HEAT AND MASS TRANSPORT INSIDE A CANDLE WICK
by
Mandhapati P. Raju
Submitted in partial fulfillment of the requirements
For the degree of Doctor of Philosophy
Dissertation Advisor: Dr. James S. T’ien
Department of Mechanical and Aerospace Engineering
CASE WESTERN RESERVE UNIVERSITY
January 2007
CASE WESTERN RESERVE UNIVERSITY
SCHOOL OF GRADUATE STUDIES
We hereby approve the dissertation of
______
candidate for the Ph.D. degree *.
(signed)______(chair of the committee)
______
______
______
______
______
(date) ______
*We also certify that written approval has been obtained for any proprietary material contained therein. TABLE OF CONTENTS
TABLE OF CONTENTS iii
LIST OF TABLES v
LIST OF FIGURES vi
ACKNOWLEDGEMENTS xii
NOMENCLATURE xiv
ABSTRACT xxii
CHAPTER 1: INTRODUCTION 1 1.1 Candle Basics 1 1.2 Candle Burning 4 1.3 Previous Work 6 1.3.1 Previous Work on Candle Flames 7 1.3.1.1 Experimental Work 7 1.3.1.2 Numerical Work 10 1.3.2 Previous Work on Two-phase Flow in Porous Media 13 1.4 Purpose and Scope of this Dissertation 21 1.5 Dissertation outline 21
CHAPTER 2: AXISYMMETRIC WICK MODELING 24 2.1 Formulation of Transport Process in Porous Media 24 2.1.1 Mathematical Formulation 24 2.1.2 Numerical Formulation 31 2.2 Multifrontal Solvers for Large Sparse Linear Systems 33 2.2.1 Introduction 33 2.2.2 Multifrontal Solution Methods 36 2.23.Benchmark Testing 40 2.3 Analysis of an Externally Heated Axisymmetric Wick 41 2.3.1 Physical Description of the Model 42 2.3.2 Sample Case Results 43 2.3.2.1 Saturation and Temperature Distribution 43 2.3.2.2 Pressure Distribution 41 2.3.2.3 Mass Distribution 44 2.3.2.4 Heat Flux Distribution 45 2.3.2.5 Variation Along the Cylindrical Surface and the Axis of the Wick 46 2.33 Mesh Refinement Studies 47 2.3.4 The Effect of Applied Heat Flux 48 2.3.5 Parametric Studies 49
iii 2.3.5.1 The Effect of Gravity 49 2.3.5.2 The Effect of Absolute Permeability 50
CHAPTER 3: GAS PHASE MODELING INCUDING RADIATION 74 3.1 Theoretical Formulation 74 3.1.1 Continuity Equation 75 3.1.2 Momentum Equations 76 3.1.3 Species Equation 77 3.1.4 Energy Equation 78 3.1.5 Boundary Conditions 79 3.2 Non-Dimensional Parameters 82 3.3 Property Values 85 3.4 Numerical Procedure 86 3.4.1 Grid Generation 87 3.4.2 Numerical Implementation 88 3.5 Gas Radiation Model 89 3.5.1 The Equation of Radiative Transfer 90 3.5.2 Numerical Solution of Discrete Ordinates Method 95 3.5.3 Discrete Ordinates Angular Quadrature 97 3.5.4 Solution of Discrete Ordinates Equation 98 3.5.5 Mean Absorption Coefficient 104 3.6 Solution Procedure 105
CHAPTER 4: RESULTS AND DISCUSSIONS 120 4.1 Candle Flame Coupled to a Porous Wick 120 4.1.1 Detailed Flame Structure at Normal Gravity and 21% O2 122 4.1.1.1 Steady State Candle Flame (Wick Length = 4 mm) 123 4.1.1.2 Self Trimmed Candle Flame 130 4.1.2 Effect of Gravity 133 4.1.3 Effect of Wick Permeability 135 4.1.4 Effect of Wick Diameter 137 4.1.5 Effect of Ambient Oxygen 137 4.1.6 Validation of results 138 CHAPTER 6: CONCLUSION 196
Recommendation for Future Work 198
BIBLIOGRAPHY 201
iv
LIST OF TABLES
Table 2.1 Porous Wick Dimensionless variables 52
Table 2.2 Table 2.2 Porous Wick Numerical Values 53
Table 3.1: Correlating equations of specific heats forO2 , N 2 ,CO2 , H 2O , and fuel. 107
Table 3.2: Gas phase property values. 108
Table 3.3: Nondimensional parameters. 109
Table 3.4: Non-dimensional governing differential equations. 110
Table 3.5: The S4 quadrature sets for axisymmetric cylindrical enclosures. 111
Table 3.6: Least-square fitting equations of Planck mean absorption coefficient for CO2 and H2O. 112
Table 4.1(a) : Effect of wick diameter on candle flame characteristics (5mm candle diameter and 21% O2) . 144
Table 4.1(b) : Effect of wick diameter on candle flame characteristics (5mm candle diameter and 21% O2) . 144
Table 4.2: Effect of wick diameter (Alsairafi, 2003) on candle flame characteristics (5mm wick length, 5mm candle diameter and 21% O2) 145
Table 4.3: Comparison with candle flame experiments in high gravity levels. 145
v
LIST OF FIGURES
Figure 1.1: Schematic of a candle flame. 23
Figure 2.1: Comparison of function residuals vs. CPU times for Newton, modified Newton and Picard’s iterative techniques. 54
Figure 2.2: Physical description of an externally heated axisymmetric wick. 55
Figure 2.3: Computational grid of an externally heated axisymmetric wick. 56
Figure 2.4: Plot of (a) saturation profiles (b) non-dimensional temperature profiles and (c) non-dimensional temperature profiles (expanded in the two-phase region) inside the porous wick for parameters shown in Table 2.2. 57
Figure 2.5: Plot of non-dimensional pressure contours: liquid pressure (top) capillary pressure (middle) and gas pressure (bottom) inside the porous wick for parameters shown in Table 2.2. 58
Figure 2.6: Plot of (a) liquid mass flux vectors (b) vapor mass flux vectors and (c) vapor mass flux vectors (expanded near the tip of the wick) inside the porous wick for parameters shown in Table 2.2. 59
Figure 2.7: Plot of (a) liquid convective heat flux vectors (b) vapor convective heat flux vectors and (c) conductive heat flux vectors inside the porous wick for parameters shown in Table 2.2. 60
Figure 2.8: Plot of saturation and temperature variation along the cylindrical surface of the wick exposed to the heat flux for parameters shown in Table 2.2 61
Figure 2.9: Plot of liquid and vapor mass flux (in r-direction) variation along the cylindrical surface of the wick exposed to the heat flux for parameters shown in Table 2.2 62
Figure 2.10: Plot of liquid mass flux (in x-direction) variation along the cylindrical surface of the wick exposed to the heat flux for parameters shown in Table 2.2 63
Figure 2.11: Plot of saturation and temperature variation along the axis of the wick for parameters shown in Table 2.2 64
Figure 2.12: Plot of liquid and vapor mass flux (in x-direction) variation along the axis of the wick for parameters shown in Table 2.2 65
vi
Figure 2.13: Comparison of (a) saturation profiles (b) pressure profiles and (c) temperature profiles for three different meshes (80x40, 80x80, 160x40) inside the porous wick for parameters shown in Table 2.2. 66
Figure 2.14 The variation of saturation at the cylindrical tip of the wick surface with the total heat supplied to the wick. 67
Figure 2.15 The variation of total mass of wax evaporated from the wick surface with the total heat supplied to the wick. 68
Figure 2.16 The variation of percentage heat that is lost to the reservoir with the total heat supplied to the wick. 69
Figure 2.17 The effect of gravity on the variation of saturation at the cylindrical tip of the wick surface with the total heat supplied to the wick. 70
Figure 2.18 The effect of gravity on the variation of total mass evaporated from the wick surface with the total heat supplied to the wick. 71
Figure 2.19 The effect of absolute permeability on the variation of saturation at the cylindrical tip of the wick surface with the total heat supplied to the wick. 72
Figure 2.20 The effect of absolute permeability on the variation of total mass evaporated from the wick surface with the total heat supplied to the wick. 73
Figure 3.1 Schematic of a candle 113
Figure 3.2: Variable grid structure for modeling candle flames for 1mm wick diameter and 5mm candle diameter 114
Figure 3.3: Schematic of radiation intensity transfer energy balance on arbitrary control volume in a participating medium. 115
Figure 3.4: Geometry and coordinate system for 2D axisymmetric cylindrical enclosure. 116
Figure 3.5: Projection of an S4 quadrature set on the μ,ξ plane using p,q numbering in r-x geometry. 117
Figure 3.6: Four types of space angle sweep direction for SN scheme. 118
Figure 3.7: Solid angle discretization of the S4 quadrature. 119
Figure 4.1: Plot of (a) pressure profiles (b) saturation profiles and (c) temperature profiles inside the porous wick coupled to a candle flame at normal gravity 146
vii
Figure 4.2: Plot of non-dimensional pressure contours: liquid pressure (top) capillary pressure (middle) and gas pressure (bottom) inside the porous wick coupled to a candle flame at normal gravity 147
Figure 4.3: Plot of (a) liquid mass flux vectors (b) vapor mass flux vectors and (c) vapor mass flux vectors (expanded near the tip of the wick) inside the porous wick coupled to a candle flame at normal gravity. 148
Figure 4.4: Plot of net heat flux supplied by the candle flame along the cylindrical surface of the wick 149
Figure 4.5: Plot of saturation and temperature variation along the cylindrical surface of the wick coupled to a candle flame at normal gravity. 150
Figure 4.6: Plot of liquid and vapor mass flux (in r-direction) variation along the cylindrical surface of the wick coupled to a candle flame at normal gravity. 151
Figure 4.7: Plot of liquid mass flux (in x-direction) variation along the cylindrical surface of the wick coupled to a candle flame at normal gravity. 152
Figure 4.8: Plot of saturation and temperature variation along the axis of the wick coupled to a candle flame at normal gravity. 153
Figure 4.9: Plot of liquid and vapor mass flux (in x-direction) variation along the axis of the wick coupled to a candle flame at normal gravity. 154
Figure 4.10: Gas temperature contours (non-dimensinalized by T∞ = 300 K ) 155
Figure 4.11: Fuel reaction rate contours (g cm-3 s-1) 155
Figure 4.12: Fuel and oxygen mass fraction contours 156
Figure 4.13: Local fuel/oxygen equivalence ratio contours 156
Figure 4.14: Carbon dioxide mass fraction contours 157
Figure 4.15: Water vapor mass fraction contours 157
Figure 4.16 Oxygen mass flux and flow field around the flame 158
p − p Figure 4.17: Isobar contours; p = ∞ 159 ρ U 2 r r
Figure 4.18: Profiles of temperature and species concentration along the symmetry line 160
viii
Figure 4.19: Effective Mean absorption coefficient distribution (cm-1 atm-1) 161
Figure 4.20: Dimensional net radiative flux vectors (W/cm2) 161
Figure 4.21: Contours of divergence of radiative heat flux (W/cm3) 162
Figure 4.22: Heat fluxes on the candle wick surface per unit radian 163
Figure 4.23: Plot of (a) saturation profiles (b) non-dimensional temperature profiles and (c) non-dimensional temperature profiles (expanded in the two-phase region) inside the porous wick for a self trimmed candle flame at normal gravity 164
Figure 4.24: Plot of non-dimensional pressure contours: liquid pressure (top) capillary pressure (middle) and gas pressure (bottom) inside the porous wick for a self trimmed candle flame at normal gravity 165
Figure 4.25: Plot of (a) liquid mass flux vectors (b) vapor mass flux vectors inside the porous wick for a self trimmed candle flame at normal gravity 166
Figure 4.26: Plot of net heat flux supplied by the candle flame along the cylindrical surface of the wick 167
