J Heat Transfer 2002 Vol 124 N5

Transactions Journal of
of the ASME Heat Transfer ®
HEAT TRANSFER DIVISION Published Bimonthly by The American Society of Mechanical Engineers
Chair, Y. JALURIA
Vice Chair, Y. BAYAZITOGLU
Past Chair, J. H. KIM VOLUME 124 • NUMBER 5 • OCTOBER 2002
Secretary, M. K. JENSEN
Treasurer, R. W. DOUGLASS
Member, R. D. SKOCYPEC TECHNICAL PAPERS
Editor, V. DHIR „2005…
Associate Editors, Conduction
C. AMON „2004…
P. AYYASWAMY „2004…
805 Optimum Design of Spacecraft Radiators With Longitudinal Rectangular
K. BALL „2004… and Triangular Fins
H. H. BAU „2003… Rizos N. Krikkis and Panagiotis Razelos
V. P. CAREY „2003…
G. CHEN „2005… Forced Convection
J. CHUNG „2005… 812 Nonstaggered APPLE Algorithm for Incompressible Viscous Flow in
G. DULIKRAVISH „2004…
Curvilinear Coordinates
A. EMERY „2005…
M. FAGHRI „2003… S. L. Lee and Y. F. Chen
J. G. GEORGIADIS „2003… Gas Turbine Heat Transfer
M. JENSEN „2004…
D. B. R. KENNING „2004… 820 Film Cooling Effectiveness and Heat Transfer Coefficient Distributions
K. KIHM „2005… Around Diffusion Shaped Holes
H. LEE „2004…
G. P. PETERSON „2003…
Y. Yu, C.-H. Yen, T. I.-P. Shih, M. K. Chyu, and S. Gogineni
V. PRASAD „2005… 828 Energy and Exergy Balance in the Process of Spray Combustion in a Gas
R. D. SKOCYPEC „2003… Turbine Combustor
S. THYNELL „2005…
P. VANKA „2005… S. K. Som and N. Y. Sharma
Radiative Heat Transfer
BOARD ON COMMUNICATIONS
Chair and Vice President 837 Reduction of False Scattering of the Discrete Ordinates Method
OZDEN OCHOA Hong-Shun Li, Gilles Flamant, and Ji-Dong Lu
OFFICERS OF THE ASME

Evaporation, Boiling, and Condensation
President, SUSAN H. SKEMP 845 Investigation of Circumferential Variation of Heat Transfer Coefficients
Executive Director, During In-Tube Evaporation for R-22 and R-407C Using Liquid Crystal
VIRGIL R. CARTER
Treasurer,
Seok Ho Yoon and Min Soo Kim
R. E. NICKELL 854 Evaporation of Water Droplets Placed on a Heated Horizontal Surface
Orlando E. Ruiz and William Z. Black
PUBLISHING STAFF
Managing Director, Engineering 864 A Cavity Activation and Bubble Growth Model of the Leidenfrost Point
THOMAS G. LOUGHLIN John D. Bernardin and Issam Mudawar
Director, Technical Publishing
PHILIP DI VIETRO 875 Rewetting of an Infinite Slab With Uniform Heating Under Quasi-Steady
Conditions
Managing Editor, Technical Publishing
CYNTHIA B. CLARK
A. K. Satapathy and R. K. Sahoo
Production Coordinator Two-Phase Flow and Heat Transfer
COLIN McATEER
881 A Natural Circulation Model of the Closed Loop, Two-Phase Thermosyphon
Production Assistant for Electronics Cooling
MARISOL ANDINO S. I. Haider, Yogendra K. Joshi, and Wataru Nakayama
Transactions of the ASME, Journal of Heat Transfer (ISSN
0022-1481) is published bi-monthly (Feb., Apr., June, Melting and Solidification
Aug., Oct., Dec.) by The American Society of Mechanical
Engineers, Three Park Avenue, New York, NY 10016. 891 Heat Transfer Characteristics of Melting Ice Spheres Under Forced and
Periodicals postage paid at New York, NY and additional Mixed Convection
mailing offices. POSTMASTER: Send address changes to
Transactions of the ASME, Journal of Heat Transfer, c/o
Y. L. Hao and Y.-X. Tao
THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS, 22
Law Drive, Box 2300, Fairfield, NJ 07007-2300. Porous Media
CHANGES OF ADDRESS must be received at Society
headquarters seven weeks before they are to be effective. 904 Directional Spectral Emittance of a Packed Bed: Influence of the
Please send old label and new address. Temperature Gradient in the Medium
STATEMENT from By-Laws. The Society shall not be
responsible for statements or opinions advanced in papers or
Dominique Baillis and Jean-François Sacadura
... printed in its publications (B7.1, Para. 3), COPYRIGHT
© 2002 by The American Society of Mechanical Engineers. For
authorization to photocopy material for internal or personal
use under those circumstances not falling within the fair use
provisions of the Copyright Act, contact the Copyright
Clearance Center (CCC), 222 Rosewood Drive, Danvers, MA
01923, tel: 978-750-8400, www.copyright.com. Request
for special permission or bulk copying should be addressed to
Reprints/Permission Department. INDEXED by Applied
„Contents continued on inside back cover…
Mechanics Reviews and Engineering Information, Inc. Canadian
Goods & Services Tax Registration #126148048.
Downloaded 04 Feb 2011 to 194.85.80.107. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
„Contents continued…
Journal of Heat Transfer Volume 124, Number 5 OCTOBER 2002
Heat Transfer Enhancement

912 R-22 and Zeotropic R-22ÕR-142b Mixture Condensation in Microfin, High-Fin, and Twisted Tape Insert Tubes
F. J. Smit and J. P. Meyer
Heat and Mass Transfer
922 Effectiveness Correlations for Heat and Moisture Transfer Processes in an Enthalpy Exchanger With Membrane
Cores
L. Z. Zhang and J. L. Niu
930 On the Potential for Homogeneous Nucleation of Salt From Aqueous Solution in a Natural Convection Boundary
Layer
Kenneth A. Smith, Marc Hodes, and Peter Griffith
Heat Transfer in Manufacturing
938 Thermal Aspects in the Continuous Chemical Vapor Deposition of Silicon
Hoseon Yoo and Yogesh Jaluria
947 Direct Chill Casting of Aluminum Alloys: Modeling and Experiments on Industrial Scale Ingots
Christopher J. Vreeman, J. David Schloz, and Matthew John M. Krane
Microscale Heat Transfer
954 Thermal and Electrical Energy Transport and Conversion in Nanoscale Electron Field Emission Processes
T. S. Fisher and D. G. Walker
963 Interface and Strain Effects on the Thermal Conductivity of Heterostructures: A Molecular Dynamics Study
Alexis R. Abramson, Chang-Lin Tien, and Arun Majumdar
TECHNICAL NOTES
971 Wavelets and the Discrete Ordinate Method for Radiative Heat Transfer in a Two-Dimensional Rectangular
Enclosure With a Nongray Medium
Ye Wang and Yildiz Bayazitoglu
975 Experimental Study on Forced Convection Heat Transfer Inside Horizontal Tubes in an AbsorptionÕCompression
Heat Pump
Li Yong and K. Sumathy
978 On the Accuracy of Beam-Averaged Interferometric Heat Transfer Measurements
D. Naylor
982 Bulk Temperature Development in Transient Heat Transfer Measurements Using Heater Foils
Jens von Wolfersdorf
985 Solutions of Radiative Heat Transfer in Three-Dimensional Inhomogeneous, Scattering Media
L. M. Ruan and H. P. Tan
988 Electric Network Representation of the Unsteady Cooling of a Lumped Body by Nonlinear Heat Transfer Modes
Francisco Alhama and Antonio Campo
992 Experimental Validation of a Combined Electromagnetic and Thermal Model for a Microwave Heating of Multi-
Layered Materials Using a Rectangular Wave Guide
P. Rattanadecho, K. Aoki, and M. Akahori
ANNOUNCEMENTS
997 Outstanding Reviewers—2001
998 Preparing and Submitting a Manuscript for Journal Production and Publication
999 Preparation of Graphics for ASME Journal Production and Publication
1000 Attention Authors: ASME Reference Format
Downloaded 04 Feb 2011 to 194.85.80.107. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Optimum Design of Spacecraft
Radiators With Longitudinal
Rizos N. Krikkis
Institute of Chemical Engineering and High
Rectangular and Triangular Fins
Temperature Chemical Processes,
P.O. Box 1414, In the present study the optimum dimensions of longitudinal rectangular and triangular
Stadiou St., Platani, 26 500 Patras, radiating fins with mutual irradiation are determined. The basic assumptions are one-
Greece dimensional heat conduction and gray diffuse surface radiation. The governing equations
are formulated by means of dimensionless variables and solved numerically in order to
Panagiotis Razelos carry out the required minimization procedure. The optimum fin dimensions, thickness and
Mem. ASME height, are presented in generalized dimensionless form and correlations are provided in
Professor Emeritus, order to assist the spacecraft thermal systems designer. The results are analyzed and
College of Staten Island, expressed in explicit correlations. Several diagrams, are also included, that give insight to
CUNY, NY the operational characteristics of the heat rejection mechanism. Moreover, special atten-
tion is given to the error analysis of the numerical methods used, since the accuracy and
the reliability of the algorithms employed for the solution of the integro-differential equa-
tion is important for the calculation of the heat dissipation at the fin base and its subse-
quent optimization. 关DOI: 10.1115/1.1497359兴
Keywords: Finned Surfaces, Heat Transfer, Radiation
Introduction optimum dimensions are obtained directly from explicit relations

without the need of interpolations from graphs or tables.
Radiating extended surfaces are widely used to reject heat from
a spacecraft, which became very important with the advent, in the
middle of the last century, of space exploration. Substantial effort Statement of the Problem
has been made to understand the fundamental energy exchange Consider a uniform density longitudinal fin in a tubular radiator
mechanisms involved in order to design radiating equipment. Bar- assembly depicted schematically in Figs. 1共a兲 and 1共b兲. The fin is
tas and Sellers 关1兴, obtained the efficiency of rectangular radiating made of material with thermal conductivity k, profile Y ⫽Y (X)
fins. Kern and Kraus 关2兴, expressed the heat dissipation and the base thickness 2w, height L, and length H. The fin base is main-
temperature distribution from a radiating fin in terms of the com- tained at constant temperature T b . Assuming one-dimensional
plete and incomplete beta functions. Liu 关3,4兴, Wilkins 关5兴, Chung heat conduction (L/wⰇ1), no heat sources or sinks in the fin,
and Nguyen 关6兴, and Kern and Kraus 关2兴, employed different radiation from all surfaces according to the Stefan-Boltzmann law,
methods to obtain the optimum dimensions of longitudinal radiat- length of the fin H much larger than L or w, (H/L,H/wⰇ1), the
ing fins. Chung and Zhang 关7,8兴, also determined the optimum steady-state temperature and radiosity must satisfy the following
dimensions of fins and fin arrays taking into consideration fin to equations:
冉冊
fin and fin to base interactions. Chung et al. 关9兴, proposed a new
d dT dS
design for space radiators and determined its optimum dimensions kA ⫽P Q (1)
in a fuzzy environment. Krishnaprakas 关10,11兴, obtained the opti- dX dX dX r
mum dimensions for fin arrays and presented the results with cor-
relations. Sunil Kumar et al. 关12兴, optimized a space radiator with
variable base temperature. Ramesh et al. 关13兴, optimized a two-
Q r 共 X 兲 ⫽J 共 X 兲 ⫺ 冕 L
X ⬘ ⫽0
J 共 X ⬘ 兲 dF dX⫺dX ⬘ (2)
dimensional tubular space radiator and gave correlations for the

optimum dimensions. Schnurr et al. 关14兴, determined and com-
pared the optimum dimensions of circular and straight fin arrays
J 共 X 兲 ⫽ ⑀ ␴ T 4 共 X 兲 ⫹ 共 1⫺ ⑀ 兲冕 L
X ⬘ ⫽0
J 共 X ⬘ 兲 dF dX⫺dX ⬘ (3)
of various profiles. The reader will find additional references in where Q r is the net radiative flux, and J is the radiosity. The
the recent review article on this subject by Aziz and Kraus 关15兴 boundary conditions are
冏
and in the new book by Kraus et al. 关32兴. dT
In the present work we determine the optimum dimensions of T 共 0 兲 ⫽T b , ⫺k ⫽ ⑀ ␴ T 4共 L 兲 (4)
constant thickness and triangular longitudinal radiating fins that dX X⫽L
includes radiant heat exchange between fins. All the surfaces in- In Eq. 共1兲, A⫽2Y (X)H is the area perpendicular to the heat flow
volved are considered gray and diffuse with constant thermal and P⫽2H is the participating perimeter in the radiative energy
properties. Sparrow et al. 关18兴 examined fins of rectangular pro- exchange. The area factor P dS/dX is equal to
file, while Karlekar and Chao 关19兴, rectangular and trapezoidal
profiles. Employing the results of the above references to deter- P dS/dX⫽2H 关 1⫹ 共 dY /dX 兲 2 兴 1/2 (5)
mine the optimum fin dimensions, one should use diagrams, Given the volume of the fin
which apparently will introduce inaccuracies 共see Aziz and Kraus
关15兴兲. In contrast, the present study proposes a method where the V⫽HA p ⫽H 冕 0
L
2Y 共 X 兲 dX (6)
Contributed by the Heat Transfer Division for publication in the JOURNAL OF
HEAT TRANSFER. Manuscript received by the Heat Transfer Division December 20, we seek the fin semi-thickness w and height L that will maximize
2001; revision received May 23, 2002. Associate Editor: B. T. F. Chung. the fin heat dissipation
Journal of Heat Transfer Copyright © 2002 by ASME OCTOBER 2002, Vol. 124 Õ 805
Downloaded 06 Dec 2010 to 129.252.86.83. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
P
dS
dX
Bi dy
⫽2 1⫹ 2
u dx 冋冉冊册 2 1/2
(12)
On the basis of the previously stated assumption regarding the one

dimensional heat conduction into the fin, the order of magnitude
of the radiation-conduction coefficient is one and Bi1/2Ⰶ1, Raze-
los and Georgiou 关17兴. Taking this into consideration Eq. 共12兲
reduces to P dS/dX⬇2. Substituting Eqs. 共8兲 and 共9兲 into Eqs. 共1兲
to 共3兲, we obtain the following expressions:
⌰ ⬙ 共 x 兲 ⫽ 关 u 2 q r 共 x 兲 ⫺y ⬘ ⌰ ⬘ 兴 /y, 0⭐x⭐1 (13)
⑀
q r共 x 兲 ⫽ 关 ⌰ 4 共 x 兲 ⫺ j 共 x 兲兴 (14)
1⫺ ⑀
j共 x 兲⫺ 冕 1
x ⬘ ⫽0
K 共 x,x ⬘ 兲 j 共 x ⬘ 兲 dx ⬘ ⫽ ⑀ ⌰ 4 共 x 兲 (15)
where in Eqs. 共13兲 and 共17兲 ⌰⬙ and ⌰⬘ represent the second and
the first derivatives with respect to x. The corresponding boundary
conditions are
⌰ 共 0 兲 ⫽1 (16)
⌰ ⬘ 共 1 兲 ⫹ ⑀ uBi1/2⌰ 4 共 1 兲 ⫽⌰ ⬘ 共 1 兲 ⫹ ⑀ B L ⌰ 4 共 1 兲 ⫽0 (17)
where B L is a dimensionless parameter:
B L ⫽h r L/k⫽LBi/w⫽uBi1/2 (18)
Using the same arguments as before, Eq. 共17兲 reduces to
⌰⬘共1兲⬇0. It is worth noticing that the problem was also solved
using both boundary conditions ⌰⬘共1兲⫽0 or Eq. 共17兲 where the
same results were obtained and therefore the optimum environ-
ment is independent of the value of B L since B L Ⰶ1. This is be-
cause ⑀⭐1, u is of O(1), ⌰ 4 (1)⬍1 and Bi1/2Ⰶ1. The kernel
function in Eq. 共15兲 takes the form
Fig. 1 „a… Schematic tubular space radiator; and „b… longitudi- 共 1⫺ ⑀ 兲 xx ⬘ sin2 共 ␥ ⫹ ␥ f 兲
nal fin geometry K 共 x,x ⬘ 兲 ⫽ (19)
2 关 x 2 ⫺2xx ⬘ cos共 ␥ ⫹ ␥ f 兲 ⫹x ⬘ 2 兴 3/2
where ␥ is the opening angle between the fins. In the above equa-
tions the physical quantities, thermal parameters and the fin di-
q f ⫽⫺kA
dT
dX
冏 X⫽0
(7)
mensions, have been combined to form a number of nondimen-
sional variables that describe the problem. The equation that
describes the profile of the fin is
Note that according to the earlier stated assumptions regarding the
length of the fin, we can take H⫽1 m without loss of generality y⫽1⫹ 共 ␭⫺1 兲 x (20)
and all quantities will be per unit length. All the above equations where ␭⫽1 corresponds to the rectangular profile and ␭⫽0.01 to
are nondimensionalized with the aid of the following dimension- the triangular. It can be seen from Eqs. 共13–15兲 that the dimen-
less variables: sionless temperature is
x⫽X/L, y⫽Y /w, ⌰⫽T/T b (8) ⌰⫽⌰ 共 x;u, ␥ , ⑀ 兲 (21)
and
The heat dissipated by the fin, after introducing dimensionless
j⫽J/ 共 h r T b 兲 , q r ⫽Q r / 共 h r T b 兲 (9) variables becomes
Furthermore, the dimensionless parameters that describe the prob- qf w Bi1/2⌰ ⬘ 共 0;u, ␥ , ⑀ 兲
lem is the radiation-conduction coefficient, Q⫽ ⫽⫺ ⌰ ⬘ 共 0;u, ␥ , ⑀ 兲 ⫽⫺
2kT b L u
h rL 2 L 2 ⫽Bi1/2D 共 u, ␥ , ⑀ 兲 (22)
u 2⫽ ⫽ 2 Bi (10)
kw w
where D⫽⫺⌰ ⬘ (0;u, ␥ , ⑀ )/u. It has been pointed out by Razelos
and the radiation Biot number Bi⫽h r w/k where h r ⫽ ␴ T 3b denotes and Georgiou 关17兴 that the condition that will economically justify
the black radiation heat transfer coefficient. With the aid of Eq. 共8兲 the use of fins is: ‘‘the ratio of heat dissipated by the fin to be
the perimetry factor, Eq. 共5兲, may be written as: much larger in comparison with the heat that would have been
冋冉冊册
dissipated from the surface 2wH, in the absence of the fin.’’ This
dS w 2 dy 2 1/2
P ⫽2 1⫹ 2 (11) ratio is the removal number and it is equal to
dX L dx
qf 2kT b Q D
Using now the earlier defined Biot number and the radiation- N r⫽ ⫽ ⫽ (23)
conduction coefficient yields: q b 共 2w 兲 ⑀ ␴ T 4b Bi1/2
806 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME
Optimization Table 1 Coefficients for the rectangular profile, Eqs. „39… and
„40…
In terms of the dimensionless variables introduced, the optimi-
zation statement given earlier becomes: Given the profile of the fin
and the desired heat dissipation determine the dimensions of the
fin that minimize its volume 共Razelos and Imre 关20兴, Razelos and
Krikkis 关16兴兲. There is also the equivalent problem stated as: For
a given fin volume we seek the dimensions that will maximize the
heat dissipation. Following Razelos and Krikkis 关16兴, a dimen-
sionless volume is introduced as
h 2r V
U⫽ ⫽uBi3/2 (24)
2C V k 2
Heat Dissipation Given
The fin volume in terms of dimensionless variables is calculated
from Eq. 共6兲 as h r kw opt 1
w *⫽ ⫽ (32)
冕 1
共 q f /T b 兲 2 4D 2opt
V⫽2Lw y 共 x 兲 dx⫽2C V Lw⫽C V V c (25) h r L opt u opt
0 L *⫽ ⫽ (33)
共 q f /T b 兲 2D opt
From Eqs. 共22兲 and 共24兲 we may observe that the dimensionless
heat dissipation and volume can be expressed as functions of u h 2r kV opt C V u opt
and Bi V *⫽ ⫽2C V w * L * ⫽ (34)
共 q f /T b 兲 3 4D 3opt
Q⫽Q 共 u,Bi; ␥ , ⑀ 兲 (26) N r* ⫽N r Q⫽D 2opt (35)
U⫽U 共 u,Bi兲 (27) Fin Volume Given
We now recast the optimization problem in terms of Q and U as:
Given U, determine the parameters u and Bi, that will maximize
Q. For a specified volume, after taking the differential of Eq. 共27兲
V⫽
w* 冉冊 k
h r V 2opt
1/3
w opt⫽
w*
共 V * 兲 2/3
⫽ 共 2C V u opt兲 ⫺2/3 (36)
and remembering that the parameters ⑀ and ␥ are specified by the

problem, we obtain L V* ⫽冉冊 hr
kV opt
1/3
L opt⫽
L*
共 V * 兲 1/3
⫽ 冉冊
u 2opt
2C V
1/3
(37)
dU⫽ 冉冊冉冊
⳵U
⳵ Bi
dBi⫹
⳵U
⳵u
du⫽0 (28)
Q *⫽
qf
T b 共 h 2r kV opt兲 1/3
⫽ 共 V * 兲 ⫺1/3⫽
D opt
共 0.25C V u opt兲 1/3
(38)
The stationary value of Q for a given U can be expressed as
* ⫽N r U 1/3
N rV opt⫽ 共 C V u opt 兲 D opt
1/3
(39)
dQ⫽
⳵Q
⳵ Bi 冉冊冉冊
dBi⫹
⳵Q
⳵u
du⫽0 (29) The important characteristic of this formulation is that the above
expressions are independent of the actual value of q f or V which-
ever is specified. For specified opening angle ␥ and emissivity ⑀
which after inserting the constrain Eq. 共28兲 is written as:
they depend on the fin’s profile. Therefore the dimensionless co-
冉冊冉冊冉冊冉冊
⳵Q
⳵ Bi
⳵U
⳵u
⫺
⳵Q
⳵u
⳵U
⳵ Bi
⫽0 (30)
efficients, Eqs. 共32–39兲, can be correlated as functions of ␥ and ⑀.
From the above eight expressions, only the parameters u opt and
D opt need to be calculated, since they all depend on u opt and D opt.
Introducing now the expressions for Q and U 关Eqs. 共22兲 and 共24兲兴 The correlations used for both the rectangular 共␭⫽1兲 and the tri-
into the above relationship yields: angular 共␭⫽0.01兲 profiles are the following:
dD 共 ⑀ u 2opt兲 ⫽ 共 u 2opt兲 s f ⫹a u exp共 ⫺b u ␥ 兲 (40)

3u ⫺D⫽0 (31) ⫺1/2
du ⑀ D opt⫽ 共 D opt兲 s f ⫺a D exp共 ⫺b D ␥ 兲 (41)
Therefore, the optimum value u opt is the root of Eq. 共31兲, and where the opening angle is in radians, and the correlation coeffi-
since in this case there is not analytical expression available, it is cients are polynomials of the emissivity for both profiles
numerically calculated. It must be emphasized here that in Eq.
共31兲 the dimensionless volume U or heat dissipation Q do not
appear in Eq. 共31兲 and the optimum values of u and D are inde-
z⫽ 兺 c⑀,
i
i
i
0.50⭐ ⑀ ⭐0.90, z⫽a u ,b u ,a D ,b D (42)
pendent of the particular values of Q or U whichever is being
given. They depend only on the fin’s profile. Thus the problem The polynomial coefficients c i in Eq. 共42兲 can be found in Table 1
defined by Eqs. 共13–15兲 with the corresponding boundary condi- for the rectangular profile and in Table 2 for the triangular profile.
tions Eqs. 共16–17兲, are solved numerically and the optimization is
performed using the univariate minimization routine DUVMIF Table 2 Coefficients for the triangular profile, Eqs. „39… and
from the IMSL library, IMSL 关21兴, with a requested absolute ac- „40…
curacy of 10⫺4 in the calculated optima. In practical applications
the designers are interested in the actual optimum fin dimensions
w opt and L opt instead of those of u opt or Biopt. This is accom-
plished, by introducing the values of u opt and L opt into Eqs. 共10兲
and 共22兲 to define certain dimensionless profile coefficients first
proposed by Razelos 关22兴. Two sets of these coefficients are
given. One when the optimization process is based on specified
heat dissipation q f , and the second when the fin volume V is
given.
Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 807
Table 3 Single fin values for Eqs. „39… and „40… obtained an estimate E n⫹1 of the local truncation error in the
formula of order 4, which may be used to adjust the step size
⌬x n . The step size control algorithm
⌬x n⫹1 ⫽0.9⌬x n 冉 TOL

储 E n⫹1 储 ⬁ 冊 1/5
(47)
is in common use and this formula is used whether or not the

tolerance TOL 共requested accuracy which is supplied by the user兲
is exceeded 储 E n⫹1 储 ⬁ . Thus in one case a rejected step is being
In correlations 共40兲 and 共41兲 the corresponding values of the recomputed and in the other case the next step is being predicted.
single fin configuration 共without fin to fin radiant interaction兲 are In practice the propagated solution of the problem is the one pro-
used 共Razelos and Kakatsios 关23兴兲. These values are shown in vided by the 5th order formula 共local extrapolation兲, as proposed
Table 3. by Shampine 关29兴. Since the function evaluations in Eq. 共13兲 are
It is worth mentioning the procedure that Karlekar and Chao not expensive a value of 10⫺8 was used for TOL as a compromise
关19兴 followed for the maximization of the heat dissipation of an between accuracy and computer time.
assembly when the volume is specified. The authors defined the The FIE Eq. 共15兲 was solved using both Simpson’s and Gauss-
dimensionless heat dissipation for the assembly as: Legendre quadrature formulas 共to ensure consistency兲 with an au-
tomatic error control algorithm, Atkinson 关30,31兴. Both methods
qT
␰⫽ (43) gave identical results 共the diagrams presented are calculated using
关共 h r T b 兲 kT b V T 兴 1/3
2
Gauss–Legendre integration formulas兲. The Nyström method is
where q T ⫽Nq f is the total heat rejection, V T ⫽NV is the total used for the discretization of the integral equation and an iterative
volume and N is the number of fins in the assembly. Substituting two-grid algorithm is used for the solution of the linear system
in the above relation we obtain: when it becomes large. The procedure begins by solving for a
discretization parameter m⫽m 0 共and l m 0 linear equations兲, with
␰⫽
Nq f
T b 共 h 2r kNV 兲 1/3
⫽N 冋
2/3
qf
T b 共 h 2r kV 兲 1/3 册 ⫽N 2/3Q * ⫽ 共 N 2 /V * 兲 1/3
(0)
an initial guess j m 0
⫽0 for the radiosity. Then the discretization
parameter m is increased until an approximate solution with suf-
(44) ficient accuracy is obtained. Given a tolerance ␦⬎0, an approxi-
It is evident from Eq. 共44兲 that maximizing ␰ is equivalent to mate solution for the radiosity j ␦ is computed which satisfies the
minimize V * for a constant N as it is done in 关19兴. Moreover, the criterion
present analysis in the paragraph entitled ‘‘OPTIMIZATION’’ is
储 j⫺ j ␦ 储 ⬁ ⬍ ␦ (48)
applicable to this problem provided that the Biot number is now
⫺6
defined by where j is the exact solution. ␦ ⫽10 was used for the computa-
冉冊 2/3 tions which resulted in a variation of the discretization parameter

UT
Bi⫽ (45) m between 32 and 128. It is worth noticing that the corresponding
Nu discretization parameter used by Karlekar and Chao 关19兴 was m
Consequently the problem of maximizing the heat dissipation of a ⫽16, while, no discretization information were given by Sparrow
fin assembly when its volume is given is equivalent to the mini- et al. 关18兴. The kernel function Eq. 共19兲, at the point (x,x ⬘ )
mization of the assembly volume when the desired heat dissipa- ⫽(0,0) becomes indefinite. To overcome this difficulty the exact
tion is given. The above statement will be used in the following value of the radiosity is employed, Sparrow et al. 关18兴, Karlekar
examples for comparison purposes. and Chao 关19兴,
⑀
j共 0 兲⫽ (49)
1⫺ 共 1⫺ ⑀ 兲兵关 1⫺sin共 ␥ ⫺120 deg兲兴 /2其
Numerical Solution
It should be noted here that the fin heat dissipation is computed
The integro-differential equation, 共13–15兲, was solved numeri- directly from the solution of the ordinary differential equation, Eq.
cally using an iterative procedure 共Modest 关24兴兲. First a tempera- 共46兲, using Eqs. 共22兲 and 共49兲. In this way the calculated solution
ture distribution is calculated and then inserted into the FIE Eq. is very accurate since Eq. 共49兲 is exact and the truncation error in
共15兲, which is solved for the radiosity and the procedure is re- the temperature derivative is of fifth order. The discrete tempera-
peated solving the ODE 共13兲 for an updated temperature distribu- ture distribution ⌰( ␰ i ), 0⭐ ␰ i ⭐1, obtained from the ODE solver
tion. Convergence is considered successful when the infinite norm and the corresponding discrete radiosity distribution j( ␰ n ), 0
of the temperature derivative is less than 10⫺6 . In order to accel- ⭐ ␰ n ⭐1, are used for the construction of a temperature and a
erate the convergence, the solution starts with the higher opening radiosity interpolating piecewise polynomials of the form:
angle first 共i.e., ␥⫽150 deg兲. The starting temperature employed is
k
the one corresponding to the single fin configuration, Razelos and 共 x⫺ ␰ i 兲 j⫺1
Kakatsios 关23兴, which is quite close to the exact one since at these ⌰ 1共 x 兲 ⫽ 兺
j⫽1
cij
共 j⫺1 兲 !
, ␰ i ⭐x⭐ ␰ i⫹1 (50)
high opening angles the mutual irradiation is less significant. The
second order, two-point BVP is transformed into a first order IVP and
with the following relations, k
共 x⫺ ␰ n 兲 j⫺1
⌰ 1 ⫽⌰, ⌰ 2 ⫽⌰ ⬘ (46) j共 x 兲⫽ 兺c
j⫽1
nj
共 j⫺1 兲 !
, ␰ n ⭐x⭐ ␰ n⫹1 (51)
which is then solved using a multi-shooting technique, Ascher
et al. 关25兴, Keller 关26兴, and a newton method for the iterations on where the breakpoint sequences ␰ i and ␰ n are provided by the
the unknown initial condition. The algorithm used is a variable ODE and the FIE solvers respectively. For the interpolants con-
step embedded Runge-Kutta algorithm of order 5共4兲, Papakostas structed above k⫽8, was used 共degree k⫺1兲 so that the right hand
et al. 关27兴, Tsitouras and Papageorgiou 关28兴. A procedure utilizing side of Eq. 共13兲 has continuous derivatives. In this way full ad-
such a pair advances the integration from x n to x n⫹1 ⫽x n ⫹⌬x n , vantage is taken of the accuracy characteristics of the ODE solver
computing at each step two approximations of the solution of since an optimum Runge-Kutta algorithm of order 5 must agree
orders 5 and 4 respectively. From the embedded form it can be with the Taylor expansion of the same order and minimize as
Fig. 2 Optimum parameter ⑀ u 2opt as a function of the opening Fig. 3 ⑀ À1Õ2D opt versus the opening angle for various emissivi-
angle for various emissivities and rectangular profile ties and rectangular profile
many as possible error coefficients of higher orders. It is therefore Example 1. Determine the optimum dimensions of the tubu-
evident that the derivative continuation is essential for an accurate lar radiator of Fig. 1, operating in free space at 0 K, where the
and reliable solution 共temperature distribution兲. Furthermore there base temperature is T b ⫽500 K. The emissivity is 0.75 and the fin
is no need to integrate over the fin surface the radiosity and the thermal conductivity is k⫽173 W/共mK). The radiator is designed
irradiation, reducing in this way the accuracy of the numerical to dissipate q T ⫽2400 W per unit tube length with N⫽6 fins of
scheme, as it is obvious that an accurate and reliable calculation of rectangular profile.
the fin heat dissipation is very important for the minimization Solution. First we calculate the opening angle ␥⫽360 deg/
process. In the solution the equivalence relation, Eq. 共30兲, is taken 6⫽60 deg. Then q f and h r are calculated as follows: q f ⫽q T /N
into consideration, that is for a given value of U an optimum
共maximum兲 value of Q is calculated which is equivalent to calcu-
late an optimum 共minimum兲 U from the given value of Q. Calcu-
lations were carried out using values of the dimensionless volume
in the range U⫽10⫺5 ⫼10⫺3 obtaining the same optimum. These
calculations were obtained to verify and numerically that the re-
sults obtained earlier who have shown that the optimum values of
u and D are independent of the values of U or Q whichever was
specified for the optimization process.
Results and Discussion

