inlet boundary velocity profile, specifically boundary-layer thickness, has an
influence on the reattachment length and the corresponding momentum and heat transfer
characteristics, such as the skin fiiction coefficient and Stanton number. Two inlet
velocity profiles are employed to examine its influence on the reattachment length. One is
the constant inlet velocity profile that was used in the grid independent study and the
other is the fully developed velocity profile generated by channel flow in Chapter V. The
fiiUy developed velocities and turbulent quantities for inlet boundary conditions are
shown in Figure 6.6, for a chaimel with height of 4H. Using two inlet velocity profiles,
the changes ofthe reattachment lengths are examined.
65
The low-Reynolds number k-e model increased the reattachment length from
x/H=5.84 to x/H=6.38, which is a 9.2% increase, by introducing the fully developed inlet
velocity profile. The velocity vectors for each case are shown in Figure 6.7.
6.4 Influence of Discretization Schemes for the Convective Term
In section 4.4, the QUICK scheme for discretization ofthe higher-order convective
term has been discussed. In this section the influence of the QUICK scheme on the
reattachment length will be discussed and compared to upwind differencing scheme for
two turbulence models. The low-Reynolds number k-s model has been shown to increase
the reattachment length from jc , IH =5.91 to x,, ///=6.38 which is a 6.9% increase, by
replacing the upwind scheme with the QUICK scheme. The velocity vector plots are
shown in Figure 6.8. Figure 6.7(b) is plotted again for better comparison with Figure 6.8.
6.5 Results without Buoyant Effects
In this section, forced convection has been studied to examine the difference in
results between the standard k-s model and the low-Reynolds number k-s model. From
the grid independent study in the previous section, 102x62 and 122x62 meshes with
different attraction factors are employed for the standard k-s model and the low-Reynolds
number k-e model, respectively. The QUICK scheme for the convective term and the
fully developed inlet velocity profiles for the inlet boimdary condition are used for this
study.
66
Velocities and temperature are plotted against the experimental measurements by
Vogel and Eaton (1985) and the turbulent quantities from two different turbulence
models are compared to each other.
6.5.1 U Velocity Profile
Velocity profiles for the velocity component U are shown in Figure 6.9 and
compared against the experimental measurement by Vogel and Eaton (1985).
Comparison is made at four representative locations in the recirculation, reattachment and
developing regions. The corresponding velocity vectors are shown in Figure 6.8 (b). The
reference velocity is measured at the inlet and is 1.08 times the mean velocity. Velocity
profiles obtained using the two turbulence models follow the trend of the experimental
velocity profiles faithfully. It can be seen from the figure that the low-Reynolds number
k-s model shows low sensitivity to the velocity change near the wall compared to the
standard k-s model and the experimental data. The discrepancy between the experimental
data and the numerical predictions for the y/H>1.0 region is due to the difference of inlet
velocity profiles between experiment and numerical computations. The boundary layer
thickness ofthe experimental inlet velocity profile is 1.1 times the step height while the
inlet velocity profile used for computation is a fiilly developed profile.
6.5.2 V Velocity Profile
Comparison of the V velocity profiles has been done for the same locations of the
U velocity profiles. In Figures 6.10 (a) and (b) which are in the recirculation region, the
magnitude of the V velocity of the standard k-s model is greater than that of the low-
67
Reynolds number k-e model because the standard k-e model has smaller recirculating
bubbles than the low-Reynolds number k-e model.
6.5.3 Reattachment Length
The reattachment length changes with the inlet profiles, turbulence models, and
with the differencing schemes. Experiment by Vogel and Eaton (1985) indicates that thd
reattachment of the current step is 6 2/3 times the step height. Table 6.1 shows the
summary of the reattachment lengths and the errors against the experimental data.
Compared to the experimental data, the standard k-e model underpredicts the
reattachment length by about 14-20% while the low-Reynolds' number k-s model
underpredicts about 4-12% ofthe same quantity.
Table 6.1 Summary ofthe Reattachment Lengths for Various Cases Model STKE
LKE
Inlet Velocity Condition Fully Developed
Constant
Fully Developed
Con\ecti\« Term QUICK QUICK upwind QUICK
Xr1/H 5.24 5.84 5.97 6.38
Xr2/H 0
0.61 0.51 0.67
Error(%) 21.4 12.44 10.49 4.35
6.5.4 Skin Friction Coefficient Comparison
The predicted skin friction coefficients are compared with the experimental data in
Figure 6.11. The pattern of the skin friction coefficients is closely related to the
reattachment length. Experimental results show the effect ofthe secondary bubble, which
makes the friction coefficient positive at the comer of the step. Numerical results,
however, do not show the clear effects ofthe secondary bubbles, although Table 6.1
shows there are secondary bubbles. The computational models overpredict the absolute
skin friction inside and outside the recirculation bubble. This means that the magnitude of
68
the velocity near the wall is overpredicted. For the standard k-e model the overprediction
can be due to the wall functions employed. Experimental results by Adams and Johnston
(1988) for the velocity profiles inside the recirculating region is compared with the
standard wall function in Figure 6.12. From the figure, it is clear that the velocities
obtained from the wall-fimctions are much higher than the experimental values. Figure
6.13 shows the vj. variations used for the two turbulence models. It is clear that yj, fot
the standard k-e model is less than 12 within the recirculation region and near the
reattachment p' int Since the velocities were prescribed from the wall function for these
points, predicted velocities, in general, would be much higher than the actual ones for the
standard k-e model. Hence, the wall function is inadequate to predict recirculating and
developing flows.
A low-Reynolds number k-e model has been employed to overcome this problem.
In low-Reynolds number k-e model, yJ, can be located anywhere theoretically. However,
Chien (1982) found that the yp should be less than 1 for an accurate prediction in
channel flow. In addition, the grid independent study for channel flow in Chapter V
supports Chien's statement. However, for practical difficulty, the fine grid system which
has a point within yJ, <l was not employed for the backward-facing step. However, the
meshes for a low-Reynolds number k-e model are designed to have smaller yJ,, e.g., less
than 10 even at the exit, compared to the standard k-e model. The standard k-e model
shows fast recovery of skin friction coefficient compared to the low-Reynolds number k-
s model and experiment. This may be due to the overprediction of the velocity near the
wall by the wall function in the developing region after the reattachment point.
69
6.5.5 Turbulent Kinetic Energy Profile
The disd-ibution ofthe turbulent kinetic energy k is shown in Figure 6.14. In Figure
6.14(a) the maximum value of k appears in the shear layer downstream of the step.
Kasaki et al. (1993) stated that the location ofthe maximum in the y-direction changes as
the flow goes downsU-eam; from the separation to just upstream of the reattachment, it
moves toward the wall, but gradually away from the wall after reattachment due to the
turbulent kinetic energy diffusion toward the outer edges ofthe shear layer. The turbulent
kinetic energy has a maximum near the reattachment point. The standard k-s model
shows faster difftision of turbulent kinetic energy than low-Reynolds number k-s model
within the recirculation region. After the reattachment point it is known that the peak
value ofthe turbulent kinetic energy decreases with the distance downstream.
6.5.6 Turbulent Kinetic Energy Dissipation Rate Profile
The mechanics of the turbulent kinetic energy dissipation rate profiles can be
explained by understanding turbulent kinetic energy profiles mentioned in the previous
section. In addition, s has a high peak near the wall which was not evident in k profiles. It
is observed that this peak value of s near the wall increases within the recirculation
region and reaches a maximum near the reattachment point as shown in Figure 6.15.