Figure 4.27: Plot of saturation and temperature variation along the cylindrical surface of the wick for a self trimmed candle flame at normal gravity 168
Figure 4.28: Plot of liquid and vapor mass flux (in r-direction) variation along the cylindrical surface of the wick for a self trimmed candle flame at normal gravity 169
Figure 4.29: Plot of liquid mass flux (in x-direction) variation along the cylindrical surface of the wick for a self trimmed candle flame at normal gravity 170
Figure 4.30: Plot of saturation and temperature variation along the axis of the wick for a self trimmed candle flame at normal gravity. 171
Figure 4.31: Plot of liquid and vapor mass flux (in x-direction) variation along the axis of the wick for a self trimmed candle flame at normal gravity. 172
Figure 4.32: Gas temperature contours for a self trimmed candle flame (non- dimensinalized by T∞ = 300 K ) 173
Figure 4.33: Fuel reaction rate contours for a self trimmed candle flame (g cm-3 s-1) 173
Figure 4.34: Fuel and oxygen mass fraction contours for a self trimmed candle flame 174
ix Figure 4.35: Local fuel/oxygen equivalence ratio contours for a self trimmed candle flame 174
Figure 4.36: Carbon dioxide mass fraction contours for a self trimmed candle flame 175
Figure 4.37: Water vapor mass fraction contours for a self trimmed candle flame 175
Figure 4.38: Oxygen mass flux and flow field around the self trimmed candle flame 176
p − p Figure 4.39: Isobar contours for a self trimmed candle flame; p = ∞ 177 ρ U 2 r r
Figure 4.40: Profiles of flame structure at the centerline for a self trimmed candle flame 178
Figure 4.41: Effective mean absorption coefficient distribution for a self trimmed candle flame (cm-1 atm-1) 178
Figure 4.42: Dimensional net radiative flux vectors for a self trimmed candle flame (W/cm2) 179
Figure 4.43: Contours of divergence of radiative heat flux for a self trimmed candle flame (W/cm3) 179
Figure 4.44: Heat fluxes on the candle wick surface per unit radian 180
Figure 4.45: Candle flames at various gravity levels for a self trimmed candle flame 181
Figure 4.46: Candle burning rate and self trimmed length of the candle (on log scale) at various gravity levels for a self trimmed candle flame 182
Figure 4.47: Candle burning rate and self trimmed length of the candle (on normal scale) at various gravity levels for a self trimmed candle flame 183
Figure 4.48: Self trimming length of the candle wick using different wick permeabilities. 184
Figure 4.49: Candle burning rates for different wick permeabilities. 185
Figure 4.50: Candle flame lengths for different wick permeabilities. 186
Figure 4.51: Maximum gas phase temperatures for different wick permeabilities. 187
Figure 4.52: Self trimming length of the candle wick for different wick diameters. 188
Figure 4.53: Candle burning rates for different wick diameters. 189
x
Figure 4.54: Maximum gas phase temperatures for different wick diameters. 190
Figure 4.55: Maximum gas phase temperature at various oxygen molar fractions 191
Figure 4.56: Self trimmed length of the candle wick at various oxygen molar fractions 192
Figure 4.57: Candle burning rate at various oxygen molar fractions 193
Figure 4.58: Fuel reaction rate contours (5.e-5 g/cm3s)at various oxygen molar fractions. 194
Figure 4.59: Comparison of flame length and flame widths at various gravity levels with the experiments (Arai and Amagai, 1993) 195
xi ACKNOWLEDGEMENTS
I would like to express my gratitude to Professor James S. T’ien for the valuable guidance, help and encouragement provided during the course of this work. He has been very patient in guiding me all along the research. His comments and suggestions have been thought provoking. It gave a lot of insight into the problem. His encouraging spirit and friendly disposition has been very inspiring.
I am grateful to Professor Krishnan V. Pagalthivarthi (IIT Delhi) for his selfless motivation in encouraging me to pursue my studies in Case Western Reserve University.
Learning Computational Fluid Dynamics under him was a rich experience for me. The credit to my strength in Fluid Mechanics and in Numerical programming goes to him.
His example has been a motivating force for taking up a career in research.
I wish to acknowledge Dr Ammar, for providing me with his candle flame gas- phase code and the relevant materials. He has been extending to me throughout my research through e-mails. His timely clarifications have helped me to smoothly progress in my work. I am thankful to Dr Amit Kumar for his cooperation and help in practical matters. I also thank my colleagues I. Feier, Sheng-Yen Hsu, Ya-Ting Tseng, K. Tolejko,
G. Mittal, A. Garg, Ravikumar, B. Han, and K. Kumar with whom I had useful discussions. I also thank Prof. Sung, Prof. Edward White, Prof. Saylor Beverly and Dr
Daniel Dietrich for serving as my examination committee members and for their valuable time and suggestions. I would also like to thank staff of the department: J. A. Stiggers, A.
Szakacs, , C. Wilson, S. Campbell and M. Marietta for their prompt help at various stages.
xii Finally I would also like to acknowledge NASA for supporting the work under the technical monitoring of Dr Daniel Dietrich.
xiii NOMENCLATURE
Notation
Gas phase parameters a Coefficient in the discretised equation
A Special function for power-law scheme
A Area (cm2)
2 Ae East surface area of a control volume (cm )
2 An North surface area of a control volume (cm )
2 As South surface area of a control volume (cm )
2 Aw West surface area of a control volume (cm ) b Constant term in the discretised equation
3 -1 -1 Bg Gas phase pre-exponential factor (cm g s ) c Speed of sound (cm s-1)
C Correction factor for the Planck mean absorption coefficient
-1 -1 Cp Dimensional gas phase specific heat (cal g K )
-1 -1 C p,r Reference gas phase specific heat (cal g K ) d geometric dimensionality (1, 2, or 3)
D Flame diameter (cm)
Dw Wick diameter (cm)
2 -1 Di Diffusion coefficient of species i (cm s )
eˆθ Unit vector in the polar direction
eˆψ Unit vector in the azimuthal direction
xiv E Gas phase activation energy (cal gmol-1)
Er Radiant energy flux fi Stoichiometric mass ratio of species i/fuel fr Weighting factor in r-direction fx Weighting factor in x-direction g Gravitational acceleration (cm s-2)
-2 ge Gravitational acceleration on Earth (cm s )
G Incident radiation (cal cm-2 s-1)
-1 hi Enthalpy of species i (cal g )
H Flame height (cm)
I Radiation intensity (cal cm-2 s-1)
J Flux influencing a dependent variable φ
L Latent heat (cal g-1)
Lr Reference length (cm)
Lei Lewis number of species i m Average burning mass fluxes (g cm-2 s-1)
-1 mt Total burning rate (mg s )
M Total number of different directions for SN scheme
-1 MWi Molecular weight of species i (g gmole ) G n Outward normal to the surface
N Order of discrete ordinates scheme p Dynamic pressure (atm)
P Pressure (atm)
xv P∞ Ambient pressure (atm)
-2 qc Conductive heat flux (W cm )
-2 qr Radiative heat flux (W cm )
r -2 qr Radiative heat flux in r-direction (W cm )
x -2 qr Radiative heat flux in x-direction (W cm )
y -2 qr Radiative heat flux in y-direction (W cm )
Q Heat of combustion (cal g-1) r Dimensional r-coordinate (cm) G r Position vector
-1 -1 Ru Universal gas constant (cal gmol K )
S Source terms
SN Discrete ordinates scheme of an order N
T Temperature (K)
Tb Boiling temperature of fuel (K)
Tm Melting temperature of fuel (K)
T∞ Ambient temperature (K)
Tr Reference temperature (K) u Axial velocity (cm s-1)
-1 UB Buoyant velocity (cm s )
-1 UD Diffusion reference velocity (cm s )
-1 Ur Reference velocity (cm s ) v Radial velocity (cm s-1)
V Volume of a control volume
xvi w Quadrature weights
-3 -1 Wi Sink or source term of species i (g cm s ) x,X Dimensional x-coordinate (cm)
Xi Mole fraction of species i (%) y Dimensional y-coordinate (cm)
Yi Mass fraction of species i (%)
α Absorptivity
2 -1 αr Reference thermal diffusivity (cm s )
β Extinction coefficient (cm-1)
γ Reflectivity
o -1 Δh f ,i Heat of formation of species i (cal g )
ε Solid emissivity
η Direction cosine
φ Dependent variable in governing equations
Φ Scattering phase function
κ Absorption coefficient (cm-1)
Γ General diffusion coefficient in the discretised equations
λ Thermal conductivity (cal cm-1 s-1 K-1)
-1 -1 -1 λr Reference thermal conductivity (cal cm s K )
μ Gas viscosity (g cm-1 s-1)
μ Direction cosine
-1 -1 μr Reference gas viscosity (g cm s )
ν Frequency
xvii
θ Polar angle
ρ Dimensional gas density (g cm-3) -3 ρr Reference gas density (g cm )
-3 ρ∞ Ambient gas density (g cm )
σ Stefan-Boltzmann constant (cal cm-2 s-1 K-4)
σs Scattering coefficient
ω Weighting factor
ξ Direction cosine
ψ Azimuthal angle
ζ Expansion factor
G Ω Solid angle in terms of ordinate direction (ξ,μ,η) (sr)
Subscripts b Black body
B Buoyant
E Neighbor in the positive x-direction on the east side f Flame i Species i max maximum min minimum
N Neighbor in the positive y-direction on the north side
P Central grid point r Reference
xviii S Neighbor in the negative y-direction on the south side w Value at wall
W Neighbor in the negative x-direction on the west side x Axial direction
Superscripts
° Old value of a variable
+ Positive component
- Negative component
Non-dimensional quantity
′ Incoming x x-direction r r-direction p p-level of μ direction cosine q q-level of ξ direction cosine
Special symbol a Absolute value of a
a1 ,a2 ,a3 ,... Largest of a1, a2, a3, …
Dimensionless numbers
Bo Boltzmann number
Da Damköhler number
Gr Grashof number
Pe Peclet number
xix
Pr Prandtl number
Ra Raleigh number
Re Reynolds number
Porous wick parameters c specific heat capacity (J/kg-K) ifg latent heat of evaporation (J/kg)
G Gibb’s phase function h enthalpy k thermal conductivity, (W/m-K)
K permeability, (m2) kr relative permeability
Mw molecular weight, (kg/kg-mole) p pressure, (N/m2)
Pe peclet number
R gas constant, (J/kg-K) s saturation
T temperature, (K) u,v velocity, (m/s) x, X Dimensional x-coordinate
α thermal diffusivity, (m2/s)
ε porosity
μ viscosity, (kg/m-s)
ρ density, (kg/m3)
xx σ surface tension, (N/m)
ζ triplet, ()pl ,,sT
σ surface tension, (N/m)
χ thermodynamic state parameter
η entropy
ζ triplet, ()pl ,,sT
Subscripts l liquid g vapor r relative c capillary eff effective m melting point
0 ambient, standard atmospheric conditions s solid
xxi
Heat and Mass Transport Inside a Candle Wick
Abstract
by
MANDHAPATI PADMANABHA RAJU
The purpose of this study is to investigate the effect of heat and mass transfer inside the porous wick on candle flame combustion. The phenomenon of self trimming that is observed in candle flame, whose wick is made of combustible material, is also analyzed. This is accomplished by modeling two-phase flow inside an axisymmetric wick and coupled with a gas-phase candle flame.