Figure 2 shows the optimum parameter ⑀ u 2opt as a function of
the opening angle for various emissivities between 0.50 and 0.90
for the rectangular profile. As the opening angle ␥→180 deg the
optimum radiation conduction coefficient becomes independent of
the emissivity. For this limiting value the view factor in Eq. 共3兲
also vanishes and the radiosity equals ⑀ ⌰ 4 . Thus in the governing
Equation 共13兲, as it is expected, there are no terms involving ra-
diation interaction and the problem reduces to a single fin con- Fig. 4 Optimum parameter ⑀ u 2opt as a function of the opening
figuration. This is a well-studied case, Wilkins 关5兴, Razelos and angle for various emissivities and triangular profile
Kakatsios 关23兴共Table 1兲 and the optimum radiation-conduction
parameter is (u 2opt) s f ⫽0.8464. This trend is a confirmation of the
consistency of the numerical solution achieved. For comparison
purposes the corresponding results of Sparrow et al. 关18兴 are in-
dicated in Fig. 2. The data have been obtained from copies of their
original figures and this circumstance explains the occasional scat-
ter in their points, which appears in our figure since no smoothing
has been applied on the data. The agreement is better at high
opening angles, while there are deviations at smaller angles. Since
the mathematical formulation of the problem is the same the dif-
ferences can only be attributed to the numerical methods em-
ployed. Figure 3 shows the optimum function ⑀ ⫺1/2D opt versus the
opening angle for various emissivities between 0.50 and 0.90 for
the rectangular profile. Again as ␥→180 deg D opt becomes inde-
pendent of the emissivity and tends to the single fin configuration
value, which is (D opt) s f ⫽0.5201. Figures 4 and 5 show the cor-
responding results for the triangular profile. For the purpose of
illustration and comparison the use of Eqs. 共32兲 to 共38兲 is demon- Fig. 5 ⑀ À1Õ2D opt versus the opening angle for various emissivi-
strated by two examples. ties and triangular profile
Table 4 Results and comparisons for example 1 two problems are equivalent, as in the works of Razelos and Imre
关20兴, Razelos and Kakatsios 关23兴, Razelos and Krikkis 关16兴. The
first important conclusion is that using the simplifying assump-
tions of the length of arc idealization and/or the insulated tip the
optimum fin dimensions and the profile coefficients defined in Eqs.
共32–39兲 depend only on the opening angle ␥, the emissivity ⑀ and
the profile of the fin and not on the specific values of Q or U.
The second important conclusion, for the problem treated by
Karlekar and Chao 关19兴, is that the maximization of the heat dis-
Table 5 Results and comparisons for example 2 sipation of a fin assembly when its volume is given is equivalent to
the minimization of the assembly volume when the desired heat
dissipation is given. Furthermore we suggest a fast and accurate
procedure suitable for all practical engineering purposes. The
present procedure can be easily adapted for computer calculations,
although there is a direct calculation from diagrams.
Nomenclature
A ⫽ surface area 关m2/m兴
Ap ⫽ fin profile area 关m2兴
Bi ⫽ (h r w/k) radiation Biot number
BL ⫽ dimensionless parameter, Eq. 共18兲
⫽2400/6⫽400 W/m and h r ⫽ ␴ T 3b ⫽5.67⫻10⫺8 (500) 3
CV ⫽ 关1/2共1⫹␭兲兴 ratio of fin volume to the volume of
⫽7.09 W/共m K). From Eq. 共42兲 and Table 1 we obtain the corre-
2
constant thickness fin
lation coefficients as a u ⫽2.018, b u ⫽1.6062 and a D ⫽0.5668, dF dx⫺dx ⬘ ⫽ view factor
b D ⫽1.2818. From Eqs. 共40兲 and 共41兲 we obtain u opt⫽1.276 and D ⫽ function defined by Eq. 共22兲
D opt⫽0.322. From Figs. 2 and 3 we obtain respectively u opt hr ⫽ ( ␴ T 3b ) black radiation heat transfer coefficient
⫽1.269 and D opt⫽0.308 which differ by 0.55 percent and 4.5 关W/共m2K兲兴
percent respectively from the correlations. Then the dimensionless
H ⫽ fin length 关m兴
semi thickness w * and height L * are calculated as w * j ⫽ dimensionless radiosity
⫽0.25/D 2opt⫽2.4112 and L * ⫽0.5u opt /D opt⫽1.9217. Now the op- J ⫽ radiosity 关W/m2兴
timum fin dimensions w opt and L opt can be obtained from Eqs. k ⫽ fin thermal conductivity 关W/共mK兲兴
共32兲 and 共33兲 respectively: w opt⫽1.26 mm and L opt⫽217 mm. K ⫽ kernel in integral equation, Eq. 共19兲
The optimum fin volume is calculated as V opt⫽2w optL opt L ⫽ fin height 关m兴
⫽546 mm3 /m. From the fin height now the value of parameter B L N ⫽ number of fins in the assembly
is obtained as: B L ⫽h r L/k⫽8.89⫻10⫺3 which provides justifica- Nr ⫽ removal number
tion for the use of ⌰⬘共1兲⫽0 instead of the boundary condition Eq. O ⫽ order of magnitude
共17兲. The results are summarized in Table 4 together with the P ⫽ perimetry 关m/m兴
corresponding results obtained by the methodologies of references qf ⫽ fin heat loss per unit length 关W/m兴
关18兴, 关19兴 and 关14兴, where an outside tube radius r i ⫽0.015 m was qr ⫽ dimensionless radiative heat flux
assumed. Q ⫽ dimensionless fin heat dissipation
Example 2. Determine the optimum fin dimensions of a ra- Qr ⫽ radiative heat flux 关W/m2兴
diator consisting of a tube fitted with uniformly spaced triangular S ⫽ arc length 关m兴
fins which is designed to dissipate q T ⫽2000 W/m when operating T ⫽ temperature 关K兴
with a base temperature of 500 K. Furthermore, k x ⫽ (X/L) dimensionless distance along fin
⫽173 W/共m K), ⑀⫽0.90, and ␥⫽72 deg 共N⫽5 fins兲. Repeat the X ⫽ distance along fin 关m兴
calculations when ⑀⫽0.50 and ␥⫽51.43 deg 共N⫽7 fins兲. y ⫽ (Y /w) dimensionless fin semi-thickness, Eq. 共20兲
Y ⫽ fin semi-thickness 关m兴
Solution. Using the same methodology as in the previous ex- u ⫽ radiation-conduction coefficient, Eq. 共10兲
ample we obtain q f ⫽2000/5⫽400 W/m, u opt⫽1.0834 and D opt U ⫽ dimensionless fin volume
⫽0.31437. Then the dimensionless semi thickness w * and height V ⫽ fin volume 关m3/m兴
L * are calculated as w * ⫽2.5296 and L * ⫽1.7231. Now the opti- Vc ⫽ (2Lw) rectangular fin volume 关m3/m兴
mum fin dimensions w opt and L opt can be obtained from Eqs. 共32兲 w ⫽ base semi-thickness 关m兴
and 共33兲 respectively w opt⫽1.3 mm and L opt⫽194.4 mm. The op- Greek Symbols
timum fin volume is calculated as V opt⫽2w optL optC V
⫽253 mm3 /m. For the second case where ⑀⫽0.50 and ␥⫽51.43 ␥ ⫽
opening angle 关deg兴
deg 共N⫽7 fins兲 the corresponding results are summarized in Table ␥f 关 2 tan⫺1(w/L)兴 triangular profile semi angle 关deg兴
⫽
5 together with the corresponding results obtained by the method- ⑀ ⫽
emissivity
ology of Karlekar and Chao 关19兴 and Schnurr et al. 关14兴, where an ⌰ ⫽
(T/T b ) dimensionless temperature
outside tube radius r i ⫽0.01 m was assumed. ␭ ⫽
tip to base fin semi-thickness ratio
␰ ⫽
dimensionless assembly heat dissipation used by Kar-
Conclusions lekar and Chao 关19兴
␴ ⫽ Stefan-Boltzmann constant
In the present study the problem of designing the optimum fin
for the tubular space radiator is solved. The governing equations Subscripts—Superscripts
are formulated by means of dimensionless variables. The optimi- 共⬘兲 ⫽ derivative with respect to x
zation is carried out for two cases, that is: First, when the heat b ⫽ fin base
dissipation is being given and the fin volume must be minimized, D ⫽ reference to function D, Eq. 共22兲
and second when the fin volume is being given and the total heat f ⫽ fin
dissipation must be maximized. It has been shown here that the opt ⫽ optimum
r ⫽ radiative transfer 关14兴 Schnurr, E. M., Shapiro, A. B., and Townsend, M. A., 1976, ‘‘Optimization of
sf ⫽ single fin configuration 共without mutual irradiation兲 Radiating Fin Arrays With Respect to Weight,’’ ASME J. Heat Transfer, 98, pp.
643– 648.
T ⫽ total 共assembly兲关15兴 Aziz, A., and Kraus, A. D., 1996, ‘‘Optimum Design of Radiating and
u ⫽ reference to conduction-radiation parameter u Convective-Radiating Fins,’’ Heat Transfer Eng., 17共3兲, pp. 44 –78.
Abbreviations 关16兴 Razelos, P., and Krikkis, R. N., 2001, ‘‘Optimum Design of Longitudinal Rect-
angular Fins with Base to Fin Radiant Interaction,’’ Heat Transfer Eng., 22共3兲,
BVP ⫽ Boundary Value Problem pp. 3–17.
FIE ⫽ Fredholm Integral Equation 关17兴 Razelos, P., and Georgiou, E., 1992, ‘‘Two-Dimensional Effects and Design
IVP ⫽ Initial Value Problem Criteria for Convective Extended Surfaces,’’ Heat Transfer Eng., 13共3兲, pp.
38 – 48.
LAI ⫽ Length of Arc Idealization 关18兴 Sparrow, E. M., Eckert, E. R. G., and Irvine, T. F., Jr., 1961, ‘‘The Effective-
ODE ⫽ Ordinary Differential Equation ness of Radiating Fins with Mutual Irradiation,’’ J. Aerosp. Sci., 28, pp. 763–
778.
References 关19兴 Karlekar, B. V., and Chao, B. T., 1963, ‘‘Mass Minimization of Radiating
Trapezoidal Fins with Negligible Base Cylinder Interaction,’’ Int. J. Heat Mass
关1兴 Bartas, J. G., and Sellers, W. H., 1960, ‘‘Radiation Fin Effectiveness,’’ ASME Transf., 6, pp. 33– 48.
J. Heat Transfer, 82, pp. 73–75. 关20兴 Razelos, P., and Imre, K., 1980, ‘‘The Optimum Dimensions of Circular Fins
关2兴 Kern, D. Q., and Kraus, A. D., 1972, Extended Surface Heat Transfer, with Variable Thermal Parameters,’’ ASME J. Heat Transfer, 102, pp. 420–
McGraw-Hill, New York. 425.
关3兴 Liu, C. Y., 1960, ‘‘On Minimum Weight Rectangular Radiating Fins,’’ J.
关21兴 IMSL, Inc., 1991, IMSL Library Reference Manual, Houston, TX.
Aerosp. Sci., 27, p. 871.
关22兴 Razelos, P., 1979, ‘‘The Optimization of Convective Fins with Internal Heat
关4兴 Liu, C. Y., 1961, ‘‘On Optimum Rectangular Cooling Fins,’’ Q. Appl. Math.,
19, p. 72. Generation,’’ Nucl. Eng. Des., 52共2兲, pp. 289–299.
关5兴 Wilkins, J. E., Jr., 1960, ‘‘Minimizing the Mass of Thin Radiating Fins,’’ J. 关23兴 Razelos, P., and Kakatsios, X. K., 2000, ‘‘Optimum Dimensions of
Aerosp. Sci., 27, pp. 145. Convective-Radiating Fins: Part I—Longitudinal Fins,’’ Appl. Therm. Eng.,
关6兴 Chung, B. T. F., and Nguyen, L. D., 1986, ‘‘Optimization of Design Param- 20共13兲, pp. 1161–1192.
eters for Radiating Longitudinal Fins of Various Geometries,’’ 24th AIAA 关24兴 Modest, M. F., 1993, Radiative Heat Transfer, McGraw-Hill, New York.
Aerospace Sciences Meeting, Reno, NV, Paper AIAA-86-0150. 关25兴 Ascher, U. M., Mattheij, R. M. M., and Russel, R. D., 1995, Numerical Solu-
关7兴 Chung, B. T. F., and Zhang, B. X., 1991, ‘‘Minimum Mass Longitudinal Fins tion of Boundary Value Problems for Ordinary Differential Equations, 2nd ed.,
with Radiation Interaction at the Base,’’ J. Franklin Inst. 328共1兲, pp. 143–161. SIAM, Philadelphia, PA.
关8兴 Chung, B. T. F., and Zhang, B. X., 1991, ‘‘Optimization of Radiating Fin Array 关26兴 Keller, H. B., 1992, Numerical Methods for Two-Point Boundary-Value Prob-
Including Mutual Irradiations Between Radiator Elements,’’ ASME J. Heat lems, Dover, New York.
Transfer, 113, pp. 814 – 822. 关27兴 Papakostas, S. N., Tsitouras, Ch., and Papageorgiou, G., 1993, ‘‘A General
关9兴 Chung, B. T. F., Zhang, B. X., and Lee, E. T., 1996, ‘‘A Multi-Objective Family of Explicit Runge-Kutta Pairs of Order 6 and 5,’’ National Technical
Optimization of Radiative Fin Array Systems in a Fuzzy Environment,’’ University of Athens, Report NA-1/93.
ASME J. Heat Transfer, 118, pp. 642– 649. 关28兴 Tsitouras, Ch., and Papageorgiou, G., 1990, ‘‘Runge-Kutta Interpolants Based
关10兴 Krishnaprakas, C. K., 1996, ‘‘Optimum Design of Radiating Rectangular Plate
on Values from Two Successive Integration Steps,’’ Computing, 43, pp. 255–
Fin Array Extending From a Plane Wall,’’ ASME J. Heat Transfer, 118, pp.
266.
490– 493.
关11兴 Krishnaprakas, C. K., 1997, ‘‘Optimum Design of Radiating Longitudinal Fin 关29兴 Shampine, L. F., 1973, ‘‘Local Extrapolation in the Solution of Ordinary Dif-
Array Extending From a Cylindrical Surface,’’ ASME J. Heat Transfer, 119, ferential Equations,’’ Math. Comput., 27, pp. 91–97.
pp. 857– 860. 关30兴 Atkinson, K. E., 1976, ACM TOMS, Vol. 2, pp. 196 –199.
关12兴 Sunil, Kumar S., Venketesh, Nayak, and Venkateshan, S. P., 1992, ‘‘Optimum 关31兴 Atkinson, K. E., 1997, The Numerical Solution of Integral Equations of the
Finned Space Radiators,’’ Int. J. Heat Fluid Flow, 14共2兲, pp. 191–200. Second Kind, Cambridge University Press, New York.
关13兴 Ramesh, N., and Venkateshan, S. P., 1997, ‘‘Optimum Finned Tubular Space 关32兴 Kraus, A. D., Aziz, A., and Welty, J., 2001, Extended Surface Heat Transfer,
Radiator,’’ Heat Transfer Eng., 18共4兲, pp. 69– 87. John Wiley, New York.
Nonstaggered APPLE Algorithm
for Incompressible Viscous Flow
S. L. Lee
in Curvilinear Coordinates
e-mail: sllee@pme.nthu.edu.tw The NAPPLE algorithm for incompressible viscous flow on Cartesian grid system is
extended to nonorthogonal curvilinear grid system in this paper. A pressure-linked equa-
Y. F. Chen tion is obtained by substituting the discretized momentum equations into the discretized
continuity equation. Instead of employing a velocity interpolation such as pressure-
Department of Power Mechanical Engineering, weighted interpolation method (PWIM), a particular approximation is adopted to circum-
National Tsing-Hua University, vent the checkerboard error such that the solution does not depend on the under-
Hsinchu 30043, Taiwan relaxation factor. This is a distinctive feature of the present method. Furthermore, the
pressure is directly solved from the pressure-linked equation without recourse to a
pressure-correction equation. In the use of the NAPPLE algorithm, solving the pressure-
linked equation is as simple as solving a heat conduction equation. Through two well-
documented examples, performance of the NAPPLE algorithm is validated for both
buoyancy-driven and pressure-driven flows. 关DOI: 10.1115/1.1484109兴
Keywords: Computational, Finite Difference, Flow, Heat Transfer, Numerical Methods
Introduction factor. Such numerical difficulties were encountered also in a