6.5.7 Turbulent Prandti Number Profile
The turbulent Prandtl number plays an important role in the prediction of the heat
transfer. For the standard k-s model a value of 0.85 was used for the turbulent Prandtl
70
number. For the low-Reynolds number k-s model Equation (3.28) first proposed by Kays
and Crawford (1993) was used to calculate the turbulent Prandtl number. Figure 6.16
shows tile contour plot of the turbulent Prandtl number predicted by the low-Reynolds
number k-e model using Equation (3.28). Basically the turbulent Prandtl number is
increased near the wall along the step and the south wall because of the increase of the
eddy viscosity due to turbulent kinetic energy production. In the same sense the rise in'
Pr, has been noticed in the shear layer downstream of the step end. The range of the
turbulent Prandti number has been examined by changing Pe, - 0 ' , /v)Pr and the plot is
shown in Figure 6.17. Pe, is a turbulent Peclet number, and is conveniently expressed as
the product of the Prandtl number and the eddy-viscosity ratio. The turbulent Prandtl
number varies between 1.7 and 0.85 as Pe, increases from 0 to 100 which can cover most
of the range for the channel flow. Figure 6.18 shows the Pr, profiles for selected
locations near the step end.
According to the Figures 6.16 and 6.18, Pr, is 0.85 for most of the recirculation
region which contradicts the experimental measurement by Pak (1999) and numerical
calculation by Rhee and Sung (2000) and Abe et al. (1995). Pak (1999) measured tiie
turbulent Prandtl number within the recirculation region using Equation (3.27) and found
that the turbulent Prandtl number is over 1 close to the wall and approaches 0.25
asymptotically away from the wall in the recirculation region. This is much less than the
conventional turbulent Prandti number, 0.85, used for boundary layer flow due to a
respectively high decrease of momentum compared to the decrease in thermal diffusion.
Rhee and Sung (2000) conducted numerical calculation using a nonlinear low-Reynolds
number k-s model to flnd that Pr, increases considerably in the recirculating region
71
owing to the enhanced eddy diffusity. They found that the turbulent Prandtl number
varies from 0 to 2.0 according to the recirculating pattern within the recirculation region.
Abe et al. (1995) found that a calculated turbulent Prandtl number with the two-equation
heat-ti-ansfer model is much higher than the conventional value of Pr,=0,85. Although
the turbulent Prandtl number plays a important role in the calculation of the heat transfer
turbulent the qualification of Prandtl number in a recirculating region is still an open'
research issue.
However, it is certain that a reasonable prediction of heat transfer with separating
and reattaching flow is impossible with a constant turbulent Prandtl number. In addition,
the empirical correlation proposed by Kays and Crawford, Equation (3.28), is not also
suitable for the recirculating flow because the correlation does not include the effects of
thermal eddy diffusivity which is used for the original calculation ofthe turbulent Prandtl
number. Equation (3.27).
6.5.8 Stanton Number Comparison
Computed Stanton numbers are compared with experimental values for non-
buoyant flow in Figure 6.19. Experimental values by Vogel and Eaton (1985) are
provided for different boundary layer thicknesses, 6/H, at the inlet. As tiie boundary layer
thickness increases, the curves of tiie Stanton numbers move downward. Numerical
predictions witii the standard k-s model, with a fully developed inlet velocity boundary
condition, underpredicts the Stanton number. This underprediction of the standard k-s
model is due to the overprediction of the wall temperature by tiie wall function. On the
contrary, with a tiirbulent Prandtl number of 0.85, the Stanton numbers were over-
72
predicted witii tiie low-Reynolds number k-e model. These overpredictions by the low-
Reynolds number k-e model have also been noticed in the work of Abe et al. (1995) with
Pr, =0.9 and the work of Rhee and Sung (2000) with Pr, =1.0. This significant
overprediction indicates that a constant Prandtl number assumption is no longer valid in
separating and reattaching flows. However, as Vogel and Eaton (1985) stated the peak
heat ti-ansfer rate for die backward-facing step occurs slightly upstream of reattachment
jK)int.
6.6 Results with Buoyant Effects
In this section, several different Richardson numbers are introduced to test the flow
patterns and corresponding characteristic in heat transfer over the back-ward-facing step
using two different turbulent models.
6.6.1 Reattachment Length
Several different Richardson numbers are introduced to test the flow patterns over
the backward-facing step. Figure 6.20 shows three representative particle traces for
results obtained using the standard k-e model. The reattachment length decreases as the
Richardson number increases. In addition, the size of the secondary bubble at the comer
of the step increases with increasing the Richardson number. This occurs because the
flow near the step moves in the main flow direction when there is significant buoyancy.
When the Richardson number is increased to 0.2, the length of the secondary bubble has
almost reached that of the main bubble. Consequently, the main bubble is lifted by the
secondary bubble and is about to be separated from the surface. The Richardson number
73
from which the main bubble is separated from the surface is referred to as the 'critical
Richardson number', hereafter.
The same ti-end has been noticed when the low-Reynolds number k-e model is
used. Compared to the results with the standard k-e model, the size of the secondary
bubble at the comer of the step with /?/„ =0 is significant, which was not shown cleariy
for the results with standard k-e model. This secondary bubble is noticed in the
experimental resuhs by Vogel and Eaton (1985) as a corner eddy. The low-Reynolds
numht'r k-i: model has relatively bigger in size recirculation bubbles, than the standard k-
e model. However, the critical Richardson number for the low-Reynolds number k-s
model is smaller than that of the standard k-s model due to the fast enlarging of the
secondary bubble as the Richard number increases.
Figure 6.22 shows the influence of the Richardson number on the reattachment
length. In Figure 15, A>, denotes the reattachment length of the main bubble and Xr^
denotes that of the secondary bubble. Two reattachment lengths approach each other
indicating the possible values of the critical Richardson numbers for two turbulence
models. The flow pattern after the critical Richardson number will be discussed in section
6.6.4.
6.6.2 Turbulent Scales
Turbulent scales such as time and length scales are important in conducting
experiment for turbulent flow. This section is dedicated to provide the quantitative ideas
that how much the time and length scales are, for the flow over backward-facing step. For
74
velocity scale, Kolmogorov velocity scale is employed for the study. The dimensions of s
and V are
£ = {lengthy litimef,
v = {length)-1 time.
Based on tiie dimensions of s and v, Kolmogorov defined a length scale as
L=(^''fe)"- (6.3)
By performing dimensional analysis using Kolmogorov velocity scale and
Kolmogorov length scale the time scale can be calculated,
T,-={vl£y'\ (6.4)
All the turbulent scales are defined by the turbulent kinetic energy dissipation rate,
e. Contour plots for the length scale and the time scale with ^/^ =0.1 are shown in Figures
6.23 and 6.24. Length scale varies between 4.62x10"'* and 6.89x10""' m with smallest
value near shear layer which defines the boundary of the main recirculating bubble after
the step end. It is also shown that length scales are relatively small within recirculating
region compared to values outside ofthe recirculating bubble. The variation of time scale
lies between 5.97x10'^ and 1.22x10'° s. Referring to the particle trace in Figure 6.20, like
the length scale, the minimum time scale exists along the shear layer after the step end.
Without buoyant effects, the length scale varies between 4.61x10"'' and 9.38x10"^ m
while the time scale varies between 5.95x10"^ and 2.46x10"° s. Maximum turbulent scales
are decreased by increase of the buoyant effects. However, it was found that the
minimum turbulent scales which are important in experimental measurement are not
affected by buoyant effects. Once the turbulent time scales are highly affected by shear
75
layer, geometry and the Reynolds number are more important factors than buoyant
effects to define the turbulent scales.