Gas-phase model has been taken from Alsairafi (2003). A finite volume method is used to solve the steady mass, momentum (Navier-Stokes), species, and energy equations, and the radiative transfer equation. The gas phase combustion process is modeled by a single-step, second-order, finite rate Arrhenius reaction. The discrete ordinates method is used to solve the radiative heat transfer with mean absorption coefficients. Flame radiation is only from CO2 and H2O, the products of combustion. The wick is modeled using volume averaged equations for two-phase flow inside an axisymmetric wick. Momentum equations for liquid and vapor flow are governed by
Darcy’s law. The liquid is drawn to the surface by capillary action and the vapor is driven in the two-phase region by vapor pressure gradient governed by thermodynamic equilibrium relations.
Before coupling with the gas-phase flame, a decoupled numerical computation of the wick phase is performed to isolate the effects of porous media parameters. This is
xxii done by applying a prescribed constant heat flux along all the exposed surfaces. The results show that the porous wick parameters like absolute permeability play an important role in determining the saturation distribution inside the wick. For the range of the wick parameters chosen for this study, gravity does not play an important role.
The next part of the study involved coupling the candle wick to the candle flame.
For candles, whose wicks are made of combustible materials, self trimming is being modeled as the burn out of the dry region. The computed results show also that for a self trimmed candle flame, the wick permeability plays an important role in determining the candle flame structure. The wick permeability affects the self trimming length of the wick and this affects the flame structure.
The effect of gravity on the self trimmed candle flame is analyzed. Trimmed wick lengths are computed as a function of gravity. The results show that the burning rate increases continuously from 0ge to 2ge and then decreases. At high gravity levels wake flame is observed with a sudden decrease in burning rate.
Calculations have also been performed for different wick diameters. Maximum flame temperature, burning rate, and flame dimensions are computed. A parametric study using different molar oxygen percentage has been made at different gravity levels.
xxiii 1
CHAPTER 1 INTRODUCTION
Combustion has been the subject of many researchers’ interest because it provides
the majority of useful energy production in residential, commercial and industrial devices. The flame, a thin zone of intense chemical reaction undergoing the process of
combustion, generally consists of a mixture of fuel and oxidizer (i.e. oxygen or air). It is
well known that flames are categorized as being either premixed flames or non-premixed
(diffusion) flames. In a premixed flame, the fuel and the oxidizer are mixed at the molecular level prior to any significant chemical reaction. In a diffusion flame, the reactants are initially separated, and reaction occurs only in the mixing zone between the fuel and the oxidizer. A typical diffusion flame is that of a burning candle, (see figure
1.1).
Candle flame studies contribute to the understanding of many fundamental
aspects like chemical reactions, diffusion processes, porous wick transport as well as their
coupled effects. Although the use of candles is common, not all the physical and chemical aspects of candle burning are sufficiently understood. Before going into the scientific aspects of a candle flame, some basic information on candles is provided.
1.1. CANDLE BASICS
The following material has been taken from the website
http://candles.genwax.com/candle_instructions/___0___how_wick.htm .
The body of a candle is comprised of a solid fuel source, usually paraffin wax. A
wick runs through the center of the body of the candle from the bottom, extending out of 2
the top. The wick, which acts as a fuel pump when the candle is burning, is generally
made of cotton fibers that have been braided together.
Candle Wax
There are two main waxes used in candle making, Paraffin Wax and Beeswax.
Paraffin wax, which is classified as a natural wax, is the most common wax used in candle making, and can be said to ultimately come from plant life.
In order to protect themselves from adverse weather conditions plants produce a
layer of wax on their leaves and stems. Material from dead plants 100-700 million years
ago accumulated in large quantities and eventually became buried beneath the surface of
the earth. After a long period of time, forces of heat and pressure turned the slowly
decaying plant material into crude oil, otherwise known as petroleum. Because of the
nature of waxes, being inert and water repellent, they were unaffected by the
decomposition of the plant material and remained intact, suspended within the crude oil.
Petroleum companies "harvest" the crude oil and process it. They refine the oil,
separating the different properties into gasoline, kerosene, lubrication oil, and many other
products. In many cases, the wax in the petroleum is considered undesirable and is
refined out. The refinery will process the wax into a clean, clear liquid, or as a solid
milky white block, and make it available to companies who may have a use for it. The
refined wax is called paraffin, which comes from the Latin "parum = few or without" and
"affinis = connection or attraction (affinity)". Basically there are few substances that will
chemically react with or bind to this type of wax. 3
A less common but more highly renowned wax for candle making is beeswax.
Classified as a natural wax, it is produced by the honeybee for use in the manufacture of
honeycombs. Beeswax is actually a refinement of honey. A female worker bee eats
honey, and her body converts the sugar in the honey into wax. The wax is expelled from
the bee's body in the form of scales beneath her abdomen. The bee will remove a wax scale and chew it up, mixing it with saliva, to soften it and make it pliable enough to work with, then attach it to the comb which is being constructed. Usually another bee
will take the piece of wax which has just been attached to the comb, chew it some more,
adding more saliva to it, and deposit it on another section of the comb. The combs are
built up, honey is deposited inside, and then the combs are capped with more wax. Since
several worker bees construct the comb at the same time, and the hive is constantly active
with other bees flying around and walking on the combs, depositing foreign matter onto
the combs, the composition of the wax becomes very complex.
Candle Wick
A candle without a wick is just a hunk of wax. The wick is what a candle is all
about. The earliest known candles were basically a wick-like material coated with tallow
or beeswax, not even resembling a candle at all. In taper candles the wick is the structure which supports the first layers of wax that create the candle. In all candles it acts as a
fuel pump, supplying liquefied wax up to the top where all of the action takes place. As a
regulator, different size wicks allow different amounts of wax up into the combustion
area providing different size flames. The wick is pretty much the most important element
of a candle. 4
The word wick comes from Old English "weyke or wicke", Anglo Saxon
"wecca", and Germanic "wieche or wicke". It is a name for a bundle of fibers that when braided or twisted together are used to draw oil or wax up into a flame to be burned in a lamp or candle.
A wick without wax around it is just a piece of string. Because the wick is fibrous and absorbent, melted wax adheres to it easily. Dipping a wick in and out of melted wax several times builds up layers of wax, sufficient enough to make a taper candle. The wick works by a principle called capillary action. Cotton fibers are spun into threads, which are bundled and braided together. The spaces between the cotton fibers, the threads, and braids act as capillaries, which cause liquids to be drawn into them. If you place a drop of water in the center of a paper towel you will see that the drop is absorbed and the wet spot expands. Where the expansion occurs is where capillary action is taking place, the candle wicking absorbs wax the same way.
Candle wicking is available in several types. Probably the most popular is the
Flat Braid, or Regular wick. Different sized wicks cause different sized flames simply because of the number of threads in the bundles. Each thread is considered a plait or ply, and a given number of ply are bundled together.
1.2. CANDLE BURNING
When a candle is lighted, the heat from the ignition source melts the wax, a heavy hydrocarbon, at the wick base. The liquid wax rises up, due to capillary action, and is then vaporized by the heat. This vaporizing wax cools the exposed wick and protects parts of it from burning out. The vaporized wax mixes with oxygen and the mixture reacts and generates heat and the process continues. 5
The heat and mass transfer taking place inside the wick is very complex. Two- phase flow regimes will exist inside the wick. All the three regimes – single phase liquid region, two-phase liquid vapor region and single phase vapor region can possibly exist inside the wick. The heat and mass transfer taking place in the two-phase region of a wick is very involved. It is of practical importance in many applications. Heat pipes work on the principle that the heat transfer coefficient is significantly reduced inside the two-phase flow in a wick. So heat pipes are used as efficient cooling devices.
In the gas-phase, heat is released from the combustion process. The hot combustion products are much less dense than the colder ambient air. In a gravity field, they rise upward and draw the oxygen to the reaction zone. This upward buoyant convective flow is the main reason that makes the flame into the so-called “tear-drop” shape.
In zero gravity environments, on the other hand, natural convection is not present and the fuel and oxidizer need to diffuse towards the reaction zone by the mechanism of molecular diffusion. The primary reason for this process to occur is the existence of concentration gradients around the flame. It is worthy to note that the diffusion transport rates in zero gravity conditions are much slower than the natural convection transport rates in normal gravity because of the absence of buoyancy-driven convection. Zero gravity flames are much less robust (in the sense of smaller reaction rate) than normal gravity flames because of the absence of the buoyant flow.
The mixing of fuel and oxidizer in the presence of a high temperature gradient means that both heat and mass transfer must be considered in addition to chemical kinetics to properly understand combustion. The processes are coupled and the 6
governing equations are nonlinear in nature. Numerical computation has become an
important tool to understand combustion phenomena. For the candle flame, it is
important to understand how the system responds to changes in parameters such as
gravity level. Interest in the candle burning has been revived in recent years. The driving force for this interest is due to the necessity of understanding flames in low gravity environment. The research conducted on the candle flame provides valuable insights on
how flames behave in microgravity which is relevant to spacecraft fire safety.
This research is mainly intended to provide a more complete simulation of a
candle flame coupled to a porous wick. The heat and mass transfer inside the wick can
control the flame shape and structure, and the extinction of flame at certain conditions.