similar formulation proposed by Peric 关10兴 for nonstaggered cur-
Many important industrial flow processes have boundaries of
vilinear nonorthogonal grid systems.
irregular shape. Heat exchangers, airfoils, turbomachines, cooling
More recently, Lee and Tzong 关11兴 developed the APPLE 共Ar-
objects, and combustors are some of the examples. For these flow
tificial Pressure for Pressure-Linked Equation兲 and NAPPLE
configurations, the Navier-Stokes equation might be efficiently
共APPLE on Nonstaggered grids兲 algorithms for Cartesian grid
solved on a Cartesian grid system with an adequate numerical
systems. In both algorithms, the pressure-linked equation is solved
scheme 关1兴 in case the boundary layer thickness is sufficiently
directly without recourse to a pressure-correction equation. The
large. However, very fine grids are generally needed when the
pressure-linked equation used in the APPLE algorithm is an alge-
boundary layer is thin. Under this situation, use of a body-fitted
braic combination of the discretized continuity equation and the
curvilinear grid system would be more practical.
discretized momentum equations. Thus, the solution satisfying the
In an early study, Rhie and Chow 关2兴 derived a pressure-
pressure-linked equation and the discretized momentum equations
correction equation for nonstaggered curvilinear nonorthogonal
is also the exact solution of the discretized continuity equation and
grids. The cross derivative terms due to grid skewness were
the discretized momentum equations. This is evidenced from the
dropped to retain the structure of the five-point approximation. To
fact that the APPLE algorithm is able to produce velocity having
ensure a strong pressure-velocity coupling, a pressure-weighted
mass-free residual. In the NAPPLE algorithm, however, an ap-
interpolation method 共PWIM兲 was proposed. This precluded the
proximation is imposed on the pressure-linked equation to circum-
checkerboard error for the pressure and allowed the SIMPLE al-
vent the checkerboard error. Due to the truncation error arising
gorithm be performed on nonstaggered grid systems. The PWIM
from this approximation, the pressure-linked equation is no longer
was extended to three-dimensional problems 关3,4兴 and unsteady
an exact combination of the discretized continuity equation and
situation 关5兴. However, the solution provided by PWIM depends
the discretized momentum equations. As a result, there is a certain
on the under-relaxation factor as pointed out by Patankar 关6兴. Fur-
level of mass residual after the solution perfectly converges. Nev-
thermore, the PWIM could produce physically impossible velocity
ertheless, the mass residual steadily approaches zero while the
when the pressure gradient has a rapid variation as remarked by
grid mesh is decreasing as demonstrated by Lee and Tzong 关11兴.
Miller and Schmidt 关7兴.
For curvilinear grids, both velocity components 共u and v 兲 are
Shyy and coworkers 关8,9兴 were one of the early groups to apply
needed at the same point when the discretized momentum equa-
computational fluid dynamics on nonorthogonal grids. Staggered
tions are substituted into the discretized continuity equation to
grid systems were used. Like in the standard SIMPLE algorithm
form a pressure-linked equation 共or pressure-correct equation兲.
on Cartesian grid system, Shyy et al. 关8兴 proposed a pressure-
This implies the need of a velocity interpolation procedure if a
correction equation that took the mass residual of the starred ve-
staggered grid system is employed. Thus, the resulting pressure-
locity 共velocity from previous iteration兲 as the only nonhomoge-
linked equation for staggered grid systems is not an exact combi-
neous term. This was done in hopes that the mass residual would
nation of the discretized equations due to the interpolation errors.
approach zero to achieve zero pressure correction after a number
Under such a situation, no zero mass residual is guaranteed. This
of iterations. Unfortunately, the mass residual never disappeared
might account for the great numerical difficulties encountered by
in their studies 关8,9兴. This led to a nonzero pressure correction and
Shyy and coworkers 关8,9兴. For nonstaggered grids, a particular
thus a changing pressure that never converged. In addition, their
treatment such as PWIM proposed by Rhie and Chow 关2兴 and the
numerical results were found to depend on the under-relaxation
interpolation procedure used by Peric 关10兴 is needed for the suc-
cess of pressure-velocity coupling as mentioned earlier. Unfortu-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division September 1, nately, these treatments are unavoidable to introduce numerical
2000; revision received April 1, 2002. Associate Editor: D. A. Kaminski. errors to the pressure-linked equation such that mass-free residual
812 Õ Vol. 124, OCTOBER 2002 Copyright © 2002 by ASME Transactions of the ASME
is not guaranteed either. The nonzero mass residual continuously
produces pressure correction (p ⬘ ⫽0) and thus gives rise to end-
less iterations.
Logically, the mass residual will not vanish for both staggered
and nonstaggered curvilinear grids due to the need of velocity
interpolation. Obviously, the principal merit of the APPLE algo-
rithm that guarantees zero mass residual on staggered Cartesian
grids will be lost when extended to curvilinear grid systems. In the
present study, the NAPPLE algorithm for Cartesian grids 关11兴 is
extended to curvilinear grids. Its performances will be examined
through two well-documented examples.
The NAPPLE Algorithm Fig. 1 A schematic coordinate transformation from „a… the
physical plane „ x , y … to „b… the computational plane „␰,␩…
The dimensionless conservation equations for heat convection
in a two-dimensional incompressible viscous flow can be written
as
⳵u ⳵v ⳵␪ ⳵␪ ⳵␪
⫹ ⫽0 (1) Pr Rec J ⫹Pr Rec U ⫹Pr Rec V
⳵x ⳵y ⳵t ⳵␰ ⳵␩
Rec 冉 ⳵u
⳵t
⫹u
⳵u
⳵x
⫹v
⳵u
⳵y
⫽⫺ 冊
⳵ p * ⳵ 2u ⳵ 2u
⳵x
⫹ 2⫹ 2
⳵x ⳵y
(2)
⫽
⳵ ␣ ⳵␪
⳵␰ J ⳵␰冉冊冉冊冉冊冉冊
⫹
⳵ ␥ ⳵␪
⳵␩ J ⳵␩
⫺
⳵ ␤ ⳵␪
⳵␰ J ⳵ ␩
⫺
⳵ ␤ ⳵␪
⳵ ␩ J ⳵␰
(9)
Rec冉⳵v
⳵t
⫹u
⳵v
⳵x
⫹v
⳵v
⳵y
⫽⫺
⳵y 冊
⳵ p * ⳵ 2u ⳵ 2u
⫹ 2⫹ 2⫹
⳵x ⳵y
Gr
Rec
共 ␪ ⫺ ␪ ref兲 where
U⫽y ␩ u⫺x ␩ v , V⫽⫺y ␰ u⫹x ␰ v (10a)
(3)
冉冊
J⫽x ␰ y ␩ ⫺x ␩ y ␰ , ␣ ⫽x ␩ ⫹y ␩ ,
2 2
␤ ⫽x ␰ x ␩ ⫹y ␰ y ␩ , ␥ ⫽x 2␰ ⫹y 2␰
⳵␪ ⳵␪ ⳵␪ ⳵ 2␪ ⳵ 2␪ (10b)
Pr Rec ⫹u ⫹v ⫽ 2⫹ 2 (4)
⳵t ⳵x ⳵y ⳵x ⳵y
The subscripts ␰ and ␩ appearing in the geometric quantities x ␰ ,
where the gravity acceleration is assigned in the ⫺y direction. x ␩ , y ␰ , y ␩ represent partial derivatives with respect to ␰ and ␩,
The variables in equations 共1兲–共4兲 have been non-dimensionalized respectively.
with the characteristic length L, characteristic velocity U c , and For convenience, the grid system in Fig. 1 is generated such
characteristic temperature difference ⌬T. The Grashof number, that the grid points on the computational coordinates are express-
the characteristic Reynolds number, and the dimensionless pres- ible as
sures are defined, respectively, by ␰ i ⫽ 共 i⫺1 兲 ⌬ ␰ , i⫽1,2 . . . ,m
(11)
g ␤ ⌬TL 3
U cL P⫺ P ref ␩ j ⫽ 共 j⫺1 兲 ⌬ ␩ , j⫽1,2 . . . ,n
Gr⫽ , Rec ⫽ , p⫽ , p * ⫽p Rec
v2 v ␳ U 2c The point P( ␰ i , ␩ j ) is numbered as the kth point with k⫽(i
(5) ⫺1)n⫹ j. Physical coordinates (x,y) and variable ␾ at point P
are simply denoted by (x k ,y k ) and ␾ k , respectively. Upon dis-
where P ref is a reference pressure. After introducing a coordinate cretizing Eq. 共7兲 at point P with a finite-difference-like method
transformation from (x,y) to 共␰,␩兲 as illustrated in Fig. 1, Eqs. such as the weighting function scheme 关12,13兴, one obtains an
共1兲–共4兲 become algebraic equation of the form
⳵U ⳵V 共 a SW 兲 k u k⫺n⫺1 ⫹ 共 a W 兲 k u k⫺n ⫹ 共 a NW 兲 k u k⫺n⫹1 ⫹ 共 a S 兲 k u k⫺1
⫹ ⫽0 (6)
⳵␰ ⳵ ␩ ⫹ 共 a N 兲 k u k⫹1 ⫹ 共 a SE 兲 k u k⫹n⫺1 ⫹ 共 a E 兲 k u k⫹n
⳵u ⳵u ⳵u ⫹ 共 a NE 兲 k u k⫹n⫹1 ⫺ 共 a P 兲 k u k
冉冊
Rec J ⫹Rec U ⫹Rec V
⳵t ⳵␰ ⳵␩ ⳵p* ⳵p*
冉冊冉冊冉冊
⫺ y␩ ⫺y ␰ ⫽共 au兲k (12)
⳵p* ⳵p* ⳵ ␣ ⳵u ⳵ ␥ ⳵u ⳵␰ ⳵␩
⫽⫺ y ␩ ⫺y ␰ ⫹ ⫹
⳵␰ ⳵␩ ⳵␰ J ⳵␰ ⳵␩ J ⳵␩ Like the conventional notation system 关11,12,14兴, the subscripts
冉冊冉冊
SW, W, NW, S, N, SE, E, NE in Eq. 共12兲 represent quantities at
⳵ ␤ ⳵u ⳵ ␤ ⳵u the grid points adjacent to point P as demonstrated in Fig. 1. More
⫺ ⫺ (7)
⳵␰ J ⳵ ␩ ⳵ ␩ J ⳵␰ detailed information on the discretized Eq. 共12兲 will be discussed
later. Note that the factor (a u ) k is the source term excluding the
⳵v ⳵v ⳵v pressure gradient.
Rec J ⫹Rec U ⫹Rec V Next, recast Eq. 共12兲 in the form
⳵t ⳵␰ ⳵␩
⫽⫺ ⫺x ␩冉冊冉冊冉冊
⳵p*
⳵␰
⫹x ␰
⳵p*
⳵␩
⫹
⳵ ␣ ⳵v
⳵␰ J ⳵␰
⫹
⳵ ␥ ⳵v
⳵␩ J ⳵␩
u k ⫽û k ⫺
共 a P兲k
1
冉 y␩
⳵p*
⳵␰
⫺y ␰
⳵p*
⳵␩ 冊 k
(13)
⫺
⳵ ␤ ⳵v
⳵␰ J ⳵ ␩冉冊冉冊
⫺
⳵ ␤ ⳵v
⳵ ␩ J ⳵␰
⫹J
Gr
Rec
共 ␪ ⫺ ␪ ref兲 (8) û k ⫽
1
共 a P兲k 冉兺 nb⫽1
8
共 a nb 兲 k u nb ⫺ 共 a u 兲 k 冊 (14)
where the subscript nb stands for the neighbors of point P. Simi- where J represents the pseudo-Jacobian (a P ) k if ␾ ⫽ p * , A⫽B
larly, discretization of Eq. 共8兲 at point P yields ⫽C⫽0, and S⫽⫺␧ˆ . Let a grid system with ⌬␰⫽⌬␩⫽1 be gen-
冉冊
erated for the problem. Next, applying the weighting function
1 ⳵p* ⳵p*
v k ⫽ v̂ k ⫺ ⫺x ␩ ⫹x ␰ (15) scheme 关12,13兴 to Eq. 共22兲 at point P(x k ,y k ), one gets an alge-
共 a P兲k ⳵␰ ⳵␩ k braic equation of the form
v̂ k ⫽
1
共 a P兲k 冉兺 8
nb⫽1
共 a nb 兲 k v nb ⫺ 共 a v 兲 k 冊 (16)
共 a SW 兲 k ␾ k⫺n⫺1 ⫹ 共 a W 兲 k ␾ k⫺n ⫹ 共 a NW 兲 k ␾ k⫺n⫹1 ⫹ 共 a S 兲 k ␾ k⫺1
⫺ 共 a P 兲 k ␾ k ⫹ 共 a N 兲 k ␾ k⫹1 ⫹ 共 a SE 兲 k ␾ k⫹n⫺1 ⫹ 共 a E 兲 k ␾ k⫹n
Substituting Eqs. 共13兲 and 共15兲 into Eq. 共10a兲, one has ⫹ 共 a NE 兲 k ␾ k⫹n⫹1 ⫽ 共 a R 兲 k (23)
U k ⫽Û k ⫺
␣ ⳵p*
冉
J * ⳵␰ k
⫹
␤ ⳵p*
J* ⳵␩ 冊冉冊 k
(17)
where
␣w ␣e
共 a W 兲k⫽ w 共 z 兲, 共 a E 兲k⫽ w 共 ⫺z e 兲 ,
V k ⫽V̂ k ⫹ 冉 ␤ ⳵p*
J * ⳵␰ k
⫺ 冊冉
␥ ⳵p*
J* ⳵␩ 冊 k
(18)
共 a S 兲k⫽
Jw f w
␥s
w 共 z 兲, 共 a N 兲k⫽
Je f
␥n
w 共 ⫺z n 兲 ,
Js f s Jn f
where Û k ⫽y ␩ û k ⫺x ␩ v̂ k , V̂ k ⫽⫺y ␰ û k ⫹x ␰ v̂ k , and (J * ) k ⫽(a P ) k .
Finally, substituting Eqs. 共17兲 and 共18兲 into the discretized conti-
nuity equation 共see Eq. 共6兲兲共 a SW 兲 k ⫽⫺2 冉冊
␤ sw
ŵ 共 z 兲 ŵ 共 z 兲 ,
J sw f s f w
冉冊
U k⫹n ⫺U k⫺n V k⫹1 ⫺V k⫺1
⫹ ⫽0 (19) ␤ nw
2⌬ ␰ 2⌬ ␩ 共 a NW 兲 k ⫽2 ŵ 共 z 兲 ŵ 共 z 兲 ,
J nw f n f w
(24)
冉冊
followed by making the approximations
␤ se
冉 ␣ ⳵p*
冊冉 ⫺
␣ ⳵p*
冊共 a SE 兲 k ⫽2 ŵ 共 z 兲 ŵ 共 z 兲 ,
J se f s f e
J * ⳵␰ k⫹n
2⌬ ␰
J * ⳵␰ k⫺n
⫽ 冉
⳵ ␣ ⳵p*
⳵␰ J * ⳵␰ 冊共 a NE 兲 k ⫽⫺2 冉冊
␤ ne
ŵ 共 z 兲 ŵ 共 z 兲 ,
J ne f n f e
冉 ␥ ⳵p*
冊冉 ⫺
␥ ⳵p*
冊
冉冊
Ck
J* ⳵␩ J* ⳵␩ ⳵ ␥ ⳵p* 共 a P 兲k⫽共 a W 兲k⫹共 a S 兲k⫹共 a N 兲k⫹共 a E 兲k⫹ ,
k⫹1 k⫺1
⫽ ⌬t
2⌬ ␩ ⳵␩ J* ⳵␩
冉 ␤ ⳵p*
冊冉 ⫺
␤ ⳵p*
冊
(20)
冉
共 a R 兲 k ⫽⫺ S k ⫹
Ck
共␾ 兲
⌬t 0 k 冊
J* ⳵␩ k⫹n
J
2⌬ ␰
* ⳵␩ k⫺n
⫽ 冉
⳵ ␤ ⳵p*
⳵␰ J * ⳵ ␩ 冊 Note that the implicit method has been applied to the unsteady
term, i.e.,
冉 ␤ ⳵p*
冊冉 ⫺
␤ ⳵p*
冊 ⳵␾ ␾⫺␾0
⫽ ␾ 0 ⫽ ␾ 共 ␰ , ␩ ,t⫺⌬t 兲
冉冊
, (25)
J * ⳵␰ k⫹1
J * ⳵␰ k⫺1 ⳵ ␤ ⳵p* ⳵t ⌬t
⫽
2⌬ ␩ ⳵ ␩ J * ⳵␰ Before the momentum Eqs. 共7兲 and 共8兲 are solved, the pressure
gradient is estimated from the guessed 共or updated兲 pressure so-
one arrives at a pressure-linked equation for the point (x k ,y k ) lution with a central difference scheme.
冉
⳵ ␣ ⳵p*
⳵␰ J * ⳵␰
⫹ 冊冉
⳵ ␥ ⳵p*
⳵␩ J* ⳵␩
⫺
⳵ ␤ ⳵p*
⳵␰ J * ⳵ ␩ 冊冉冊 In Eq. 共24兲, the weighting functions w f (z) and ŵ f (z), and the
grid Peclet numbers z w , z e , z s , z n are defined, respectively, by
冉冊
z
⳵ ␤ ⳵p* w f 共 z 兲⫽ ⬇ 关 0,共 1⫺0.1兩 z 兩兲 5 兴 ⫹ 关 0,z 兴
⫺ ⫽␧ˆ (21a) 1⫺exp共 ⫺z 兲
⳵ ␩ J * ⳵␰ (26)
z
Û k⫹n ⫺Û k⫺n V̂ k⫹1 ⫺V̂ k⫺1 ŵ f 共 z 兲 ⫽ ⬇ 共 2⫹0.3332z 2 ⫹0.0172z 4 兲 ⫺1
␧ˆ ⫽ ⫹ (21b) exp共 z 兲 ⫺exp共 ⫺z 兲
2⌬ ␰ 2⌬ ␩
and
It is interesting to note that the coefficient (a P ) k acts as a pseudo-
Jacobian for the pressure-linked equation if Eq. 共21兲 is regarded as A wJ w A eJ e B sJ s B nJ n
z w⫽ , z e⫽ , z s⫽ , z n⫽ . (27)
a heat conduction equation with heat generation. For a Cartesian ␣w ␣e ␥s ␥n
grid system 共␰ ⫽x and ␩ ⫽y兲, the pressure-linked Eq. 共21兲 reduces
to Eq. 共23兲 of Lee and Tzong 关11兴. where 关 a,b 兴 stands for the greater of a and b. For simplicity, the
arithmetic mean of properties at two adjacent grid points is used
to approximate the mean value over the interval between them.
Solution Method
For instance,
Equations 共7), (8兲, 共21兲, and 共9兲 constitute a system of partial
differential equations for the physical quantities u, v , p, and ␪. All A w ⫽ 共 A k⫺n ⫹A k 兲 /2, J w ⫽ 共 J k⫺n ⫹J k 兲 /2,
(28)
of these equations possess a common form, i.e., ␣ w ⫽ 共 ␣ k⫺n ⫹ ␣ k 兲 /2, ␤ sw ⫽ 共 ␤ k⫺n⫺1 ⫹ ␤ k 兲 /2
C
⳵␾
⳵t
⫹A
⳵␾
⳵␰
⫹B
⳵␾
⫽
⳵ ␣ ⳵␾
⳵ ␩ ⳵␰ J ⳵␰
⫹
⳵ ␥ ⳵␾
⳵␩ J ⳵␩
⫺ 冉冊冉冊冉冊
⳵ ␤ ⳵␾
⳵␰ J ⳵ ␩
System of algebraic equations defined by Eqs. 共23兲 and 共24兲 can
be efficiently solved with the SIS solver 关15兴. The convergence
冉冊
problem due to high grid aspect ratios encountered by Peric 关10兴
⳵ ␤ ⳵␾ can be remedied by using the SIS solver instead of the SIP solver
⫺ ⫹S (22)
⳵ ␩ J ⳵␰ 关16兴. This point has been well clarified by Lee 关15兴.
It is quite straightforward to show that the weighting factors for
the pressure-linked Eq. 共21兲 reduce to
␣w ␣e ␥s
共 a W 兲k⫽ , 共 a E 兲k⫽ , 共 a S 兲k⫽ ,
共 J*兲w 共 J*兲e 共 J*兲s
␥n
共 a N 兲k⫽ ,
共 J*兲n
␤ sw ␤ nw
共 a SW 兲 k ⫽⫺0.5 , 共 a NW 兲 k ⫽0.5 ,
共 J * 兲 sw 共 J * 兲 nw
(29)
␤ se ␤ ne
共 a SE 兲 k ⫽0.5 , 共 a NE 兲 k ⫽⫺0.5 ,
共 J * 兲 se 共 J * 兲 ne
共 a P 兲k⫽共 a W 兲k⫹共 a S 兲k⫹共 a N 兲k⫹共 a E 兲k , 共 a R 兲 k ⫽⫺ 共 ␧ˆ 兲 k
Suppose the velocity (u, v ) on the boundary ␰⫽0 is known such
that the contravariant velocity U k⫺n with n⫹1⭐k⭐2n can be
determined from Eq. 共10a兲. This gives rise to
Û k⫺n ⫽U k⫺n and 共 J * 兲 k⫺n ⫽ 共 a P 兲 k⫺n ⫽⬁ (30)
from Eq. 共17兲, and thus
␣w
共 a W 兲k⫽ ⫽0 (31) Fig. 2 Geometry description and a grid system for example 1
共 J*兲w
As a result, no pressure boundary condition is needed on a bound-
ary where the velocity is known. This is one of the major differ-
ences between the pressure-linked Eq. 共21兲 and the conventional Nevertheless, the fictitious time step ⌬t could be assigned infinity
pressure Poisson equation. Although both equations are of the to reach the steady-state solution directly 共see Eq. 共24兲兲 in case the
Poisson type, the former comes directly from the continuity equa- Reynolds number and the Rayleigh number are not large. The
tion, while the latter is essentially a divergence of the momentum results will be compared with that based on the stream-vorticity
equation. Their numerical characteristics thus are quite different formulation as well as the available numerical solutions and ex-
as remarked by Lee and Tzong 关11兴. perimental data from previous investigations.
It is very important to note that, to obtain a pressure solution
without checkerboard error, the present study employs the ap-
proximation 共20兲 instead of using a velocity interpolation proce- Example 1 Natural Convection in the Annulus Between Two
dure such as the PWIM 关2兴 and a similar method 关10兴. This pre- Horizontal Cylinders. Consider two horizontal cylinders of ra-
cludes the possibility of resulting a spurious velocity arising from dius R i and R 0 . Center of the inner cylinder lies above that of the
improper velocity interpolation. In addition, unlike in conven- outer cylinder by a distance E v as shown in Fig. 2. The surfaces of
tional numerical methods 关2,8 –10兴, the solution based on the both cylinders are maintained at their own uniform temperatures
present method does not depend on the under-relaxation factor. 共T i and T o 兲 such that a natural convection prevails in the annulus
This is the most important feature of the present method. Due to between them. Let the characteristic velocity be assigned as U c
the truncation error arising from this approximation, however, the ⫽ ␣ t /L. This leads to Rec⫽1/Pr and thus Gr/Rec⫽Ra. The char-
discretized pressure-linked Eqs. 共23兲 and 共29兲 is no longer an acteristic length L, the eccentricity e v , and the dimensionless tem-
exact combination of the discretized continuity Eq. 共19兲 and the perature ␪ are defined by
discretized momentum equations as mentioned earlier. A certain
level of mass residual thus could remain after the solution per- T⫺T o T⫺T o
L⫽R o ⫺R i , e v ⫽E v /L, ␪⫽ ⫽ (32)
fectly converges. Nevertheless, the mass residual is expected to ⌬T T i ⫺T o
disappear when the grid size approaches zero.
The associated boundary conditions are
u⫽0, ⳵ v / ⳵ x⫽0, ⳵ ␪ / ⳵ x⫽0 on the y-axis
Performances of the NAPPLE Algorithm
Two well-known examples are employed in this section to ex- u⫽0, v ⫽0, ␪ ⫽1 on the inner cylinder (33)
amine the performances of the NAPPLE algorithm. Example 1, a u⫽0, v ⫽0, ␪ ⫽0 on the outer cylinder
natural convection in the annulus between two horizontal cylin-
ders, is conducted to study the performance of the NAPPLE algo- In the present study, steady-state solutions are obtained for the
rithm in the presence of buoyancy term. In example 2, the prob- three cases having the parameters (e v ,Ra)⫽(0.652, 4.80⫻104 ),
lem of fluid flow through an axisymmetric constricted tube is used (0, 4.70⫻104 ), (⫺0.623, 4.93⫻104 ), while R o /R i ⫽2.6 and Pr
to investigate the performance of the NAPPLE algorithm when ⫽0.706. The numerical procedure is similar to that for the Carte-
the flow is driven by a pressure difference. It is important to note sian grid system 关11兴. Once the temperature solution is available,
that the coefficient matrix formed with Eq. 共23兲 is no longer di- the equivalent thermal conductivity along the boundaries is evalu-
agonally dominant for steady-state problems due to the grid skew- ated from
ness 共␤⫽0兲. This might lead to a numerical difficulty in achieving
a convergent solution as reported by Shyy et al. 关8兴 and Peric 关10兴. Ri Ro ⳵␪ Ro Ro ⳵␪
共 k eq 兲 i ⫽ ln ⫺ , 共 k eq 兲 o ⫽ ln ⫺
In the present study, both examples 1 and 2 are treated as unsteady L Ri ⳵n L Ri ⳵n
problems with made-up initial conditions, although they are actu- (34a)
冉冊冉冊
ally at steady-state. Such a treatment could produce a strictly di-
agonally dominant coefficient matrix for the discretized momen- ⳵␪ ␤ ⳵␪ ␥ ⳵␪
⫽⫺ ⫹ (34b)
tum equations, and thus gains a good numerical stability. ⳵n J ⳵␰ J ⳵␩
Fig. 3 Influence of grid mesh on the equivalent thermal con- Fig. 4 Comparison of k eq among the numerical predictions
ductivity k eq and the experimental result for „ e v ,Ra…Ä„0.652,4.8Ã104 …
where n is a coordinate normal to the boundary specified by positive 共counterclockwise兲 circulation flow near ␾⫽0 deg. It
␩⫽constant. When the boundary grid is orthogonal 共␤⫽0兲, the moves the isotherms downward and thus decreases the tempera-
first term on the right-hand-side of Eq. 共34b兲 vanishes. ture gradient on the outer cylinder. However, the Benard instabil-
For the case of e v ⫽0.652, three grid systems with 67⫻41, 109 ity 共occurs when heated from below兲 might have brought about
⫻61, and 133⫻81 grid points are employed. These grid systems three-dimensional vortices in Kuehn and Goldstein’s experiment
are generated with the weighting function scheme proposed by 关19兴. Under such a situation, strength of the little positive circu-
Hsu and Lee 关13兴. The boundary values of the control functions lation flow could be weakened. As a result, a larger heat transfer
required for orthogonal boundary grids with desired grid meshes rate was observed in the experiment.
are determined along all of the boundaries. The control functions Comparisons of the resulting equivalent thermal conductivity
inside the computational domain are then interpolated with among the present results and that from previous investigations
关17–20兴 are shown in Figs. 6 and 7 for the cases of (e v ,Ra)
⳵2P ⳵2P ⳵ 2Q ⳵ 2Q ⫽(0, 4.70⫻104 ) and (⫺0.623, 4.93⫻104 ), respectively. Again,
⫹ ⫽0, ⫹ ⫽0 (35)
⳵␰ 2 ⳵ ␩ 2 ⳵␰ 2 ⳵ ␩ 2 excellent performances of the NAPPLE algorithm are confirmed.
The resulting grid system of 67⫻41 points is illustrated in Fig. 2.
Based on the three grid systems 67⫻41, 109⫻61, and 133⫻81,
influence of the grid meshes on the equivalent thermal conductiv-
ity k eq is revealed in Fig. 3 for the case (e v ,Ra)⫽(0.652, 4.80
⫻104 ). From Fig. 3, the grid system of 133⫻81 points is found
adequate for this case. In this connection, the mass residual re-
duces monotonously as the mesh size reduces, similar to that in
the Cartesian coordinates 关11兴. To examine the accuracy of the
NAPPLE algorithm, this same problem was solved with the
stream-vorticity formulation. The weighting function scheme
关12,13兴 was employed to discretize the partial differential equa-
tions. The grid mesh was reduced until grid independence was
achieved. For convenience, results based on the stream-vorticity
formulation will be referred to as ‘‘SV results’’ in the present
study. Figure 4 shows both NAPPLE 共133⫻81兲 and SV 共109⫻61兲
results of k eq . The available numerical and experimental results
from previous investigations are plotted in Fig. 4 also for com-
parison. From Fig. 4, one observes an excellent agreement be-
tween the NAPPLE and SV results. This evidences the accuracy
of the NAPPLE algorithm. The SV results from Projahn et al. 关17兴
poses numerical errors from the coarse grids 共61⫻21兲 as well as
the central difference scheme used by them. The interpolation
procedure proposed by Hwang 关18兴 seems to under-predict the
heat transfer coefficient at a great amount.
It is very interesting to note that in the region of ␾⬍30 deg on
the outer cylinder there are great discrepancies between the ex-
perimentally measured 关19兴 and the numerically predicted k eq .
This might be attribute to the possibility of the Benard instability
occurring in the experiment. To clarify this point, the isotherms
and streamlines from the present NAPPLE and SV results are Fig. 5 Isotherms and streamlines for the case „ e v ,Ra…
depicted in Fig. 5. As observable from Fig. 5, there is a little Ä„0.652,4.80Ã104 …
Fig. 6 Comparison of k eq among the numerical predictions
and the experimental result for „ e v ,Ra…Ä„0,4.70Ã104 …
Example 2 Fluid Flow Through an Axisymmetric Con- Fig. 8 „a… Flow configuration, „b… grid system and streamlines,
stricted Tube. In their experimental study on a fluid flow and „c… isobars with increments 0.1 and 0.02 for example 2
through a locally constricted tube, Young and Tsai 关21兴 employed
an axisymmetric constriction with the cosine shape
⳵ 共 rU 兲 ⳵ 共 rV 兲
1 ⫹ ⫽0 (37)
r⫽r s ⫽ 共 2⫺cos共 ␲ x/x 0 兲兲 (36) ⳵␰ ⳵␩
6
⳵u ⳵u ⳵u
to simulate an arterial stenosis in the portion ⫺x 0 ⭐x⭐x 0 as Rec J ⫹Rec U ⫹Rec V
shown in Fig. 8共a兲, where the cylindrical coordinates (x,r) has ⳵t ⳵␰ ⳵␩
冉冊冉冊冉冊
been normalized with the tube diameter D 0 in the portion ( 兩 x 兩
⳵p* ⳵p* 1 ⳵u ⳵u ⳵ ␣ ⳵u
⬎x 0 ) without constriction. The pressure drop ⌬ P across the con- ⫽⫺ r ␩ ⫺r ␰ ⫺ ⫺x ␩ ⫹x ␰ ⫹
striction was measured at x⫽⫾l. The length l was held at l⫽8 ⳵␰ ⳵␩ r ⳵␰ ⳵␩ ⳵␰ J ⳵␰
冉冊冉冊冉冊
for all tests to insure fully developed flow in both sites. In the
⳵ ␥ ⳵u ⳵ ␤ ⳵u ⳵ ␤ ⳵u
present study, this same flow configuration is solved with the ⫹ ⫺ ⫺ (38)
NAPPLE algorithm. ⳵␩ J ⳵␩ ⳵␰ J ⳵ ␩ ⳵ ␩ J ⳵␰
Let the characteristic length and the characteristic velocity be
⳵v ⳵v ⳵v
assigned as L⫽D 0 and U c ⫽ 冑⌬ P/ ␳ , respectively. After introduc- Rec J ⫹Rec U ⫹Rec V
ing a coordinate transformation from (x,r) to 共␰,␩兲, the governing ⳵t ⳵␰ ⳵␩
冉冊冉冊
equations become ⳵p* ⳵p* ⳵v ⳵v
1 v
⫽⫺ ⫺x ␩ ⫹x ␰ ⫺ ⫺x ␩ ⫹x ␰ ⫹J 2
⳵␰ ⳵␩ r ⳵␰ ⳵␩ r
⫹ 冉冊冉冊冉冊冉冊
⳵ ␣ ⳵v
⳵␰ J ⳵␰
⫹
⳵ ␥ ⳵v
⳵␩ J ⳵␩
⫺
⳵ ␤ ⳵v
⳵␰ J ⳵ ␩
⫺
⳵ ␤ ⳵v
⳵ ␩ J ⳵␰
(39)
where (u, v ) is the velocity in the cylindrical coordinates (x,r).
The corresponding pressure-linked equation based on the
NAPPLE algorithm is
⳵ ␣r ⳵p*
⳵␰ J * ⳵␰冉⫹ 冊冉
⳵ ␥r ⳵p*
⳵␩ J* ⳵␩
⫺
⳵ ␤r ⳵p*
⳵␰ J * ⳵ ␩ 冊冉冊
⫺ 冉冊
⳵ ␤r ⳵p*
⳵ ␩ J * ⳵␰
⫽␧ˆ (40a)
r k⫹n Û k⫹n ⫺r k⫺n Û k⫺n r k⫹1 V̂ k⫹1 ⫺r k⫺1 V̂ k⫺1

␧ˆ ⫽ ⫹ (40b)
2⌬ ␰ 2⌬ ␩
⳵ u 共 x,0兲 / ⳵ r⫽0, v共 x,0兲 ⫽0, u 共 x,r s 兲 ⫽0, v共 x,r s 兲 ⫽0,
⳵ u 共 ⫺l,r 兲 / ⳵ x⫽0, v共 ⫺l,r 兲 ⫽0, p * 共 ⫺l,r 兲 ⫽Rec , (41)
Fig. 7 Comparison of k eq among the numerical predictions
and the experimental result for „ e v ,Ra…Ä„À0.623,4.93Ã104 … ⳵ u 共 l,r 兲 / ⳵ x⫽0, v共 l,r 兲 ⫽0, p * 共 l,r 兲 ⫽0
Fig. 10 Comparison of separation point x s and reattachment
Fig. 9 Comparison of separation point x s and reattachment point x r among the numerical predictions and the experimental
point x r among the numerical predictions and the experimental result for x 0 Ä2
result for x 0 Ä1
the experimental difficulties in delineating the reattachment point

as remarked by Young and Tsai 关21兴 and Deshpande et al. 关22兴.
Note that the only parameter in the problem 共37兲–共41兲 is For the case of x 0 ⫽2, the separation point x s and the reattach-
U c L 冑⌬ P/ ␳ D 0
ment point x r based on the NAPPLE algorithm and the SV for-
Rec ⫽ ⫽ (42) mulation are revealed in Fig. 10. The available numerical predic-
v v tions from Deshpande et al. 关22兴, Karki 关23兴, Melaaen 关4兴, and
Once the solution (u, v ,p) corresponding to a given pressure drop Liou et al. 关25兴 as well as the experimental observation from
⌬ P 共or Rec兲 is obtained, the Reynolds number can be evaluated Young and Tsai 关21兴 are shown also in Fig. 10 for comparison.
from Again, an excellent agreement is found to exist between the
present NAPPLE and SV results. Due to the use of a coarse grid
Re⫽
4Q
␲vD0
⫽8 Rec 冕 0
rs
urdr (43)
mesh, the SV results from Deshpande et al. 关22兴 over-predicts the
reattachment point x r when the Reynolds number is large. Other-
wise, satisfactory agreements are observable among the numerical
where Q is the volume flow rate through the tube. predictions and the experimental result. This validates the perfor-
In this example, two models of stenoses 共i.e., x 0 ⫽1 and x 0 mance of the NAPPLE algorithm.
⫽2兲 are investigated. Grid systems with the desired grid size and As a final note, it is mentioned that the coefficient matrix of the
skewness 共orthogonal兲 along all of the boundaries are generated discretized pressure-linked equation in example 1 is singular be-
with the weighting function scheme 共see Hsu and Lee 关13兴, and cause of (a P ) k ⫽⬁ along all of the boundaries. This implies the
Eq. 共35兲兲. Figure 8共b兲 reveals one of the resulting grid systems. need of an artificial pressure with a pressure level to ensure the
The numerical solution based on the NAPPLE algorithm is refined existence and uniqueness of the pressure solution for the updated
by reducing the grid size until grid independence is achieved. The velocity 关11兴. In example 2, the weighting factor (a P ) k has a finite
resulting streamlines and isobars for x 0 ⫽1 and Rec⫽100 are il- value on both upstream and downstream boundaries (x⫽⫾1).
lustrated in Fig. 8共b兲 and 8共c兲, respectively. Increment of the Thus, the Dirichlet boundary conditions
streamlines is one tenth of the flow rate in Fig. 8共b兲, while that of
p * 共 ⫺l,r 兲 ⫽Rec , p * 共 l,r 兲 ⫽0 (44)
the isobars is 0.1 and 0.02, respectively, in the upper and lower
portion of Fig. 8共c兲. As observable from Fig. 8共b兲, the separation can be imposed there. In this case, no artificial pressure is needed
point x s and the reattachment point x r can be clearly identified. when the pressure-linked equation is solved.
Figure 9 shows the separation point x s and the reattachment
point x r obtained with the NAPPLE algorithm for x 0 ⫽1 at vari-
ous Reynolds numbers. Results based on the stream-vorticity 共SV兲 Conclusion
formulation and that from previous numerical 关4,22–25兴 and ex- In the present study, the NAPPLE algorithm is extended to
perimental 关21兴 investigations are presented also in Fig. 9 for nonorthogonal curvilinear grid system for incompressible viscous
comparisons. The numbers (m⫻n) appearing in Fig. 9 denote the flow. A pressure-linked equation is derived from the discretized
numbers of grid points used in each numerical investigations. As governing equations. In the course of the derivation, a particular
in example 1, the stream-vorticity equations were discretized with treatment is adopted for a strong pressure-velocity coupling. No
the weighting function scheme 关12,13兴. The SV results were grid- velocity interpolation is needed such that the solution does not
independent. depend on the under-relaxation factor. This is the most important
From Fig. 9, an excellent agreement is found to exist between feature of the present method. In the present method, solving the
the present NAPPLE and SV results. The SV results from Desh- pressure-linked equation is as simple as solving a heat conduction
pande et al. 关22兴 seems to possess a considerable numerical error equation. Due to the truncation error, the pressure-linked equation
due to a coarse grid mesh used by them. On the separation point is not an exact combination of the discretized continuity equation
x s , all of the numerical predictions agree with the experimental and the discretized momentum equations. Thus, there is a certain
observation from Young and Tsai 关21兴. For the reattachment point level of mass residual after the solution perfectly converges. Nev-
x r , however, a great discrepancy between the numerical predic- ertheless, the mass residual approaches zero when the grid mesh is
tions and the experiment result is seen. This might be attributed to reduced. Through two well-documented examples, performance
of the NAPPLE algorithm on non-orthogonal curvilinear grid sys- References
tem is validated. Due to its simplicity and accuracy, the NAPPLE 关1兴 Lee, S. L., and Lin, D. W., 1997, ‘‘Transient Conjugate Heat Transfer on a
algorithm is believed to have good performances for fluid flow Naturally Cooled Body of Arbitrary Shape,’’ Int. J. Heat Mass Transf., 40, pp.
and heat transfer problems. 2133–2145.
关2兴 Rhie, C. M., and Chow, W. L., 1983, ‘‘Numerical Study of the Turbulent Flow
Past an Airfoil With Trailing Edge Separation,’’ AIAA J., 21, pp. 1525–1532.
关3兴 Melaaen, M. C., 1992, ‘‘Calculation of Fluid Flow With Staggered and Non-
staggered Curvilinear Nonorthogonal Grids—The Theory,’’ Numer. Heat
Acknowledgment Transfer, Part B, 21B, pp. 1–19.
关4兴 Melaaen, M. C., 1992, ‘‘Calculation of Fluid Flow With Staggered and Non-
The authors would like to express their appreciation to the Na- staggered Curvilinear Nonorthogonal Grids—A Comparison,’’ Numer. Heat
tional Science Council of Taiwan for the financial support of this Transfer, Part B, 21B, pp. 21–39.
关5兴 Zang, Y., Street, R. L., and Koseff, J. R., 1994, ‘‘A Non-Staggered Grid,
work through the project NSC 89-2212-E007-032. Fractional Step Method for Time-Dependent Incompressible Navier-Stokes
Equations in Curvilinear Coordinates,’’ J. Comput. Phys., 114, pp. 18 –33.
关6兴 Patankar, S. V., 1988, ‘‘Recent Developments in Computational Heat Trans-
fer,’’ ASME J. Heat Transfer, 110, pp. 1037–1045.
关7兴 Miller, T. F., and Schmidt, F. W., 1988, ‘‘Use of a Pressure-Weighted Interpo-
Nomenclature lation Method for the Solution of the Incompressible Navier-Stokes Equations
on a Non-Staggered Grid System,’’ Numer. Heat Transfer, 14, pp. 213–233.
Gr ⫽ Grashof number, g ␤ ⌬TL 3 / v 2 关8兴 Shyy, W., Tong, S. S., and Correa, S. M., 1985, ‘‘Numerical Recirculating
J ⫽ Jacobian 共10b兲 Flow Calculation Using a Body-Fitted Coordinate System,’’ Numer. Heat
Transfer, 8, pp. 99–113.
J* ⫽ pseudo-Jacobian 关9兴 Braaten, M., and Shyy, W., 1986, ‘‘A Study of Recirculating Flow Computa-
k eq ⫽ equivalent thermal conductivity 共34兲 tion Using Body-Fitted Coordinates: Consistency Aspect and Mesh Skew-
P,p ⫽ pressure, and dimensionless pressure 共5兲 ness,’’ Numer. Heat Transfer, 9, pp. 559–574.
关10兴 Peric, M., 1990, ‘‘Analysis of Pressure-Velocity Coupling on Nonorthogonal
p* ⫽ dimensionless pressure, p Rec Grids,’’ Numer. Heat Transfer, Part B, 17B, pp. 63– 82.
Pr ⫽ Prandtl number, ␯ / ␣ t 关11兴 Lee, S. L., and Tzong, R. Y., 1992, ‘‘Artificial Pressure for Pressure-Linked
(r,x) ⫽ cylindrical coordinates Equation,’’ Int. J. Heat Mass Transf., 35, pp. 2705–2716.
Ra ⫽ Rayleigh number, Pr Gr 关12兴 Lee, S. L., 1989, ‘‘Weighting Function Scheme and Its Application on Multi-
dimensional Conservation Equations,’’ Int. J. Heat Mass Transf., 32, pp. 2065–
Re ⫽ Reynolds number 2073.
Rec ⫽ characteristic Reynolds number, U c L/ v 关13兴 Hsu, K., and Lee, S. L., 1991, ‘‘A Numerical Technique for Two-Dimensional
T ⫽ temperature Grid Generation With Grid Control at All of the Boundaries,’’ J. Comput.
t ⫽ dimensionless time 共normalized with L/U c ) Phys., 96, pp. 451– 469.
关14兴 Patankar, S. V., 1980, Numerical Heat Transfer and Fluid Flow, Hemisphere,
(U,V) ⫽ contravariant velocity 共10a兲 New York, Chap. 6.
(u, v ) ⫽ velocity in Cartesian coordinates 关15兴 Lee, S. L., 1989, ‘‘A Strongly Implicit Solver for Two-Dimensional Elliptical
(û, v̂ ) ⫽ pseudo-velocity in Cartesian coordinates Differential Equations,’’ Numer. Heat Transfer, Part B, 16B, pp. 161–178.
Uc ⫽ characteristic velocity 关16兴 Stone, H. L., 1968, ‘‘Iterative Solutions of Approximations of Multidimen-
sional Partial Differential Equations,’’ SIAM 共Soc. Ind. Appl. Math.兲 J. Numer.
w f (z) ⫽ weighting function, z/(l⫺exp(⫺z)) Anal., 5, pp. 530–558.
ŵ f (z) ⫽ weighting function, z/(exp(z)⫺exp(⫺z)) 关17兴 Projahn, U., Rieger, H., and Beer, H., 1981, ‘‘Numerical Analysis of Laminar
z ⫽ grid Peclet number Natural Convection Between Concentric and Eccentric Cylinders,’’ Numer.
Heat Transfer, 4, pp. 131–146.
Greek 关18兴 Hwang, Y. H., 1994, ‘‘Arbitrary Domain Velocity Analyses for the Incompress-
␣, ␤, ␥ ⫽ geometric parameters 共10b兲 ible Navier-Stokes Equations,’’ J. Comput. Phys., 110, pp. 134 –149.
关19兴 Kuehn, T. H., and Goldstein, R. J., 1978, ‘‘An Experimental Study of Natural
␣t ⫽ thermal diffusivity Convection Heat Transfer in Concentric and Eccentric Horizontal Cylindrical
⌬P ⫽ pressure difference Annuli,’’ ASME J. Heat Transfer, 100, pp. 635– 640.
⌬T ⫽ characteristic temperature 关20兴 Kuehn, T. H., and Goldstein, R. J., 1976, ‘‘An Experimental and Theoretical
Study of Natural Convection in the Annulus Between Horizontal Concentric
␧ˆ⫽ pseudo-dilation Cylinders,’’ J. Fluid Mech., 74, pp. 695–719.
␪⫽ dimensionless temperature 关21兴 Young, D. F., and Tsai, F. Y., 1973, ‘‘Flow Characteristics in Models of Arterial
v ⫽ kinematic viscosity Stenoses—I. Steady Flow,’’ J. Biomech., 6, pp. 395– 410.
共␰,␩兲 ⫽ computational coordinates 关22兴 Deshpande, M. D., Giddens, D. P., and Mabon, R. F., 1976, ‘‘Steady Laminar
␾ ⫽ a function or an angle Flow Through Modeled Vascular Stenoses,’’ J. Biomech., 9, pp. 165–174.
关23兴 Karki, K. C., 1986, ‘‘A Calculation Procedure for Viscous Flows at All Speeds
Subscripts in Complex Geometries,’’ Ph.D. thesis, University of Minnesota.
关24兴 Kadja, M., 1987, ‘‘Computation of Recirculating Flow in Complex Domains
ref ⫽ reference With Algebraic Reynolds Stress Closure and Body Fitted Meshes,’’ Ph.D. the-
e,s,w,n ⫽ surfaces of control volume, Fig. 1 sis, University of Manchester.
关25兴 Liou, T. M., Chen, S. H., and Chen, L. T., 1995, ‘‘Numerical Simulation on
␰, ␩ ⫽ derivatives with respective to ␰ and ␩ Fluid Flow Through a Locally Constricted Tube 共in Chinese兲,’’ Proceedings of
i ⫽ inner the Third National Conference on Computational Fluid Dynamics, Taiwan, pp.
o ⫽ outer 139–148.
Film Cooling Effectiveness and
Heat Transfer Coefficient
Distributions Around Diffusion
Y. Yu1
C.-H. Yen2
Shaped Holes
T. I.-P. Shih3 Presented in this paper is an experimental study focusing on the effects of diffusion
hole-geometry on overall film cooling performance. The study consists of three different
M. K. Chyu but closely related hole shapes: (1) Shape A: straight circular hole with a 30 deg inclined
injection, (2) Shape B: same as Shape A but with a 10 deg forward diffusion, and (3)
Department of Mechanical Engineering,
Shape C: same as Shape B with an additional 10 deg lateral diffusion. The blowing ratios
University of Pittsburgh,
tested are 0.5 and 1.0. The density ratio is nominally equal to one. Measurements of the
Pittsburgh, PA 15261
experiment use a transient liquid crystal technique that reveals local distributions of both
film effectiveness (␩) and heat transfer coefficient (h). The data obtained indicate that
Shape C with a combined forward and lateral diffusion produces a significant increase in
S. Gogineni ␩ and decrease in h as compared to Shape A, the baseline case. These improvements
Innovative Scientific Solution, Inc.,
combined yield an about 20 percent to 30 percent reduction in heat transfer or thermal
Dayton, OH 45440
load on the film protected surface. Shape B, with forward diffusion only, shows a much
less significant change in both film effectiveness and overall heat transfer reduction than
Shape C. However, it has the lowest heat transfer coefficient in the vicinity of the injection
hole among all the three hole-shapes studied. A flow visualization study using pulsed laser
sheet-light reveals that Shape A and Shape B inherit quite similar flow structures. The
coolant injected out of Shape C flows much closer to the protected wall than that of Shape
A and Shape B. 关DOI: 10.1115/1.1418367兴
Keywords: Convection, Cooling, Film Cooling, Heat Transfer, Turbines
Introduction ratio elevates the linear momentum of the injectant, which in turn,
lifts the coolant away from the surface. This phenomenon is un-
Film cooling is one of the major schemes for protection of
desirable due to the penetration of injection into the mainstream
turbine airfoils from thermal deterioration imposed by the hot gas
apparently, permitting the hot gas to flow under the coolant jet
stream. Turbine engines with effective cooling permit a higher gas
close to the surface for protection.
temperature at the turbine inlet, which in turn, improves the ther-
In addition to film effectiveness, the heat transfer coefficient 共h兲
mal efficiency and performance of the system. Since the coolant
can also be affected by the coolant-mainstream interaction. The
that facilitates the film cooling is extracted from the compressor,
focus of interest in the past has been placed primarily in the region
poor management of the cooling air can be otherwise detrimental
sufficiently downstream from the injection hole, where the mag-
to the engine overall efficiency. Therefore, the design of an effec-
nitude of heat transfer coefficient appears to be uninfluenced by
tive cooling system with a minimum amount of coolant expended
the presence of injection. However, the situation is expected to be
requires a sufficient database as well as fundamental knowledge
quite different in the vicinity of an injection hole. The interaction
of the transport phenomena associated with film cooling.
Heat transfer pertaining to film cooling has been actively inves- between the injection and the mainstream can drastically alter the
tigated in the gas turbine community for nearly three decades. flow conditions in the neighboring region, causing the magnitude
One key parameter in measuring the film cooling performance is of heat transfer coefficient to be either substantially higher or
the film effectiveness 共␩兲, which is the dimensionless form of the lower than that without injection. Eriksen and Goldstein 关3兴 earlier
adiabatic wall temperature, i.e., ␩ ⫽(T aw⫺T m )/(T f ⫺T m ). While indicated that the heat transfer coefficient with injection is lower
the values of ␩ vary with the actual flow conditions, optimal film than that without injection, because the injection thickens the
effectiveness for coolant injected through discrete round holes oc- boundary layer. However, this notion only prevails in the cases of
curs when the blowing ratio 共M兲 is about 0.5 or the momentum low blowing ratios. Hay et al. 关4兴 later reported that the ratio of
ratio 共I兲 is about 0.25 关1,2兴. Such an optimal blowing ratio phe- heat transfer coefficient between the cases with and without injec-
nomenon is closely related to the interaction between the injected tion (h/h o ) increases with the blowing ratio. The maximum ratio
coolant and the mainstream. For cases with low blowing ratios, reaches about 1.35 with an injection of 35 deg inclined to the
better film protection over the hot surface is expected with an mainstream. The active momentum exchange between the partici-
increase in the amount of the coolant injected to the mainstream. pating streams along with excessive shear generated in the region
However, excessive coolant in conjunction with higher blowing is responsible for the increased heat transfer coefficient.
Primarily for more effective downstream cooling, film cooling
1
MTU Aeroengine Design Inc., Rocky Hill, CT 06067.
in actual turbine engines often uses a coolant flow rate substan-
2
Currently at General Motor, Milford, MI 48380 tially greater than that corresponding to the optimal film effective-
3
Currently at Department of Mechanical Engineering, Michigan State University, ness condition. As a result, an increase in the blowing ratio, pos-
East Lansing, MI 48824 sibly with an elevated heat transfer coefficient near the injection
HEAT TRANSFER. Manuscript received by the Heat Transfer Division February 26, hole, can in fact, increase the overall heat load on the airfoil. In an
2000; revision received May 25, 2001. Associate Editor: J.-C. Han. attempt to reduce the momentum of the injected flow, while still
maintaining a sufficient amount of coolant flow for proper down- row of conical holes. They observed that an increase in the span-
stream cooling, one possible approach is to alter the geometry by wise injection angle can in fact, impose a higher level of heat load
enlarging the flow passage near the film hole exit. Film holes of on the airfoil, as the heat transfer coefficient is higher and the film
such nature are commonly known as shaped-diffusion holes, effectiveness is lower compared to the streamwise-directed holes.
shaped holes or diffusion holes. The present study evaluates the performance of two different,
Film cooling pertaining to shaped holes has been little studied but closely related, shaped diffusion holes, along with a direct
until recently. In one of the earliest studies, Goldstein et al. 关5兴 comparison to the corresponding baseline case of standard circular
reported a significant increase in film effectiveness in the near holes under the same flow conditions. Figure 1 schematically il-
hole region as well as an improved lateral spread of the coolant. lustrates the three hole configurations, labeled as Shape A, B, and
The test geometry in their study is a 10 deg expansion over a row C. Shape A is the baseline circular injection with a 30 deg inclined
of circular holes inclined 35 deg relative to the mainstream. More angle relative to the freestream. The hole inlet diameter 共D兲 is
recently, Sen et al. 关6兴 and Schmidt et al. 关7兴 examined the per- about 6.4 mm 共0.25 in.兲 and the length is 10 times the diameter.
formance of forward-expanded holes with compound-angle injec- Although such a length-to-diameter ratio is significantly greater
tion. Their data suggest that diffusion shaped-holes show particu- than that in the modern day turbine engines, it permits a study
larly better performance with high blowing and momentum ratios. focused exclusively on the effects of hole-exit shape without
Wittig et al. 关8兴, Gritsch et al. 关9,10兴 examined both the film ef- much influence of the inlet conditions. Also shown in Fig. 1, the
fectiveness and heat transfer coefficient of two fan-shaped holes test section consists of five holes spaced three-dimensional in the
with about 14 deg lateral expansion. Haven et al. 关11兴 performed spanwise direction. Shape B has all the same geometric param-
detailed flow visualization around three different diffusion holes eters as Shape A except that the hole has a 10 deg forward,
and assessed their effects on the film effectiveness. Most recently, streamwise diffusion initiated at approximately 0.8 hole-inlet di-
Chen et al. 关12兴 studied the compound-angle injection through a ameter below the test surface. Shape C has all the features of
Shape B, but with an additional 10 deg diffusion in the lateral
direction. Viewing from the top toward the test surface, Shape B
has an elongated oval opening, while Shape C has a tear-drop
profile. Performance characterization of each hole geometry is
made by measurements of both the film effectiveness and heat
transfer coefficient using a transient liquid crystal imaging tech-
nique. In addition, pulsed laser-sheet flow visualization is also
performed to complement the heat transfer results.
Transient Liquid Crystal Technique for Film Cooling