6.6.3 Skin Friction Coefficient Comparison
Figure 6.25 shows the variation of the skin friction coefficient with different
Richardson numbers. With a high Richardson number the flow is accelerated near the hot-
wall. The acceleration in die flow creates a high friction coefficient along the south wall
with constant heat flux. The friction factor remains almost positive throughout the entire
length of the south wall indicating that the secondary bubble has almost reached the size
of tiie primary bubble at /?/„=0.2 and /?/„=0.175 for the standard k-s model and the low-
Reynolds number k-s model, respectively. In Figure 6.25(b), the c/ curve with
Ri^ =0.175 for the low-Reynolds number k-s model shows fluctuation within the
recirculation region. This means that the secondary bubble becomes unstable due to the
interaction writh the main recirculation bubble and is about to create the vortex shedding.
The most significant difference between the two results is the magnitude of the friction
coefficients predicted by the two different turbulence models. The cf predicted by the
standard k-e model is more than double that of the cf by the low-Reynolds number k-s
model. This could be explained by the wall function functions used in the standard k-s
model which overpredict the velocity within the recirculation region and results in fast
recovery of the velocity in the developing region after the reati:achment point. However,
the friction coefficient in buoyany-affected flow is closely related with heat transfer.
Therefore, it is too early to say which model predicted better results in the skin friction
coefficient without any reliable data.
76
6.6.4 Temperature Contour Comparison
The size and pattern of recirculation bubbles are important factor to characterize
the temperature pattern after the step end. The buoyancy-affected temperature contours
obtained by two turbulent models are plotted in Figures 6.26 and 6.27. Figure 6.26(a)
shows that temperature contour has been rotated in clockwise direction due to the main
recirculating bubble. In (b) and (c) of Figure 6.26, the secondary bubble moves the hot
temperature region away from the step end as Richardson number increases due to the
increase of the secondary bubble in magnitude. The same trend has been noticed in
Figure 6.27 obtained by the low-Reynolds number k-e model.
Zero-gradient temperature boundary condition has been imposed to the exit
boundary condition. If the thermal layer is developing even at the exit the boundary
condition may create numerical error for the calculation of the thermal field. Magnified
temperature contours at the exit are shown in Figure 6.28. Temperature gradient near exit,
x/H=22, was calculated to examine the validity of non-zero boundary condition for the
temperature at the exit. Temperature gradients for cases (a) and (c) in 6. 28 were found to
be 1.67x10'^ and 3.33x10'^, respectively. Although the temperature gradient had been
increased as the buoyant effect was increased the magnitude of the temperature gradient
were close to the given zero-gradient temperature boundary condition at the exit.
6.6.5 Stanton Number Comparison
In Figure 6.29, computed Stanton numbers for different Richardson numbers are
compared with each other. For the standard k-e model, Stanton numbers for each case are
77
highly dependent on the structure of tiie recirculation bubble within the recirculation
region. After tiie reattachment point, the Stanton numbers mainly depend on the
Richardson number. It is found that the flow with a higher Richardson number has a
higher Stanton number in the region after the main bubble.
However, the Stanton numbers predicted by the low-Reynolds number k-e model
decrease as Uie Richardson number increases. This is a different trend compared to the'
results shown by tiie standard k-e model. The Stanton number tends to increase slightly
with the increase of buoyant effects in turbulent flow (Osborne &. Incropera, 1985). This
nonphysical prediction can be due to two different reasons.
One ofthe reasons for the low prediction ofthe Stanton number with the increase
of the Richardson number is the constant turbulent Prandtl number discussed in section
6.5.7. As mentioned, the turbulent Prandtl number should be defined by the ratio of the
eddy viscosity to the thermal eddy diffusivity. However turbulent heat transfer modeling
with a constant turbulent Prandtl number adopts one representative turbulent Prandti
number for the calculation ofthe whole domain, including the recirculation region. If the
turbulent Prandtl number is fixed as a constant for a flow with recirculating flow, thermal
eddy diffusivity should be proportional to the eddy viscosity regardless of its physical
values. Figure 6.30 shows the effect of the constant turbulent Prandtl number on the
Stanton numbers.
The second reason for the low prediction ofthe Stanton number for buoyant flow is
due to the non-dimensional distance from the hot wall, y]>. Figure 6.31 shows the y].
variation along the hot wall with various Richardson numbers. A high Richardson
number provides a high y], away from the step end. As discussed in Chapter V for
78
channel flow, the low-Reynolds number k-e model requires that the yp should lie with
tiie viscous sublayer, if possible >';<1. If this condition is not satisfied, the low-Reynolds
number k-e model generates inaccurate predictions because of the collapse of turbulent
quantities near the wall, as shown in Chapter V. However, in practice, it is difficult to
keep all .i> within such a range along all the boundary walls and obstacle placed in the
sti-eam. Especially when tiiere is a buoyancy effect , the orthogonal grid system cannot
catch up witii the velocity development along the heated wall. Therefore, an adaptive grid
system will be necessary when tiie low-Reynolds number k-e model is employed to
predict the buoyant flow or developing flow in a complex geometry.
6.6.6 Vortex Shedding
In Figure 6.22, the influence of the Richardson number on the reattachment length
is shown discussed. As mentioned in the section 6.6.1, two recirculating bubbles interact
with each other and become unstable creating vortex shedding after the critical
Richardson number. For the standard k-e model the critical Richardson number was
found to be 0.225. For the range of such a large Richardson number, pattems of two
recirculating bubbles have changed their form along time showing unsteady flow patterns
while the flows with small Richardson numbers give only steady-state solution. As the
secondary recirculation bubble grows, the main bubble becomes unstable and is divided
into two small bubbles creating periodic vortex shedding. In the same marmer, the main
bubble breaks the secondary bubble into two, creating vortex shedding with small
magnitude. Figure 6.32 shows the bubble pattems for a flow with a Richardson number
of 0.25, which produces vortex shedding. The period of vortex shedding was computed as
79
5.35s, approximately, based on the reattachment length in Figure 6.32. Figure 6.32(e)
shows the instant moment at which the flow is not reattached on the hot wall at time
3.99s, which makes the friction coefficients positive for the whole range along the wall.
The variation of the reattachment of the two recirculation bubbles for a flow with
/?/„=0.25 is shown in Figure 6.33.
The prediction by low-Reynolds number k-e model, after the critical Richardson'
mmiber, is shown in Figure 6.34. Interaction between the two bubbles is relatively small
compared to tiie result using the standard k-s model. The main recirculation bubble is all
consistently separated from the heated surface by the secondary bubble, and the center of
the main bubble is lifted up and down creating a neglectable vortex'. Unlike the standard
k-s model, the vortex pattem predicted by the low-Reynolds number k-s model changes
randomly without a clear period due to minor interaction between the two bubbles.
Cetegen et al. (1998) and Jiang and Luo (2000) correlated the oscillation frequency
with the Richardson number for the puffing phenomenon of an axisymmetric thermal
plume. Investigation of the relation between the Richardson number and the Strouhal
number, Sh„ =flllu, for the resuhs predicted by the standard k-s model was conducted
for this study. Frequencies ofthe vortex shedding were evaluated for the range of 0.225 <
Ri < 0.6 where f is the frequency of the vortex shedding.
Eleven point three (11.3) m/s of velocity and 3.8xl0"^m of step height have been
used to calculate the Strouhal numbers. The result is shown in Figure 6.35 with the
correlation, Sh = 0.048 /°'™ • It is also found that friction coefficients and the Stanton
numbers grew larger with increasing Richardson numbers regardless of vortex shedding.