When a candle burns, it slowly consumes the wax from the wax shoulder. As the level of wax comes down, the length of the exposed wick is increased. This process is generally very slow compared with the processes occurring in the gas and porous phases. Therefore in modeling the candle burning, a quasi steady approximation can be assumed with wick length treated as a parameter. As the length of the exposed wick increases, the surface exposed to the heat of candle flame increases and on the other hand, more capillary action is required for the liquid wax to reach the surface of the wick. At some point of time, the tip of the wick dries up. This causes the temperature of the wick to increase sharply at the tip of the wick. If the wick is made of pyrolyzable materials, then the dry portion of the wick will be consumed (burnt out). This phenomenon is referred to as the self trimming of the wick. A self trimming candle regulates its wick length by a balance of
the above two processes. 7
1.3. PREVIOUS WORK
In this section, previous works on the study of candle flame and the study two- phase flows inside porous media are reviewed. The first part will cover the candle flame studies and the second part will cover the porous media studies.
1.3.1 PREVIOUS WORK ON CANDLE FLAMES
1.3.1.1 EXPERIMENTAL WORK
Candles have been a focus of attention of scientific study for hundreds of years.
In the 19th century, English scientist Michael Faraday, who discovered many principles of
electricity, delivered one of the most famous of his Christmas lectures for children called
“The Natural History of a Candle” (Faraday, 1988). His observations have served as the basis for lessons in taking observations in combustion.
Although the processes that occur in a candle flame are complex, the setup of a
candle experiment is very easy. The candle flame is an excellent example of wick stabilized diffusion flame. Many researchers have chosen candle flame to understand a
wide range of combustion phenomena. For example, an early work by Lawton and
Weinberg (1969) examined the effect of magnetic fields on flame deflection using a candle. Chan and T’ien (1978) performed experimental work on commercial candles to study the spontaneous flame oscillation phenomena. Buckmaster and Peters (1986) and
Maxworthy (1999) have both studied stability and flickering phenomena for diffusion flames using a candle for a model diffusion flame.
The influence of hyper-gravity (gravity greater than that on earth) on candle
diffusion flames has been reported by Villermaux and Durox (1992) using a 6 meter
diameter centrifuge. They were able to vary the gravity up to seven times the gravity of 8
earth. Above 7ge, the candle flame becomes extinct. They found that the flame length and the candle burning rate both decreases as the gravity level increases. Another experimental investigation to study the behavior of candle flames in high gravitational field has been reported by Arai and Amagai (1993). They varied the gravity level by using a spin tester in the range from 1 to 14ge. There results showed that both the candle flame length and width were monotonically reduced with increasing gravity. In both the works there is no mention about the wick length of the candle and whether it reached the self trimming length.
Bryant (1995) investigated the effects of gravity level on the heat release rate for a candle flame under an imposed low-speed forced flow. The experiment was performed on board an aircraft flying repeated parabolic trajectories. He reported that the rate of heat release is flow rate dependent, decreases significantly in microgravity, and changes insignificantly under elevated-gravity conditions. Oostra et al. (1996) measured the soot production of a candle during microgravity and normal gravity condition during a parabolic flight. Their measurements predict a lower candle burning rate with higher soot production in microgravity. Amagai et al. (1997) investigated the effect of variable gravity using a gaseous butane diffusion flame issued from a tube in a centrifuge with a rotary arm of radius 0.9 meter. In their experimental work, they found that both the length and the width of butane diffusion flames reduced with an increase of gravity level for a constant feed rate. In addition, oscillations and blow off of the flame were observed at high gravity level.
Recent microgravity works sponsored by NASA have been a major source for fundamentally improving the science of combustion. Microgravity experimental data on 9
candle combustion has been obtained by several different techniques, all of which depend
on the experiment achieving a state of free-fall. A number of studies have used drop
towers where a few seconds of free-fall condition can be generated, i.e. between 2 to 5
seconds. Such facilities make it possible to conduct brief periods of microgravity research on earth. An experimental investigation of candle flame ignition behavior in drop tower has been reported by Ross et al. (1991). The experiments conducted at atmospheric pressure, under 19%-25% O2 concentration, and in nitrogen- or helium-
diluted environments. They found that the visible blue candle flame assumes a
hemisphere shape relatively quickly (about 1 second) after the drop. Although the blue
flame shape and size respond to the ignition or g-variation transient quickly, the color of
the flame continuously changes throughout the entire test duration. Thus it suggests that
more time is required to obtain the true steady state.
Besides the drop towers, microgravity research also took place using parabolic
aircraft maneuver which provide additional time of reduced-gravity environment.
-2 However, the g-jitter in the aircraft is too high (∼10 ge) to obtain a steady flame. The
candle flame fluctuates quite extensively in the airplane experiments. Using Space
Shuttle to obtain extended time in buoyant and weakly buoyant atmospheres, Dietrich et
al. (1994) investigated the candle flames behavior in microgravity. The experimental
results showed that the candle flame was spherical and bright yellow, presumably from
soot, immediately after ignition. After a few seconds, the yellow disappeared and the
flame became blue and nearly hemispherical with a large stand-off distance from the
wick. The flame luminosity decreased continuously with time until extinction. The
extinction typically occurred around one minute. They also reported that the mass of 10
liquefied wax grew continuously without dripping off, contrary to a candle burning in normal gravity. It has been observed, however, while the flame behavior was quasi-
steady during the majority of burning time, axisymmetric flame oscillations developed
near extinction due to oxygen depletion in the finite size combustion chamber. The term
“quasi-steady” needs to be explained. Physically, the flame size response is related to the
diffusion time scale (order of seconds) which is much lower than the depletion time scale of oxygen (order of minutes) in the experimental chamber. This means that the flame size and shape will respond in a quasi-steady manner to the oxygen molar fraction of the ambient even though this oxygen molar concentration is changing with time.
A second series of flight experiment on microgravity candle flame was carried out
in the Mir Station. In the Mir station the oxygen mole fraction is higher than that in the
Shuttle (between 0.23 to 0.25). Also the candle cage has more open area to facilitate oxygen diffusion from the ambient to the flame. Dietrich et al. (2000) showed that the candles onboard Mir with the largest wick diameters had the shortest flame lifetime where as the candles with the smallest wick diameters had the largest lifetimes. All of the candles in the Mir tests burned longer than those on Shuttle. The above examples indicate that candle flames onboard the space shuttle and Mir both oscillated prior to extinction with periodic increases and decreases of flame surface area.
1.3.1.2 NUMERICAL WORK
In addition to experimental effort, limited numerical works to model candle
flames have been conducted, all in zero gravity. To the best knowledge of the author,
there are two different candle flame numerical models. The first model is by Shu (1998),
Shu et al. (1998) and the other model is by Chang (see Dietrich et al., 2001). These two 11
different numerical models predict some candle flame behavior in zero gravity
environments. Both calculations utilize a single-step, finite rate, gas-phase chemical
reaction model in a frictionless flow. The main difference between the two models is the
candle and wick geometry. In Shu’s model, a spherical wick and conical candle body
were used and the equations were formulated using spherical coordinates. The model by
Chang used a more realistic wick and candle geometry and the equations were formulated
using cylindrical coordinates. Both models included a wick surface radiation loss term
and flame radiation is represented as a pure heat loss.
The work by Shu (1998) and Shu et al. (1998) were primarily focused on
computing the flame characteristics, the flammability limits, the effects of flame radiative
loss, and the effect of fuel and oxygen Lewis numbers. The model predicts a stable
steady-state of candle flame in an infinite ambient. They also reported that decreasing the oxygen concentration, prior to extinction, raises the flame base relative to the porous sphere without significant changes in the position of the flame top. Unlike oxygen Lewis
number, the numerical experiments reveal no flame properties dependence on the fuel
Lewis number except for the fuel vapor profile. Their model also predicts near-flame
oscillation similar to those observed in the experiments but there were some differences.
The modeling work by Chang (Dietrich et al., 2001), used the grid generation
technique and a body-fitted coordinate system to fit the more realistic wick and candle
shape. Instead of the explicit scheme used by Shu (1998), an implicit time-marching
scheme was developed and it shortened the computational time needed to reach steady
state and made it possible to examine the transient extinction phenomena near the limit.
There are two shortcomings (or limitations) on the candle modeling work by Shu and 12
Chang. The first is the assumption of potential flow. This assumption simplifies the
momentum equations but has the shortcoming of not being able to satisfy the no-slip
boundary condition on the solid surface. It also cannot handle buoyant induced flow so
gravity effect can not be studied. The second limitation is on the treatment of radiation.
In these models, flame (gas-phase) radiation is treated as a simple heat loss term without
knowing the detailed distribution of radiation heat flux. In previous modeling work on
solid fuel combustion (e.g. Rhatigan et al., 1998), part of the flame radiation becomes the
energy feedback to the solid pyrolysis process. A more rigorous treatment of radiative
heat transfer is thus needed to verify whether the simple heat loss model is adequate or
not.
A more detailed gas-phase modeling has been done by Alsairafi (2003). A
simplified two-dimensional axisymmetric flow around a realistic candle has been
numerically simulated. A finite volume method is used to solve the steady state Navier
stokes equations in conjunction with species and energy equation, and the radiative
transfer equation. The gas phase combustion process is modeled as a single step, second
order, finite rate Arrhenius reaction. The discrete ordinate method is used to solve the radiative transfer equation. The main assumption used in this work is the candle wick is
assumed to be a solid coated with liquid fuel all along its surface. This evades the
necessity to model the heat and mass transport inside the candle wick and its coupling
effect on the candle flame. They reported a significant match between their simulated
results and the experimental results. However, there is a major discrepancy even in the
qualitative trends of the candle burning rate with the imposed gravity level. Experiments
report a monotonic decrease in the burning rate with increasing gravity levels but the 13
numerical results indicate an increase in burning rate upto 3ge and a decrease in burning rate with further increase in gravity level. The studies by Alsairafi (2003) indicate the necessity to model the detailed heat and mass transport inside the wick to achieve a more realistic numerical simulation of candle burning.
1.3.2 PREVIOUS WORK ON TWO-PHASE FLOW IN POROUS MEDIA
Simulation of a porous wick involves modeling two phase flow with phase change inside the porous media. Two phase flow in porous media involves many features which distinguishes it from the single phase flow through a porous medium. The flow can be distinguished as either co-current or countercurrent flow depending on the direction of the flow of the two phases. In two phase flow through porous media, there are three possible saturation regimes that can exist. The porous media may be completely saturated with one phase. This is called the complete saturation regime. The porous media may have the lowest possible saturation with one phase. This is called the pendular regime. In this regime, one of the phases occurs in the form of pendular bodies throughout the porous media. These pendular bodies do not touch each other, so there is no possibility of flow for that phase. The porous media exhibits an intermediate saturation with both phases. This regime is known as funicular regime. Earlier it was thought that in this regime, both liquid and vapor phases flowed simultaneously through channels with gas moving in the inner core and the liquid in the annulus between the gas and the solid channel walls [Scheidegger, 1974]. Flow visualizations have shown that the gas and the liquid flow through their own network of channels [Dullien, 1979].