Measurements
The use of liquid crystal coating as a surface temperature indi-
cator for heat transfer imaging has been quite popular, especially
in the gas turbine community. When the technique is operated in a
transient mode, the change of liquid crystal color provides the
temporal variation of local surface temperature over the test re-
gion of interest. The local heat transfer coefficient can be inferred
by assuming the one-dimensional, transient conduction model pre-
vailing in the solid substrate underneath the test surface. The heat
transfer coefficient is defined in a general sense, i.e., h⫽q/(T w
⫺T r ), where T r is the reference temperature of the system. Such
a model can be represented in the following governing equation,
with z being the coordinate pointing into the substrate:
⳵ 2T ⳵T
k ⫽␳Cp (1)
⳵ 2z ⳵t
⫺k
⳵T
⳵z
冏 z⫽0
⫽h 共 T w ⫺T r 兲 (2)
T 兩 z⫽⬁ ⫽T i (3)
T 兩 ␶ ⫽0 ⫽T i (4)
The solution of the above equation leads to a relationship
among the heat transfer coefficient, the thermal properties of the
substrate, the initial and reference temperatures of the system, and
the time-varying surface temperature.
T w ⫺T i
T r ⫺T i
⫽1⫺exp
h 2 ␣␶
k2 冋册冋册
erfc
h 冑␣␶
k
(5)
While the above approach is generally valid for resolving heat

convection problems, it is, however, unsuitable for film cooling.
Film cooling is known to be a three-temperature problem, as its
thermal transport is governed by the temperatures of two partici-
pating streams and the surface exposed to the mixture of those
streams. As a contrast to conventional two-temperature problems,
Fig. 1 Shape hole geometry and test section the reference temperature T r in Eq. 共5兲 is no longer a known
quantity and needs to be solved simultaneously with the heat in the substrate is expected to be insignificant. The ‘‘hot test,’’ on
transfer coefficient. T r is also termed the adiabatic wall tempera- the other hand, has a strong influence on the outcome of the heat
ture, T aw , and has a dimensionless appearance as the film effec- transfer coefficient. The injectant and the mainstream air during
tiveness, as previously mentioned. The overall implication of this this test are heated to nearly the same level, about 40⬃50 C
is that a second relation similar to Eq. 共5兲 needs to be established. higher than the initial temperature of the substrate. Except for the
The technique based on transient liquid crystal imaging for region immediate adjacent to the injection hole and sufficiently
three-temperature, film cooling problems was first introduced by below the surface, the temperature gradients along the lateral di-
Vedula and Metzger 关13兴. They delineated the approach of deter- rections 共perpendicular to the z-axis, see Fig. 1兲 are expected to be
mining ␩ and h by using the same test setup and a transient test far lower than that along the z-direction in the substrate. There-
with natural boundary conditions. Ekkad et al. 关14,15兴 and Chen fore, the one-dimensional model is mostly valid throughout the
et al. 关12兴 used the same method for studying film cooling with entire test section, provided that the test duration is not overly
compound-angle injection. Yu and Chyu 关16兴 extended the method long.
further for four-temperature problems involving two injection of
different temperatures. The fundamental principle of Vedula and Experimental Apparatus and Procedures
Metzger’s approach is based on the notion that the turbulent con- The test section consists of a Plexiglas rectangular duct, with
vection is predominantly controlled by flow dynamics rather than 102 mm 共4 in.兲 across the span and 51 mm 共2 in.兲 in height. For
thermal conditions. Hence the second relation can be obtained by each shape, five coolant injection holes with a 30 deg inclination
performing two different, but closely related, heat transfer tests are located near the upstream portion of the test surface. The test
under the same flow conditions. The resulting equations for the surface is coated with a thin layer of thermochromic liquid crystal
two unknowns; i.e., h and T aw , are 共TLC兲 about 0.1 to 0.3 mm thick. An airbrush is used to spray the
T w ⫺T i1
T aw⫺T i1
⫽1⫺exp
h 2 ␣␶ 1
k2 冋册冋册
erfc
h 冑␣␶ 1
k
, (6)
coating. The temperature calibrated for the particular TLC used in
the present study is 38.4°C, 39.8°C and 43.5°C for red, green and
blue, respectively. Primarily because of its distinct intensity
冋册冋冑册
strength, the green color displayed by TLC is used as the surface
T w ⫺T i2 h 2 ␣␶ 2 h ␣␶ 2
⫽1⫺exp erfc . (7) temperature tracer during a transient measurement.
T aw⫺T i2 k2 k Figure 2 shows a schematic of the overall test setup. A labora-
In reality, one of the two tests, so-called the ‘‘hot test,’’ uses a tory compressor supplies both the mainstream and film flows.
film injection close to, but slightly lower than, the heated main- Flow rates of the compressed air are measured by standard ASME
stream; while the ‘‘cold test’’ uses an unheated or slightly heated orifices. After metering each stream is routed through a tubular
injection. A key criterion for selecting these test temperatures is to in-line heater controlled by an auto-transformer, and its tempera-
ensure that the two equations, Eqs. 共6兲 and 共7兲, are well coupled, tures can be accurately set to a desired level. Downstream of the
so that when combined they yield a solution for the two un- heater, the two flows are initially diverted away from the test
knowns. section by a three-way ball valve until the actual test starts.
True step changes in both the mainstream and injection tem- A transient test is initiated by switching both ball valves simul-
peratures during an actual test are usually not possible, hence the taneously to introduce the main flow and film injection into the
reference temperature which appears as a boundary condition in test section. The switching of these valves is made only when the
Eq. 共2兲 can vary with time. This fact is accounted by modifying flow rates and temperatures of both streams have reached steady
the solution of Eqs. 共6兲 and 共7兲 with the use of superposition and state. Initiation of the test also triggers an automated data acqui-
Duhamel’s theorem. Decomposing the temporal variation into a sition system for recording thermocouple readings at flow inlets as
series of finite-step changes, these two equations are transformed well as exits. Simultaneously, a CCD camera starts to record the
to the following form: video images of the TLC coated on the test surface. The video
provides the follow-on data reduction procedure with the lap time
N of TLC changing from colorless to green at every pixel of the
T w ⫺T i ⫽ 兺 U 共 ␶ ⫺ ␶ 兲 ⌬T
i⫽1
i r , (8) viewing domain. The post-run data reduction procedure uses cus-
tom developed software run in a personal computer 关18兴. The test
where cases include two blowing ratios 共M ⫽0.5 and 1.0兲 for Reynolds
numbers 共based on hole inlet-diameter兲 from ReD⫽2300 to 4300.
冋
U 共 ␶ ⫺ ␶ i 兲 ⫽1⫺exp
k2 册冉
h 2␣共 ␶ ⫺ ␶ i 兲
erfc
k
冊
h 冑␣ 共 ␶ ⫺ ␶ i 兲
. (9)
The corresponding Reynolds number based on the hydraulic di-
ameter of the housed channel is Re⫽25,000 and 46,000, respec-
Since ⌬T r is related directly to the variations of the mainstream

temperature, T m , and injection temperature, T c , by the definition
of film effectiveness ␩,
⌬T r ⫽ 共 1⫺ ␩ 兲 ⌬T m ⫹ ␩ ⌬T c . (10)
Substituting Eq. 共10兲 into Eq. 共8兲, the unknowns for the present
system become the film effectiveness, ␩, and the heat transfer
coefficient, h.
In view of the geometrical complexity, the heat transfer charac-
teristic in the close vicinity of an injection hole is understandably
multi-dimensional in reality. Any amount of heat diffused deviated
from the direction normal to the substrate surface 共z-axis, Fig. 1兲
is considered to be an error in the present transient measurement.
While an attempt to systematically investigate such an issue has
been reported recently 关17兴, the error in the present system may
not be as significant as it is generally perceived. The film effec-
tiveness is determined predominantly by the ‘‘cold test,’’ where
temperatures of the injectant and the Plexiglas substrate are very
close throughout the test. Hence the effect of lateral heat transfer Fig. 2 Test apparatus
Fig. 3 Spanwise-average film effectiveness
tively. The density ratio in the present study is nominally equal to

Fig. 4 Spanwise distribution of film effectiveness for non-
one, though the actual value during a transient test varies between shaped circular holes at X Õ D Ä3.0, M Ä0.5, and ReDÄ4300
0.95 and 1.1. Hence the existing blowing ratio is essentially the
flow velocity ratio between the injection and the mainstream. For
purpose of validation and comparison, the case with no film in-
jection (M ⫽0) was also performed. Under such a non-injecting
condition, the boundary layer at the location of injection exhibits
fully turbulent characteristics, with a thickness about 8 mm and 6
Fig. 5 Streamwise distribution of film effectiveness: „a… M Ä0.5; „b… M Ä1.0; „c… spanwise average, M Ä0.5; and „d… spanwise
average, M Ä1.0.
mm for Re⫽25,000 and 46,000, respectively. The turbulence in-
tensity in the mainstream measured by a hot-wire anemometer is
about 1.2 percent.

Most of the results presented below are in the form of film
effectiveness 共␩兲 and heat transfer coefficient 共h兲. The uncertainty
analysis with a 95 percent confidence level in the measured pa-
rameters is based on the method of Kline and McClintock 关19兴.
Both ␩ and h measured in the present liquid-crystal-based system
depend strongly on the temperatures of the test surface, inlet, and
exit. The uncertainties of these temperature readings are about
⫾0.2 C, which correspond to 0.5 percent, 0.3 percent, and 0.7
percent at these key locations, respectively. In addition, the uncer-
tainty of both ␩ and h is also affected by the errors in the
temperature-color calibration of liquid crystal, the lap-time for the
liquid crystal reaching the designated color, and the thermal dif-
fusivity of the substrate. The errors in these three parameters are
estimated at 0.5 percent, 0.4 percent, and 1 percent, respectively.
Another major error contributing to the uncertainty is the error in
measuring the flow rate or Reynolds number in the air supply
system, which is about 4 percent. Combining all these errors leads
to the overall 7 percent uncertainty for both ␩ and h.
Film Effectiveness. Although the present TLC imaging tech-
nique is capable of revealing detailed distributions and contour
data over the entire viewing domain, most local data are presented
along two key streamwise paths for direct comparison. One of the
paths is along the centerline of the test section and directly behind
an injection hole, and the other is along the mid-span of two
adjacent injection holes. These line data are good representatives
of the transport features over the entire measurement domain, as
the overall distributions in both ␩ and h are very symmetric across
the span of the test section. Figure 3 illustrates a sample compari-
son of the present film-effectiveness data with the similar results
Fig. 6 Laser-sheet flow visualization
published recently by Ekkad et al. 关14兴. Although their film holes
are spaced relatively wider 共4D兲 and hole inclined angle is differ-
ent 共35 deg兲, the comparison is very favorable. To demonstrate the
level of data symmetry in the present experiments, Fig. 4 shows forward expansion only, exhibits a notable but unsubstantial im-
the spanwise variation of local film effectiveness at X/D⫽3.0 for provement in film effectiveness as compared to the baseline case,
all three hole configurations, M ⫽0.5 and ReD⫽4300. At this near- Shape A. This implies that, at least for simple, non-compound-
hole location, there is little evidence of coolant spread laterally. angle injection, lateral expansion may be a more effective mecha-
Figure 5 shows the effect of hole shaping on the streamwise nism than the forward expansion in terms of shaping the injection
distribution of film effectiveness. The origin of X, the streamwise holes.
coordinate, locates at the leading edge of an injection hole. Since The quantitative measure of film effectiveness is closely related
the influence of Reynolds number is rather insignificant overall, to its inherent jet flow behavior for each hole geometry. Figure 6
only the case with Re⫽2350 is presented here. Shape C, which shows flow visualization photos 共turn 90 deg in photo兲 using a
has both forward and lateral expansion, consistently results in the laser sheet illumination along the centerline for M ⫽1.0 and ReD
highest film effectiveness among the three hole shapes studied. As ⫽2350. To facilitate the visualization experiment, the injected
shown in Figs. 5共c兲 and 共d兲, its spanwise-averaged ␩ is approxi- flow is seeded with 1 ␮m alumina particles, which form the white
mately 50 percent higher than the baseline case, Shape A. The traces in the photos. Image recording uses a Kodak ES1.0 camera
enhancement appears to be more significant with M ⫽1.0 than which has a CCD imager with 1000 by 1000 pixels. It is evident
with M ⫽0.5, particularly in the near-hole region, say X/D⬍8. that the flow injected from Shape C remains the closest to the wall
The aforementioned averaged characteristics are consistent with among all the three cases studied. This is an indication that proper
the local data along the centerline, directly behind an injection hole-shaping reduces the jet momentum for penetration to the
hole. As seen in Figs. 5共a兲 and 共b兲, the magnitude of local ␩ along freestream as well as increases the coolant concentration near the
the centerline for Shape C is approximately 30 percent and 50 wall. In addition, the scales of flow structure appear to be the
percent higher than the corresponding values of Shape A, for M smallest for Shape C, implying more effective flow diffusion and
⫽0.5 and 1.0, respectively. mixing. On the other hand, the jets from both Shape A and Shape
As a typical feature of discrete-hole film cooling, local values B exhibit strong lift-off, hence relatively poor wall coverage as
of ␩ along the mid-span are much lower than those along the well as film effectiveness is expected. In fact, these two cases
centerline in the region sufficiently close to the injection hole. The display quite comparable flow features. As a result, their magni-
coolant just ejected out of the film cooling holes provides virtually tudes of film effectiveness are also very similar. Similar life-off
no coverage in the region between neighboring holes and then effect with non-shaped, circular hole was reported earlier by Lig-
spreads out laterally toward downstream. Hence ␩ along the two rani et al. 关1,2兴.
streamwise paths displays completely opposite trends and con-
verges eventually to the same level at a location sufficiently Heat Transfer Coefficient. Figure 7 shows the streamwise
downstream. Overall, Shape C shows a better lateral coverage distributions of the local heat transfer coefficient for various hole
than the other cases, especially when the Reynolds number is shapes. The particular cases shown here are M ⫽0.5 and 1.0 at
relatively low. As an intermediate case, Shape B, which features a ReD⫽2350. Hole geometry has a pronounced influence on the
Fig. 7 Streamwise distribution of heat transfer coefficient: „a… M Ä0.5; „b… M Ä1.0; „c… spanwise average, M Ä0.5; and „d…
spanwise average, M Ä1.0.
local heat transfer coefficient, particularly in the region X/D⬍6. shear in the near-hole region. Shape C, on the other hand, repre-
Regardless of the nature of hole shaping, the heat transfer coeffi- sents the opposite extreme, with the thinnest boundary layer and
cient in this region is always lower along the centerline than that lowest level of shear generation. Shape B is expected to be mid-
along the mid-span. In fact, the magnitude of h along the center- level of both effects, which when combined ultimately results in
line are the lowest for any given hole geometry. Although the the lowest heat transfer coefficient along the centerline. For the
difference in the centerline h among the three hole shapes is rather case with a higher blowing ratio, M ⫽1.0, Shape A exhibits a
modest or even within the range of measurement uncertainty, the sharp rise of heat transfer coefficient directly behind the injection
forward-diffusion hole, i.e., Shape B, consistently exhibits the hole, as shown in Fig. 7共b兲. Strong local shear generation appar-
lowest value. On the other hand, Shape A, which is the non- ently overwhelms the effect of boundary layer thickening in this
shaped, baseline case, exhibits the highest value. The heat transfer region.
coefficients along the mid-span for all three cases studied are rela- Figures 7共c兲 and 7共d兲 show the streamwise distribution of
tively comparable and, in fact, are similar to that over the test wall spanwise-average heat transfer coefficient normalized by its coun-
without injection. All the local data collapse to more or less a
terpart without injection (h o ). The value of h o was obtained using
single curve sufficiently downstream to the injection holes.
the same experimental setup, but without injection. The test under
The magnitude of heat transfer coefficient around a film-
cooling hole is influenced by two competing factors. First, it de- this condition is a conventional ‘‘two-temperature-driven convec-
creases with a thickening boundary layer induced by injection tion problem’’ and it requires only a ‘‘hot-test’’ to reveal the heat
added flow mass and linear momentum normal to the wall. transfer coefficient. The results in Fig. 7共c兲 indicate that discrete-
Thicker boundary layer presents more heat convection resistance hole film cooling with M ⫽0.5, in average, enables a 20 to 30
between the mainstream and the film-protected wall. Second, the percent reduction of heat transfer coefficient compared to h o in
value of heat transfer coefficient may increase with enhanced flow the near hole region. Due mainly to excessive interaction between
shear induced by interaction between the injection and the main- the injection and the mainstream, the situation can be drastically
stream. The first effect appears to be the predominant feature near different when the blowing ratio is sufficiently high. As evidenced
the centerline, particularly when the blowing ratio or the jet exit in Fig. 7共d兲 for M ⫽1.0, the h/h o data for Shape A exceed unity
velocity is relatively low. The second effect with flow shear is around X/D⫽3 and in certain regions further downstream. This
expected to prevail along the side-edge of an injection hole, which implies that, depending on the nature of flow interaction between
is responsible for the high heat transfer coefficient along the mid- the participating streams, implementation of film cooling can be
span. In view of the flow structure revealed in Fig. 6, Shape A either favorable or detrimental to the overall thermal protection
induces the greatest boundary layer thickening as well as flow from the viewpoint of heat transfer coefficient. Nevertheless, the
Fig. 8 Heat flux ration: „a… M Ä0.5; „b… M Ä1.0; „c… spanwise average, M Ä0.5; and „d… spanwise average, M Ä1.0.
information revealed in Fig. 7共d兲 confirms that one of the major values of q/q o than with M ⫽1.0. This observation is consistent
effects of hole shaping is the reduction of heat transfer coefficient, with the general notion that the optimal cooling performance with
particularly for high-blowing-ratios. injection through non-shaped, discrete circular holes occurs when
M ⬃0.5.
Heat Transfer Relative to Uncooled Situation. As men- As a result of favorably combining film effectiveness and heat
tioned earlier, to estimate the wall heat transfer rate with film transfer coefficient, Shape C with diffusion in both forward and
cooling requires the knowledge of both heat transfer coefficient lateral directions produces the lowest heat transfer, by a margin of
共h兲 and reference temperature 共T r or T aw兲, i.e., 20 to 30 percent lower than Shape A. The most significant im-
q⫽h 共 T aw⫺T w 兲 . (11) provement lies in the region near the injection hole with a higher
blowing ratio, M ⫽1.0. Shape B with forward expansion only
T aw is related to the film effectiveness 共␩兲 via the definition of yields a relatively moderate heat transfer reduction compared to
␩ ⫽(T aw⫺T m )/(T c ⫺T m ), as previously mentioned. According to Shape A. However, it is notably less effective than Shape C.
Metzger et al. 关20兴, Sen et al. 关6兴 and Ekkad et al. 关14,15兴, the
ratio of heat flux on a film-protected surface to the corresponding
baseline value without film cooling can be expressed by Conclusions
The transient liquid crystal technique has been further demon-
q/q o ⫽ 共 h/h o 兲共 1⫺ ␩ / ␸ 兲 , (12)
strated as an effective means of experimentally examining film
where ␸ is the overall cooling effectiveness given by ␸ ⫽(T w cooling performance in great detail. The technique is unique in its
⫺T m )/(T c ⫺T m ). Typical values of ␸ in actual engines range capability of revealing both the film effectiveness and heat trans-
from 0.5 to 0.7. The present study uses ␸⫽0.6, as the mean value fer coefficient simultaneously. The present study marks the first
of the actual range. attempt of using the transient liquid crystal technique to system-
Figure 8 gives both the local and spanwise-averaged distribu- atically evaluate the effects of film hole shaping as compared to
tions of q/q o along the streamwise coordinate for different blow- the reference case with a row of five straight circular holes ori-
ing ratios. According to Figs. 8共c兲 and 共d兲, all the three cases ented 30 deg relative to the mainstream direction 共Shape A兲. Two
studied, in average, provide the surface with a certain degree of closely related, diffusion shaped holes, one with a 10 deg forward
thermal protection, as the values of spanwise-averaged q/q o are expansion 共Shape B兲 and the other with a 10 deg expansion in
always less than one. In addition, the case with M ⫽1.0 appears to both forward and lateral directions 共Shape C兲 are evaluated and
be more effective overall in reducing the local heat flux than that compared. The heat transfer results are substantiated by a flow
with M ⫽0.5. The only exception lies in the region of X/D⬍5, visualization study using a pulsed laser sheet with alumina seed-
where Shape A and Shape B with M ⫽0.5, in fact, have lower ings in the injection. Summarized below are key conclusions.
Shape C with both forward and lateral diffusion produces the Subscript
most significant effect on the film cooling performance in terms of aw ⫽ adiabatic wall
hole shaping. It gives the highest film effectiveness as well as D ⫽ based on jet inlet diameter
overall heat transfer reduction among all the three hole-geometries Dh ⫽ based on hydraulic diameter of test channel
studied. The margin of improvement for these two performance c ⫽ coolant at hole exit
measures is about 30 to 50 percent, depending on the blowing i ⫽ initial
ratio and location. While its heat transfer coefficient is not overall m ⫽ mainstream
the lowest, the magnitude is substantially lower than that of the o ⫽ without injection
baseline, Shape A, and somewhat close to that of Shape B. r ⫽ reference
Shape B with forward-expansion and without lateral diffusion w ⫽ wall
shows a moderate performance improvement compared to Shape
A and a notably worse performance than Shape C. It inherits the Superscript
lowest heat transfer coefficient in the near-hole region among all - ⫽ spanwise average
the three cases studied. This outcome appears to be a compromise
of two competing effects, i.e., boundary layer thickening and References
injection-mainstream shear interaction. Its film effectiveness, on 关1兴 Ligrani, P. M., Wigle, J. M., Ciriello, S., and Jackson, S. M., 1994, ‘‘Film-
the other hand, resembles greatly to that around Shape A. This Cooling From Holes With Compound Angle Orientations: Part 1—Results
Downstream of Two Staggered Rows of Holes with 3d Spanwise Spacing,’’
observation further attests to the fact that the flow features around ASME J. Heat Transfer, 116, pp. 341–352.
these two shaped holes are quite similar, as evidenced in the laser- 关2兴 Ligrani, P. M., Wigle, J. M., and Jackson, S. M., 1994, ‘‘Film-Cooling from
sheet flow visualization also performed in this study. Holes With Compound Angle Orientations: Part 2—Results Downstream of a
Spatial variations in both film effectiveness and heat transfer Single Row of Holes with 6d Spanwise Spacing,’’ ASME J. Heat Transfer, 116,
pp. 341–352.
coefficient are very significant regardless of hole shaping. The 关3兴 Eriksen, and Goldstein, 1974, ‘‘Heat Transfer and Film Cooling Following
range of variation across a measurement span often exceeds ten Injection Through Inclined Tubes,’’ ASME J. Heat Transfer, 96, pp. 239–245.
folds for both variables. As expected, the local film effectiveness 关4兴 Hay, N., Lampard, D., and Saluja, C. L., 1985, ‘‘Effects of Cooling Films on
always reveals the highest magnitudes along the centerline and the the Heat Transfer Coefficient on a Flat Plate With Zero Mainstream Pressure
Gradient,’’ ASME J. Eng. Gas Turbines Power, 107, pp. 104 –110.
lowest along the mid-span for a given streamwise location down- 关5兴 Goldstein, R. J., Eckert, E. R. G., and Burggraf, F., 1974, ‘‘Effects of Hole
stream to the injection hole. On the contrary, the heat transfer Geometry and Density on Three-Dimensional Film Cooling,’’ Int. J. Heat Mass
coefficient generally has its greatest values around the side-edge Transf., 17, pp. 595– 607.
of an injection hole. Such an elevated h is attributable to excessive 关6兴 Sen, B., Schmidt, D. L., and Bogard, D. G., 1996, ‘‘Film Cooling with Com-
pound Angle Holes: Heat Transfer,’’ ASME J. Turbomach., 118, pp. 800– 806.
flow shear generated locally, due mainly to the injection- 关7兴 Schmidt, D. L., Sen, B., and Bogard, D. G., 1996, ‘‘Film Cooling with Com-
mainstream interaction. As a result, the mid-span, which is closer pound Angle Holes: Adiabatic Effectiveness,’’ ASME J. Turbomach., 118, pp.
to the side-edge of an injection hole, generally has the higher heat 807– 813.
transfer coefficient than the centerline. However, Shape A with a 关8兴 Wittig, S., Schultz, Gritsch, M., and Thole, K. A., 1996, ‘‘Transonic Film-
Cooling Investigations: Effects of Hole Shapes and Orientations,’’ ASME Pa-
relatively high blowing ratio reveals a sharp rise in h immediately per 96-GT-222.
behind the injection hole. Again, the strong shear generated be- 关9兴 Gritsch, M., Schultz, A., and Wittig, S., 1998, ‘‘Adiabatic Wall Effectiveness
neath the lifted jet may be responsible for such an phenomenon Measurements of Film-Cooling Holes with Expanded Exits,’’ ASME J. Tur-
关21兴. bomach., 120, pp. 549–556.
关10兴 Gritsch, M., Schultz, A., and Wittig, S., 1998, ‘‘Heat Transfer Measurements
of Film-Cooling Holes with Expanded Exits,’’ ASME Paper 98-GT-28.
关11兴 Haven, B. A., Yamagata, D. K., Kurosaka, M., Yamawaki, S., and Maya, T.,
Acknowledgment 1997, ‘‘Anti-Kidney Pair of Vortices in Shaped Holes and Their Influence on
Film Cooling Effectiveness,’’ ASME Paper 97-GT-45.
Support of this work, in part, is under a DOE Advanced Gas 关12兴 Chen, P-H., Ai, D., and Lee, S.-H., 1998, ‘‘Effects of Compound Angle Injec-
Turbine System Research program sponsored by the South Caro- tion on Flat-Plate Film Cooling Through a Row of Conical Holes,’’ ASME
lina Energy Research Center. Paper 98-GT-459.
关13兴 Vedula, R. J., and Metzger, D. E., 1991, ‘‘A Method for the Simultaneous
Determination of Local Effectiveness and Heat Transfer Distributions in
Three-Temperature Convection Situations,’’ ASME Paper 91-GT-345
Nomenclature 关14兴 Ekkad, S. V., Zapata, D., and Han, J. C., 1997, ‘‘Heat Transfer Coefficients
Over a Flat Surface With Air and CO2 Injection Through Compound Angle
Cp ⫽ fluid specific heat Holes Using a Transient Liquid Crystal Image Method,’’ ASME J. Turbom-
ach., 119, pp. 580–586.
D ⫽ injection hole inlet diameter 关15兴 Ekkad, S. V., Zapata, D., and Han, J. C., 1997, ‘‘Film Effectiveness over a Flat
Dh ⫽ channel hydraulic diameter Surface With Air and CO2 Injection Through Compound Angle Holes Using a
h ⫽ convective heat transfer coefficient Transient Liquid Crystal Image Method,’’ ASME J. Turbomach., 119, pp. 587–
k ⫽ thermal conductivity of test surface 593.
关16兴 Yu, Y., and Chyu, M. K., 1998, ‘‘Influence of Gap Leakage Downstream of the
M ⫽ blowing ratio⫽( ␳ V) c /( ␳ V) m Injection Holes on Film Cooling Performance,’’ ASME J. Turbomach., 120,
q ⫽ wall heat flux pp. 541–548.
Re ⫽ Reynolds number based on D h ,⫽V c D h / ␮ 关17兴 Kassab, A. J., Divo, E., and Chyu, M. K., 1999, ‘‘A BEM-Based Inverse
ReD ⫽ Reynolds number based on film hole inlet diameter, Algorithm to Retrieve Multi-dimensional Heat Transfer Coefficients from
⫽V c D/ ␮ Transient Temperature Measurements,’’ BETECH99, Proc. 13th International
Boundary Element Technology Conference, Computational Mechanics, Las
T ⫽ temperature Vegas, June 8 –10, 1999, C. S. Chen, C. A. Brebbia, and D. Pepper, eds.,
V ⫽ mean flow velocity Billerica, MA.
X ⫽ streamwise coordinate 关18兴 Chyu, M. K., and Ding, H., 1997, ‘‘Heat Transfer in a Cooling Channel With
Vortex Generators,’’ ASME J. Heat Transfer, 119, p. 206.
Y ⫽ spanwise coordinate 关19兴 Kline, S. J., and McClintock, F. A., 1953, ‘‘Describing Uncertainties in Single-
Greek Symbol Sample Experiments,’’ Mech. Eng. 共Am. Soc. Mech. Eng.兲, 75, pp. 3– 8.
关20兴 Metzger, D. E., Carper, H. J., and Swank, L. R., 1968, ‘‘Heat Transfer with
␣ ⫽ test surface thermal diffusivity Film Cooling Near Nontangential Injection Slots,’’ J. Eng. Power, 90, pp.
␤ ⫽ injection angle 157–163.
关21兴 Metzger, D. E., and Larson, D. E., 1986, ‘‘Use of Melting Point Surface
␩ ⫽ film cooling effectiveness Coating for Local Convective Heat Transfer Measurements in Rectangular
␮ ⫽ fluid dynamic viscosity Channel Flows with 90 Degree Turns,’’ ASME J. Heat Transfer, 108, pp.
␶ ⫽ time 48 –54.
Energy and Exergy Balance in the
Process of Spray Combustion in a
Gas Turbine Combustor
A theoretical model of exergy balance based on availability transfer and flow availability
S. K. Som in the process of spray combustion in a gas turbine combustor has been developed to
e-mail: sksom@mech.iitkgp.ernet.in evaluate the total thermodynamic irreversibility and second law efficiency of the process
at various operating conditions, for fuels with different volatilities. The velocity, tempera-
N. Y. Sharma1 ture and concentration fields in the combustor, required for the evaluation of the flow
availabilities and process irreversibilities, have been computed numerically from a two
Department of Mechanical Engineering phase separated flow model of spray combustion. The total thermodynamic irreversibility
Indian Institute of Technology in the process of spray combustion has been determined from the difference in the flow
Kharagpur 721 302, India availability at inlet and outlet of the combustor. The irreversibility caused by the gas
phase processes in the combustor has been obtained from the entropy transport equation,
while that due to the inter-phase transport processes has been obtained as a difference of
gas phase irreversibilities from the total irreversibility. A comparative picture of the
variations of combustion efficiency and second law efficiency at different operating con-
ditions for fuels with different volatilities has been made to throw light on the trade off
between the effectiveness of combustion and the lost work in the process of spray com-
bustion in a gas turbine combustor. 关DOI: 10.1115/1.1484393兴
Keywords: Combustion, Energy, Exergy, Heat Transfer, Second Law, Sprays
Introduction in the process of spray evaporation based on the exergy loss in the
process. Dunbar and Lior 关11兴 calculated the sources of combus-
The combustion of liquid fuel spray has widespread applica-
tion irreversibilities with hydrogen and methane flames and found
tions in gas turbines, internal combustion engines, rocket motors that approximately one-third of the useful energy of the fuel is
and industrial furnaces. The research in the area of spray combus- destroyed during the combustion process.
tion is going on over the last few decades and a host of articles, The objective of the present work is to furnish a comparative
including the exhaustive reviews of Faeth 关1,2兴 and Sirignano picture of the effectiveness in the quantitative conversion of en-
关3–5兴, are available in the literature. However, all the works were ergy and the effectiveness in preserving the quality 共work poten-
centered around in predicting the different aspects of combustion tial兲 of energy in a spray combustion process in a gas turbine
performance like combustion efficiency, temperature and species combustor. This is made through a comparison of combustion
concentration distribution within the combustor and emission efficiency and second law efficiency for fuels with different vola-
characteristics. Thus the basic motivation of the studies, already tilities at various operating conditions of the combustor. The rela-
done in the field, remained to identify the operating parameters to tive role of gas phase and interphase transport processes towards
achieve a high combustion efficiency with a low liner temperature the loss of exergy has also been analyzed. In the first part of the
and a relatively uniform exit temperature distribution for turbo- work, a basic flow and combustion modeling of a typical spray
machinery consideration. combustion process is described. In the second part, a theoretical
From the viewpoint of energy economy, an efficient spray com- model of exergy balance is developed to evaluate the second law
bustion process should be guided not only by the combustion efficiency with the help of the data on velocity, temperature and
efficiency of the process but also by its second law efficiency that species concentration fields in the combustor provided by the
gauges the effectiveness in preserving the quality of energy. spray combustion model in the first part of the work.
Hence the exergy analysis of a spray combustion process to evalu-
ate the second law efficiency provides additional thermodynamic
information to determine the locations, types and magnitude of Theoretical Formulation
losses. A minimization of the losses by suitable choice of operat- Physical Model. The physical problem refers to the evapora-
ing parameters leads to an optimum design of the process for an tion and combustion of a continuously injected liquid fuel spray in
overall energy economy. a can type combustor 共Fig. 1兲, typical of a gas turbine combustion
The information regarding the thermodynamic irreversibilities process. The air supply to the combustor is split among the swirler
and exergy loss in the process of spray combustion is not available at the entry and through two radial jets in the form of secondary
in the literature till date. While the work of Dash et al. 关6兴 pro- and dilution air. Fuel spray is injected from an atomizer located at
vides information regarding the thermodynamic irreversibilities in the hub of the swirler. The problem is assumed to be axi-
droplet evaporation, the works of Dash and Som 关7兴, Puri 关8兴, and symmetric.
Hiwase et al. 关9兴, provide similar information in case of droplet
combustion. Som and Dash 关10兴 studied the second law analysis Flow and Combustion Modeling. A numerical model of the
spray combustion process within the combustor has been devel-
1
On leave from Dept. of Mech & IP Engg, Manipal Institute of Technology,
oped to determine the velocity, temperature and species concen-
MAHE, Manipal, India. tration fields along with the combustion efficiency.
HEAT TRANSFER. Manuscript received by the Heat Transfer Division May 9, 2001; Numerical Computation of Spray Combustion. The com-
revision received April 11, 2002. Associate Editor: H. Lee. putation is based on a typical two phase Eulerian-Lagrangian for-
2 2 ⳵U j
p ⬘ ⫽p⫹ ␳ k⫹ ␮ eff (4)
3 3 ⳵x j
Turbulent Kinetic Energy (␬).
⳵
⳵t
共 ␳␬ 兲⫹
⳵
⳵xi
共 ␳ U i␬ 兲⫽
⳵ ␮t ⳵␬
⳵xi ␴k ⳵xi 冉冊
⫹ P⫺ ␳ ␧⫹Ṡ ␬ (5)
where, turbulent kinetic energy production rate P is given by
P⫽ ␮ t 冉 ⳵x j
⫹ 冊
⳵Ui ⳵U j ⳵Ui 2 ⳵Ui
⫺
⳵xi ⳵x j 3 ⳵xi
␳␬⫹␮t
⳵Ui
⳵xi 冉冊 (6)
Fig. 1 Physical model Turbulent Kinetic Energy Dissipation Rate (␧).