80
Finally, tiiere is a question as to whether the buoyant-induced vortex shedding is
desirable for heat ti-ansfer. Figure 6.36 shows the Stanton number variation for different
times during tiie vortex shedding. The Stanton number for flow with /?/„=0.2 has been
plotted together as a reference. It shows that the Stanton number has increased, in the
recirculation region, as well as after the recirculation region, while the other flows
without the vortex shedding shows that the Stanton number in the recirculation region is
decreased as tiie Richardson number is increased. Fluctuation of the Stanton number and
friction factor have been also noticed right after the recirculation region as shown in
Figures 6.36 and 6.37, respectively.
81
•
•
' lMJ(y) • V=0
• T=To 1
H • t
ck=0 >
, m 4H
qw=o
^dU dV dT
8K 8K dx
'
q^=CQnst
261 1
=0
Figure 6.1 Problem Description and Boundary Conditions for Turbulent Flow
82
Figure 6.2. Grid Configuration for the Turbulent Flow (LKE Model, 122x62 Grid)
.—\—t'^ti—^±~^1i "irf—"ttl" 1
x/H
Figure 6.3. Partial View ofthe Grid Configuration near Step End
83
-0.50
3-0
.LKE( 123*93)"'
.LKE(122*62)
.LKE(85*43) '
x/H=5.0
0.00 0.50
U/Um
1.00 1.50
Figure 6.4 Grid Independent Solution for U velocity at x/H=5.5 Using Low-Reynolds Number k-s Model
84
5C
.LKE(123*93)
.LKE(122*62)
.LKE(85*43)
-0.15 -0.12 -0.09 -0.06 -0.03 0.00
V/Um
Figure 6.5 Grid Independent Solution for V velocity at x/H=5.5 Using Low-Reynolds Number k-s Model
85
4.0
3.0
5 2.0
1.0
0.0
0.0 1.0 U/Um
(a) U Profile
n
2.0
4.0
-1.0E-03 O.OE+00 1.0E-03
V/Um
(b) V Profile
(c) k Profile (d) s Profile
Figure 6.6 Profiles for Inlet Boundary Conditions
86
O.S 1 1.5 2 2.5 i 3.5 4 4.5 5 5.5 6 6.5
(a) Low-Reynolds Number k-e Model with Constant Inlet Velocity Profile(.x,i IH =5.84)
f --
-
= ^ t n ,
\ . I c \ \ \ \ \ s >.' \' V > ,'' r ' / ! ^ ^^
4 l i m ^ ^ ^ ^ ^ ^
- . — ^ r ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^
1' ' '|f ' V^f~^T^'%^^^ '\ " h ^ 0.5 1 l.S 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
(b) Low-Reynolds Number k-s Model with Fully Developed Inlet Velocity Profile(;c,., ///=6.38) Figure 6.7 Influence ofthe Inlet Velocity Profile on Reattachment Length
87
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
(a) Low-Reynolds Number k-e Model with Upwind Scheme( .v , / H =5.97)
0-5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
(a) Low-Reynolds Number k-s Model with QUICK Scheme(x„ ///=6.38)
Figure 6.8 Influence ofthe Schemes for Discretization of Convective Term on Reattachment Length
88
% Experiment(Vogel) .<!—LKE -A—STKE
x/H=2.20
(a)
« Expenmerit(Vogel) H>-LKE .^—STKE
-0.2
x/H=6.67
0.2 0.6 U/Uref
X
9 ExperlmenKVogel) j K3—LKE -A—STKE
-0.2 0.2 0.6 U/Uref
(b)
Experimeht(\7ogel j
.IKE
J
-0-i-»
x/H=8.87
-0.2 0.2 0.6 U/Uref
(0 ("> Figure 6.9 U Profiles for Reference Conditions
89
X
-0.20 -0.10 0.00 0.10 0.201
V/Um
-0.20
(a)
0.20
0.201
(b)
X
-0.20 0.20
(c) (d)
Figure 6.10 V Profiles for Reference Conditions
90
4-O
Figure 6.11 Comparison of Model Results for Skin Friction Coefficient to Vogel's Experiment (6 / i / = 1.1)
91
15
10
+ 3
0
/ • • • / •
/ • • ^ / •
/ • y/ \ y+=u+
y^ Law of the wall ^^^.^"•^ • Experlment(Adams) x/Xr =
•""""^ • Experiment(Adams) x/Xr = 1 , 1 1 X
•
0.67 0.45
10
y+ 100
Figure 6.12 Mean Velocity Profiles in Recirculating Region(Re=38000, ER=1.67)
92
Figure 6.13 y+ variations along the south wall
93
X "5»
0.00 0.02 0.04 0.06
k/U'
(a)
ref
X
0.00 0.02 0.04 0.06
2 ref k/U^
X
0.00 0.02 0.04 0.06i
k/U" ref
(b)
0.00 0.02 0.04 0.06
k/U^ef
(C) (d)
Figure 6.14 k Profiles for Reference Conditions
94
0.00 0.01 0.02 0.03 0.04'
e/U^ ref
(a)
0.00 0.01 0.02 0.03 0.04|
0.00 0.01 0.02 003 0.041
^'^ ref
(b)
0.00 0.01 0.02 0.03 0.04|
e/U^ref
(c) (d)
Figure 6.15 s Profiles for Reference Condhions
95
1.2S 1.5
Figure 6.16 Contour for Turbulent Prandtl Number Using Eq. (3.28)
0 50 100
Figure 6.17 Range of Turbulent Prandti Number Using Eq. (3.28)
96
0.85
Pr. 1.35
Pr. 1.85
(a) (b)
0.85 1.35
Pr.
1.85i
x/H=1.23
-A-A-0.85 1.35 1.85:
(c) (d)
Figure 6.18 Turbulent Prandtl Number Profiles Variation near the Step End Using Eq. (3.28)
97
CO
0.006
0.005
0.004
0.003
0.002 -
0.001
* ff
STKE(Prt=0.85) • LKE(Prt=0.85)
-X— Abe(Prt=0.9) A Rhee(Prt=1.0)
10 15
X/H
Figure 6.19. Comparison of Stanton Number to Experimental Results with Various Inlet Boundary Layer Thickness
98
x/H
a) /?/„=()
b) ^/^=0.1
c) Ri„=0.2
Figure 6.20. Particle Trace for Mixed Turbulent Convection Using standard k-s Model
99
a) Riff=0
x/H b) Riff =0.1
c) Riff =0.175
Figure 6.21. Particle Trace for a Mixed Turbulent Convection Using Low-Reynolds number k-s Model
100
Xr1(STKE)^ Xi2(STKE) Xrl(LKE) Xr2(LKE)
0 0.05 0.1 0.15 0.2 0.25
Ri
Figure 6.22. Influence ofthe Richardson Number on Reattachment Length
101
V .