Two-phase flow in porous media is involved in the following classes of problems
1) Drying of porous materials 14
2) Heat pipe applications
3) Burning over porous wick surfaces
The drying problem is essentially transient in nature. It has been a subject of
interest since 1920’s. Lewis (1921) suggested that drying takes place primarily through a diffusion mechanism. Around this time, the soil scientists and chemists were attempting to explain the movement of moisture in porous media in terms of surface tension forces or by capillary action. Hougen et. al. (1940) made an extensive survey of the importance of capillary action in the drying process. They found out that capillary action can play an important role in the movement of moisture during the drying of a porous solid. Krischer
(1940) was the first to identify the importance of energy transport in a drying process.
Phillip and Devries (1957) included the effects of capillary flow and vapor transport, and
incorporated the thermal energy equation into the set of governing equations that describe
the drying process. Luikov (1975) also published similar equations for the heat and mass
transfer in porous media. Whitaker (1977) made a rigorous formulation of the theory of
drying based on the well known transport equations for a continuous media. These
equations were volume averaged to provide a rational route to a set of equations
describing the transport of heat and mass in porous media. These equations are limited by
the restrictions and assumptions that he had used while deriving the volume averaged
equations.
Even though Luikov (1975) published the full set of governing equations
describing the heat and mass transport in porous media, the equations are nonetheless
difficult to solve because of its complexities. A large number of parameters are required
to solve the equations. There have been efforts to study different aspects of the heat and 15
mass transfer by simplifying the equations using suitable assumptions. Initial efforts
involved consideration of only the vapor transport inside the porous media. The capillary
action on the liquid is neglected. Cross (1979) used the momentum and energy equations
for the vapor flow, neglecting the convection terms in the energy equation. He solved
analytically for the maximum pressure that is build up at the dry-wet interface. As drying
takes place, the evaporation takes place at the interface between the dry and the wet
regions. A high pressure is build up at this interface to generate the necessary driving
force for the transport of vapor in the dry region. Dayan (1981) extended this to include
the convective term in the energy equation. They have analyzed an intensely heated porous space using a transient model to obtain the transient pressure, temperature and moisture distributions. Dayan and Glueker (1982) analyzed the same problem using an explicit time marching technique and incorporating the liquid and vapor transport. They have assumed that the migration of liquid phase is primarily governed by pressure gradients generated by the vapor. They were able to get reasonable predictions for the temperature, pressure and moisture distribution during the drying of cement structure.
During intense heating, a region near the heated surface dries out. Once such a region develops, evaporation takes place exclusively at the dry-wet interface. The evaporation process leads to the pore pressurization and subsequent filtration of all the pore constituents towards both the heated surface and the inner wet zone of the concrete. In the above analysis, the effect of capillary pressure and gravity were neglected.
In drying, three different regions have been observed experimentally (Rahli et. al.,
1997) in the porous media. In the initial stages of drying, the temperature inside the
porous media is below the saturation temperature. Two regions are observed at this stage. 16
A two-phase region is observed near the heating surface and a liquid region is present deep inside the porous media. As drying proceeds, the surface exposed to heating becomes completely dried up and single phase vapor region is formed near the heating surface which penetrates deep into the media as time proceeds. The temperature in the single phase vapor region exceeds the saturation temperature. Three zones have been found to exist. Near the heated surface, a vapor saturated zone is observed. Adjacent to this zone and extending into the medium is a two phase region, which is dominated by capillarity and vapor transport. Ahead of this zone, the medium is saturated with the liquid phase.
In most cases, the capillary pressure and gravity play an important role in the two phase flow inside porous media. In the absence of any forced flow, the liquid is transported primarily due to capillary action and gravity forces. Kaviany and Mittal
(1987) analyzed the two phase flow inside porous media, where liquid is governed by capillary forces. They have analyzed the convective drying of a porous slab in the funicular state initially saturated with liquid. The liquid is driven by capillary pressure gradient and the vapor is driven by the partial pressure of the evaporating species generated by the temperature gradient established inside the porous slab. The transport of non-condensable gas and the effect of gravity are neglected. The calculations are performed only in the funicular regime, i.e. till the appearance of dry patches on the surface. They found good agreement between their experimental results and the predicted results for the drying rate, surface temperature and the average saturation up to the time of first appearance of dry patches on the surface (critical time). Rogers and Kaviany
(1991) extended this work to include the evaporative-penetration front. As the porous 17
slab is dried, the surface saturation decreases as time proceeds. After some period termed
as the critical time, dry patches appear on the surface. When dry patches appear on the
surface, the evaporative front starts penetrating into the porous slab. The effect of gravity
and the surface tension non-uniformities are included. They have also included the
transport of non-condensable gases into the porous slab. The speed of the evaporative
front and the mass transfer rate during this regime were predicted. A significant drop in
the drying rate is observed in the evaporative front regime. This is the result of the high
resistance to heat and mass transport in the dry region.
A heat pipe application is another major area which involves heat and mass
transport in the two-phase region of a porous substance. A heat pipe is a simple device
that can quickly transfer heat from one point to another. Heat pipe application is based on the high heat transfer rate obtained due to the evaporation-condensation mechanism
taking place inside the porous wick involving phase changes. Experimental studies
conducted by Hansen (1970) showed an enhancement of heat transfer over that of
conduction in saturated porous media. The experimental data obtained by Somerton et. al.
(1974) on effective thermal conductivity of steam-water saturated porous materials is
several times larger than those of the same medium saturated only with the liquid phase.
These classes of problems have been treated as steady state systems. In the heat
pipe applications, the porous wick is enclosed and the liquid and the vapor re-circulate
inside the porous wick. There is a counter-current transport of liquid and vapor in the
two-phase region. The liquid and the vapor regions above and below the two-phase
region are essentially stationary. A one-dimensional, steady state analytical model (heat
pipe problem) has been developed by Udell (1985) to study the effects of capillarity, 18
gravity and phase change. Their results predict the increase of heat transfer due to the
combined effect of evaporation, convection and condensation inside the porous media.
The heat pipe effect is also observed in two phase geothermal reservoirs. Bodvarsson
(1994) used a two dimensional porous slab model with a non-uniform heat flux at the bottom. Their results show very efficient heat transfer in the vapor dominated zone consistent with the observations in natural geothermal reservoirs.
Burning over porous wick surfaces also involves heat and mass transport in a
porous wick. The influence of capillarity on the combustion behavior and the effects of
the properties of porous beds on the combustion characteristics make this study unique.
Ignition and transition to the flame spread over the ground soaked with spilled
combustible liquid are of interest to researchers and engineers because they have
important implications in terms of fire safety. Kaviany and Tao (1988) had done some
experiments and numerical calculations on the burning of liquids supplied through a
wick. They have analyzed the burning of a porous slab which is initially saturated with
liquid fuel, during the funicular regime. The liquid is driven by capillarity and the vapor
is driven by the vapor pressure gradient. The effect of gravity and the transport of non-
condensable gases have been neglected. The effect of surface saturation, relative
permeability and vapor flow rate on critical time (time during which the surface become
first dried up) has been studied.
An experimental study was conducted by Kong et. al. (2002) to investigate the
effects of sand size and sand layer depth on the burning characteristics of non-spread
diffusion flames of liquid fuel soaked in porous beds. A porous sand bed initially soaked
with methanol is ignited and the burning characteristics of the flame are studied. The 19 flame temperature profiles, location of vapor/liquid interface, vapor region moving speed, combustion duration time, fuel consumption were studied in the experiments. The results indicate that the fuel consumption rate increased rapidly during the beginning stages. As time proceeded, the formation of vapor region increased the resistance to vapor transport, resulting in a decreasing trend in the fuel consumption rate. Their results confirm that the resistance to vapor transport is the controlling mode when a dry region is formed in the porous bed. There is an abrupt increase in the temperature profile in the dry vapor phase region. In the liquid region, the temperatures are close to the boiling point of the fuel.
They have also studied the effect of sand sizes and the sand layer depths on the burning characteristics of the fuel.
Modeling Two-phase flows in Porous Media
Numerical analysis of multidimensional two-phase flow including phase change in porous media is intrinsically complicated. One reason is the strongly nonlinear and coupled nature of the governing equations for the two-phase flow. Another fundamental difficulty lies in the presence of moving and irregular interfaces between the single and two-phase subregions in a domain of interest. The location of such an interface is not known a priori and must be determined by the coupled flows in adjacent regions.
Primarily three different numerical approaches have been used by researchers – separate flow model, enthalpy model and thermodynamic equilibrium model.
Traditionally separate flow model (Bear, 1972) has been used in which separate equations for the two phases are formulated and the interface between the different regions is explicitly tracked. The explicit tracking of moving interface involves complex coordinate mapping or numerical remeshing (Ramesh and Torrance, 1990). To reduce 20
the complexities of separate flow models, Wang and Beckermann (1993) developed two-
phase mixture model based on enthalpy formulation. The separate equations for the two phases are combined to form a single set of equations for the mixture. In this way, the number of governing differential equations to be solved is reduced by almost half and the rest are replaced by algebraic relations. Based on the enthalpy values of the mixture, the state of the system inside the porous wick is identified. The inherent assumption in this model is that the two-phase region is at a constant temperature (equal to the boiling temperature of the liquid) and hence there is no phase change taking place inside the two- phase region. Phase change is taking place only at the interfaces between the two-phase and the single phase regions. Usually in many physical systems, either condensation or evaporation does occur inside the two-phase region and the temperature of the two-phase region does slightly vary. This can only be accounted by invoking thermodynamic equilibrium relations. Recently, Benard et al. (2005) extended the enthalpy modeling approach of Wang et. al. (1993) by incorporating the equilibrium thermodynamic relations, which determines the thermodynamic state of the system and also accounts for phase change taking place inside the two-phase region.
Raju (2004) developed a steady state one-dimensional stagnation point diffusion
flame stabilized next to a porous media. The detailed one-dimensional heat and mass
transport inside the porous wick has been studied. In this study the liquid in the porous
media is assumed to be driven by capillary action and the vapor is driven by the vapor
pressure gradient induced by the temperature gradient (based on Classius-Clayperon
equilibrium relationship) inside the porous media. These studies reveal that the liquid and 21
the vapor flow counter currently inside the two-phase region. It has also been found out that the steady state solution exists only for stretch rates below a critical value.