⳵
⳵t
共 ␳␧ 兲⫹
⳵
⳵xi
共 ␳ U i␧ 兲⫽
⳵ ␮t ⳵␧
⳵xi ␴␧ ⳵xi 冉 ␧
冊
␧2
⫹c 1␧ P⫺ ␳ c 2␧ ⫹Ṡ␧
␬ ␬
(7)
mulation of an axi-symmetric gas-droplet flow with a standard ␬-␧
model for the gas phase turbulence and a global reaction kinetics Constants for standard ␬-␧ model are taken as,
for the gas phase reaction of fuel vapor with oxygen. The body c ␮ ⫽0.09, c 1␧ ⫽1.44, c 2␧ ⫽1.92, ␴ k ⫽1.0, and ␴ ␧ ⫽1.3
force and buoyancy force are neglected in the analysis, so also the
Energy.
virtual mass force and Besset force are not considered due to high
density ratio between the phases. Droplets do not take part in ⳵ ⳵
radiative energy exchange, which is solved purely as a gas phase 共 ␳ c gp T g 兲 ⫹ 共 ␳ c gp U i T g 兲
⳵t ⳵xi
phenomenon.
The adoption of standard ␬-␧ model has been made despite the
fact that many researchers observe some shortcomings in the abil-
ity of the ␬-␧ model in predicting a swirling flow field quantita-
⫽
⳵
⳵xi 冉
␳ c gp ␣ eff
⳵Tg
⳵xi
⫺
⳵xi 冊
⳵ q ri
⫹Ṡ E ⫹c gp T g Ṡ (8)
tively. However, there is no conclusive information available in The Energy source term, Ṡ E has two components—one of which
the literature regarding the accurate adaptability of a suitable is the energy absorbed by the liquid droplets during their
modification of the ␬-␧ model for a confined swirling flow. The heating-up period calculated from the inter-phase transport and
models like ASM and RSM are either too complex or equally poor the other is the energy generated due to chemical reaction.
in predicting a strong swirling flow. Moreover, it is expected that The energy addition due to combustion is determined in con-
the accuracy in predicting the gross flow parameters like combus- sideration of a single step, irreversible, global reaction between
tion efficiency and second law efficiency will not be much af- the fuel vapor and oxygen following a finite rate chemistry as,
fected with the adoption of a standard ␬-␧ model. Therefore the
standard ␬-␧ model has been considered for the solution in the ␭ F F⫹␭ O O 2 →␭ P P
present work. The reaction in the combustion chamber is either kinetic or turbu-
The axi-symmetric spray model is a simplified assumption. lent diffusion controlled. The kinetic controlled reaction rate is
However, it predicts a fairly accurate results in the primary zone determined following an Arrhenius-type equation with the corre-
of fuel rich region, though it fails to generate accurate quantitative sponding kinetic parameters taken from the work of Westbrook
results in the secondary zone due to blowing of secondary and and Dryer 关12兴, as
dilution air through radial holes. The simplification of axi-
symmetric model provides at least an understanding in recogniz-
ing the pertinent controlling parameters and their qualitative influ-
ences on the irreversibility rate and second law efficiency of a
␻˙ k ⫽B ␳ 2
C jC O2
M f M O2
exp ⫺ 冉冊
E
RT
(9)
spray combustion process which is the prime objective of the In a turbulent diffusion flame, the rate of combustion may be
present work. Moreover, it can be argued that, since the major part guided by the rate of inter mixing on a molecular scale of fuel and
of irreversibilities occurs in the primary zone due to droplet va- oxygen eddies. The rate of turbulent diffusion reaction is given by
冋冉冊册
porization, chemical reaction and the sharp temperature and con-
centration gradients, the quantitative results related to the thermo- ␳ ␧ C O 2 cC P
␻˙ d ⫽A min C f , , (10)
dynamic aspects may not be far from the actual ones. Mf ␬ ␥ 1⫹ ␥
Gas Phase Conservation Equations. The average gas phase where,
conservation equations are as follows:
␥⫽␹oM o /␹ f M f
Continuity.
The empirical constants A and c are as suggested by Magnussen
⳵␳ ⳵ et al. 关13兴. Thus the energy source term due to chemical reaction
⫹ 共 ␳ U i 兲 ⫽Ṡ (1) will be,
⳵t ⳵xi
˙ k ,␺˙ d)兴
⌬H R M f 关 min(␻
Momentum. where, ⌬H R is the enthalpy of reaction. The radiative energy ex-
⳵
⳵t
共 ␳Ui兲⫹
⳵
⳵x j
共 ␳ U i U j 兲 ⫽⫺
⳵p⬘
⫹
⳵
⳵xi ⳵x j 再冉
␮ eff
⳵Ui ⳵U j
⳵x j
⫹
⳵xi 冊冎 change within the gas phase is evaluated by neglecting the influ-
ence of the droplets, and assuming the gas phase to be gray ab-
sorbing emitting medium. The radiative transfer equation is solved
following a first order moment method, 共Ozisik 关14兴兲 which re-
⫹Ṡ M i ⫹ṠU i (2)
duces the integro-differential equation of radiative energy transfer
where, into a differential equation as shown below.
k2 ⳵ q ri
␮ eff⫽ ␮ ⫹c ␮ ␳ (3) ⫽ ␬ 关 4 ␲ I b 共 T 兲 ⫺G 兴 (11)
␧ ⳵xi
where the irradiation G is related to q ri as where p v and p are the partial pressure of fuel vapor at droplet
surface and total pressure of the local gas phase, respectively.
1 ⳵G
q ri ⫽⫺ (12) Droplet Temperature.
3␬ ⳵xi
In the light of the assumption of a gray gas radiation model, the dT d dm d
m d c dp ⫽h ␲ d 2 共 T g ⫺T d 兲 ⫹ ⌬H v (18)
absorption coefficient ␬ is replaced by Planck’s mean absorption dt dt
coefficient, ␬ P . The local value of ␬ P is computed as
where, ⌬H v is the enthalpy of vaporization of the liquid fuel at
␬ P⫽C H2 O␬ PH2 O共 T 兲 ⫹C CO2 ␬ PCO2 共 T 兲 (13) droplet temperature. Mass transfer coefficient ␤ and heat transfer
coefficient h in Eqs. 共17兲 and 共18兲, respectively, are evaluated
The values of ␬ P (T) as a function of temperature, for the radia- from the standard correlations by Ranz and Marshall 关17兴 and
tively participating gases CO2 and H2 O, are taken from Sparrow Dash et al. 关6兴 as
and Cess 关15兴
Individual Species Conservation. Nu共 1⫹B 兲 ⫽2⫹0.6 Re0.5
d Pr
0.33
(19)
⳵ ⳵ ⳵
冉
⳵C j
冊 Sh共 1⫹B 兲 ⫽2⫹0.6 Re0.5 0.33
d Sc (20)
共 ␳C j 兲⫹ 共 ␳ U iC j 兲⫽ ␳ D eff ⫹Ṡ C j ⫹ṠC j (14)
⳵t ⳵xi ⳵xi ⳵xi where B is the Transfer Number. Eqs. 共15兲, 共17兲, and 共18兲 are
The species conservation equation is solved for fuel vapor, oxy- solved for V di , m d , and T d respectively, with appropriate initial
gen, carbon-di-oxide and water vapor while nitrogen concentra- conditions.
tion is obtained by difference. The conservation equation for each The initial drop size distribution of liquid fuel spray is assumed
species contains a source term S C j given by to follow a realistic four-parameter Rosin-Rammler distribution
function given by,
S C j ⫽⫺ ␻ M f for fuel vapor
exp共 ⫺bd ni 兲 ⫺exp共 ⫺bd nmax兲
S C j ⫽⫺ ␻ M f ␥ for oxygen G ⬘共 d i 兲 ⫽ (21)
exp共 ⫺bd min
n
i 兲 ⫺exp共 ⫺bd max i 兲
n
S C j ⫽⫺ ␻ M f 共 1⫹ ␥ 兲 where, G ⬘ (d i ) is the mass fraction of the spray having diameter

above d i . The dispersion parameter n is taken as 3, as recom-
for the product 共 carbon—dioxide and water vapor兲 mended by Mugele and Evans 关18兴.
where the rate of reaction ␻ is equal to min(␻k ,␻d).
The additional source term SC j is zero for all species except the
fuel vapor. This takes care of the fuel mass evaporated from the
droplets.
Method of Solution
The gas phase conservation equations 共Eqs. 共1兲, 共2兲, 共5兲, 共7兲,
Generation of Droplet Phase Information. The velocity, 共8兲, and 共14兲兲 were solved simultaneously by an explicit finite
mass and temperature history of all droplet classes along their difference computing technique developed by Hirt and Cook 关19兴
trajectories are obtained from the respective conservation equa- following the original MAC 共Marker and Cell兲 method due to
tions on a Lagrangian frame as follows. Harlow and Welch 关20兴. The space derivatives of the diffusion
Droplet Velocity. terms were discretized by the central differencing scheme, while
the advection terms were discretized by a hybrid differencing
dV di ␲ 2 scheme based on the local Peclet number 共Pe兲 associated with the
md ⫽ ␳ d C D 兩 V gi ⫺V di 兩共 V gi ⫺V di 兲 (15) cell. The conservation equations for the droplets 共Eqs. 共15兲, 共17兲
dt 8
and 共18兲兲 were solved by fourth order Runge-Kutta method with
Drag coefficient C D is computed following the standard drag law appropriate initial conditions. A variable sized adaptive grid sys-
by Clift et al. 关16兴 tem was considered and the variations in the size of the grids were
The effect of gas phase turbulence on the droplet motion is simu- made smoothly. A numerical mesh of 89⫻31 grid nodes was used
lated using a stochastic approach. The instantaneous gas phase after several numerical experiments, which showed that further
velocity (V gi ) is obtained by computing the fluctuating velocity refinement in grids in either direction did not change the result
component from the turbulent kinetic energy in consideration of 共maximum change in velocity or any scalar variable in the carrier
isotropic turbulence, and using a normally distributed random phase兲 by more than 2 percent. The choice of time increment ⌬t
number ␨, as was made to ensure stability in the computation in accordance
冑
with the criteria of cell transit time of fluid due to convection and
2␬ diffusion respectively. Often a more stringent restriction was re-
V gi ⫽U i ⫹␵ (16)
3 quired to have a converged solution and was fixed by trail and
error in the computation.
Droplet Mass. The total air flow to the combustor was splitted among the
dm d swirler, and two radial jets according to a ratio of 5:7:8 following
⫽⫺ ␳ m ␤ ␲ d 2 共 C f s ⫺C f 兲 (17) Cameron et al. 关21兴. Air flow entering the combustor through the
dt swirler was considered to be in a plug flow mode with a typical
where ␳ m and C f s , are the density of the gas phase and the mass solid body type rotation imparted by helicoidal vane-swirler.
fraction of fuel vapor respectively at the droplet surface. In con- Therefore, a uniform axial velocity distribution and a linear tan-
sideration of thermodynamic equilibrium at the droplet surface, gential velocity distribution of air were considered at the inlet
C f s , is calculated from the vapor pressure at droplet temperature plane of the combustor. The temperature distribution of air at the
inlet plane was also considered to be uniform. A zero axial gradi-
as ent was prescribed at the outlet for all the variables. Standard
1 logarithmic law of wall was considered for the near wall region.
冉冊
C s⫽ For radiation calculation, the end planes were assumed to be ra-
Mo p
1⫹ ⫺1 diatively adiabatic and Marshak boundary condition was applied
M f pv at the solid combustor wall. The fuels were considered to be n
Table 1 Fuel volatility characteristics
hexane (C6 H14), kerosene(C10H20) and n dodecane(C12H26). The

volatility characteristics of the fuels considered are as shown in Fig. 3 Comparisons of predicted axial velocity and tempera-
Table 1. ture distributions with experimental results of Khalil et al. †24‡
„—present computation, 䊊 Khalil et al.…: „a… z Õ D Ä0.55; „b… z Õ D
Combustion Efficiency. In consideration of a global reaction Ä0.55; „c… z Õ D Ä3.6; and „d… z Õ D Ä3.6.
kinetics and without any possibility of the existence of liquid fuel
droplet at the exit, the combustion efficiency for a given inflow of
fuel is determined from the fuel vapor concentration at the com-
bustor exit as, ever it is observed from Figs. 2共b兲 and 共d兲 that the present numeri-
冉冕冊
cal results and the computations of Chang and Chen, being in
R
agreement with each other, do not conform well with the experi-
2␲ ␳ e U z e C f e rdr
0
mental results of Dellenback for the tangential velocity distribu-
␩ c ⫽1⫺ (22) tion at far down stream from the inlet (z⫽4.0) and mainly near
ṁ f the axis. This was explained by Chang and Chen in the light of
Calibration of the Numerical Model. The accuracy of the under-prediction of the length of central toroidal recirculation
quantitative and even the qualitative trends of the predicted results zone by the standard ␬-␧ model.
relating to combustion efficiency, process irreversibilities, and A possible comparison of axial velocity and temperature distri-
second law efficiency depends mainly on the accuracy with which butions predicted by the present model in case of an axi-
the velocity, temperature and species concentration fields in the symmetric swirling flow in a dump combustor has been made with
process of spray combustion have been determined from the nu- the experimental work of Khalil et al., 关24兴 under a similar situa-
merical computation of the present model. A comparison of the tion 共Figs. 3共a–d兲兲. Due to a lack of information on spray param-
results predicted from the present model with those of earlier eters in the reported experimental work of Khalil et al., the spray
works in identical situations is not always possible, since the in- parameters considered for the comparison are chosen after several
formation available in those works do not give explicitly all the trials taken over the mean drop diameter and spray cone angle.
input parameters required to generate the output data of the Since the fuel used in the experiments of Khalil et al., was kero-
present model for the purpose of comparison. However a possible sene, the same fuel has been chosen in the present model for the
comparison of the present results in case of an axisymmetric purpose of comparison. It is observed that the predictions of axial
sudden expansion isothermal swirling flow, and also for a reac- velocity components agree fairly well with the experimental re-
tive swirling flow in a dump combustor has been made with sults at an upstream location within the combustor 共Fig. 3共a兲兲,
the relevant empirical and computational results available in the while a deviation between the two results is observed mainly near
literature. the axis at a section far downstream from the inlet 共Fig. 3共c兲兲. This
In case of an isothermal flow, the comparison of the present discrepancy may be attributed to the fact that the standard ␬-␧
results with the computational work of Chang and Chen 关22兴 and model under-predicts the length of the central toroidal recirculat-
the experimental results of Dellenback 关23兴 under the similar situ- ing zone in the swirling flow field. Moreover, the assumption of
ations are shown in Figs. 2共a兲–共d兲. Figures 2共a兲 and 共c兲 show a an axi-symmetric model is less justified at a far downstream loca-
fair agreement of the predicted axial and tangential velocity pro- tion because of the introduction of secondary air through radial
files near the inlet region (z⫽0.75) with the computations of holes. The temperature distributions predicted by the present
Chang and Chen and the empirical values of Dellenback. How- model show a fair agreement with the empirical results both at
upstream and downstream locations of the combustor 共Figs. 3共b兲
and 3共d兲兲. However, the predicted temperature distribution at the
downstream location shows an increasing trend near the wall, un-
like that obtained in the experimental results of Khalil et al. This
may possibly be attributed to the fact that the assumption of stick
model for the droplet evaporation results in a zone of burning near
the wall in the downstream part of the combustor.
Exergy Model of Spray Combustion

The exergy analysis of the problem has been made on the basis
of flow availability to the system comprising the entire combustor
共Fig. 1兲, considering the outer wall to be adiabatic. Since the
words ‘‘availability’’ and ‘‘exergy’’ are synonymous in their ther-
modynamic implications, both of them will be referred to in the
Fig. 2 Comparisons of predicted axial and tangential velocity remaining part of the text, depending upon the appropriate uses at
distributions, with Chang and Chen †22‡ and Dellenback †23‡ places following the general convention. It can be written for the
„—Present computation, 䊊 Chang and Chen, and Ã Dallen- conservation of exergy of the system that,
back… „a… z Õ D Ä0.75; „b… z Õ D Ä4.0; „c… z Õ D Ä0.75; and „d… z Õ D
Ä4.0. Ȧ in⫽Ȧ e ⫹İ (23)
where, A in and A e are the rates of flow availability coming in and ā inf
going out of the system, respectively, and i is the rate of thermo- a inf ⫽ (27)
dynamic irreversibility within the system. Mf
Determination of Flow Availabilities. The flow availabilities where, M f the molecular weight of the fuel.
of different species at inlet and outlet of the combustor have been The flow availability at the combustor exit can be expressed as,
冕冉兺冊
evaluated with respect to an exergy reference thermodynamic R
state of P r ⫽101.35 kN/m2 , T r ⫽298.15 K with mole fractions Ȧ e ⫽2 ␲ C ej a ej ␳ ge U z e rdr (28)
O CO H O
of the constituents as x r 2 ⫽0.2035, x r 2 ⫽0.0003, and x r 2 0 j
⫽0.0303 as recommended by Moran and Shapiro 共25兲.

The flow availability at the inlet, being associated with the in- where, C ej is the mass fraction of the j th species and U ze is the
flow of air and fuel to the system, can be written as, local axial gas velocity at the exit plane.
The specific flow availability of each species at a point in the
Ȧ in⫽ṁ a ina ain⫹ṁ f ina inf (24) exit plane (a ej ) is calculated based on local pressure and tempera-
ture in the similar fashion as done for inlet. Both chemical and
where, thermo-mechanical availabilities for fuel vapor, and only thermo-
ṁ a in ⫽ mass flow rate of air into the combustor mechanical availability for the other species are considered
a ain ⫽ specific flow availability of air at inlet
Estimation of Process Irreversibilities. The total irrevers-
ṁ f in ⫽ mass flow rate of fuel into the combustor ibility in a typical two phase gas-droplet flow in the combustor is
a inf ⫽ specific flow availability of fuel at inlet composed of two parts, namely, 共i兲 the irreversibility due to the
The specific flow availability of air is determined from the follow- transport processes and chemical reaction in the continuous gas
ing equation: phase, İ g and 共ii兲 the irreversibility due to the inter-phase trans-
ports 共transport processes between the discrete evaporating drop-
a ain⫽ 共 h ain⫺h r 兲 ⫺T r 共 s ain⫺s r 兲 (25a) lets and their local ambience兲 İ d . Therefore, it can be written as,
where,
İ⫽İ g ⫹İ d ⫽T r 共 Ė g ⫹Ė d 兲
冕
(29)
a
T in
h ain⫺h r ⫽ c ap 共 T 兲 dT (25b) where Ė g is the entropy generation rate due to transport processes
Tr
and chemical reaction in the gas phase and Ė d is the entropy
s ain⫺s r ⫽ 冕 Tr
a
T in c ap 共 T 兲
T
dT⫺
R̄
Ma 冕 p in
pr
a
dP
P
(25c)
generation rates due to inter-phase transport processes of discrete
evaporating droplets.
The total irreversibility İ is determined from Eq. 共23兲. To assess
The molar flow availability 共comprising both chemical and the relative weightage of İ g and İ d in İ, any one of these two 共İ g
thermo-mechanical components兲 of a generic hydrocarbon fuel and İ d 兲 has to be determined independently. The entropy genera-
C a H b undergoing a single step global reaction tion İ g due to gas phase transport process and chemical reaction is
冉冊
determined from the general entropy transport equation in a con-
b b tinuous field of compressible Newtonian fluid. The entropy gen-
C a H b ⫹ a⫹ O →aCO2 ⫹ H2 O
4 2 2 eration rate per unit volume, ė g at a point in the gas phase can be
written, following Hirschfelder et al. 关26兴 as,
can be written, following Moran and Shapiro 关25兴 as,
冉冉冊
ā inf ⫽ h̄ inf ⫹ a⫹
b O2
4
CO b HO
h r ⫺ah̄ r 2 ⫺ ¯h r 2
2 冊 ė g ⫽
⌬: ␴ ⫺J q ⵜT g 兺共 ⫺J m j ⵜ ␮ c j 兲
Tg
⫹
共 Tg兲2
⫹
Tg
冉冉冊
⫺T r s̄ inf ⫹ a⫹
b O
4
b
s̄ 2 ⫺as̄ CO2 ⫺ ¯s H2 O
2 冊 (26a) ⫹
兺共 ⫺s̄ j J m j ⵜT g 兲
T g ⫹
1
Tg
兺 ␭ j␮ c j␻
˙ (30)
where,
where, ␴ and ⌬ are the stress and rate of strain tensors respec-
h̄ inf ⫽⌬h̄ Ff ⫹ 冕 Tr
T in
f
c̄ pf 共 T 兲 dT (26b)
tively, J q the heat flux per unit area and J m j , ␮ c j , s̄ j are the mole
flux per unit area, chemical potential and partial molal entropy of
the j h species respectively. The first term in Eq. 共30兲 is due to fluid
O
h̄ r 2 ⫽0 (26c) friction, the second term is due to conduction and radiation of
heat, the third term pertains to mass transfer, the fourth term arises
h̄ r
CO2
⫽⌬h̄ F
CO2
(26d) from the coupling between heat and mass transfer and the fifth
term is due to chemical reaction
H O H O The total gas phase entropy generation is, therefore, calculated
h̄ r 2 ⫽⌬h̄ F 2 (26e)
as,
s̄ inf ⫽s̄ rf ⫹ 冕 Tr
f
T in C pf 共 T 兲
T
dT⫺
R̄
Mf 冕 p in
pr
f
dP
P
(26f) Ė g ⫽ 冕冕冕 ᭙
ė g d᭙ (31)
O O
s̄ O2 ⫽s̄ r 2 ⫺R̄ ln共 x r 2 兲 (26g) where, ᭙ is the volume of the domain of physical processes. Each
CO CO
term of the Eq. 共8兲 is expanded in a cylindrical coordinate system
s̄ CO2
⫽s̄ r 2 ⫺R̄ ln共 x r 2 兲 (26h) to determine ė g under the present situation, with the help of gas
H O H O
phase velocity, temperature and concentration fields calculated
s̄ H2 O⫽s̄ r 2 ⫺R̄ ln共 x r 2 兲 (26i) from numerical computations of spray combustion.
The specific flow availability of the fuel is calculated from the Second Law Efficiency. The second law efficiency of the
molar availability as, combustor can be expressed as,
increases with pressure for the higher volatile fuel 共n hexane兲,
Ȧ e İ while it decreases with pressure for the lower volatile fuels 共kero-
␩ II ⫽ ⫽1⫺ (32)
Ȧ in Ȧ in sene and n dodecane兲.