•I" =F ^ 2 - 3 4 5 6 7 8
Figure 6.23. Contour Plot for tiie Kolmogorov Length Scale {Riff =0.1, 10 contour between 4.62x10"^ and 6.89x10'^)
10
J-3
- .2
1L^ -h + + + + ^
2 3 4 5 6 7 8 Figure 6.24. Contour Plot for the Kolmogorov Time Scale {Riff=0.\, 10 contour between 5.97x10"^ and 1.22x10"°)
10
102
o o
u
10
8
6
4
2
0
-2
-4
10
8
6
4
«- 2
0
-2
.Ri=0.0(STKE) • Ri=0.1(STKE) i • RI=0.2(STKE) Vogel
x/H
(a) Standard k-s Model
-i»-Ri=0.0(LKE) _o_Ri=0.1(LKE) -A-Ri=0.175(LKE)
• Vogel
(1 ^^«5ft(|&§*^ 10 15
x/H
(b) Low-Reynolds number k-s Model
Figure 6.25. Influence ofthe Richardson Number on Skin Friction Coefficient
103
^•N
\ .
aaep —~^^^ asfitea
3 x/H
(a)/?/V,=0
6 3 x/H
(b) i?//,=0.1
1 2 3 4 5 x/H
(c)i?/;,=0.2
Figure 6.26 Buoyancy-affected Temperature Contour for Recirculating Region Obtained by Standard k-s Model
6
104
• ^ " • ^ " " " ^ ^ ^ — - ^ - — . ^
\ \ \ \
3 x/H
(a)/?/„=0
\ \ \ \ \
i^M^
x/H
(b) Riff =0.1
1 2 3 4 5 x/H
(c) Riff =0.175
Figure 6.27 Buoyancy-affected Temperature Contour for Recirculating Region Obtained by Low-Reynolds Number k-s Model
105
T-T;,=O.OO
22 24 x/H (b)
26
22 26 24 26 22 24 x/H x/H (c) (d)
Fig. 6. 28 Temperature Contour Plots at Exit Boundary for Two Turbulent Models: (a) Standard k-s Model(Ri=0); (b) Low-Reynolds Number k-s Model(Ri=0); (c)
Standard k-s Model(Ri=0.1); (d) Low-Reynolds Number k-s Model(Ri=0.1)
106
4.E-03
3.E-03
S 2.E-03
1.E-03
O.E+00 1
:-g^B5^ •
# ^ ^ ^ ^ r ^ jsf"V-o- Ri=0.d(gTOr*'*«=9*,fca
/ l i __Ri=0.1(STKE) fSjr - «^Ri=0.2(STKE) W^ • Vogel a — —.—...—-—.
10 ^ 15
x/H
(a) Standard k-s Model
(b) Low-Reynolds Numer k-s Model
Figure 6.29 Influence ofthe Richardson Number on Stanton Numbner
107
10
X/H
15 20
{a) Riff =0.0
Prt=0.35
^Prt=0.45
Prt=0.65
Prt=0.85
(c)i?i;,=0.175
Figure 6.30 Effects of Turbulent Prandtl Number on Stanton Number
108
0 10 15 x/H
20 25
Figure 6.31 Effects of Richardson Number on Wall Normal Distance for Low-Reynolds Number k-s Model
109
1 2
(a) time=0
(c) time=2.8s
(f) time=5.35s
Figure 6.32 Vortex Shedding after the Backward-Facing Step with Riff =0.25 using Standard k-s Model
110
0
Xrl
Xr2
8
time(s)
Figure 6.33 Reattachment Variation ofthe Flow with Riff =0.25 along the time
111
1 2 3
(b) time=0.71s
1 2 3
(c) time= 1.24s
1 2 3
(d) time= 1.92s
2 3
(e) time=2.65s
2 3
(f)time=3.22s
Figure 6.34 Vortex Interaction Between Two Bubbles after the Backward-Facing Step with Riff =0.20 using Low-Reynolds Number k-s Model
112
(O
0.044
0.042
0.04
0.038
0.036
0 034 .
^^
—o^
-- — - -
o^y""^ y^— _ _.
<>
^ ^ Sh = 0.048 Rl°^^^^ —
: o Sh Power (Sh)
0.2 0.3 0.4 RI
0.5
Figure 6.35 Effect of Richardson Number on Strouhal Number of Vortex Shedding Using Standard k-s Model
0.6
113
I . .
(O
0.004
0.003
0.002
0.001
0.000
X/H
Figure 6.36 Stanton Number Change during Vortex Shedding Using Standard k-s Model
114
o o o
O
X/H
Figure 6.37 Cf Change during Vortex Shedding Using Standard k-s Model
115
CHAPTER VII
CONCLUSIONS AND RECOMMENDATIONS
In this study, the low-Reynolds number k-s model is employed to study turbulent
buoyant flow to overcome the shortcoming of the standard k-s model in predicting the
reattachment length of separating and reattaching flow.
A constant turbulent Prandtl number was used for the standard k-s model and the
varying turbulent model proposed by Kays and Crawford (1993) was used for the low-
Reynolds number k-s model. An algorithm was introduced to find the normal distance of
the damping functions of tiie low-Reynolds number k-s model. The'QUICK scheme was
also used as an alternative of the upwind scheme to enhance the discretization of the
convective terms of the governing equations. The results obtained by the standard k-s
model and the low-Reynolds number k-s model were presented for channel flow and the
backward-facing step flow against experimental data. In addition, the influence of
buoyancy effects was discussed for the channel flow and backward-facing step flow. The
required conditions for the low-Reynolds number k-s model for accurate results were
discussed.
7.1 Conclusions
From the present study, the following conclusions were made concerning the
usage of k-s models to predict separating buoyant flow:
1. The reattachment length for the backward-facing step has been improved by 17%
by replacing the turbulent models of the standard k-s model with tiie low-Reynolds
116
number k-e model. Conventional log-laws cannot be applied to complex flows with
separations. Therefore, it is recommended to use the low-Reynolds number k-s model for
a recirculating flow.
2. The correlation of the turbulent Prandtl number proposed by Kays and Crawford
(1993), which varies between 0.85 and 1.75, did not provide significant effects on
recirculating flow because the correlation was originally developed for channel flow,'
which is an attached flow. With tiie constant turbulent Prandtl number, the low-Reynolds
number k-s model overpredicted the Nusselt number, while the standard k-s model
underpredicted the Nusselt number for backward-facing step flow.
3. In backward-facing step flow, buoyant effects reduced the size of the main
recirculating bubble while increasing that of the secondary bubble. The changes in the
recirculating bubble affected the skin friction coefficient and the Stanton number along
the heated wall.
4. Vortex shedding occurred when the Richardson number exceeded the critical
Richardson number. With a standard k-s model vortex shedding enhanced heat transfer
characteristics due to the fluctuations.
5. A low-Reynolds number k-s model showed physically wrong results when buoyant
effects were included. It was found that grid configuration plays a significant role when
the low-Reynolds number k-s model is used. For channel flow, yl along the wall
boundary conditions should be less than 1 to obtain accurate results. The same rule
should be applied to backward-facing step flow.
117
7.2 Recommendations
The following recommendations are made for improving the accuracy of the
prediction for buoyant flow with the low-Reynolds number k-s model.
1. A varying turbulent Prandtl number, based on a two-equation model for thermal
field, needs to be employed rather than a constant turbulent Prandtl number or varying
turbulent Prandtl nimiber based on an empirical correlation, to solve eddy diffusivity
directiy for heat wdthout prescribing tiie turbulent Prandtl number.
2. An adaptive grid system should be developed to obtain y*^ along the walls, or the
obstacles, for the low-Reynolds number k-s model. In addition, the stability ofthe multi-
block grid system should be studied to secure stable convergence.
118
REFERENCES
Abe, K, Kondoh, T. and Nagano, Y., 1994, "A new turbulence model for predicting fluid flow and heat ti-ansfer in separating and reattaching flows - 1 . Flow field calculations," Int. J. Heat Mass Transfer, Vol. 37, No. 1, pp. 139-151.