1.4. PURPOSE AND SCOPE OF THIS DISSERTATION
The purpose of this thesis is to provide a more complete picture of the candle
flame behavior which is coupled to a porous wick. A more realistic picture is obtained by considering the role of heat and mass transport inside the candle wick. The thesis aims at improving our understanding of wick stabilized candle flames by addressing the following issues
(1) The role of heat and mass transport inside the candle wick
(2) Develop a model for simulating the self trimming of a candle wick
(3) The role of gravity on the shape and size of self trimmed candle flames.
(4) The effect of different wick diameters on burning rate and flame temperature.
The present study is primary computational. The computational results are
generated by solving the full Navier-Stokes equations with a one-step finite rate chemical
reaction rate in the gas phase and two-phase flow equations inside the porous wick.
1.5. DISSERTATION OUTLINE
This dissertation describes numerically the heat and mass transfer taking place
inside the burning candle wick and its effect on the candle flame structure. The first
Chapter of this dissertation describes the background related to candle burning and two-
phase flow inside porous media. A brief review on previous work in this field is
presented. 22
Chapters 2 and 3 include the theoretical and numerical aspects of the current
computational model. In Chapter 2, the governing equations for the fluid flow describing the mass conservation, energy conservation, inert gas species conservation, capillary and thermodynamic relations inside the wick. The numerical method adopted for solving the discretized equations is also included. The description of using multifrontal solvers in this context is presented. Different variants of Newton solvers are also presented.
Preliminary analysis of two-phase flow inside the wick is done by applying a constant heat flux all along the wick surface.
In Chapter 3, the governing equations for the gas phase fluid flow and combustion
reaction are presented using mass, momentum and energy conservation equations. The
pressure-velocity coupling is handled by the SIMPLER algorithm. The numerical method
adopted for solving the discretized equations is also included.
The numerical test results are presented in Chapter 4 of this dissertation. The gas phase model is coupled with the two-phase axisymmetric wick model. The detailed flow structures in the gas phase and in the wick are presented. The phenomenon of self trimming of candle wick is modeled as a burn out of the dry region. Parametric analysis has been done to study the effect of gravity, absolute permeability of the wick, wick diameter and ambient oxygen percentage on the candle flame structure and the burning rate.
Finally, in Chapter 5, a summary will be given and some future work is
suggested.
23
Flow direction
Products of combustion
conduction
Fuel vapor radiation g x
r
oxidizer
Figure 1.1 Schematic of a Candle Flame 24
CHAPTER 2 AXISYMMETRIC WICK MODELING
2.1 FORMULATION OF TWO PHASE FLOW INSIDE POROUS MEDIA
Two phase flow in porous media typically consists of three phases: the solid phase, the liquid phase and the gas phase. In addition to the transport of individual phases, there is a phase change process involved inside the porous media (but no chemical reaction is assumed). Treatment of individual phases from the well known point equations of continuum physics is rather complicated and computationally expensive.
Hence volume averaging technique described by Whitaker (1977) is used to provide a rational route to a set of equations describing the transport of heat and mass in a porous media.
Candle wicks are usually cylindrical in shape although other shapes are also used.
The gas phase and the wick characteristics are assumed to be symmetric around the angular direction and hence the present wick is modeled for an axisymmetric geometry.
2.1.1 MATHEMATICAL FORMULATION
The constitutive equations for the solid, liquid and gaseous phases are volume averaged (Whitaker, 1977). The equations are written in cylindrical coordinates for two phase flow.
In addition to the assumptions of Whitaker (1977), the following additional assumptions are made in this study.
1. The liquid phase is assumed to be continuous. 25
2. The flow is assumed to be laminar. Darcy’s law is assumed to be valid both for
the liquid and gas phases.
3. The transport of non-condensable gases inside the porous wick is neglected. The
term “non-condensable gases”, refers to all the gases other than the fuel vapor
(e.g. ambient air, combustion products like CO2, H2O etc.)
4. Radiative heat transfer inside the porous media is neglected.
5. The surface tension of the liquid is assumed to be constant. It does not vary with
temperature.
6. The vapor is locally in thermal equilibrium with the liquid and the thermodynamic
Gibbs phase equilibrium relations are assumed to be valid in the two-phase
region.
7. Although phase change can occur, there is no chemical reaction in the porous
media.
8. The properties of the wick, like thermal conductivity, permeability etc is assumed
to be isotropic. In principle, the thermal conductivity and permeability can be
different in axial and longitudinal directions.
9. Deformation of the wick material due to thermal stresses or bending of the wick
during burning of the candle is neglected.
10. Ideal gas law is assumed to be valid for the vapor phase
The resultant simplified volume averaged equations are written down for an axisymmetric, steady state, two-phase flow inside a porous wick. 26
Continuity equation:
∂∂11 ∂ ∂ ()ρρlluurvrv++()gg () ρll +() ρgg =0 . (2.1) ∂∂x x rr ∂ rr ∂
The individual terms represent the net mass flux of liquid and vapor at a point in x and r directions respectively.
Momentum equations:
The momentum equations are given by Darcy’s law.
kKrl ⎛⎞∂ Pl ugll=−−⎜⎟ρ , μl ⎝⎠∂x
kKrg ⎛⎞∂ Pg uggg=−−⎜⎟ρ , μg ⎝∂x ⎠ (2.2-5) kKrl ⎛⎞∂ Pl vl =−⎜⎟, μl ⎝⎠∂r
kKrg ⎛⎞∂ Pg vg =−⎜⎟. μg ⎝⎠∂r
The liquid phase is treated as incompressible and the gaseous phase is treated as an ideal gas.
Energy equation:
∂∂11 ∂⎛⎞⎛∂TT∂∂⎞ ()ρρlllhu++ gg h u g () r() ρρlll hv + ggg h v =⎜⎟⎜ keff +rkeff ⎟. (2.6) ∂∂x rr ∂x⎝⎠⎝∂ x rr∂∂ r⎠
Using the relations hcThll== and gg cTi+ fg, equation (2.6) becomes
∂∂11⎛⎞∂∂ρρllurv ll ()ρρlllcu++ gg c u g T() r() ρρlll cv + ggg c v T +−− i fg⎜⎟= ∂∂xrr⎝⎠ ∂ xr∂r (2.7) ∂∂⎛⎞⎛TT1 ∂ ∂ ⎞ ⎜⎟⎜kreff + keff ⎟ ∂∂∂xxrrr⎝⎠⎝ ∂ ⎠
The first two terms on the left hand side of equation 2.7 represent the convective heat transport of liquid and vapor in the x and r direction. The third term represents the heat 27 source term due to phase change taking place between the liquid and the vapor. The right
hand side of this equation represents the conductive heat transfer. keff represents the effective thermal conductivity.
Capillary and permeability relations
The gas pressure is related to the liquid pressure using the capillary relation
Pscg( ) = P− Pl. (2.8)
The capillary pressure is related to the saturation given by Leverett’s function (Leverett,
1941),
σ ⎡ 23⎤ Pssc =−1.42() 1−− 2.12 () 1+ 1.26 () 1−s. (2.9) ()K / ε 1/2 ⎣ ⎦
The relative permeability of the porous media is given by the following approximation
(Bau and Torrence, 1982)
kskrl = , rg =−( 1 s) . (2.10 a-b)
The non-dimensionalization of the porous wick variables are carried out according to the variables indicated in table 2.1. The non-dimensionalized variables are indicated by a
‘hat’ symbol on the top of the variable.
Non-dimensionalized equations:
∂∂11 ∂ ∂ ()uurvrvˆˆˆˆˆˆlg+++()ρˆg() l()ρˆ gg=0 , (2.11) ∂∂xˆˆ x rr ∂ˆˆ rr ∂ˆ 28
⎛⎞∂Pˆ uPekˆˆ=−−l g, lrl⎜⎟ ⎝⎠∂xˆ Pek⎛⎞∂ Pˆ ugˆˆ=−−rg⎜⎟ g ρˆ , ggμˆˆ⎜⎟∂x g ⎝⎠ (2.12-15) ⎛⎞∂Pˆ vPekˆ =−l , lrl⎜⎟ ⎝⎠∂rˆ Pek⎛⎞∂ Pˆ vˆ =−rg⎜⎟ g . g μˆˆ⎜⎟∂r g ⎝⎠
∂∂11⎛⎞∂∂urvˆˆˆ∂⎛⎞⎛∂TTˆˆ1∂∂⎞ ˆˆˆˆˆˆˆ ˆˆˆˆ ˆ ll ˆˆˆ ()ucuTrvcvTil+++ρρ g g g ()()l g g g +−−=+ fg ⎜⎟⎜⎟⎜keff rkeff ⎟ ∂∂xˆˆrrˆ⎝⎠ ∂x ˆ rˆ∂ rˆˆˆ∂ x⎝⎠⎝∂ x rrˆ∂ˆ∂ rˆ⎠ (2.16)
The effective thermal conductivity of the wick is function of saturation given by the expression (Udell and flitch, 1985)
ˆˆ kseff =+(1 − sk) s , (2.17)
ˆ where ks is the thermal conductivity of the solid wick material.
Conditions for Phase Transition:
The equilibrium thermodynamic state of candle wax (single phase liquid, single phase vapor, two-phase) can be determined for given liquid and gaseous pressures and temperature conditions, using the vapor pressure equilibrium data obtained from the thermodynamic Gibbs phase relationships (Benard et. al. 2005).
GhTˆ =−ˆ ˆηˆ , llll (2.18-19) ˆ ˆ ˆ GhTg =−gggηˆ , where G is the Gibbs potential per unit mass of the corresponding phase and symbol
“hat” denotes the non-dimensional value. h and η are respectively the enthalpy and the 29 entropy of the corresponding phase. The dimensional expressions for the enthalpy and entropy are given by
hcTTll=−( 0 ),
higfgg=+ cTT() −0 , (2.20-23) ηll= cTTlog() /0 ,
ηgfgg=+iTc/00 log() TTR / − log() PPg /0 .
The thermodynamic equilibrium relations based on the minimization of Gibbs function
(Saad, 1966) are given as follows
ˆˆ State 1. GGlg< , no vapor phase is present ( s =1 )
ˆˆ State 2. GGlg= , liquid and vapor are in equilibrium ( 01< s < )
ˆˆ State 3. GGl> g, no liquid phase is present ( s = 0 )
(2.24-26)
Equations 2.24-26 simply mean that whichever phase has the least Gibbs phase potential will dominate over the other phase. Equations 2.18-23 in conjuction with the phase equilibrium condition (Eq. 2.25) yield the well known, Classius-Clayperon equation.
Boundary conditions:
(1) Base of the wick
The wick is immersed in the candle wax pool, which can be assumed to be at its melting temperature.
ˆˆ sT=1, = Tm
(2) Cylindrical surface of the wick 30
Depending on the thermodynamic state of the wick on the surface, the boundary conditions will vary. In the case of either pure liquid or pure vapor, all the heat flux imposed on the surface is conducted into the wick. In the case of two-phase region, part of the heat supplied is used for evaporating the liquid on the surface of the wick and part of the heat is conducted into the wick.