Influences of Inlet Pressure on Combustion Efficiency and Influences of Inlet Swirl on Combustion Efficiency and Sec-
Second Law Efficiency for Fuels With Different Volatilities. ond Law Efficiency for Fuels With Different Volatilities. It is
It is observed from Table 2共a兲 that an increase in fuel volatility observed from Table 2共a兲 that the combustion efficiency is unin-
increases the combustion efficiency only at a higher pressure for fluenced by the inlet swirl to the flow of air at its lower pressure of
any given inlet temperature and swirl number. The success for 100 kPa., while it decreases with an increase in inlet swirl when
complete combustion depends on the penetration of fuel droplets, the pressure is increased to 1000 kPa. The purpose of imparting
their rate of vaporization and mixing of fuel vapor with air. At low swirl to incoming primary air, is to cause an on-axis recirculatory
pressures, for a given temperature, all these three physical pro- flow near the upstream region of the primary zone to enhance the
cesses are relatively fast and combustion efficiency is usually very rates of droplet vaporization and mixing of fuel vapor in air, along
high. Therefore, the change in the rate of droplet vaporization with the subsequent stabilization of flame. But on the other hand,
with fuel volatility does not make any marked influence in com- the recirculatory flow field in the inlet region of the combustor,
bustion efficiency. reduces the penetration of droplets to further downstream in the
However, on the other hand, when the pressure in the combus- primary zone for a more effective droplet vaporization and burn-
tion chamber is high, at a given inlet temperature, there occur 共i兲 ing. The relative role of these two counter-weighing effects de-
a reduction in droplet penetration due to increased density of am- pends on the range of inlet air pressure. At a lower pressure, these
bient air, 共ii兲 a reduction in the rate of droplet vaporization be- two counter-weighing effects nullify each other in producing any
cause of a reduction in mole fraction of fuel vapor at droplet change in combustion efficiency. However, when the pressure in
surface, and 共iii兲 a reduction in the diffusion of fuel vapor in the combustion chamber is high, the droplet penetration is as such
surrounding air due to a decrease in fuel-air diffusivity. Therefore low. A further reduction in droplet penetration due to an increase
for a given fuel, an increase in combustion pressure for a fixed in inlet swirl counter weighs the favorable influence on droplet
inlet temperature always reduces the combustion efficiency. Under vaporization and mixing of fuel vapor, and finally results in a
this situation, the favorable effects of enhanced droplet vaporiza- lower combustion efficiency.
tion and its mixing with ambient air, due to an increase in fuel The inlet swirl causes a marginal increase in second law effi-
volatility is felt sharply through a marked increase in the combus- ciency at a lower inlet pressure of 100 kPa., while it causes a
tion efficiency. considerable decrease in the second law efficiency at a higher inlet
While the combustion efficiency decreases, the second law ef- pressure of 1000 kPa. At any inlet pressure and swirl, the second
ficiency increases with an increase in pressure at a given tempera- law efficiency increases with a decrease in fuel volatility. It is
ture. The increase in second law efficiency with pressure can be most important to note that at higher inlet pressure, an increase in
attributed to simultaneous increase in flow availability of incom- inlet swirl decreases both the combustion efficiency and second
ing air at a high pressure and less process irreversibility due to law efficiency for all the fuels. While the reduction in combustion
reduced droplet vaporization. The reduction in total irreversibility efficiency is due to the lack of effective droplet penetration in the
(İ), is manifested mostly through a marked reduction in the irre- primary zone, the reduction in second law efficiency is mainly
versibility rate due to inter-phase transport processes İ d , with an because of an increase in İ g due to enhanced gas phase transport
increase in pressure 共Table 2共b兲兲. The gas phase irreversibility (İ g ) processes at increased swirl. Therefore an increase in inlet swirl at
Table 2 Influences of inlet pressure and swirl on combustion efficiencies and

process irreversibilities for fuels with different volatilities „ṁ a inÄ0.1 kgÕs, A s
Ä0.02 m2 , A d Ä0.04 m2 , T inÄ600 K, ReinÄ52400, ␺Ä80°, SMDi Ä52 ␮ m, A Õ F
Ä60…: „ a … comparison of combustion efficiency and second law efficiency „%…;
and „ b … process irreversibilities „JÕs….
Table 3 Influences of spray cone angle on combustion efficiencies and pro-
cess irreversibilities for fuels with different volatilities „ṁ a inÄ0.1 kgÕs, A s
Ä0.02 m2 , A d Ä0.04 m2 , PinÄ100 kPa TinÄ600 K, ReinÄ52400, S Ä0.76, SMDi
Ä52 ␮ m, A Õ F Ä60…: „ a … comparison of combustion efficiency and second law
efficiency „%…; and „ b … process irreversibilities „JÕs….
a higher inlet pressure for the process of spray combustion is not It is observed from Table 3共a兲 that the second law efficiency
thermodynamically meaningful from the view point of conserva- ( ␩ II ) decreases with an increase in spray cone angle 共␺兲 for all
tion of both the quantity and quality of energy. fuels. Again at a given ␺, the value of ␩ II increases with a de-
crease in fuel volatility. The decrease in the value of ␩ II with ␺ is
Influence of Spray Parameters on Combustion Efficiency due to the increase in total thermodynamic Irreversibility, İ of the
and Second Law Efficiency for Fuels With Different Volatili- process 共Table 3共b兲兲 for a given flow availability at the inlet. How-
ties. ever an interesting picture is obtained, if we look into the variation
of different irreversibility components, İ g , and İ d with ␺ for fuels
Influence of Spray Cone Angle. It is observed from Table 3共a兲 of different volatilities. It is observed from Table 3共b兲 that for the
that, the combustion efficiency for a higher volatile fuel 共n hex- higher volatile fuel 共n-hexane兲, an increase in ␺ from 60 deg to
ane兲 is very high and increases marginally with the spray cone 100 deg increases the irreversibility rate, İ d due to inter-phase
angle. For fuels of lower volatility 共kerosene and n dodecane兲, the
transport processes, while the irreversibility rate İ g , contributed
combustion efficiency is very low at a lower spray cone angle of
by the gas phase, remains almost the same. In case of lower vola-
60 deg. However when the spray cone angle of the lower volatile
tile fuels 共kerosene and n-dodecane兲, an increase in ␺ from 60 deg
fuels is increased from 60 deg to 80 deg, there is a considerable
increase in combustion efficiency followed by a slight decrease to 80 deg increases the value of İ g considerably while the value of
with a further increase in spray cone angle to 100 deg. This can be İ d is reduced. But with a further increase in ␺ from 80 deg to 100
physically explained as follows. deg, the value of İ g is reduced while the value of İ d is consider-
At a lower spray cone angle of 60 deg, the penetration of fuel ably increased. This can be explained physically as follows.
droplets is relatively high. The poor rate of droplet vaporization, For a higher volatile fuel, an increase in the value of ␺ causes
in case of fuels with lower volatility, may lead to a situation where larger radial dispersions of fuel droplets and involves more air
a class of vaporizing droplets may reach the downstream location circulating within the spray in enhancing the rate of droplet va-
beyond the primary zone where burning is less conducive due to a porization and thus causes an increase in the value of İ d . In case
lower temperature because of excess air. This does not happen in of lower volatile fuels, an increase in the value of ␺ produces two
case of a higher volatile fuel because of higher rate of fuel vapor- contrasting effects in the rate of droplet vaporization. With an
ization. When the spray cone angle is increased 共␺⫽80 deg, 100 increase in the value of ␺, a large class of droplets strikes the wall
deg兲 most of the droplet classes, for all fuels, strike the combustor and vaporizes while sticking to the wall. Therefore the rate of
wall due to their large radial dispersions. According to the as- inter-phase transport processes is reduced due to a reduction in the
sumed stick model, droplets striking the wall stick and undergo a droplet surface area available for such processes. On the other
complete vaporization there itself. This causes an effective mix- hand, an increase in the value of ␺, reduces the undue droplet
ture of air and fuel vapor within the primary zone and hence penetration. The droplets under this situation, are retained within
results in an increase in combustion efficiency. The increase is an effective burning region in the primary zone, and are exposed
substantial for a lower volatile fuel because many of the droplets to a much steeper temperature and concentration gradients at the
which could have escaped the primary zone before complete va- interface, for which the rates of inter-phase transport processes are
porization in case of lower spray cone angle, have in fact been increased. When the value of ␺ is increased from 60 deg to 80
trapped within the primary zone due to wall sticking at higher deg, the influence of wall sticking of droplets in the rate of inter-
spray cone angle. For a higher volatile fuel, the situation is a little phase transport processes dominates in reducing the value of İ d ,
different. Due to a rapid rate of fuel vaporization, almost all the while for an increase in the value of ␺ from 80 deg to 100 deg, the
droplets are trapped within the primary zone for their complete influence of droplet penetration on inter-phase transport rates
vaporization at all the spray cone angles. However, an increase in dominates in increasing the value of İ d .
spray cone angle involves more combustion air within the spray to
enhance the process of mixing between the fuel vapor and air and Influence of Initial SMD. It is observed from Table 4共a兲, that
thus produces a slight increase in combustion efficiency. the combustion efficiency increases for all fuels when the initial
Table 4 Influences of initial SMD on combustion efficiencies and process ir-
reversibilities for fuels with different volatilities „ṁ a inÄ0.1 kgÕs, A s Ä0.02 m2 ,
A d Ä0.04 m2 , PinÄ100 kPa TinÄ600 K, ReinÄ52400, S Ä0.76, SMDi Ä52 ␮ m,
A Õ F Ä60…: „ a … comparison of combustion efficiency and second law efficiency
„%…; and „ b … process irreversibilities „JÕs….
SMD (SMDi ), is increased from 25 ␮m to 50 ␮m. However, with ent operating conditions to throw light on the trade off between
a further increase in the value of SMDi from 50 ␮m to 75 ␮m, the the effectiveness of combustion and the lost work due to thermo-
combustion efficiency suffers a marginal decrease for all the fuels. dynamic irreversibilities in the process of combustion. The major
With an increase in SMDi , the rate of droplet vaporization per observations are as follows:
unit mass is reduced but at the same time, the spray penetration is An increase in fuel volatility increases combustion efficiency
increased. In the lower range of SMDi the spray penetration is only at a higher pressure for any given inlet temperature and swirl.
more important from the viewpoint of combustion efficiency. An The combustion efficiency is uninfluenced by the inlet swirl at its
increase in combustion efficiency takes place so long the penetra- lower pressure, while it decreases with an increases in inlet swirl
tion of the droplets in the gas phase increases, but the droplets when the pressure of inlet air is high. The second law efficiency
remain almost within the primary zone where burning rate of fuel increases with a decrease in fuel volatility and an increase in
is more effective. For still higher penetration, droplets may reach combustor pressure. At a higher combustor pressure, an increase
to a downstream location beyond the primary zone, where burning in inlet air swirl decreases both the combustion efficiency and
is less conducive due to lower temperature because of additional second law efficiency.
air. This results in a reduction in combustion efficiency. When the spray cone angle is increased, the combustion effi-
It is further observed from Table 4共a兲 that the second law efficiency increases drastically for a lower volatile fuel, followed by
ciency decrease with an increase in initial SMD of the fuel spray. a marginal decrease, while for a higher volatile fuel, the influence
The decrease in the value of ␩ II with the SMDi is attributed to an of spray cone angle on combustion efficiency is relatively mar-
increase in total process irreversibility İ, while the flow availabil- ginal. The second law efficiency decreases with an increase in
ity at the inlet remains the same, since the incoming species 共air spray cone angle for all the fuels considered.
and fuel vapor兲 and their thermodynamic states remain the same. The combustion efficiency increases with an increase in initial
It is interesting to observe from Table 4共b兲 that for a higher SMD of fuel spray up to a value of 50 ␮m, but a further increase
volatile fuel 共n hexane兲, an increase in SMDi increases the Irre- in initial SMD causes a decrease in combustion efficiency. The
versibility İ d due to inter-phase transport processes but reduces second law efficiency shows a monotonic decreasing trend with
irreversibility İ g due to gas phase transport processes; but it is just initial SMD.
the opposite in case of fuels with lower volatilities. The increase The optimum design of the combustor is based on a trade off
in the value of İ d with the SMDi for a higher volatile fuel, can be between ␩ c and ␩ II for an overall energy economy. The relative
physically attributed to the enhanced rates of inter-phase transport weightage of ␩ c and ␩ II to be assigned in the process of optimi-
processes and vaporization of coarser droplets because of their zation depends on the relative saving in the cost of energy quan-
better penetration and mixing in the gas phase. While, on the other tity over the energy quality for a specific application. This is a task
hand, the enhanced inter-phase transports and better mixing of of energy management in practice where the information provided
fuel vapor in the ambience makes the gas phase more homoge- by the present paper will serve as the fundamental inputs.
neous and brings about a slight decrease in the value of İ g due to
gas phase transport processes. In case of lower volatile fuels, Nomenclature
larger penetration of fuel droplets, beyond the primary zone with
an increase in SMDi reduces the rate of inter-phase transport pro- Ȧ ⫽ rate of flow availability
cesses and hence in the value of İ d . An increase in the value of İ g As ⫽ slot area of secondary air inlet
takes place accordingly. Ad ⫽ slot area of dilution air inlet
a ⫽ specific flow availability
b ⫽ size parameter of Rosin-Rammler function
Conclusion cp ⫽ specific heat at constant pressure
A comparative study on the variations in combustion efficiency Cfs ⫽ mean fuel vapor mass fraction at the droplet surface
and second law efficiency along with the various components of Cj ⫽ mean mass fraction of j th species
exergy losses of a spray combustion process in a gas turbine com- d ⫽ droplet diameter
bustor has been made for fuels with different volatilities at differ- di ⫽ droplet diameter of i th class of droplet
d min i ⫽ minimum diameter of i th droplet class Superscripts
d max i ⫽ maximum diameter of i th droplet class d ⫽ droplet phase
D ⫽ combustor diameter f ⫽ fuel
D eff ⫽ effective mass diffusivity g ⫽ gas phase
Ė ⫽ entropy generation rate
h ⫽ enthalpy of formation References
h̄ ⫽ molar enthalpy 关1兴 Faeth, G. M., 1983, ‘‘Evaporation and Combustion of Sprays,’’ Prog. Energy
⌬h f ⫽ enthalpy of formation Combust. Sci., 9, pp. 1–76.
⌬h̄ f ⫽ molar enthalpy of formation 关2兴 Faeth, G. M., 1987, ‘‘Mixing, Transport and Combustion in Sprays,’’ Prog.
Energy Combust. Sci., 13, pp. 293.
İ ⫽ rate of Irreversibilities 关3兴 Sirignano, W. A., 1983, ‘‘Fuel Droplet Vaporization and Spray Combustion
I b (T) ⫽ black body radiation intensity at a temperature, T Theory,’’ Prog. Energy Combust. Sci., 9, pp. 291.
M ⫽ molecular weight 关4兴 Sirignano, W. A., 1986, ‘‘The Formulation of Spray Combustion Models:
m ⫽ mass Resolutions Compared to Droplet Spacing,’’ ASME J. Heat Transfer , 108, pp.
633.
ṁ f t ⫽ rate of fuel injection 关5兴 Sirignano, W. A., 1988, ‘‘An Integrated Approach to Spray Combustion Model
Nu ⫽ Nusselt number Development,’’ Combust. Sci. Technol., 58, pp. 231.
p ⫽ pressure 关6兴 Dash, S. K., Sengupta, S. P., and Som, S. K., 1991, ‘‘Transport Processes and
Associated Irreversibilities in Droplet Evaporation,’’ Am. Inst. Aeronaut. As-
Pr ⫽ Prandtl number tronaut. J. Thermo Physics and heat Transfer, 5共3兲, pp. 366 –371.
q ri ⫽ radiative heat flux in x i direction 关7兴 Dash, S. K., and Som, S. K., 1991, ‘‘Transport Processes and Associated
R ⫽ radius of the combustor Irreversibilities in Droplet Combustion in a Convective Medium,’’ Int. J. En-
ergy Res., 15, pp. 603.
Re ⫽ Reynolds number 关8兴 Puri, I. K., 1992, ‘‘Second Law Analysis of Convective Droplet Burning,’’ Int.
r ⫽ radial location J. Heat Mass Transf., 35, pp. 2571.
s ⫽ entropy 关9兴 Hiwase, S. D., Datta, A., and Som, S. K., 1998, ‘‘Entropy Balance and Exergy
s̄ ⫽ molar entropy Analysis in the Process of Droplet Combustion,’’ J. Phys. D, 31, pp. 1601.
关10兴 Dash, S. K., and Som, S. K., 1993, ‘‘Thermodynamics of Spray Evaporation,’’
S ⫽ Swirl number J. Phys. D, 26, pp. 574.
Ś ⫽ gas phase mass source term due to droplets 关11兴 Dunbar, W. R., and Lior, N., 1994, ‘‘Sources of Combustion Irreversibility,’’
Sc ⫽ Schmidt Number Combust. Sci. Technol., 103, pp. 41.
Sh ⫽ Sherwood Number 关12兴 Westbrook, C. K., and Dryer, F. L., 1981, ‘‘Simplified Reaction Mechanisms
for the Oxidation of Hydrocarbon Fuels in Flames,’’ Combust. Sci. Technol.,
Ś C ⫽ gas phase species conservation equation source term 27, pp. 31– 45.
Ś E ⫽ gas phase energy conservation equation source term 关13兴 Magnussen, B. F., Hjertager, B. H., 1977, ‘‘On Mathematical Modeling of
due to evaporation and combustion of liquid droplets. Turbulent Combustion With Special Emphasis on Soot Formation and Com-
bustion,’’ Sixteenth Symposium (International) on Combustion, The Combus-
Ś M ⫽ gas phase momentum source term due to momentum tion Institute, pp. 719–727.
exchange with the droplets 关14兴 Ozisik, M. N., 1973, Radiative Heat Transfer, Wiley, New York.
SMD ⫽ Sauter Mean Diameter 关15兴 Sparrow, E. M., Cess, R. D., 1978, Radiation Heat Transfer, Hemisphere,
Washington DC.
t ⫽ time 关16兴 Clift, R., Grace, J. R., and Weber, M. E., 1978, Bubbles, Drops and Particles,
T ⫽ mean temperature Academic Press, New York.
Ui ⫽ mean velocity in x i direction 关17兴 Ranz, W. E., and Marshall, W. R., Jr., 1952, ‘‘Evaporation From Drops: Part
Uz ⫽ mean axial velocity II,’’ Chem. Eng. Prog., 48, pp. 173–180.
关18兴 Mugele, R. A., and Evans, H. D., 1951, ‘‘Droplet Size Distribution in Sprays,’’
x ⫽ mole fraction Ind. Eng. Chem., 43, 1317–1324.
Vi ⫽ velocity in x i direction 关19兴 Hirt, C. W., and Cook, J. L., 1972, ‘‘Calculating Three Dimensional Flows
y ⫽ molar concentration of species Around Structures and over Rough Terrain,’’ J. Comput. Phys., 10, pp. 324 –
z ⫽ axial location 341.
关20兴 Harlow, F. H., and Welch, J. E., 1965, ‘‘Numerical Computation of Time
Greek Letters Dependent Viscous Incompressible Flow of Fluid with a Free Surface,’’ Phys.
Fluids, 8共12兲, pp. 2182–2190.
␣ eff ⫽ effective thermal diffusivity 关21兴 Cameron, C. D., Brouwer, J., Wood, C. P., and Samuelson, G. S., 1989, ‘‘A
␮ eff ⫽ effective viscosity Detailed Characterization of Velocity and Thermal Fields in a Model Can
␮ t ⫽ eddy viscosity Combustor With Wall Jet Injection,’’ ASME J. Eng. Gas Turbines Power, 3, pp.
␳ ⫽ gas phase density 31– 40.
关22兴 Chang, K. C., and Chen, C. S., 1993, ‘‘Development of a Hybrid k-␧ Turbu-
␺ ⫽ spray cone angle lence Model for Swirling Recirculating Flows Under Moderate to Strong Swirl
␭ ⫽ stoichiometric coefficient Intensities,’’ Int. J. Numer. Methods Fluids, 16, pp. 421.
关23兴 Dallenback, P. A., 1986, ‘‘Heat Transfer and Velocity Measurements in Turbu-
Subscripts lent Swirling Flow Through an Abrupt Axi-symmetric Expansion,’’ Ph.D the-
sis, Arizona State University, Tempe, A2.
i ⫽ initial/index of tensor notation 关24兴 Khalil, K. H., El Mahallawy, F. M., and Moneib, H. A., 1977, ‘‘Effect of
j ⫽ index of tensor notation Combustion Air Swirl on the Flow Pattern in a Cylindrical Oil Fired Furnace,’’
in ⫽ inlet Sixteenth Symposium (International) on Combustion, The Combustion Insti-
e ⫽ exit tute., pp. 135–141.
关25兴 Moran, M. J., and Shapiro, H. N., 1988, Fundamentals of Engineering Ther-
f, F ⫽ fuel modynamics, John Wiley, New York.
P ⫽ product 关26兴 Hirschfelder, J. C., Curtiss, C. F., and Bird, R. B., 1954, Molecular Theory of
r ⫽ reference quantitity Gases and Liquids, John Wiley, New York.
Hong-Shun Li Reduction of False Scattering of
Gilles Flamant
e-mail: gilles.flamant@imp.cnrs.fr
the Discrete Ordinates Method
A novel method, the Double Rays Method (DRM), is proposed to capture the discontinu-
Institut de Science et de Génie des Matériaux
ous nature of the radiation intensity, in order to reduce false scattering of the discrete
et Procédés,
ordinates method (DOM). Numerical tests demonstrate that the DRM successfully re-
IMP-CNRS, BP 5-Odeillo,
moves false scattering in all the two-dimensional test problems discussed in this paper.
66125 Font-Romeau Cédex,France
The effect of false scattering on the computational results in two-dimensional situations is
investigated with the DRM. False scattering plays a double role: when the boundary emits
radiation in a limited number of directions, or when the irradiation comes from a limited
Ji-Dong Lu number of directions, it produces a smeared intensity field and radiative heat flux distri-
National Laboratory of Coal Combustion,
bution, and thus must be removed. In the case of diffuse boundary, however, false scat-
Huazhong University of Science & Technology,
tering plays a useful role and thus should be retained. 关DOI: 10.1115/1.1495518兴
Wuhan 430074, P.R. China
Keywords: Heat Transfer, Modeling, Numerical Methods, Radiation, Scattering
1 Introduction field. Several scholars employed the Fredholm integral form of the
equation of transfer 关22,23兴 for this purpose. Unfortunately, as
In the last two decades, the discrete ordinates method 共DOM兲
analyzed in Refs. 关10兴 and 关21兴, the discontinuity cannot be com-
has become more and more popular in radiative heat transfer com-
pletely eliminated by this equation. Since its solution often re-
putations because it has several desirable features: easy coding,
quires an evaluation of a singular kernel, hence, false scattering
computational economy, good accuracy, flexible marching proce-
may occur. The purpose of this section is just to define a method
dures, etc. Up to now, the DOM has been successfully applied to
that can effectively handle discontinuous boundary conditions. In
various radiation problems, such as radiation transfer in Cartesian,
addition, as concerns the spatial integration of the RTE, false scat-
cylindrical, and spherical coordinates, curvilinear system, com-
tering cannot be eliminated by the cell-averaged schemes 共namely,
plex or irregular geometry, combined heat transfer, and non-gray
the diamond and step schemes, etc.兲关18,24兴, and thus other meth-
problems, to name but a few. The efficiency of the DOM is con-
ods must be found. In this work, the discrete ordinate interpolation
stantly improved; new methods for selecting the discrete ordinate
method 共DOIM兲关9兴 is used, even though the DOIM still suffers
quadratures 关1–7兴 and for solving radiation transfer equations
from false scattering due to its unsuitable treatment of the discon-
关8 –15兴 have been proposed. Another kind of DOM, the finite-
tinuous intensity field.
volume method 关16,17兴, was also proposed and has found wide
applications.
The DOM, however, still has two shortcomings: ray effect and 2.1 Basic Idea of the DRM: Improved Treatment of Dis-
false scattering 关18兴. To eliminate ray effect, Ramankutty and continuous Boundary Conditions. For ease of presentation, let
Crosbie 关19兴 used a modified differential approximation, in which us begin the analysis from the example problem 2 in Ref. 关18兴:
the radiation intensity is divided into two parts. For the same consider a square enclosure with four black walls and a non-
purpose, more recently, Baek et al. 关20兴 used a combined Monte- participating medium. The bottom half of the left wall is hot, and
Carlo and Finite-Volume Method. However, as discussed by its top half and the other three walls are cold. The objective of this
Thynell 关21兴, false scattering is much more difficult to handle, and test is to compute the radiation distribution for the radiation com-
additional approaches for eliminating false scattering should be ponent, which is at an angle ␪⫽60 deg to the grid lines, and
developed. Pessoa-Filho and Thynell 关10兴 proposed an approxi- parallel to the x-y plane, as shown in Fig. 1. The shadowing
mate technique to eliminate false scattering, but this method is region is the region with an intensity of unity, and the intensity is
different from the standard DOM. The differencing scheme in zero in the rest of the enclosure. Let us consider the computation
Ref. 关8兴 and the hybrid scheme in Ref. 关15兴 can also partially of the intensity at point D, I D . As shown in Fig. 1, this beam
reduce false scattering. To our knowledge, since Chai et al. 关18兴 comes from point B, which is located between two nodal points A
first presented a detailed analysis on ray effect and false scattering and C. According to the DOIM 关9兴, to compute I D it is necessary
associated with the DOM, there has been no report of successful to determine the intensity at its upstream point B, I B . This inten-
removal of such false scattering. The purpose of this study is to sity can be obtained with linear interpolation between nodal points
propose a new method, the Double Rays Method 共DRM兲, to cap- A and C 关9兴:
ture the discontinuous nature of the radiation intensity field, in
order to remove or minimize false scattering. Then the DRM is x B⫺x A
I B⫽I A⫹ 共 I C⫺I A兲 (1)
used to investigate how false scattering influences the computa- x C⫺x A
tional results.
The intensity at point C, I C , is already known (I C⫽0), the re-
maining problem is to determine I A . If the real boundary condi-
2 Reduction of False Scattering tion is I A⫽0, then, I B⫽0. Since the medium is transparent, then
As analyzed by Chai et al. 关18兴, false scattering occurs once the I D⫽I B ⫽0. This is identical with the physically real distribution
spatial discretization is performed over domains where the inten- 共as in Fig. 1, I D is equal to zero兲. However, if the real boundary
sity is discontinuous. Thus, to eliminate false scattering, one condition is I A⫽1, and Eq. 共1兲 is still used to calculate I B , then,
should consider the discontinuous nature of the radiation intensity obviously, I B is not equal to zero, and thus I D is not equal to zero
either. As a result, false scattering occurs.
In a similar test problem in Ref. 关9兴, Cheong and Song assumed
HEAT TRANSFER. Manuscript received by the Heat Transfer Division October 6, that I A⫽0.5 共i.e., average value of unity and zero兲, but this still
2000; revision received May 9, 2002. Associate Editor: A. Majumdar. resulted in false scattering 关9兴. Note that 0.5 is correct: if a min-
Fig. 1 Test problem and the method for interpolation
iature radiometer is put at a position close to point A and on the