Abe, K, Kondoh, T. and Nagano, Y., 1995, "A new turbulence model for predicting fluid flow and heat ti-ansfer in separating and reattaching flows - II. Thermal field calculations," Int. J. Heat Mass Transfer, Vol. 38, No. 8, pp. 1467-1481.
Abu-Mulaweh, H. 1., Chen, T. S., and Armaly, T. S., 1996, "Measurements of turbulent natural convection flow over a vertical backward-facing step," Proc. ASME Heat Transfer Division, International Mechanical Engineering Congress and Exposition, HTD-Vol.338, Vol. 2, pp.313-321.
Abu-Mulaweh, H. 1., Chen, T. S., and Armaly, T. S., 1999, "Turbulent natural convection flow over a backward-facing step," Experimental Heat Transfer, Vol. 12, pp.295-308.
Acharya, S., Dixit, G., and Hou., Q., 1993, "Laminar mixed convection in a vertical channel wotii a backstep:A Benchmark Sttidy," ASME HTD Vol.258, pp.11-20.
Adams, E. W. and Johnston, J. P., 1988, "Flow structure in the near-wall zone of ttirbulent separated flow," AIAA J. Vol. 26, pp. 932-939.
Blackwell, B. F., Kays, W. M. and Moffat, R. J., April 1972, Report HMT-16, Thermosciencec Division, Department of Mechanical Engineering, Standford University.
Blackwell, B. F. and Armaly, B. F., 1993, "Computational Aspects of Heat Transfer Benchmark Problems," ASME HTD, Vol. 258., pp.1-10.
Blackwell, B. F. and Pepper, D. W.(editors), 1992, "Benchmark problems for heat transfer codes," ASME HTD vol.222.
Boussinesq, J., 1977, "Essai sur la tiieorie des eaux courantes(Essay on the theory of water flow)." Memoires Academic de Science (Parie), 23(1): 1-680.
Cetegen, B. M., Dong, Y., and Soteriou, M. C , 1998, "Experiments on stability and oscillatory'behavior of plannar buoyant plumes," Phys. Fluids, Vol. 10(7), pp. 1658-1665.
Chen, C. J. And Jaw, S. Y., 1998, "Fundamental of turbulence modeling," P. 4, Taylor
and Francis.
119
Chien, K. Y., 1982, "Predictions of channel and boundary-layer flows with a low-Reynolds-number turbulence model," AIAA J. Vol. 20, pp.33-38.
Chopin, T. R., 1993, "Mixed convection flow and heat transfer in a vertical backward facing step using tiie FLOTRAN CFD program," ASME HTD. Vol. 258., pp.21-28.
Choudhury, D. and Woolfe, A. E., 1993, "Computation of Laminar Forced and Mixed Convection in a Heated Vertical Duct with a Step," ASME HTD, Vol. 258.
Cochran, R. J., Horstiiian, R. H., Sun, Y. S., and Emery, A. F., 1993, "Computation of Laminar Forced and Mixed Convection in a Heated Vertical Duct with a Step," ASME HTD, Vol. 258, pp. 37-47
Craft, T.J., Launder, B. E., and Suga, K. 1993, "Extending the Applicability of Eddy Viscosity Model Through tiie Use of Deformation Invariant and Non-Linear Elements," in Proc. 5"" Int. lAHR Symp. On Refined Flow Modeling and Turbulence Measurements, Paris, France.
Driver, D. M., and Seegmiller, H. L., "Features of a Reattaching Turbulence Shear Layer in Divergent Channel Flow," AIAA J. Vol. 23, pp. 163-171.
Eckert, E. R. G., 1950, Introduction to the Transfer of Heat and Mass, pp. 158-164, McGraw-Hill, New York.
Eckert, E. R. G., and Jackson, T. W., 1950, NACA(now NASA), TN2207, Washington.
Ede, A. J., 1967, Advances in Heat Transfer, Vol.4, pp. 1-64, Academic Press, New York.
Fujii, T., 1959, "Experimental studies of free convection heat transfer," Bulletin of JSME, Vol. 2, No.8, pp.555-558.
Fujii, T. and Fujii, M, 1976, Int. J. Heat and Mass Transfer, Vol. 19, pp. 121-122.
Garg, R.P., Ferziger, J.H., Monismitii, S.G., and Koseff, J.R., 2000, "Stably stratified turbulent channel flows. I. Stratification regimes and turbulence suppression Mechanism," Physics of Fluids, Vol. 12, pp.2569.
George, W. K. And Capp, S. P., 1979, "A theory for natural convection turbulent boundary layers next to heated surfaces," Int. J. Heat Mass Transfer, Vol. 22, pp.813-826.
Gibson, M. M., Verriopoulos, C. A., 1984, Experiment in Fluids, Vol. 2, pp. 73-80.
Gibson M M., Vertiopoulos, C. A. and Nagano Y., 1982, Measurements in the heated turbulent boundary layer on a mildly curved convex surface," In turbulent Shear Flows 3, Edited by L. J. S. Bradbury et al., pp.80-89.
120
Giel, P.W. and Schmidt, F. W., 1986, "An experimental study of high Rayleigh number natural convection in an enclosure," in Tien, C. L., Carey, V. P., and Ferrell, J. K., (eds). Heat Transfer, vol.4, pp. 1459-1464.
Godaux, R. and Gebhart, B., 1974, "An experimental study ofthe transition of natural convection flow adjacent to a vertical surface," Int. J. Heat and Mass Transfer, Vol.17, pp.93-107.
Heindel, T.J., Ramadhyani, S., and Incropera, F.P, 1994, "Assessment of turbulence models for natural convection in an enclosure," Numerical Heat Transfer, Part B (Fundamentals), Vol. 26, pp.147.
Henkes, R. A. W. M. And Hoogendoom, C. J., 1990, "Numerical determination of wall functions for the turbulent natural convection boundary layer". Int. J. Heat and Mass i lonsfer. Vol. 33, No.6, pp.1087-1097.
Henkes, R. A. W. M., 1990, Natural-convection boundary layer, Ph.D. Thesis, Delft University.
Henkes, R. A. W. M., Van Der Vlugt, F. F., and Hoogendoom, 1991, " Natural convection flow in a square cavity calculated with low-Reynolds-number turbulence models". Int. J. Heat Mass Transfer, Vol. 34, pp.37-7-388.
Hinze, J. O., 1975, Turbulence, New York, McGraw-Hill.
Hollingsworth, D. K, Kays, W. M. and Moffat, R. J., Setember 1989, Report HTM-41, Thermal sciences Division, Department of Mechanical Engineering, Stanford University.
Hong, B., Armaly, B. F., and Chen, T. S., 1993, "Mixed convection in a laminar, vertical, backward-facing step flow:Solution to a benchmark program," ASME HTD, Vol. 258., pp.57-62.
Humphreys, W. W. and Welty, J. R., 1975, "Natiiral convection witii mercury in a uniformly heated vertical channel during unstable laminar and transitional flow," AIChE J. Vol. 21, pp.268-274.
Iglessias, I., Humphrey, J. A. C , abd Giralt, F., 1993, "Numerical calculation of two-diineiisional buoyancy-assisted flow past a backward-facing step in a vertical channeir ASME HTD, Vol. 258., pp.63-72.
Inagaki T and Kitamura, K, 1996, "Turbulent heat transfer with combined forced and nattiral convection along a vertical flat plate (Effect of Prandtl Number)," Heat Transfer-Japanese Research, Vol. 19, p. 70
121
Inagaki, T., 1995, "Heat transfer and fluid flow of turbulent natural convection along a vertical flat plate with backward-facing step," Exp. Heat Transfer, Vol. 7, pp.285-
Incropera, F. P. and DeWitt, D. P., 1996, Fundamentals of Heat and Mass Transfer, 4" edition, Wiley, New York, pp.490-491
Issa, R. I., 1986, "Solution ofthe inplicitly discretized fluid flow equations by operator splitting," J. Comput. Phys., Vol. 62, pp.40-65.