ˆ ˆ ⎛⎞∂T Liquid region: sv==1, ˆˆlfe 0, qk =−ff⎜⎟ ⎝⎠∂rˆ
⎛⎞ˆ ˆ ∂T ˆ ˆˆ Two-phase region: 0< ˆ ˆ ⎛⎞∂T Vapor region: spppqk==−0, ˆˆˆlcfe0 (0), ˆ =−ff⎜⎟ ⎝⎠∂rˆ (3) Tip of the wick Similar to that of the boundary conditions on the cylindrical surface. ˆ ˆ ⎛⎞∂T Liquid region: sv==1, ˆˆlfe 0, qk =−ff⎜⎟ ⎝⎠∂rˆ ⎛⎞ˆ ˆ ∂T ˆ ˆˆ Two-phase region: 0< ˆ ˆ ⎛⎞∂T Vapor region: spppqk==−0, ˆˆˆlcfe0 (0), ˆ =−ff⎜⎟ ⎝⎠∂rˆ (4) Symmetry line Symmetry boundary conditions are imposed along the symmetry line 31 2.1.2. NUMERICAL FORMULATION: The continuity, momentum and energy equations are discretized using finite difference approximation. This results in a set of non-linear discrete balance equations. These equations are coupled with the inequalities resulting from the phase equilibrium relationships. The system is solved using Newton’s method. The thermodynamic state of each grid block is updated at each iteration of the method. In this method there is no need to separately track the interface between the single phase and the two-phase regions. Equations 2.11-2.16 are discretized using finite difference approximation and combined appropriately with the non-dimensionalized forms of equations 2.8-2.9 to form 2 equations for the variables Psl , and T. The equations are closed by using the thermodynamic equilibrium relationships (equations 2.24-2.26). The thermodynamic relationships are incorporated in the iteration scheme as described below. For each grid node i, we set a thermodynamic state parameter χi ( 1 for pure liquid , 2 for liquid-vapor equilibrium, 3 for pure vapor) which is determined based on the previous guess value of saturation distribution, ζ ilii= ( p ,,sTi) the triplet of unknowns and Bs()ζ i ,1=− 1, ˆˆ B()ζ il,2=−GGg , (2.27-29) Bs()ζ i ,3= . Note here that B is the thermodynamic equation which depends on the thermodynamic state χi of the system at any given grid point. The only speciality of this 32 formulation is that the equation B is different at each grid node depending upon the thermodynamic state at that grid point χi . Based on the thermodynamic relationships described in the previous section, the following relationships hold if χζii== 1, then BB( ,1) 0, and ( ζ i ,2) < 0, ⎧⎪B()ζ i ,1< 0 if χζii== 2, then B() ,2 0, and ⎨ , (2.30-32) ⎩⎪B()ζ i ,3> 0 if χζii== 3, then BB() ,3 0, and () ζ i ,2> 0. The solution procedure is described as follows Let ζ ()n , χ (n) , be the values at the beginning of (n+1) th iteration and α be the under relaxation parameter. First ζ (n+1) is computed from Newton’s step. JF()ζχ()nn,, () () δζ() n=− ( ζχ() nn ()), (2.33-34) ζζαδζ(nn+1 )=+ () ()() n. Next the thermodynamic variable is updated at each grid node by the following relations ()nn()++11() n if χζii=> 1 and if B( ,2) 0, then χ i= 2; ()nn()++11()n if χζii=> 2 and if B() ,1 0, then χ i= 1; ()nn()+1()n+1 if χζii=< 2 and if B() ,3 0, then χ i= 3; ()nn()++11()n if χζii=< 3 and if B() ,2 0, then χ i= 2; ()nn+1 () otherwise χχii= . The linear system of equations 2.33 is solved using a direct sparse solver UMFPACK, which is based on a multifrontal technique. This solution procedure is outlined in the next section. Since the system of equations is highly non-linear, very high 33 under relaxation parameter of 0.001 is imposed to ensure smooth convergence of the variables. 2.2 MULTIFRONTAL SOLVERS FOR LARGE SPARSE LINEAR SYSTEMS This section describes the implementation of sparse direct solvers based on multifrontal techniques for solving the highly non-linear equations resulting from the two-phase flow equations inside the porous wick. The implementation of standard solvers like ADI solvers failed to produce convergence. Implementation of direct solution techniques based on sparse Gaussian elimination (specifically the multifrontal technique) resulted in a stable convergent solution and henceforth developed in this study. A modified Newton’s method is implemented, which when coupled with a multifrontal solver resulted in significant reduction in the computational time. The description of multifrontal technique is beyond the scope of this work but an attempt is made to give a brief introduction of the sparse matrix storage techniques and frontal solution techniques. Finally, the efficiency of the multifrontal solver is tested for a differential cavity benchmark problem. 2.2.1 INTRODUCTION Solving general sparse linear systems can be accomplished either by direct solution methods or by iterative solution techniques. Earlier direct solution method was often preferred to iterative methods in real applications because of their robustness and predictable behavior. Iterative solvers were often special-purpose in nature and were developed with certain applications in mind. For Finite Volume problems, the ADI method (Peaceman and Rachford, 1955) and implicit Stone (Stone, 1965) algorithms are 34 preferred because of their high computational efficiency and very low memory requirements compared to the direct solvers. Over the years in 1960’s and 1970’s, there has been significant development in the solutions of large linear systems. Techniques were developed to take advantage of the sparsity to design special direct methods that can be quite economical. The frontal method (Irons, 1970 and Hood, 1976) is a variant of Gaussian elimination and makes use of the sparsity pattern for matrices resulting from the discretization of PDE’s. Later Duff (1984), and Duff et al. (1986) extended the frontal techniques for solving any general sparse matrices. Over the recent years, there has been significant improvement in the development of multifrontal direct solvers which take advantage of the sparsity pattern and the Level 3 BLAS routines to enhance the computational speed on high power computing architectures. Before going into the discussion of solution techniques for sparse linear systems, it is important to discuss the different sparse storage schemes for storing the non-zero elements of a sparse matrix. Sparse Matrix Storage Schemes In order to take advantage of the large number of zero elements, special schemes are required to store sparse matrices. The main goal is to represent only the nonzero elements and to be able to perform the common matrix operations. Only the most popular schemes (Duff et al., 1986) are presented. The following material is taken from Saad, (2003) and is presented here for completeness. Compressed row format 35 A(n*n) with nz nonzero entries is represented using three arrays ax(nz), aj(nz), ai(n+1). The entries are entered row wise. Example: n=5, nz = 12 [1 2 3 ] [ 4 5 ] [6 7 8 ] [9 10 ] [ 11 12 ] row pointer array, ai(n+1) = [ 1 4 6 9 11 13 ] element value array, ax(nz) = [ 1 2 3 4 5 6 7 8 9 10 11 12 ] element column index array, aj(nz) = [ 1 3 4 2 5 1 3 4 1 4 3 5 ] Compressed column format A(n*n) with nz nonzero entries is represented using three arrays ax(nz), ai(nz), aj(n+1). The entries are entered column wise. Example: n=5, nz = 12 [1 2 3 ] [ 4 5 ] [6 7 8 ] [9 10 ] [ 11 12 ] Column pointer array, aj(n+1) = [ 1 4 5 9 11 13 ] element value array, ax(nz) = [ 1 6 9 4 2 7 10 11 3 8 5 12 ] element row index array, ai(nz) = [ 1 3 4 2 1 3 4 5 1 3 2 5 ] Triplet format 36 A(n*n) with nz nonzero entries is represented using three arrays ax(nz), ai(nz), aj(nz). Any nonzero entry is identified by a triplet (ax,ai,aj), where ax is the value of the entry, ai is the row entry and aj is the column entry Example: n=5, nz = 12 [1 2 3 ] [ 4 5 ] [6 7 8 ] [9 10 ] [ 11 12 ] element value array, ax(nz) = [ 1 2 3 4 5 6 7 8 9 10 11 12 ] element row index array, ai(nz) = [ 1 1 1 2 2 3 3 3 4 4 5 5 ] element column index array, aj(nz) = [ 1 3 4 2 5 1 3 4 1 4 3 5 ] 2.2.2 MULTIFRONTAL SOLUTION METHODS Sparse direct solution methods involve band width algorithms, unifrontal and multifrontal techniques. The frontal techniques are much efficient compared to the band width algorithms and are widely used in the solving matrices resulting from finite element formulations. This section gives a brief introduction to unifrontal and multifrontal techniques which are direct methods for solving the linear equation Ax= B where A can be a symmetric, unsymmetric, definite or indefinite matrix. Unifrontal method is a derivative of the classical gaussian algorithm. Although it is developed for finite element applications, it can be used in other fields also. It uses a small dense sub-matrix called “frontal matrix’’. Original matrix is read sequentially and frontal matrix is filled with the rows and columns. All-rows must be chosen to be fully-summed i.e. there are no further 37 contributions to come to the rows. When the frontal is filled with such rows, pivot(s) are chosen from the fully-summed columns and basic gaussian elimination start. LU factors are stored in RAM or in disk. Shur complement for the part that cannot be eliminated further is calculated. Pivot row(s) and column(s) are deleted from the frontal to accommodate new elements coming from original matrix. Having built the new frontal, previous steps are repeated. Frontal matrices can be rectangular and doesn't have to be the same size. But important thing is that, there is no more than one frontal at the same time. Since the same working array is used throughout the factorization, the method is memory-efficient. Partial pivoting among the fully-summed columns may be applied to preserve numerical stability. Unifrontal method is competitive with the best band matrix routines. Frontal method is first described by Irons (1907) with the article ``A frontal solution program for finite element analysis''. Method is primarily designed for finite element analysis and for symmetric and positive definite matrices. Later, Hood (1976) modified the method for unsymmetric matrices. Again the algorithm was designed to solve finite element problems. In both papers, performance and memory management issues are compared with band matrix routines. Frontal method is found to be efficient especially in the case where the bandwidth is large. Band matrix routines require too much memory for large bandwidths so factorization must be preceded by an ordering phase to obtain smaller bandwidths. In 1983, ``Multifrontal Method'' was derived by Duff and Reid (1983) in which more than one frontal matrix is used. This method was first designed for symmetric and undefinite matrices but later modified for unsymmetric and definite matrices. If there is 38 more than one frontal matrix at the same time unifrontal method is called Multifrontal method. The description of multifrontal algorithm is beyond the scope of this work, but a brief overview is presented here. Since couple of frontals exists, there are some dependency relations between them. These relations are analyzed to build the assembly tree (factorization tree). This tree roughly says which frontal must be handled first and which one is last. Then from the leaf nodes to the root, factorization starts. First, leaf nodes are factorized and assembled to the parents i.e. Shur complements are calculated and contribution blocks are sent to upper nodes. Since, there is more than one frontal matrix at the same time, management of these sub-matrices, calculating shur complements and assembling them to the upper part of the tree is a difficult task. However, it gives the chance for doing these calculations in a parallel environment. Additionally, multifrontal method is more powerful then unifrontal method in the case of the matrices of which bandwith is very large. A parallel version of multifrontal