direct line A-E 共actually, it is a plane兲 to measure the intensity at Fig. 2 The double rays method
point A, within the ‘‘pencil’’ solid angle that the radiometer re-
ceives, half is filled with the radiation with the intensity of unity,
and the other half is zero; as a result, the ‘‘measured value’’ is 0.5.
However, keep in mind that Eq. 共1兲 can give a prediction of sec- but note that the intensity field has a step change at y⫽0 and y
ond order precision only in a continuously changing intensity ⫽3⌬y. Namely, the intensity is zero at y⫽0, and unity at y
field. If the variation of the intensity is non-continuous, Eq. 共1兲 ⫽0 ⫹ ; it is unity at y⫽3⌬y, and zero at y⫽3⌬y⫹0. Thus, when
may produce serious error. If I A⫽1, the intensity field has a step the diamond spatial differencing scheme is used to extrapolate the
change at point A: at point 共0,0兲, the intensity is equal to unity, but boundary intensity at y⫽⌬y, then the intensity at y⫽0 ⫹ , instead
at point (0 ⫹ ,0) 共here ‘‘0 ⫹ ’’ means a positive infinitesimal, as in of at y⫽0, should be employed. Similarly, the same treatment is
the following兲, it becomes zero. Thus, if Eq. 共1兲 is still used to applied to all nodes at y⫽3⌬y to extrapolate the boundary inten-
calculate I B , it will produce an incorrect result; consequently, sity at y⫽4⌬y. In this way, the final result obtained with the
false scattering occurs. diamond scheme is also identical with the physically real solution:
Based on the above analyses, the basic idea of the DRM is to no overshoot, no negative intensity 共since it is very easy to under-
conduct the interpolation between point (0 ⫹ ,0) and point C, since stand the above analysis, this result is not shown here for brevity,
the intensity is continuous in this region, namely the reader may refer to Fig. 4共a兲 in Ref. 关18兴 for details兲.
x B⫺x A
I B⫽I Aright⫹ 共 I C⫺I Aright兲 (2) 2.2 General Situations: The Double Rays Method „DRM….
x C⫺x A In the above two example problems, there is only one discrete
where I Aright is the intensity at point (x A ⫹0,y A ), i.e., at point direction, and the medium is non-participating. A general situation
(0 ⫹ ,0). In this way, no matter how much the intensity at point in a two-dimensional rectangular enclosure 共participating me-
(0,0) is, I B is always equal to zero; thus I D⫽I B⫽0. This is iden- dium, more discrete directions than one兲 is shown in Fig. 2. For
tical with the physically real solution. Similarly, to calculate I H , it illustrative purposes, we assume there is only one discontinuous
is necessary to determine I F . The interpolation is conducted be- boundary point, A, namely, the boundary condition has a step
tween point G and point (x E⫺0,y E) 共here ‘‘⫹/⫺0’’ means to add/ change at point A. The straight lines, A-1, A-2, and A-3, are the
subtract an infinitesimal, as in the following兲, instead of between projection on x-y plane of the discrete directions in the first octant
points G and D, because the intensity is non-continuous in the emitting from point A 共only three directions are shown for illus-
region between points G and D. Namely trative purpose; note that the discrete directions in the fifth octant
coincide with the above straight lines because of the symmetry in
x F⫺x G z-direction兲. Each straight-line represents two rays 共the present
I F⫽I G⫹ 共 I Eleft⫺I G兲 (3) method is named the Double Rays Method for this reason兲, i.e.,
x E⫺x G
one ray starts from point (x A⫺0,y A), while the other one starts
where I Eleft is the intensity at point (x E⫺0,y E), as in Fig. 1. It is from point (x A⫹0,y A). For example, the straight-line A-1 repre-
equal to unity, since the medium is non-participating. Thus I F sents the following two rays 共projection on the x-y plane兲: the first
⫽1, and I H⫽I F⫽1, also the physically real solutions. Following one starts from point (x A⫺0,y A), passes points (x M⫺0,y M), (x R
the above method, one node by one node, all the intensities can be ⫺0,y R), (x V⫺0,y V) in sequence, and reaches point (x 1 ⫺0,y 1 ).
calculated. The final computational result is identical with the For ease of reference, this ray is hereafter labeled Ray A-1a.
physically real distribution, without any false scattering occurring. Meanwhile, the second ray starts from (x A⫹0,y A), passes points
Now the above principle is applied to the example problem 1 in (x M⫹0,y M), (x R⫹0,y R), (x V⫹0,y V) in sequence, and reaches
Ref. 关18兴. The situation is the same as the example problem in Fig. point (x 1 ⫹0,y 1 ). Hereafter, this ray is labeled Ray A-1b. The
1, but the radiation direction is normal to the left wall. In Fig. 4 of purpose of the double rays is to capture the discontinuous nature
Ref. 关18兴, it is shown that the diamond scheme produces over- of the radiation intensity field, which is due to the discontinuous
shoot and negative intensities. Here, the diamond scheme is used, boundary at point A and to the discretization of the angular vari-
able of the DOM. Note that the discrete direction in the second Table 1 Method for the interpolation in Fig. 2
octant emitting from point A are not shown in Fig. 2 for clarity,
but they are treated in the same way.
How does the DRM capture the discontinuous nature of the
intensity field? Let us take the above Rays A-1a and A-1b as an
example. They start from points (x A⫺0,y A) and (x A⫹0,y A), re-
spectively. Since point A is a discontinuous boundary point, then
I 1Rleft⫽I 1Rright (I 1Aleft and I 1Aright are the intensities in A-1 direc-
tion at point (x A⫺0,y A) and (x A⫺0,y A), respectively; the rest
may be inferred by analogy兲. Thus, at any position on Ray A-1a,
say at point (x R⫺0,y R), the intensity in A-1 direction is not equal
to the corresponding intensity at its ‘‘twin’’ point (x R⫹0,y R), i.e.,
I 1Rleft⫽I 1Rright . By this way is recorded the step change of the
radiation intensity across the straight A-1. Meanwhile, the region
in which an interpolation is conducted is always limited at one
side of the straight line A-1; for example, the Eq. 共5兲 in the fol-
lowing. In this way, the discontinuous nature of the radiation in-
tensity in the discrete direction A-1 that is due to the discontinu-
ous boundary point A is respected.
Attention is now focused on the procedure for solving the ra-
diation transfer equations. To solve the discrete transfer equations,
an iterative procedure must be used. As aforementioned, in the
present work, the DOIM 关9兴 is still used to solve the equations.
Thus, all the procedures, such as the marching direction of the
iteration, and so on, are the same as the DOIM 关9兴 except for two
differences. The first one is that the regions of linear interpolation
may be different from the DOIM: in the DOIM the region of
interpolation is always between two nodal points, whereas in the
present work, the region of interpolation is subject to the DRM.
The second difference is that in the DOIM, the discontinuous
nature of the radiation across the straight lines A-1, A-2, and A-3
is not considered, whereas in the present work it is considered. nately, since the medium is participating, this intensity is un-
Moreover, the intensities on all the double rays, such as Rays A-1a known, and thus it is necessary to determine it with iteration. This
and A-1b mentioned above, will be computed with iteration. For leads to the aforementioned second difference between the DOIM
the sake of brevity, in the following analyses, only these two 关9兴 and the present work. Concerning this difference, the most
differences are analyzed in details; the reader may refer to 关9兴 for important question is how to compute the intensities on the double
details about the DOIM. rays the straight lines A-1, A-2, and A-3 stand for.
Consider the first difference mentioned above. For example, to The general iterative procedures are the following 共take the two
use the DOIM to predict the intensity in A-1 direction at point T in rays the straight line A-1 represents as the example兲. First, to
Fig. 2, I 1T , it is necessary to calculate the intensity at its upstream predict the intensities at the points on the aforementioned Ray
point Q, I 1Q , with the interpolation. According to the DOIM 关9兴, A-1a, the iterative direction is along its propagating direction:
the linear interpolation is conducted between nodal points P and S from point A to point 1. For example, to predict the intensity in
the A-1 direction at point (x R⫺0,y R), one must compute the in-
x Q⫺x P
I 1Q⫽I 1P⫹ 共 I 1S⫺I 1P兲 (4) tensities for all the discrete directions at point (x R⫺0,y R) and its
x S⫺x P upstream point (x M⫺0,y M). At the upstream point, the intensity
in the A-1 direction has already been predicted; consequently, the
Then, I 1T is determined with the exponential scheme 关9兴. How- interpolation is unnecessary. The radiation intensity in A-2 direc-
ever, since the intensity is discontinuous in the region between tion experiences a step change at point L, thus the linear interpo-
nodal points P and S 共the radiation intensity distribution experi- lation is conducted between points (x L⫺0,y L) and N, as listed in
ences a step change at point R兲, the above interpolation will result Table 1, namely:
in false scattering. Thus in the DRM, the region of interpolation is
changed to be between points (x R⫺0,y R) and P, as listed in x M⫺x N
Table 1, I 2M⫽I 2Mright⫽I 2Mleft⫽I 2N⫹ 共 I 2Lleft⫺I 2N兲 (7)
x L⫺x N
x Q⫺x P
I 1Q⫽I 1P⫹ 共 I 1Rleft⫺I 1P兲 (5) However, the radiation intensity in A-3 direction is continuously
x R⫺x P changing between nodal points N and J, thus the linear interpola-
tion is conducted between them
where I 1Rleft is the intensity in A-1 direction at point (x R
⫺0,y R). However, note that if there is not a discontinuous change x M⫺x N
in radiation intensity between the two nodal points, then the inter- I 3M⫽I 3Mright⫽I 3Mleft⫽I 3N⫹ 共 I J⫺I 3N兲 (8)
polation is still conducted between them. For example x J⫺x N
The intensities in the discrete directions in the second, third, and
y K⫺y P fourth octants at point (x M⫺0,y M) are obtained with the linear
I 3K⫽I 3P⫹ 共 I 3U⫺I 3P兲 (6)
y U⫺y P interpolation similar to Eq. 共8兲.
Similarly, the intensities in all other discrete directions at point
Since the intensity in A-3 direction is continuously changing be- (x R⫺0,y R) can also be obtained with linear interpolation between
tween points P and U 共as in Fig. 2, the straight line A-3 does not points P and S. Then the intensity in A-1 direction at point (x R
pass through this region兲. ⫺0,y R) can be predicted with the exponential scheme
In the above Eq. 共5兲, obviously, before I 1Q is predicted, one
must know the intensity at points (x R⫺0,y R), I 1Rleft . Unfortu- I 1R⫽I 1Mexp共 ⫺ ␤ av⌬s 兲 ⫹S av / ␤ av关 1⫺exp共 ⫺ ␤ av⌬s 兲兴 (9)
⌬s⫽ 共 y R⫺y M兲 / ␩ 1 (10) Table 2 Discrete ordinates for LSO S 4 -approximation
where ␤ av and S av are the average values of the corresponding

parameters at points M and R, and the latter is determined using
Eq. 共11兲 introduced in the following. In this way, one points by
one point one can predict the intensity in the A-1 direction at all
the points on the Ray A-1a.
Then, the same procedures are repeated to predict the intensities
in A-1 direction at all the points on Ray A-1b. But note that each
point on Ray A-1b shares the same intensities in all the discrete
directions except for A-1 direction with its ‘‘twin’’ point on Ray
A-1a, and they also share the same incident radiation and source the y-direction dimensionless heat flux distribution on the bottom
function. For example, the point (x R⫹0,y R) on Ray A-1b and its wall. In the following analyses, only LSO S 2 - and
‘‘twin’’ point (x R⫺0,y R) on Ray A-1a share the same incident S 4 -approximations are employed, because it is easy to know the
radiation and source function for A-1 direction: numerically exact solutions of the discrete ordinate equations of
these two quadrature sets 共needless to day, these exact solutions
兺
are free of false scattering兲. For S 2 set, the numerically exact
G⫽2w 1 共 I 1Rleft⫹I 1Rright兲 /2⫹ 2w i I 1Rleft (11) solution is that the heat flux is zero everywhere except for the
i⫽2,n
region between x⫽2 to x⫽3 where it is equal to 0.5; this is due
冋
S 1 ⫽ ␬ I bR⫹ 共 ␴ s /4␲ 兲 2⌽ 1 w 1 共 I 1Rleft⫹I 1Rright兲 /2
to the discrete direction of ␰ ⫽ ␩ ⫽⫺0.5 关18兴. To facilitate the
presentation, the direction cosines of LSO S 4 are listed in Table 2.
The numerically exact solution of the S 4 set will also be equal to
⫹ 兺
i⫽2,n
2⌽ i w i I 1Rleft 册 (12)
zero everywhere except for three regions, i.e., x⫽(2,3), 共4.0277,
5.0277兲, 共5.9773, 6兲, in which the heat flux is equal to 0.09863,
0.30275, and 0.30275, respectively. This is due to the three dis-
Note that, for the situation in Fig. 2. crete directions coming from the hot section, namely, the first and
second directions in the third octant 共see Table 2兲, and the second
I iRleft⫽I iRright 共 i⫽2,n 兲 (13) direction in the fourth octant 共␰⫽0.2958759, ␩⫽⫺0.9082483兲.
Figure 3 illustrates the heat flux distributions on the bottom
where w i is the weight of the discrete direction A-i, and n is the wall obtained with the DRM. To facilitate the comparison, solu-
total number of the discrete directions in the first to fourth octants tions obtained with the step scheme by a 26⫻26 points grid are
共note the symmetry of two-dimensional problems兲. It is easy to also illustrated in the figure. As shown in Fig. 3, the solutions
understand the above Eqs. 共11兲–共13兲 by recalling the discussion in obtained with the DRM using S 2 and S 4 -approximations are iden-
subsection 2.1. tical with the numerically exact solutions of the discrete ordinate
Then the same procedures are repeated to predict the intensities equations in the above paragraph 共the numerically exact solution
on all other double rays emitting from point A 共in all other discrete for S 2 -approximation is not shown in the figure for clarity兲. On
directions in the first and second octants兲. After all these intensi- the contrary, the solution using the step scheme exhibits a con-
ties on the double rays are predicted, the above interpolating tinuous heat flux distribution due to false scattering. However, the
method 共some typical regions of interpolation are listed in Table physically exact solution is a continuous distribution without any
1兲 is used to predict the nodal intensities in all the discrete direc- bump. Thus, from this standpoint, false scattering is helpful to
tions, and the iterative procedures are the same as the DOIM 关9兴. remedying ray effect; of course, this does not mean that ray effect
Then the above procedures are repeated again and again until the is eliminated, as pointed out in Ref. 关18兴. In fact, obviously, the
iteration precision is satisfied. Finally, the discontinuous changes physically exact solution should exhibit the maximum heat flux at
in heat flux at points 1, 2, and 3 can also be computed, whereas x⫽5.5 共not shown in Fig. 3 for clarity兲. Yet, the above two solu-
the DOIM 关9兴 cannot provide these data. tions with S 2 and S 4 -approximations and the step scheme do not
2.3 Accuracy of the DRM. Since the DRM uses a linear
interpolation, and the linear DOM is also a scheme of second
order precision 关9兴, thus the numerical precision of the DRM is of
second order too.
3 Results
Two cases are solved and compared with physically exact so-
lutions: participating medium and non-participating medium, in
order to test the DRM. Before beginning the discussion, note the
difference between ‘‘NUMERICALLY exact solution of the discrete
ordinate equations’’ and ‘‘PHYSICALLY exact solution.’’ The former
is the exact solution satisfying the discrete ordinate equations,
while the latter is the real solution of the physical problem. Ob-
viously, the former may not be equal to the latter, since the dis-
crete ordinate equations are merely the discrete representation of
the continuous directional variation of the radiation intensity.
3.1 Non-Participating Medium and Diffuse Boundaries

„Test Problem 3…. Here, the example problem 3 in Ref. 关18兴 is
chosen, since it is a very illustrative one. Consider a two-
Fig. 3 Dimensionless heat flux distribution on the bottom wall
dimensional rectangular enclosure with three black walls and non- for the example problem 3 in Ref. †18‡, obtained with — — —
participating medium, the top and bottom walls’ length is six, and LSO S 2 and the step scheme, — LSO S 4 and the step scheme,
the two sidewalls’ length is three. A section of the top wall is hot — – LSO S 2 and the DRM, – – LSO S 4 and the DRM, 䊏 numeri-
共from x⫽5 to x⫽6), while the rest of the top wall and the other cally exact solution of the discrete ordinate equations of LSO
three walls are cold. The objective of this problem is to compute S4
Table 3 Dimensionless heat flux on the top wall for test prob-
lem 4 obtained with the DRM
predict this result; in fact the solution obtained with the

S 4 -approximation and the step scheme displays a dip at about x
⫽5.5.
If one uses the example problem 4 in Ref. 关18兴 to further test
the DRM, the results are also identical with the numerically exact
solutions of the discrete ordinate equations. These results prove
that, in the case of non-participating medium, the DRM can thor-
oughly remove false scattering.
3.2 Participating Medium and Diffuse Boundary „Test Fig. 4 Dimensionless heat flux distribution on the top wall for
Problem 4…. Consider a purely scattering uniform medium in a test problem 4: „a… using LSO S 4 approximation and 101Ã101
two-dimensional black square enclosure, the bottom wall is hot grid points; and „b… using LSO S 8 approximation and 101Ã101
while the other three are cold, the lengths of the walls are one, and grid points. obtained with the DRM, — obtained with the step
the scattering coefficient of the medium is unity; hereafter, this scheme, 䊉 physically exact solution †22‡
problem is referred to as test problem 4. This particular case is
selected because the physically exact solution is available in the
literature, for example, Ref. 关22兴. It has also been used as a bench- A-2, whose direction cosines ( ␰ 2 , ␩ 2 , ␮ 2 ) are equal to
mark by many researchers 关1,2,4,10,19兴. 共0.2958759, 0.9082483, 0.2958759兲, i.e., the second discrete di-
First, the dimensionless heat flux distributions on the bottom rection in Table 2. The sharp drop at x⫽0.67423 is due to radia-
wall predicted with the DRM using LSO S 4 and tion beam B-5, whose direction cosines ( ␰ 2 , ␩ 2 , ␮ 2 ) are equal to
S 8 -approximations and a 101⫻101 points grid are compared with 共0.2958759, 0.9082483, 0.2958759兲. In fact, the intensity can be
the physically exact solution 关22兴, see Table 3. The iterative pre- divided into two parts 关19兴, namely
cision of these results is 1.0⫻10⫺15 共using the double precision of I i ⫽I wi ⫹I mi (14)
FORTRAN code, as for all the following results兲 and the corre-
sponding iterative numbers are 23 and 24 times for LSO S 4 and Here subscript ‘‘i’’ stands for the ith discrete direction, I wi is the
S 8 -approximations respectively. As listed in Table 3, with an indirect intensity component due to the emission and reflection at
crement of the order of S N-approximation, the predicted heat flux the walls, and I mi is the intensity due to the contribution of the
approaches the physically exact solution. The highest relative er- medium by emission of scattering. Consider the radiation beam
ror of the results with S 8 -approximation is 0.78 percent 共at x A-2: it reaches point 2 on the top wall after traveling an optical
⫽0.0 and x⫽1.0, i.e., the two discontinuous boundary points兲. path of 1/␩ 2 共note that the medium is uniform兲. Thus when it
This good agreement confirms the numerical precision of the reaches the top wall, its direct intensity component becomes
DRM.
I w2 ⫽I 0 exp共 ⫺1/␩ 2 兲 (15)
In this test problem, predicting the heat flux distribution on the
top wall is a severe test. Figure 4 shows the dimensionless heat with
flux distribution obtained with the DRM using LSO S 4 and
S 8 -approximations, as well as the results obtained with the step I 0 ⫽ ␴ T 41 / ␲ (16)
scheme. Figure 5 illustrates the corresponding dimensionless heat here T 1 is the temperature of the bottom wall. I w2 contributes to
flux in y-direction along the centerline normal to the bottom wall. the y-direction dimensionless heat flux on the top wall:
In Fig. 4共a兲 the solution using the DRM and S 4 -approximation
exhibits a sharp rise or drop of 0.1006737 at x⫽0.32577 and x q w2 ⫽w 2 ␩ 2 I w2 / 共 ␴ T 41 兲 ⫽ 共 ␲ /3兲 •0.9082483•I 0
⫽0.67423, respectively. Ray tracing analysis reveals that these
•exp共 ⫺1.0/0.9082483兲 / 共 ␴ T 41 兲 ⫽0.1006737 (17)
sharp changes are due to ray effect. As shown in Fig. 6, the
boundary has a step change at points A and B. The sharp rise in As a result, this produces a sharp rise of 0.1006737 in the dimen-
heat flux at x⫽0.32577 on the top wall is due to radiation beam sionless heat flux at point 2 (x 2 ⫽0.32577, y 2 ⫽1) on the top wall.
Fig. 6 Analysis on the ray effects in Figs. 4„a… and 5„a…
S 8 -approximation, just one direction, ␰ ⫽ ␩ ⫽ ␮ ⫽0.5773503, its

weight 共⫽0.4617179兲 makes up 29.4 percent of the total. Needless
to day, this enhances ray effect.
3.3 Participating Medium and Diffuse Boundary „Test

Problem 5…. In the above two subsections, the step scheme and
the DRM are used. In this subsection, the linear DOIM 关9兴 will be
used to further examine the effect of false scattering. For ease of
Fig. 5 Dimensionless heat flux in y-direction along the center- presentation, we first present some explanations about false scat-
line normal to the bottom wall for test problem 4: „a… using LSO
tering associated with these two spatial schemes. It has been
S 4 approximation and 101Ã101 grid points; and „b… using LSO
shown that the DOIM also ‘‘suffers from’’ false scattering 关9兴.
S 8 approximation and 101Ã101 grid points. obtained with the
DRM, — obtained with the step scheme. However, we found that, for two-dimensional problems, the
DOIM does not produce false scattering in the discrete directions
that satisfy ␰ / ␩ ⫽⌬x/⌬y. Here ⌬x and ⌬y are the grid size in x
and y-directions, respectively. But the DOIM produces false scat-
The result from Eq. 共17兲 is surprisingly identical with the afore- tering in all other directions. If ⌬x⫽⌬y, then the DOIM does not
mentioned value predicted by the DRM 共note that in the DRM the produce false scattering in all the directions that satisfy ␾⫽45 deg
intensity is not separated into the aforementioned two compo- 共‘‘selective false scattering’’兲. This point will not be demonstrated
nents兲. This identity is achieved in all other discrete directions,
without exception; no matter what quadrature set is used. This
identity not only demonstrates that false scattering has been com-
pletely eliminated in this test problem, but also indicates the DRM
is numerically precise 共here the relative error for q w2 is zero兲.
In Fig. 5共a兲, the heat flux using S 4 -approximation has a sharp
drop at y⫽0.16 and y⫽0.5. Obviously, these bumps are also due
to the ray effects: as shown in Fig. 6, the bump at y⫽0.16 共i.e., at
point C兲 is due to the beams A-3 and B-6, while the bump at y
⫽0.5 共i.e., at point D in Fig. 6兲 is due to the beams A-1 and B-4.
Meanwhile in Fig. 5共b兲, the heat flux using S 8 -approximation ex-
hibits more bumps; the drop at y⫽0.5 is the strongest one. The
reason for these bumps is that, among the ten discrete directions in
the first octant of the LSO S 8 quadrature set 共this set is available
in many references, for example, Ref. 关3兴兲, two are located on the
plane ␾⫽␲/4. Moreover, the sum of the weights of these two
directions is 0.6329538, which corresponds to 40.3 percent of the
total weight of the first octant 共i.e., ␲/2兲! As a result, this produces
a strong variation at y⫽0.5. Moreover, the top wall cannot ‘‘see’’
such an amount of radiation except at the two corners 共as a con-
sequence, the heat flux distribution in Fig. 4共b兲 turns upwards at
x⫽0 and x⫽1); this is obviously not very reasonable. Further
examination of other quadrature sets reveals that though the stan-
dard discrete ordinate quadrature sets have the advantage of rota-
tional symmetry, too high weights are concentrated on the plane
of ␾⫽␲/4. For example, in the quadrature set of Fig. 7 Test problem 5
Figure 8共a兲 reveals that even if a scheme does not produce false
scattering in only one or two discrete directions, ray effect may be
strong. Thus Fig. 8 proves again that false scattering effectively
reduces unwanted bumps due to ray effect; in other words, it plays
a useful role. Thus it is better to retain false scattering in the case
of diffuse boundary conditions.
3.4 Other Boundary Conditions. Now let us analyze test
problem 4 and the results in Fig. 4共a兲 from an opposite point of
view: assume that the hot bottom wall is not a diffuse surface but,
on the contrary, that it REALLY emits radiation in the same direc-
tions as the discrete directions of the S 4 -approximation 共i.e., 12
directions in the half solid angle range of 2␲兲, and that the radia-
tive heat flux in each direction is the same as the flux in the
corresponding direction of the S 4 -approximation for diffuse sur-
face. Then, in this case, the physically real heat flux distribution
on the top wall will be close to the result obtained with the DRM
in Fig. 4共a兲. In other words, the physically exact solution REALLY
has a bump at x⫽0.32577 and x⫽0.67423. However, the result
obtained with the step scheme does not show these bumps; on the
contrary, the step scheme smears these sharp changes 共see Fig.
4共a兲兲. In this case, false scattering becomes a shortcoming and
must be removed. The results in Figs. 3, 4共b兲, 5, and 7 can also be
analyzed in the same way; all these analyses give the same con-
clusion as above. In fact, this is also the conclusion stated in Ref.
关18兴.
4 Conclusion
From the above studies, the following conclusions are drawn:
1. In all the test problems discussed in this paper 共two-
dimensional兲, with non-participating or participating me-
dium兲, the DRM effectively captured the discontinuous na-
ture of the intensity field that is due to the discontinuous
boundary points and to the discretization of the angular vari-
Fig. 8 Dimensionless heat flux on the right wall for test probable, and false scattering was thoroughly removed using the
lem 5: „a… using the DOIM †9‡ and a 50Ã50 grid; „b… using the
DRM.
step scheme and a 50Ã50 grid „— using LSH S 10 quadrature
set †3‡; - - - - - using LSO S 8 quadrature set; 䊉 exact solution 2. Previously, false scattering was thought to be a ‘‘shortcom-
„using the Monte Carlo Method… ing’’ 关9,10,15,18,21,24兴; however, the above analyses show
that false scattering plays a double role. In the case of dif-
fuse boundary, it plays a useful role, and makes the compu-
tational results close to the physically exact solutions. Thus,
here, since it can be easily verified from the data in Ref. 关9兴. On it will be better to retain it. However, when the boundary
the contrary, the step scheme produces false scattering in all the emits radiation in a limited number of directions, or when
discrete directions 关18,24兴. the irradiation comes from a limited number of directions
Now, consider the test problem 5 shown in Fig. 7. The situation 共e.g., collimated irradiation兲, it produces smeared intensity
is similar to test problem 4, but only a section 0.2 long at the field and radiative heat flux distribution, thus it really be-
center of the bottom wall is hot. This problem was also used as a comes a shortcoming, and it must be removed. Under these
test case in Ref. 关19兴. Figure 8 shows the dimensionless heat flux circumstances the DRM predicts more accurately than other
distributions on the right wall for test problems 5, obtained with methods.
LSO S 8 and LSH S 10 quadrature sets 关3兴 using the linear DOIM
关9兴 and the step scheme, as well as the physically exact solution
obtained with the Monte Carlo Method. When the DOIM scheme
is applied, the intensities at x⫽0.4 and x⫽0.6 are treated as Acknowledgment
Cheong and Song 关9兴 did 共i.e., assumed to be 0.5兲. As shown in The authors are grateful to AFCRST 共Sino-French Association
Fig. 8共a兲, the result with LSO S 8 and the DOIM exhibits a sharp for Scientific and Technological Research兲 for their financial aid.
change at y⫽0.4 and y⫽0.6, respectively. Ray tracing analysis
reveals that this is due to two reasons. The first one is ray effect: Nomenclature
the effect of the two rays located on the plane ␪⫽45 deg 共here, G ⫽ dimensionless incident radiation
⌬x⫽⌬y) and the too high weights associated with them 共see the I ⫽ dimensionless radiation intensity
analyses in subsection 3.2兲. The second one is selective false scat- q ⫽ dimensionless heat flux
tering: although the DOIM produces false scattering in all other S ⫽ dimensionless source function
directions, it does not produce false scattering in the directions w ⫽ weights in quadrature
that satisfy ␾⫽45 deg. On the contrary, since the step scheme ␤ ⫽ extinction coefficient
produces false scattering on all the discrete directions 关18,24兴, it ⌽ ⫽ scattering phase function
greatly reduces the fluctuations in the heat flux 共see Fig. 8共b兲兲 and ␾ ⫽ azimuthal angle
the results are closer to the physically exact solution than the ␴s ⫽ scattering coefficient
results in Fig. 8共a兲. Especially for the results with LSH S 10 set, the ␨,␩,␮ ⫽ direction cosines with respect to x, y, z-axes, respec-
bumps at y⫽0.4 and 0.6 in Fig. 8共a兲 are not observed in Fig. 8共b兲. tively
Subscript Discrete Ordinates in Complex Geometry,’’ Rev. Gen. Therm., 37共3兲, pp. 440–
449.
共capital letters兲 ⫽ at corresponding points in Fig. 1 or Fig. 2 关12兴 Mohamad, A. A., 1998, ‘‘Local Analytical Discrete Ordinate Method for the
b ⫽ black body Solution of the Radiative Transfer Equation,’’ Int. J. Heat Mass Transf., 39共9兲,
pp. 1859–1864.
关13兴 Fiterman, A., and Ben-Zvi, R., 1999, ‘‘Dots: Pseudo-Time-Stepping Solution
References of the Discrete Equations,’’ Numer. Heat Transfer, Part B, 35共1兲, pp. 163–183.
关1兴 Fiveland, W. A., 1987, ‘‘Discrete Ordinate Methods for Radiative Heat Trans- 关14兴 Selcuk, N., and Kirbas, G., 2000, ‘‘The Method of Lines Solution of the
fer in Isotropically and Anisotropically Scattering Media,’’ ASME J. Heat Discrete Ordinates Method for Radiative Heat Transfer in Enclosures,’’ Numer.
Transfer , 109, pp. 809– 812. Heat Transfer, Part B, 37共1兲, pp. 107–125.
关2兴 Truelove, J. S., 1987, ‘‘Discrete-Ordinates Solutions of the Radiative Transport 关15兴 Kim, Il-K., and Kim, W.-D., 2001, ‘‘A Hybrid Spatial Differencing Schemes
Equation,’’ ASME J. Heat Transfer, 109共4兲, pp. 1048 –1051. for Discrete Ordinates Method in 2D Rectangular Enclosures,’’ Int. J. Heat
关3兴 Fiveland, W. A., 1991, ‘‘The Selection of Discrete Ordinate Quadrature Sets Mass Transf., 4共4兲, pp. 575–586.
for Anisotropic Scattering,’’ in Fundamental of Radiation Heat Transfer, 关16兴 Raithby, G. D., and Chui, E. H., 1990, ‘‘A Finite-Volume Method for Predict-
ASME HTD-Vol. 160, ASME, New York, pp. 89–96. ing a Radiative Heat Transfer in Enclosures with Participating Media,’’ ASME
关4兴 El Wakil, N., and Sacadura, J. F., 1992, ‘‘Some Improvements of the Discrete J. Heat Transfer , 112共2兲, pp. 415– 423.
Ordinates Method for the Solution of the Radiative Transport Equation in 关17兴 Chai, J. C., Lee, H. S., and Patankar, S. V., 1994, ‘‘Finite-Volume Method for
Multidimensional Anisotropically Scattering Media,’’ in Developments of Ra- Radiation Heat Transfer,’’ J. Thermophys. Heat Transfer, 8共3兲, pp. 419– 425.
diative Heat Transfer, ASME HTD-Vol. 203, ASME, New York, pp. 119–127. 关18兴 Chai, J. C., Lee, H. S., and Patankar, S. V., 1993, ‘‘Ray Effect and False
关5兴 Thurgood, C. P., Pollard, A., and Becker, H. A., 1995, ‘‘The T N Quadrature Set Scattering in the Discrete Ordinates Method,’’ Numer. Heat Transfer, Part B,
for the Discrete Ordinate Method,’’ ASME J. Heat Transfer, 117共4兲, pp. 1068 – 24共2兲, pp. 373–389.
1070. 关19兴 Ramankutty, M. A., and Crosbie, A. L., 1997, ‘‘Modified Discrete Ordinates
关6兴 Koch, R., Krebs, W., Wittig, S., and Viskanta, R., 1995, ‘‘Discrete Ordinates Solution of Radiative Transfer in Two-Dimensional Rectangular Enclosures,’’
Quadrature Scheme for Multidimensional Radiative Transfer,’’ J. Quant. Spec- J. Quant. Spectrosc. Radiat. Transf., 57, pp. 107–140.
trosc. Radiat. Transf., 53共4兲, pp. 353–372. 关20兴 Baek, S. W., Byun, D. Y., and Kang, S. J., 2000, ‘‘The Combined Monte-Carlo
关7兴 Li, B.-W., Yao, Q., Cao, X.-Y., and Cen, K.-F., 1998, ‘‘A New Discrete Ordi- and Finite-Volume Method for Radiation in a Two-Dimensional Irregular Ge-
nate Quadrature Scheme for Three-Dimensional Radiative Heat Transfer,’’ ometry,’’ Int. J. Heat Mass Transf., 43共13兲, pp. 2337–2344.
ASME J. Heat Transfer , 120共2兲, pp. 514 –518. 关21兴 Thynell, S. T., 1998, ‘‘Discrete-Ordinates Method in Radiative Heat Transfer,’’
关8兴 Liu, F., Becker, H. A., and Polland, A., 1996, ‘‘Spatial Differencing Schemes Int. J. Eng. Sci., 35共12–14兲, pp. 1651–1675.
of the Discrete Ordinates Method,’’ Numer. Heat Transfer, Part B, 30共1兲, pp. 关22兴 Crosbie, A. L., and Schrenker, R. G., 1984, ‘‘Radiative Transfer in a Two-
23– 43. Dimensional Rectangular Medium Exposed to Diffuse Radiation,’’ J. Quant.
关9兴 Cheong, K.-B., and Song, T.-H., 1997, ‘‘An Alternative Discrete Ordinates Spectrosc. Radiat. Transf., 31共4兲, pp. 339–372.
Method With Interpolation and Source Differencing for Two-Dimensional Ra- 关23兴 Thynell, S. T., and Özisik, N. N., 1987, ‘‘Radiation Transfer in Isotropically
diative Transfer Problem,’’ Numer. Heat Transfer, Part B, 32共1兲, pp. 107–125. Scattering, Rectangular Enclosures,’’ J. Thermophys. Heat Transfer, 1共1兲, pp.
关10兴 Pessoa-Filho, J. B., and Thynell, S. T., 1997, ‘‘An Approximate Solution to 69–76.
Radiative Transfer in Two-Dimensional Rectangular Enclosures,’’ ASME J. 关24兴 Chai, J. C., Patankar, S. V., and Lee, H. S., 1994, ‘‘Evaluation of Spatial
Heat Transfer, 119共4兲, pp. 738 –745. Differencing Practices for the Discrete Ordinates Method,’’ J. Thermophys.
关11兴 Sakami, M., and Charrette, A., 1998, ‘‘A New Differencing Scheme for the Heat Transfer, 8共1兲, pp. 140–144.
Investigation of Circumferential
Variation of Heat Transfer
Coefficients During In-Tube
Seok Ho Yoon
Evaporation for R-22 and R-407C
Ph.D. candidate
Min Soo Kim