Iwai, H.. Nakabe, K., Suzuki, K. and Matsubara, K., 1999, "Numerical simulation of buoyancy-assisting, backward-facing step and heat transfer in a rectangular duct," Heat Transfer-Asian Research, Vol.28, #1, pp.58-76.
JayatiUaka, C. L. V., 1969, "The influence of Prandti number and surface roughness on t.ic resistance of tiie laminar sublayer to momentiam and heat transfer," in : U. Grigull and E. Hahne (eds). Process in Heat and Mass Transfer 1, Pergamon Press, N. Y.
Jiang, X. and Luo, K.H., 2000, "Direct numerical simulation ofthe puffing phenomenon of an axisymmetric thermal plume," Theoretical and Computational Fluid Dynamics, Vol. 14, pp.55.
Jones, W. P. and Launder, C. J., 1972, "The prediction of laminarization with a two-equation model of turbulence," Int. J. Heat Mass Transfer, Vol. 15, pp.301-314.
George, W. K., and Capp, S. P., 1979, "A theory for natural convection turbulent boimdary layer next to heated vertical surfaces," Int. J. Heat and Mass Transfer, Vol. 22, pp.813-826.
Kasagi, N., Matsunaga, A., and Kawara, S., 1993, "Turbulence measurement in a separated and reattaching flow over a backward-facing step with the aid of three-dimensional particle tracking velocimetry," Joumal of wind engineering and industrial aerodynamics, Vol. 46 & 47, pp.821-829.
Kays, W. M. and Crawford, M. E., 1993, Convective Heat and Mass Transfer, McGrawhill
Kim, J., Moin, P. and Moser, R., 1987, "Turbulence statistics in fully developed channel ' flow at low Reynolds number," Joumal of Fluid Mechanics, Vol. 177, pp. 133-166.
Kim, J., 1990, "Collaborative testing of turbulence models," Data Disk No. 4.
Kiris I. M. 1994, High-order accuracy and multigrid acceleration for two-dimensional 'flow computations, Ph.D. Thesis, Texas Tech University.
122
Kishinami, K., Saito, H. and Suzuki, J., 1995, "A combined forced and free laminar convective heat transfer fi-om a vertical plate with coupling of discontinuous surface heating," Int. J. of Numerical Methods for Heat and Fluid Flow, Vol. 5, No. 9, pp.839-851.
Kishinami, K., Saito, H., Suzuki, J., and Ali, A. H. H., Umeki, H. and Kitano, N., 1998, "A fundamental study of combined free and forced convective heat transfer from a vertial plate followed by a backward step," Intemational Journal of Numerical Methods for Heat and Fluid Flow, Vol. 8, No. 6, pp.717-736.
Kolmogorov, A. N., 1941, "Local structure of turbulence in an incompressible fluid at very high Reynolds numbers," Dokl.Akad. Nauk (Report of Academy of Science) USSR 30(4):299-303.
Kwak, C. E and Song, T. H., 1998, "Experimental and numerical study on natural convection from vertical plates with horizontal rectangular grooves," Int. J. Heat and Mass Transfer, Vol. 41, No. 16, pp.2517-2528.
Lam, C. K. G. and Bremhorst, K., 1981, "A modified form ofthe k-s model for low-Reynolds-number hirbulence model," AIAA J. Vol.20, pp.33-38.
Launder, B. E. And Spalding, D. B., 1974, "The numerical computation of turbulent flows," Computer Method in Applied Mechanics and Engineering 3: 269-89.
Launder, B. E. and Sharma, B. 1., 1974, "Application ofthe energy-dissipation model of turbulent to the calculation of flow near a spinning disc," Letters in Heat and Mass Transfer, Vol. I,pp.l31-138.
Leonard, B. P., 1979, "A Stable and Accurate Convective Modeling Procedure Based on Quadratic Upsfream Interpolation," Compu. Meth. Eng., Vol. 19, pp. 59-98.
Leonard, B. P., MacVean M. K., and Lock, A. P., 1993, "Positivity-Preserving Numerical Schemes for Multidimensional Advection," NASA Technical Memorandum 106055, ICOMP-93-05.
Nagano, Y.and Kim, C , 1988, "A two-equation model for heat transport in wall turbulent shear flows," Transactions ofthe ASME. Joumal of Heat Transfer, Vol. 110, pp.583.
Nagano, Y., and Tagawa, M., 1990, "An improved k-s model for boundary layer flows," Trans.'ASME, J. Fluid Engrg., Vol. 112, pp. 33-39.
Navier M 1827, Memoire sur les Lois du Mouvement des Fluides (Memorir on tiie theory of fluid motion). Memoires Academic de Science ( Momoir of academy of science), Paris 6:389-416.
123
Osborne, D. G, and F. P. Incropera, 1985, "Experimental study of mixed convection heat transfer for transitional and turbulent flow between horizontal, parallel plates," Int. J. Heat Mass Transfer, Vol. 28, pp.1337.
Pak, J, 1999, A study of the thermal law of the wall for separated flow caused by a backward facing step, Ph.D. Thesis, Texas Tech University.
Parameswaran, S. V., 1985, Finite volume equations for fluid flow based on nonortiiogonal velocity projections, Ph.D. thesis. University of London.
Parameswaran, S., Srinivasan, A., and Sun, R., 1992, "Numerical aerodynamic simulation of steady and transient flows around two-dimensional bluff bodies using the nonstaggered grid system," Numerical Heat transfer. Part A, Vol. 21, pp.443-461.
Patankar, S. V., 1980, Numerical Heat Transier and Fluid Flow, Hemisphere, Washington, D. C.
Prandti, L., 1925, "Uber ide ausgebildete turbulenz (On the fully developed turbulence)," ZAMM (Joumal of Applied Mathematics and Mechanics) 5: 136-39.
Raithby, G. D., 1976, "Skew upstream differencing schemes for problems involving fluid flow," Computer Methods in Applied Mechanicas and Engineering, Vol. 9, pp. 153-164.
Reynolds, W. C, Kays, W. M. and Kline, S. J. Kline, 1958, NASA Memo 12-1-58W, Wachington, December.
Rhee, G. H. and Sung, H. J., 2000, "A nonlinear low-Reynolds number heat tiransfer model for turbulent separated and reattaching flows," Int. J. Heat and Mass Transfer, Vol. 43, pp.1439-1448.
Rodi, W., 1984, Turbulence Models and Their Application in Hydraulics, pp.28-34.
Sanchez, J. G. and Vradis, G. C , 1993, "Mixed Convection Heat Transfer over a Backward-Facing Step," ASME HTD, Vol. 258, pp.105-113.
Seo E R., and Parameswaran, S., 2000, "Buoyant effects on fluid flow and heat transfer ' in separating flows," ASME 2000 Fluids Engineering Division Summer Meeting,
Boston, Massachusetts, June 11-15.
Singh S Biswas G., and Mukhopadhyay, A., 1998, "Effect of thermal buoyancy on the flow through a vertical channel with a built-in circular cylinder," Numerical Heat Transfer, Part A, Vol. 34, pp.769-789.
124
Spalart, P. R., 1988, "Direct Simulation of a Turbulent Boundary Layer up to RCg =1410", Joumal of Fluid Mech., Vol. 187, p. 61.