method is also introduced in by I. S. Duff (1986). Parallel implementation is based on the elimination tree concept. There is more than one frontal matrix in multifrontal approach and these sub-matrices also called nodes are dependent on each other. If one draw the dependency graph it will be observed that the graph will be a tree. Since leaf nodes of the tree are independent of each other, they can be handled separately. This is the first step of parallelization. At later steps, other nodes are factorized by different processors and so on. In the present work, a multifrontal solver (Davis et al., 1997, 1999 and 2004) termed as the unsymmetric multifrontal package (UMFPACK v2.2) is used. UMFPACK is a set of routines for solving unsymmetric sparse linear systems, Ax=b, using the 39 Unsymmetric MultiFrontal method and direct sparse LU factorization. UMFPACK v2.2 (Davis et al., 1999) is a set of fortran subroutines. UMFPACK relies on the Level-3 Basic Linear Algebra Subprograms (dense matrix multiply) for its performance. This code works on Windows and many versions of Unix (Sun Solaris, Red Hat Linux, IBM AIX, SGI IRIX, and Compaq Alpha). The present code is being run on a windows machine (Intel Pentium 4 machine) using optimized GOTO BLAS library available for free download from the internet. Using GOTO BLAS has enhanced the performance of UMFPACK significantly. Direct solution methods for solving set of non-linear equations usually involve a Newton’s linearization step to convert the set of non-linear equations into a set of linear equations which is then updated during each iterative step. The disadvantage of using direct solution methods is that the solution of set of linear equations during each iteration takes a huge amount of computational time. Moreover it requires a huge memory. There has been significant improvement in the available of large RAM in the present market. In addition, the onset of 64 bit machines has made it possible to access more than 2GB RAM. Thus it has been possible to solve even large scale problems using direct solvers. The development of efficient direct sparse solvers has significantly reduced the computation time for solving a linear system. The most time consuming step in solving the linear system is the factorization step. In the present work, a modified Newton’s method is used which can save a significant amount of computational time. Newton step: Jx()nnδ () =− F() n, () (2.35-36) x()nn+1 =+xx ()αδ ()n. 40 Modified Newton step: Jx()0 δ ()nn=− F(), () (2.37-38) x(nn+1 )=+xx ()αδ ()n, where F is the function residual, J is the Jacobian matrix consisting of the derivatives of the function residual, x and δ x are the variable and the increment in the variables to be solved, n is the iteration number and α is the under relaxation parameter. In the modified Newton’s step, the Jacobian matrix is not updated. It is calculated only during the first iteration and the same Jacobian matrix is reused. The advantage with Modified Newton’s method is that the left hand matrix needs to be factorized only during the first iteration. The factors calculated by UMFPACK solver during the first iteration are stored for reuse in the subsequent iterations. Since the factorization is the most expensive step, by skipping the factorization step in the subsequent iterations, the subsequent iterations become very cheap. Although the rate of convergence is reduced compared to the Newton’s step, the savings in computational time per iteration overrules the decrease in convergence and hence significant amount of saving is obtained. 2.2.3 BENCHMARK TESTING To test the performance of sparse direct solvers over the standard SIMPLE technique used for solving Navier-Stokes equations, a differential cavity problem is chosen as a benchmark problem and the total computational times are compared. Differential Cavity problem ∂U ∂V + = 0, (2.39) ∂X ∂Y ∂U ∂U ∂P ⎛ ∂2U ∂2U ⎞ U +V −= + Pr⎜ + ⎟ , (2.40) ⎜ 2 2 ⎟ ∂X ∂Y ∂X ⎝ ∂X ∂Y ⎠ 41 ∂V ∂V ∂P ⎛ ∂2V ∂2V ⎞ U +V −= + Pr⎜ + ⎟ + PrTRa , (2.41) ⎜ 2 2 ⎟ ∂X ∂Y ∂Y ⎝ ∂X ∂Y ⎠ ∂T ∂T ⎛ ∂2T ∂2T ⎞ U +V = ⎜ + ⎟. (2.42) ⎜ 2 2 ⎟ ∂X ∂Y ⎝ ∂X ∂Y ⎠ The boundary conditions for the driven cavity problem can be expressed as U = V = ,0 allat other boundaries T = ,1 X = ,0 T = ,0 X = ,1 ∂T = ,0 ,0 and YY == .1 ∂Y Figure 2.1 compares the function residuals FF= for the Direct solvers ( R ∞ ) (Newton, Picard and Modified Newton) and the SIMPLE method. Figure 2.1 indicates that the direct solvers are more efficient compared to the SIMPLE method. Modified Newton’s method is found to extremely efficient compared to all the other methods. This is because of the skipping the factorization step after the first iteration which leads to a significant amount of savings in computational time. 2.3 ANALYSIS OF AN EXTERNALLY HEATED AXISYMMETRIC WICK In a realistic candle, both the candle flame and the wick are coupled to each other. The heat from the candle flame evaporates the candle wax, providing the driving force for the liquid to rise up through capillary action. The wax evaporated from the surface provides the fuel for the candle flame. In this way the fuel supplied by the wick and the heat supplied by the candle flame are coupled together. The heat and mass transfer taking place inside the porous wick is very complex due to the presence of both liquid and vapor inside the wick. To gain sufficient insight into the physics of two-phase flow inside the wick, the wick is first analyzed separately 42 by decoupling the wick from the flame. A constant heat flux is imposed along the cylindrical surface and at the tip of the wick. The detailed structure and the flow patterns inside the porous wick are analyzed. Later, the wick is coupled to the candle flame and the detailed study of gas phase flame characteristics and porous flow fields is done in chapter 4. 2.3 1 PHYSICAL DESCRIPTION OF THE MODEL The wick is now treated as being heated from a constant external heating source. The heat flux is uniformly distributed along the cylindrical surface of the wick and on the tip surface of the wick (Figure 2.2). Experiments have been conducted by Zhao and Liao (2000) to study the heat transfer characteristics of a capillary-driven flow in a porous structure heated with a permeable heating source at the top. Their experimental set up is essentially one-dimensional in nature. This present physical situation is different from their experimental set up but their experiments reveal certain essential characteristics of two-phase flow inside a wick. In this present set up, the wick is dipped in a liquid wax pool. The level of the wax pool is assumed to be constant. The wax pool is at the melting point temperature of the wax (330 K). A constant heat flux is applied along the cylindrical surface and on the wick tip. A wick of length 5mm and diameter 1mm is chosen for the present study. The two-phase flow inside the wick is simulated. Figure 2.3 shows the computational grid used for the axisymmetric wick model. The grid size is chosen as 80x40. Grid clustering is used in both the x and r directions. 43 2.3.2 SAMPLE CASE RESULTS The porous wick parameter chosen are shown in the Table 2.2. A constant heat flux of 810× 4 W/m2 is applied on the wick as described above. The detailed structure inside the porous wick is presented below. 2.3.2.1 Saturation and Temperature Distribution Figure 2.4(a) shows the saturation profiles inside the porous wick. The figure indicates that there are two regions inside the wick – single phase liquid region and two- phase vapor liquid region. The contour line s=1 demarcates the two-phase region and the pure liquid region. The saturation is lowest on the cylindrical corner of the wick reaching a value of 0.825. As the wick is receiving heat on both the cylindrical surface and the tip surface from the external heating source, the evaporation causes a decrease in the saturation at the surface of the wick. Figure 2.4 (b) shows the non-dimensional temperature distribution inside the wick. The base of the wick is at the melting point (323 K) of the liquid wax and on the wick surface, where there is evaporation; the temperature is at its boiling point (620 K). Fig. 2.4 (b) shows the presence of temperature gradients near the base of the wick indicating the heat lost to the wax pool. Figure 2.4 (c) shows the temperature contours inside the two-phase region of the wick. Notice that the temperature variation inside the two-phase region is very small. Still this slight temperature variation causes a significant variation in the vapor pressure distribution inside the two-phase region given by the equilibrium phase relations (Eq. 2.25). 2.3.2.2 Pressure Distribution Figure 2.5 (a) shows the non-dimensional liquid pressure distribution inside the wick. This drives the liquid from the base of the wick to the surface of the wick. As the 44 wick is receiving heat on both the cylindrical surface and the tip surface from the external heating source, the evaporation causes a decrease in the saturation at the surface of the wick. This causes a decrease in liquid pressure, given by capillary relations, on the surface (both the cylindrical surface and the wick tip surface) of the wick. Pressure gradients are present along the length of the wick and in the radial direction indicating liquid motion in both the directions (refer fig. 2.4 (a)). Figure 2.5 (b) shows the non- dimensional capillary pressure distribution inside the wick. Capillary pressure is a function of the saturation (refer Eq. 2.9). Hence the capillary pressure distribution is qualitatively similar to that of the saturation distribution. At the interface between the liquid and the two-phase region, the capillary pressure is zero. Figure 2.5 (c) shows the non-dimensional vapor pressure distribution inside the wick. The vapor pressure in the two-phase region is a function of the temperature as given by the Gibbs phase relations (Eq. 2.25). Hence the vapor pressure distribution variation is qualitatively similar to the temperature variation inside the two-phase region. Figure 2.5 (c) shows sharp vapor pressure gradients in the r-direction near the cylindrical surface of the wick. 2.3.2.3 Mass flux Distribution Figure 2.6 (a) shows the liquid mass flux vectors indicating the flow of liquid inside the wick. The liquid is drawn from the base and it comes out of the wick along the cylindrical surface and the tip of the wick. As the liquid is evaporated along the cylindrical surface, the liquid mass flux along the length of the wick decreases. Figure 2.6 (b) shows the vapor mass flux distribution inside the wick. Vapor motion is very intricate and also interesting. Vapor is being driven by the temperature gradients inside the two-phase region. Since the temperature gradients (related to the pressure gradient 45 through the equilibrium relationship Eq. 2.25) are directed into the wick, the vapor moves into the wick. The liquid and the vapor move counter currently to each other. This is also found in one-dimensional wick model [Raju, 2004] and is confirmed by experimental results [Zhao and Liao (2000)]. The vapor eventually condenses at the interface between the liquid and the two-phase region. 2.3.2.4 Heat flux Distribution Figure 2.7 shows the heat flux distribution inside the porous wick. There is convective heat flux due to both liquid and vapor motion and there is also conductive heat flux due to temperature gradient. The heat flux vectors can be expressed as follows