Speziale, C. G., 1991, "Analytical methods for the development of reynolds stress closures in turbulence," Ann. Rev. Fluid Mech. Vol. 23, pp. 107-157.
Stokes, G. G., 1845, "On the theories of intemal friction of fluids in motion," Transactions of tiie Cambridge Philosophical Society 8:287-305.
Stone, H.L., 1968, "Iterative solution of implicit approximations of multidimensional partial differential equations," SI AM Journal on Numerical Analysis, Vol. 5, pp.530.
To, W. M. and Humphrey, J. A. C, 1986, "Numerical simulation of buoyant, turbulent flow-I. Free convection along a heated, vertical, flat plate". Int. J. Heat Mass Transfer, Vol. 29, pp. 573-592.
Thangam, S. and Speziale, C. G., 1992, "Turbulent flowpast a backward-facing step: A critical evaluation of two-equation models," AIAA J. Vol. 30, pp. 1314-1320.
Thompson, J. F., Warsi, Z. U. A., and Mastin, C. W., 1985, Numerical Grid Generation-Foundations and Applications, North-Holland, New York.
Tsuji, T. and Nagano, Y. 1988, "Characteristics of a turbulent natural convection boundary layer along a vertical flat plate," Int. J. Heat and Mass Transfer, Vol. 31, pp. 1723-1734.
Tsuji, T. and Nagano, Y. 1989, "Velocity and temperature measurements in a natural convection boundary layer along a vertical plate," Experimental Thermal and Fluid Science, Vol. 2, pp.208-215.
Vogel, J C. and Eaton, J. K., 1985, "Combined heat transfer and fluid dynamic measurements downstream of a backward-facing step," J. Heat Transfer, Vol. 107, 922-929.
Westphal, R. V., and Johnson, J. P., 1984, "Effects of initial conditions on turbulent reattachment downstream of a backward-facing step," AIAA J., Vol. 22, No. 12, pp. 1727-1732.
Wu, H. and P. S. Pemg, 1998, "Heat transfer augmentation of mixed convection through ' vortex shedding from an inclined plate in a vertical channel containing heated
blocks," Numerical Heat Transfer, Part A, Vol.33, pp.225-244.
Yang R J and Aung, W., 1985, Equation and coefficients for turbulent modeling, in Kakac, S., Aung,' W., and Viskanta, R. (eds.), A Natural Convection Fundamentals and Applications, pp'.259-30. Hemisphere, Washington, DC.
125
Yin. Y., Nagano, Y., and Tsuji, T., 1990, "Numerical prediction of turbulent convection boundary lines," Heat Transfer : Japanese Rresearch. Vol.19, #6.pp. 584-601.
Yuan, X, Moser, A, and Suter, P., 1993, "Wall functions for numerical simulation of turbulent natural convection along vertical plates," Int. J. Heat and Mass Transfer, Col. 36, No. 18, pp.4477-4485.
126
APPENDIX
DERIVATION OF GOVERNING EQUATIONS
IN A GENERAL COORDINATE SYSTEM
127
A.I Transformation of Governing Equation
U-momentum equation is selected to be transformed from the equation in physical
domain into the equation in ti-ansformed domain.
d{pU) d , d
dp d St dx
(dU dU dy ^^<#
dV dU — + —
^ dx dy
(A.l) + pgp(r-ro)
l.I Transform Relations for Gradient
Transform relations are well defined by Thomson et al. (1985). For 2-dimensional
domain the gradient can be defined in conservative form and non-conservative form as
follows:
Conservative form
Non-conservative form
A =-kA->'4/nl J
/,=-j[-V^+Vn]
where the Jacobian ofthe transform is defined
J=Xi,y^-Xr,yi,
(A.2)
(A.3)
(A.4)
(A.5)
(A.6)
128
Using the ti-ansform relations each term ofthe U-momentum Equation (A.l) can be
transformed into the term for computational domain.
A. 1.2 Transient Term
Transient term will keep its original form if the time derivative is understood to be at
fixed point in the transformed space.
d{pU) _ d{pU)
dt dt (A.7)
A. 1.3 Convective Term
Convective terms can be transformed using Equations (A.2) and (A.3).
. ;5 1 ±{pU') + -l-{pUV) J dx dy =>
[y^pU^X -{y,pU'hj[-kpUv\A^,pUv\[ (A.8)
where G^ = pUy^ -pVx^,G^ = pVx^ -pUy^
A. 1.4 Pressure Term
-^=>-j[ynP^-y^p^ (A.9)
A. 1.5 Diffusive Term
dx
(dU dU => —
J
|(>'.^.^[7k^.-3^.^l4k^^->'^^nl
1 - I - —
J dx\ y^y-eff
j[y,U,-y,uy-j[y,U,-y,U,] (A.10)
129
d_
dy J
+ — J
(A. l l )
A. 1.6 Buoyancy term
pg^iT-T,)=>pg^(T-T„) (A.12)
Final transformed equation can be arranged as follows.
d{pU) 1 d . *W'^^Hi,M-^^ Peff _ dU], \ d fPeff dU^ -8u + — I, J " d^J jdt] J " ^ - drj
^ J d^ d (Peff „ dU\, \ d (Peff dU
-^12-r- +-7^r-J dT]) J drj
where tiie covariant metric components for 2-D are defined
J "^'^ d^
(A.13)
+ 5,
^11 =Xr^+y, 1 •
^12 = - ( j f 4 ^ n + n > ' T i ) '
gn=yl+xl
The detailed description of covariant metric components, g ,y, are given in Thomson
etal . (1985).
A.2 Governing Equation in General Form of Differential Equations
in General Coordinate System
130
J ^"a^j a^
^ J dt]
Peff d6
_1__5_ /V^ a ^ 1 a r;/e# „ d(i> ^ J drj
\
J ' a? +5.
(A.14)
A.3 Source functions for each equation
A.3.1 U-Momentum Equation
5=--^ ^''de, -^anJ y a ^ l y
r 2 dji
dk y:-:^-yi-^r^
dU_ an
^}_d_\\^
J dk\ J
dV '-y^i;{-'-y^^
dV^ 1 d \V;sf • + —
d\)\ J dx\\ J
dV^
2dU dU ^' an ^ " a
^7l^fxh^-^^^^^l^^^^«^-^«>
(A. 15)
A.3.2 V-Momentum Equation
5.. = — dp dp
I "a^ ^a7j ^ 1 a |/^e#^
y a ^ J
dU
1 a / e# J d4[ J
1 a f/^.#
2 aF " d^ ^ ^ dT]
dU^
dU dU dV\\ 1 d \Peff(
j a ^ y
aF a^ - ^ ^ ^ ' 7
ar 5^
(A.16)
A.3.3 k Equation
Sk=P.
+ Pt [Ll^^u, .x,uy-[y/, -y,v,^-pe-gP^'-j[-^.T,.x,T,]
(A.17)
131
A.3.4 e Equation
k
-^c\-Pt k J
£ p, 1
- C — (A. 18)
A.3.5 Energy Equation
Sr=0 (A.19)
A.4 The Associated Voscosities for each Equation
A.4.1 U-Momentum Equation
Peff=P + Pt (A.20)
A.4.2 V-Momentum Equation
Peff=P + Pt (A.21)
A.4.3 k Equation
Peff=P^Pt'^k (A.22)
A.4.4 s Equation
Peff^P^^^tl^E (A.23)
132
A.4.5 Energy Equation
p,ff=klCp^p,l?x, (A.24)
133