Using Liquid Crystal
e-mail: minskim@snu.ac.kr
Heat transfer coefficients during evaporation in a horizontal smooth tube for R-22 and
R-407C (R-32/125/134a, 23/25/52 wt.%) have been measured using thermochromic liquid
School of Mechanical and crystal. Focus has been put on the circumferential variation of heat transfer coefficients at
Aerospace Engineering, several cross-sections of the test tube with inner diameter of 11.3 mm for several vapor
Seoul National University, qualities of refrigerant. The inner wall temperatures were calculated by one dimensional
Seoul 151-742, Korea heat conduction equation from the measured outer wall temperatures, which were ob-
tained using an image processing technique with thermochromic liquid crystal (TLC). The
relation between measured temperature and color information (Red-Green-Blue values) of
thermochromic liquid crystal was calibrated by a neural network method. Results show
that circumferential variation of heat transfer coefficients for R-22 is quite large with the
highest heat transfer coefficient at the top of the tube. For zeotropic mixture of R-407C,
similar trend has been observed with less difference between the heat transfer coefficients
at the top and bottom than that of R-22. 关DOI: 10.1115/1.1484110兴
Keywords: Evaporation, Heat Transfer, Image Processing, Refrigeration
1 Introduction suring points using thermocouples, and conduction loss through

the thermocouple wire tends to reduce the accuracy of the mea-
Environmental issues of global warming and ozone depletion
surement. The circumferential heat transfer coefficients differ
by CFC and HCFC refrigerants have driven us to find and develop
around the circumference mainly because of the liquid film thick-
new alternative refrigerants. Extensive studies on alternative re-
ness change due to gravity, flow patterns, and composition shift
frigerants have already been carried out for their thermophysical
for zeotropic refrigerant mixtures. Even though circumferential
properties and their performances in refrigerators, air-
heat transfer variation will not fully explain these phenomena, it is
conditioners, heat pumps, etc. Since two-phase heat transfer is a
quite important to investigate the circumferential heat transfer co-
main mechanism of energy transfer in these systems, investigation
efficient change, and to compare this with the locally averaged
of evaporation and condensation heat transfer has been carried out
heat transfer coefficient from 3 or 4 point measurements.
and enhancement of heat transfer has been constantly pursued
Thermochromic liquid crystals exhibit all colors of the visible
which, in turn, will improve system performance. In designing a
spectrum for specified temperature range. Recently, quantitative
new machine with new alternative refrigerants, it is a prerequisite
measurements using liquid crystals were carried out in a convec-
to get information on two-phase heat transfer for new refrigerants.
tive heat transfer study by Camci et al. 关17兴, and it is found that a
Until now, several experiments and theoretical studies have
very accurate quantitative approach is possible when a linear hue
been done for in-tube evaporation of new alternative refrigerants.
versus temperature relation for a certain range is used. It is note-
Measurements of heat transfer coefficients along test tubes have
worthy that Matsumoto et al. 关18兴 used a neural network method
been the most frequent studies 共Wang et al. 关1兴; Shin et al. 关2兴;
in their calibration to get a relation between RGB signals and
Kim et al. 关3兴; Choi et al. 关4兴; Kattan et al. 关5兴兲, and several cor-
temperatures. In this study, neural network method is also used for
relations based on experimental data have been proposed for the
color-to-temperature transformation, because an accurate calibra-
purpose of heat exchanger design and analysis 共Gungor and Win-
tion curve over a wide temperature range can be obtained by
terton 关6兴; Jung et al. 关7兴; Kandlikar 关8,9兴; Kattan et al. 关10兴兲.
When the circumferential variations of heat transfer coefficients at considering a strong non-linear relation between color images
one cross section of the tube are considered, there are some stud- 共RGB signals兲 and temperatures when compared with the hue-
ies based on the measurements at several positions around the temperature method. The advantage of this technique is the pos-
circumference 共Ross et al. 关11兴; Jung et al. 关12兴; Yoshida et al. sibility to get two dimensional temperature distributions over the
关13兴; Niederkrüger et al. 关14兴; Boyd et al. 关15兴; Shin et al. 关16兴; circumference and for some axial length simultaneously.
Kattan et al. 关5兴; Choi et al. 关4兴兲. In most cases, it is quite usual to The main objective of this study is to investigate a local cir-
measure tube wall temperatures at 3 or 4 locations of test section cumferential variation of heat transfer coefficients during in-tube
共i.e., top, one or two sides, and bottom兲 using thermocouples and evaporation. It includes a comparison of the measured data with
calculate the average value to get a local heat transfer coefficient existing heat transfer correlations. In this study, an imaging tech-
as a representation of heat transfer coefficient at one axial posi- nique using thermochromic liquid crystal 共TLC兲 is used to mea-
tion. However, there always are limitations on the number of measure an almost continuous variation of outer wall temperatures
during in-tube evaporation of pure R-22 and zeotropic refrigerant
mixture of R-407C 共R-32/125/134a, 23/25/52 wt %兲. Circumfer-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division March 27, ential variation of heat transfer coefficients for these two refriger-
2000; revision received April 1, 2002. Associate Editor: T. Y. Chu. ants at several vapor qualities is provided.
2 Wall Temperature Measurement Using Liquid
Crystal
Image Processing System. First of all, wall temperature mea-
surement using liquid crystal is described in this section. Since the
color of liquid crystal varies uniquely as temperature changes, it is
required to get color information of the liquid crystal to be incor-
porated with the temperature of the wall of a test section. As the
liquid crystal film was attached on the outer wall of the test tube
in a very thin layer while keeping the tube temperature near the
ambient air, it is quite acceptable that the temperature of liquid
crystal is very close to that of the wall. Color image processing
technique is used for obtaining color information of liquid crystal,
which means colors of liquid crystal that appear on the outer wall
of the test section are observed by a CCD camera to get red-green-
blue 共RGB兲 information at the spot of interest. The CCD camera
sends analog 共NTSC兲 signals to the frame grabber in a computer,
which converts analog signals to digitized image of 640⫻480
pixels. However, only 5⫻5 pixels around the spot of interest
which is perpendicular to the camera angle have been used for
color to temperature conversion, where each pixel has digitized
RGB values ranging from 0 to 255. For the measurement at other
Fig. 1 Relative magnitude of RGB values as a function of
positions around the circumference, camera has been rotated by temperature
keeping a perpendicular orientation to the tube surface. As an
illumination system, halogen lamp of 100 W with ring-type opti-
cal fiber was used.
ability to learn the relation between given inputs and their corre-
Calibration. To measure a temperature using thermochromic sponding outputs as stated by Kimura et al. 关19兴. The structure of
liquid crystal, the relation between the color of liquid crystal and neural network used in this study is depicted in Fig. 2共b兲. The
the temperature of the test section should be known. A test section weighting coefficients (w k ) and offset 共c兲 between two neurons
for calibration was made of a quartz tube, where thermochromic are optimized to get best results for the normalized calculated
liquid crystal film and copper foil were attached on the inner temperature, ␪ C from the measured temperature, ␪ M , where the
wall of the quartz tube. Five thermocouples were attached on the normalized temperature is shown in Eq. 共3兲.
copper foil inside the quartz tube. Single phase water flows
through the test tube at the maximum rate of about 0.2 kg/s so that T⫺T min
␪⫽ (3)
the temperature change in the test section would be negligible T max⫺T min
while maintaining a designated temperature. Thermochromic
liquid crystal used in this study is a cholesteric-type that has an where T max is the maximum event temperature of 20°C and T min
event temperature range from 10°C 共red兲 to 20°C 共blue兲. The represents the minimum event temperature of 10°C. The opti-
lighting source during the calibration is positioned 25 deg off mized weighting coefficients and offsets are listed in Table 1,
from the CCD camera viewing direction, which is the same as based on normalized RGB values and normalized reference tem-
in the main experiments to minimize the effects of illumination
angle to the test surface on temperature measurement. When
steady state is reached, the image capturing with CCD camera and
the temperature measurement using thermocouples are carried out
simultaneously.
Color to Temperature Transformation. Typical example of
RGB values for liquid crystals in this study versus measured tem-
peratures is presented in Fig. 1. RGB values obtained from the test
as shown in Fig. 1 are then normalized with their mathematical
sum to represent the observer-independent color values as stated
by Camci et al. 关17兴. The normalized RGB values are defined as
in the following Eq. 共1兲.
R G B
r⫽ , g⫽ , b⫽ (1)
R⫹G⫹B R⫹G⫹B R⫹G⫹B
Instead of calibrating RGB values obtained from the captured im-
age to the measure temperature, normalized RGB values (rgb)
are used for the calibration. In this study, neural network method
is introduced to formulate the relation between color information
and temperature. An exemplary neuron in the neural network is
shown in Fig. 2共a兲. With given inputs of x k ’s, the output f (s) of
the neuron is given by Eq. 共2兲;
3
1
f 共 s 兲⫽
1⫹exp共 ⫺u 0 s 兲
, s⫽ 兺 x w ⫹c
k⫽1
k k (2)
where c is an offset for the neuron, w k ’s are weighting coeffi- Fig. 2 „a… An exemplary neuron; and „b… neural network used
cients, and u 0 is a constant, which is given as 0.5. A neuron has an for calibration
Table 1 Optimized weighting coefficients and offsets of the chiller, heat exchanger for subcooling the refrigerant, pre-heater,
neural network in this study mass flow meter 共Oval, D012S-SS-200兲, and heat transfer test
section. In addition to these, CCD camera 共Panasonic, KR-222兲
for capturing color images, illumination system and computer
with frame grabber 共Matrox, Meteor兲 become part of the test
setup. The magnetic gear pump circulates the subcooled liquid
from the liquid receiver to the pre-heater, and the pre-heater is
installed to adjust the inlet quality of the refrigerant to the desired
value. The mass flow meter is installed before the pre-heater to
measure the flow rate of the refrigerant in the liquid phase.
Figure 5 represents the schematic diagram of the heat transfer
test section. Seamless stainless steel tube is used as a heat transfer
test section with inner and outer diameters of 11.3 and 12.7 mm,
respectively. Thermochromic liquid crystal is attached on the
outer wall of the test tube where black paint is painted between
the outer wall and the layer of thermochromic liquid crystal. The
test section is surrounded by quartz tube for insulation, since the
thermal conductivity of quartz tube is about one tenth of that of
stainless steel tube. In addition to this, the temperature difference
between the surroundings of the test section and the outer wall of
the tube has been kept within 2°C. The heat is generated from the
tube since AC voltage has been directly applied to the test tube
and the amount of heat transferred to the refrigerant at the test
section is measured by a powermeter. Pressure taps are drilled at
the inlet and outlet of the test section to measure the pressures of
the refrigerant.
Test Conditions. Mass flux of refrigerant flow and heat flux
to the test section are major factors which affect evaporative heat
transfer of refrigerant. Test section inlet temperature of the refrig-
erant was also varied to investigate the influence of refrigerant
temperature on heat transfer. The experiment in the high vapor
quality region above 0.6 has not been performed to protect ther-
mochromic liquid crystal from a burn-out. Test conditions in this
study are shown in Table 2 for R-22 and R-407C.
Data Reduction. Prior to the measurement, a steady state
should be reached while operating the chiller to condense the
refrigerant that was evaporated in the main test section. During
the test, refrigerant temperatures at the inlet and outlet of the test
section were measured together with the pressures at the same
positions. Mass flow rate of refrigerant and heat flux to the test
section were also measured, and all the measured values were
transmitted to personal computer through multi-channel recorder
using GPIB interface. Data were saved into a file during the
steady state operation when the maximum deviations of tempera-
ture, mass flux, and pressure from the mean values were less than
0.1°C, 3 kg/m2s, and 2 kPa, respectively. The saturation tempera-
Fig. 3 Comparison of the measured temperatures with the cal- tures of refrigerants were calculated from the measured pressure
culated values using a neural network by using a modified Carnahan-Starling-DeSantis equation of state
共Huber et al. 关20兴兲. For mixtures, vapor quality, which was ob-
tained from the energy balance in the preheater section, was ad-
peratures that are used for calibration. Once the optimal values of ditionally used to calculate the local temperature together with the
weighting coefficients and offsets are obtained, neural network pressure at the test section. The difference between the measured
system can calculate a normalized temperature using normalized temperature of refrigerant and the calculated temperature from
RGB values. The calibration results are shown in Fig. 3 and it is pressure measurement was less than 0.15°C. The colors of the
clear that the calculated temperatures agree very well with the liquid crystal appearing at the outer wall of the test section were
measured temperatures. The maximum deviation between the cal- observed by the CCD camera, and when steady state was reached,
culated temperatures and the measured temperatures is 0.3 K and the images were saved for an analysis.
rms of the deviation is about 0.1 K. Heat transfer coefficient is defined as in Eq. 共4兲.
3 Experiment
q⬙
Experimental Apparatus. Experimental apparatus for h⫽ (4)
T wi ⫺T sat
evaporative heat transfer study is designed to get circumferential
distribution of heat transfer coefficients inside a horizontal smooth
tube from the wall temperature measurement using thermochro- where q ⬙ is the heat flux to the test section and T wi is the inner
mic liquid crystal which is attached outside of the test tube. The wall temperature calculated by one dimensional heat conduction
schematic diagram of experimental apparatus is shown in Fig. 4. equation from measured outer wall temperatures which were ob-
The test rig is composed of magnetic gear pump 共Tuthill, tained with thermochromic liquid crystal using an image process-
TXCM26MCNK兲 for refrigerant circulation, liquid receiver, ing technique.
Fig. 4 Schematic diagram of experimental apparatus for measuring circumferential heat transfer coefficient
the heat rate generated in the test tube made of stainless steel
when the electric current was applied to the tube.
For comparison of the measured heat transfer coefficient in the
single phase flow, Dittus-Boelter’s correlation 共Dittus and Boelter
关21兴兲 for single phase heat transfer was used. Absolute average
deviation of 4.1 % was obtained when calculated with Eq. 共6兲 for
100 data points.
兺冏冏
N
1 h C,i ⫺h M ,i
E h⫽ ⫻100共 % 兲 (6)
N i⫽1 h C,i
where h M is the measured single phase heat transfer coefficient,
Fig. 5 Cross sectional view of the test section and h C is the calculated heat transfer coefficient using Dittus-
Boelter’s correlation. In measuring the single phase heat transfer
Table 2 Test conditions in this study for circumferential mea- coefficient, the refrigerant temperature was measured directly by
surement of heat transfer coefficient the thermocouple, and the tube wall temperature was obtained
from the color information using the liquid crystal.
Circumferential Variations of Inner Wall Temperature With

Respect to Quality. Circumferential variations of inner wall
temperatures and heat transfer coefficients at several vapor quali-
ties for R-22 and R-407C were obtained. Figures 6 and 7 show
inner wall temperatures and refrigerant temperatures at several
circumferential locations for R-22 and R-407C. Circumferential
variation of inner wall temperatures together with refrigerant tem-
perature at quality near 0.25 are shown for R-22 and R-407C in
Fig. 6 with mass flux G of 200 kg/m2s and heat flux q ⬙ of 3.6
kW/m2. The inner wall temperature at the top is lower than that at
the bottom, because of the different liquid film thickness caused
by gravity for both cases of R-22 and R-407C. For R-22 as shown
4 Results and Discussion in Fig. 6共a兲, the liquid film at the top is thinner than that at the
Energy Balance Test. Energy balance tests in the test section bottom, which represents that the inner wall temperature at the top
have been done to check the accuracy in heat flux measurement. will be lower if we assume the liquid-vapor interface temperature
The basic idea is to compare the heat given to the refrigerant with is uniform and the heat generation is also uniform along the tube.
the heat gain by the refrigerant, which is R-22 in this case. Only However, for the case of R-407C, the inner wall temperature is
liquid phase flow was used in energy balance test for simplicity higher than that of R-22, which means the wall superheat differ-
and easiness to measure the temperature difference before and ence at the top and bottom for R-407C is not so great as that of
after the test section. Absolute average of deviation calculated by R-22. This is mainly because of the composition difference in the
Eq. 共5兲 was 0.7 % for 100 data points. local liquid film of refrigerant. If zeotropic refrigerant mixtures
are considered, more volatile component will easily evaporate,
兺冏冏
N
1 Q C,i ⫺Q M ,i which means there exists a concentration gradient near the inter-
E Q⫽ ⫻100共 % 兲 (5) face. In the liquid phase near the interface, there is a higher con-
N i⫽1 Q C,i
centration of less volatile component, which will generally reduce
where Q M is the measured heat rate using a wattmeter, and Q C is the evaporation of more volatile component. As a result, the tem-
the sensible heat rate calculated by the mass flow rate and temperature at the interface differs at the top and at the bottom. When
perature change of refrigerant in the test section. Actually, Q M is this effect of preferential evaporation is purely considered, the
Fig. 6 Circumferential variation of inner wall temperatures
Fig. 7 Circumferential variation of inner wall temperatures
at quality near 0.25 for R-22 and R-407C: „a… R-22; G
at quality near 0.50 for R-22 and R-407C: „a… R-22; G
Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; P satÄ670 kPa; x Ä0.26; and
Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; P satÄ670 kPa; x Ä0.46; and
„b… R-407C; G Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; PsatÄ760 kPa;
„b… R-407C; G Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; P satÄ760 kPa;
x Ä0.24
x Ä0.54
interface temperature at the top tends to be lower than that at the variation of inner wall temperature for R-407C is quite decreased
bottom 共Jung et al. 关12兴兲. However, in practical point of view, it is when compared with that for R-22, which represents the inner
very hard to measure or estimate the local interface temperature, wall temperature for R-407C is also affected by a preferential
therefore the equilibrium temperature for a given pressure and a evaporation of more volatile component in the refrigerant mixture.
mass quality is generally used for the determination of heat trans-
fer coefficient. Circumferential Variation of Heat Transfer Coefficient With
Circumferential variations of inner wall temperatures at vapor Respect to Vapor Quality. Circumferential variation of heat
quality near 0.5 for R-22 and R-407C are shown in Fig. 7 with transfer coefficient versus vapor quality for R-22 are shown in
mass flux G of 200 kg/m2s and heat flux q ⬙ of 3.6 kW/m2. As Fig. 8 when saturation pressure is 670 kPa, mass flux is 200
quality increases, the liquid film thickness becomes thinner and kg/m2s, and heat flux is 3.6 kW/m2. For R-22, the heat transfer
more uniform along the circumference. Therefore, the difference coefficient at the top of the test tube is greater than that at the
between the inner wall temperature at the top and that at the bottom, because of the different liquid film thickness along the
bottom is reduced for both cases of R-22 and R-407C. But still, circumference. When vapor velocity increases due to evaporation
the difference between these two positions exists due to the effect in the test section, the liquid film thickness becomes thinner and
of gravity on the liquid film especially for R-22. A circumferential more uniform along the circumference until it reaches a dryout
It should be noted that the average of circumferential tempera-
ture distribution is quite different from the local heat transfer co-
efficient measurement using wall-attached thermocouples nor-
mally at the top, bottom, or side positions of the tube. Concerning
the measurement of heat transfer coefficients around the circum-
ference using thermocouples, there are some data in the literature
previously listed, however, it is very hard to compare directly with
them mainly due to the different test conditions. One of the ac-
ceptable methods is to compare the circumferentially-averaged
heat transfer coefficient with that obtained by a 4-point 共top, bot-
tom, and two sides兲 average value using the corresponding data
obtained in this study with liquid crystals. This is quite reasonable
since the temperature measured by liquid crystal has been cali-
brated with the thermocouple measurement with a maximum de-
viation of 0.3 K 共rms deviation of 0.1 K兲 as shown in Fig. 3. For
R-22, the difference between the circumferentially-averaged value
and the 4 point average is about 5 %, and for R-407C it is about
2 %.
Comparison With Several Correlations. Figure 10 shows
circumferential variation of heat transfer coefficients at specified
test conditions and it is compared with in-tube heat transfer cor-
Fig. 8 Circumferential variation of heat transfer coefficients relations 共Gungor and Winterton 关6兴; Kandlikar 关8兴; Kattan et al.
at several vapor qualities of R-22 „G Ä200 kgÕm2 s; q ⬙
Ä3.6 kWÕm2 ; P satÄ670 kPa…
point. Therefore, heat transfer coefficient increases as quality in-

creases in the test range, which is thought mostly in the annular
flow regime and the difference between the heat transfer coeffi-
cient at the top and that at the bottom is accordingly reduced.
Circumferential variation of heat transfer coefficient for
R-407C is shown in Fig. 9 for the refrigerant pressure of 760 kPa,
mass flux of 200 kg/m2s, and heat flux of 3.6 kW/m2. Circumfer-
ential variation of heat transfer coefficient for R-407C is quite
different from that of R-22. The heat transfer coefficient for
R-407C at the top is slightly higher than that at the bottom, which
is due to the combined effects of the film thickness change and the
preferential evaporation at the interface. As vapor quality in-
creases, the local heat transfer coefficients along the circumfer-
ence are elevated due to higher mean velocity of the refrigerant.
At low quality region, the ratio of heat transfer coefficients at the
top and at the bottom is greater compared with the case at high
quality region.
Fig. 10 Circumferential variation of heat transfer coefficients

and comparison with correlations for R-22 and R-407C: „a…
Fig. 9 Circumferential variation of heat transfer coefficients R-22; G Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; P satÄ670 kPa; x Ä0.46;
at several vapor qualities of R-407C „G Ä200 kgÕm2 s; q ⬙ and „b… R-407C; G Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; P sat
Ä3.6 kWÕm2 ; P satÄ760 kPa… Ä760 kPa; x Ä0.54
关10兴兲 for R-22. For R-407C, the correlations of Gungor and Win- erage deviations between the values from the correlation and the
terton 关6兴, and that of Kattan et al. 关10兴 are used together with a measured values relative to the measurement result for R-407C
correlation for refrigerant mixtures proposed by Thome and Sha- were 0.2 % and 29.2 %, and the absolute average deviation is 11.4
kir 关22兴. In Fig. 10共a兲, a very high value of heat transfer coeffi- % and 29.2 % for Gungor and Winterton 关6兴, and Kattan et al.
cient is obtained especially at the top during the horizontal in-tube 关10兴, respectively.
evaporation of R-22 共mass flux of 200 kg/m2s; heat flux of 3.6 The comparison of test results with the correlations mentioned
kW/m2; pressure of 670 kPa; quality of 0.46兲, and the maximum above for several qualities up to about 0.55 is shown in Fig. 11,
deviation of 88 % from the average is observed at this point. The where one data point represents the average of 36 heat transfer
average deviation between the average of measured circumferen- coefficients measured along the circumference. The RMS 共root
tial heat transfer coefficient and the correlation by Gungor and mean square兲 deviation between the locally averaged heat transfer
Winterton 关6兴 relative to the average of measured values is ⫺9.4 coefficient from the measurement for R-22 and the correlation by
% and absolute average deviation is 21.7 %. The average devia- Gungor and Winterton 关6兴 is 22.2 %. That by Kandlikar 关8兴 and by
tion between the average of measured circumferential heat trans- Kattan et al. 关10兴 is 28.1 % and 36.2 %, respectively. For R-407C,
fer coefficient and the correlation by Kandlikar 关8兴 using the fluid the correlation by Kattan et al. 关10兴 shows RMS deviation of 21.7
dependent parameter of 1.0 for stainless steel tube is ⫺4.0 % and %, and that by Gungor and Winterton 关6兴 exhibits 27.2 % relative
absolute average deviation is 22.8 %. The average deviation from to the circumferentially averaged heat transfer coefficient. From
the correlation of Kattan et al. 关10兴 relative to the experimental the comparison in the test range of this study, the correlation by
value is 30.6 %, and the absolute average deviation is 39.9 %. Gungor and Winterton 关6兴 and that by Kattan et al. 关10兴 is quite
Figure 10共b兲 shows measured heat transfer coefficients for acceptable in predicting local heat transfer coefficients.
R-407C 共mass flux of 200 kg/m2s; heat flux of 3.6 kW/m2; pres-
sure of 760 kPa; quality of 0.54兲, the values from the correlations
by Gungor and Winterton 关6兴, and by Kattan et al. 关10兴. The av- 5 Concluding Remarks
Evaporative heat transfer characteristics for R-22 and R-407C
in a horizontal smooth tube have been experimentally investi-
gated. The circumferential variation of heat transfer coefficient
was focused on, and thermochromic liquid crystal was used to
measure the outer wall temperature of the test tube. Heat transfer
coefficients along the circumference were obtained for several va-
por qualities. Experimental results were compared with a correla-
tion for heat transfer coefficient proposed by Gungor and Winter-
ton 关6兴, Kandlikar 关8兴, and Kattan et al. 关10兴. Some conclusions
from this study are drawn as follows.
1 Circumferential variation of outer wall temperature and cor-
responding heat transfer coefficient for R-407C was quite differ-
ent from that for R-22. The heat transfer coefficient for R-22 at the
top of the test tube is much greater than that at the bottom because
of the different liquid film thickness due to gravity. However, the
heat transfer coefficient for R-407C at the top of the test tube is
slightly higher than that at the bottom, which is due to the com-
bining effects of liquid film thickness and the preferential evapo-
ration of more volatile component near the vapor-liquid interface.
There has been a difference between the circumference averaged
heat transfer coefficient and the average value based on the mea-
sured data at the top, bottom, and two sides.
2 As the vapor quality increases, the heat transfer coefficients
generally increase, and the ratio of heat transfer coefficient at the
top and that at the bottom is reduced. This reduction is quite
evident at higher vapor quality for both pure R-22 and zeotropic
refrigerant mixture, R-407C. Quite amount of an increase in heat
transfer coefficient with respect to quality was found at the bottom
of the tube.
3 Measured experimental data were compared with the existing
correlations proposed by Gungor and Winterton 关6兴, Kandlikar
关8兴, and Kattan et al. 关10兴. The RMS deviation between locally
averaged heat transfer coefficient and that from the correlation by
Gungor and Winterton 关6兴 was 22.2 % for R-22 and 27.2 % for
R-407C in the vapor quality range of this study. When Kandlikar’s
correlation 关8兴 was used for R-22, the RMS deviation is 28.1 %.
The RMS deviation of the correlation by Kattan et al. 关10兴 is 36.2
% for R-22, but if the lower quality region is not considered, the
predictions become much better. For R-407C, it shows 21.7 % of
RMS deviation.
4 In calculating the heat transfer coefficients using the outer
wall temperature measurement during in-tube evaporation, the
Fig. 11 Variation of average heat transfer coefficients with re- thermochromic liquid crystal film is one of good tools to measure
spect to quality and comparison with correlations for R-22 and circumferential or even two dimensional wall temperature distri-
R-407C: „a… R-22; G Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; P sat bution. However, caution should be given to the measurement for
Ä670 kPa; and „b… R-407C; G Ä200 kgÕm2 s; q ⬙ Ä3.6 kWÕm2 ; the curved surface, i.e., outer wall of the tube, and the color-to-
P satÄ760 kPa temperature calibration process.
冉冊冉冊冉冊冉冊冉冊
Acknowledgments 2
Bh 2
BV 2
BI 2
B T wi 2 B T sat
bias limit ⫽ ⫹ ⫹ ⫹
This work has been supported by Korea Science and Engineer- h V I ⌬T ⌬T
ing Foundation 共1999-1-304-006-3兲 and by the Brain Korea 21
冉冊冉冊冉冊冉冊冉冊
2 2
project of the Ministry of Education. Ph 2
PV 2
PI 2 P T wi P T sat
precision limit ⫽ ⫹ ⫹ ⫹
h V I ⌬T ⌬T
Nomenclature
1 Uncertainty of power measurement,
c ⫽ offset in Eq. 共2兲
U q ⬙ B V /V⫽0.02 % from rms voltage measurement
G ⫽ mass flux 共kg/m2s兲
h ⫽ heat transfer coefficient 共kW/m2K兲 BI
P ⫽ pressure 共kPa兲 ⫽共0.22 ⫹0.12 兲 0.5⫽0.22 %
I
Q ⫽ transferred heat 共kW兲
q⬙ ⫽ heat flux 共kW/m2兲 rms current measurement 0.2 %
RGB ⫽ red, green, blue Clamp meter accuracy 0.1 %
rgb
T
E
⫽
⫽
⫽
red, green, blue 共normalized value兲
temperature 共°C兲
average deviation
P q⬙
q⬙
⫽ 冉冉冊冉冊冊
PV
V
2
⫹
PI
I
2 0.5
⫽1.21 %
w ⫽ weighting coefficient in Eq. 共2兲 2 Uncertainty of inner wall temperature measurement, U T wi

x ⫽ vapor quality
␪ ⫽ normalized temperature, defined in Eq. 共3兲 BTwi
⫽共52⫹0.052 ⫹1.02 兲 0.5⫽5.1 %
Subscripts ⌬T
M ⫽ measured value T-type thermocouple error 0.1°C
C ⫽ calculated value DC voltage measurement 0.05 %
wi ⫽ inner wall Color-temperature conversion 1.0 %
sat ⫽ saturated state
h ⫽ heat transfer coefficient P T wi
⫽2.5 %
⌬T
Appendix A Uncertainty Analysis 3 Uncertainty of saturation temperature measurement, U T sat
Uncertainty evaluation of heat transfer coefficient is performed Uncertainty of saturation temperature is determined as shown
in accordance with a 95 % confidence interval to satisfy ANSI/ in the following equation.
冏
ASME PTC 19.1. dT
Heat transfer coefficient is determined as shown in the follow- UTsat⫽ ⫻U p
ing equation. dP sat
q⬙ B P sat P P sat
h⫽ ⫽0.25 %, ⫽0.3 %
T wi ⫺T sat P sat P sat
The total experimental uncertainty in the measurement is
B T sat P T sat
classified by bias error and precision error. The 95 % confidence ⫽4.3 %, ⫽5.18 %
uncertainty of heat transfer coefficients is calculated from as fol- ⌬T ⌬T
lows. 4 Calculation of uncertainty of heat transfer coefficient
U h ⫽ 冑B 2h ⫹ P 2h After evaluating the bias and precision errors, the uncertainty of
2 2
heat transfer coefficient is calculated as follows.
冑冉冊冉冊冉冊冉冊
Bh 2
B q⬙ 2 B T wi B T sat
bias limit ⫽ ⫹ ⫹ Bh BV 2
BI 2 B T wi 2
B T sat 2
h q⬙ ⌬T ⌬T
⫽ ⫹ ⫹ ⫹
2 2 h V I ⌬T ⌬T
Ph 2
P q⬙ 2 P T wi P T sat
precision limit ⫽ ⫹ ⫹ , ⫽ 共 0.22 ⫹0.222 ⫹5.12 ⫹4.32 兲 0.5⫽6.68 %
h q⬙ ⌬T ⌬T
⌬T⫽T wi ⫺T sat , where heat flux supplied to test section, q ⬙ , is
VI cos ␾
Ph
h
⫽ 冑冉冊冉冊冉冊冉冊
PV
V
2
⫹
PI
I
2
⫹
P T wi
⌬T
2
⫹
P T sat
⌬T
2
q ⬙⫽
A ⫽ 共 1.21 ⫹2.5 ⫹5.18 兲 ⫽5.88 %
2 2 2 0.5
冑冉冊冉冊
Because power factor, cos ␾ is near to 1 and heat transfer area, A, 2 2
Uh Bh Ph
is constant, the bias limit and precision limit of heat flux is as ⫽ ⫹ ⫽ 冑6.682 ⫹5.882 ⫽8.9 %
follows. h h h
Bias limit 冉冊冉冊冉冊
B q⬙
q⬙
2
⫽
BV
V
2
⫹
BI
I
2
References
关1兴 Wang, C. C., Kuo, C. S., Chang, Y. J., and Lu, D. C., 1996, ‘‘Two-Phase Flow
Precision limit 冉冊冉冊冉冊
P q⬙
q⬙
2
⫽
PV
V
2
⫹
PI
I
2 Heat Transfer and Friction Characteristics of R-22 and R-407C,’’ ASHRAE
Trans., 102, Part 1, pp. 830– 838.
关2兴 Shin, J. Y., Kim, M. S., and Ro, S. T., 1996, ‘‘Correlation of Convective
Boiling Heat Transfer in a Horizontal Tube for Pure Refrigerants and Refrig-
Therefore, uncertainty of heat transfer coefficient, U h , is deter- erant Mixtures,’’ Korean Journal of Air-Conditioning and Refrigeration Engi-
mined as follows. neering, 8共9兲, pp. 254 –266.
关3兴 Kim, S. H., Kwak, K. M., Bai, C. H., and Chung, M., 1997, ‘‘An Experimental
Uh
h
⫽ 冑冉冊冉冊 Bh
h
2
⫹
Ph
h
2 Study on Nucleate Boiling of Ternary Refrigerant R407C,’’ Korean Journal of
Air-conditioning and Refrigeration Engineering, 9共3兲, pp. 276 –283.
关4兴 Choi, T. Y., Kim, Y. J., Kim, M. S., and Ro, S. T., 2000, ‘‘Evaporation Heat
Transfer of R-32, R-134a, R-32/134a, and R-32/125/134a inside a Horizontal 关14兴 Niederkrüger, M., Steiner, D., and Schlünder, E.-U., 1992, ‘‘Horizontal Flow
Smooth Tube,’’ Int. J. Heat Mass Transf., 43, pp. 3651–3660. Boiling Experiments of Saturated Pure Components and Mixtures of R846-
关5兴 Kattan, N., Thome, J. R., and Favrat, D., 1998, ‘‘Flow Boiling in Horizontal R12 at High Pressures,’’ Int. J. Refrig., 15共1兲, pp. 48 –58.
Tubes: Part 2-New Heat Transfer Data for Five Refrigerants,’’ ASME J. Heat 关15兴 Boyd, R. D., Smith, A., and Turknett, J., 1995, ‘‘Two-Dimensional Wall Tem-
Transfer, 120, pp. 148 –155. perature Measurements and Heat Transfer Enhancement for Top-Heated Hori-
关6兴 Gungor, K. E., and Winterton, R. H. S., 1987, ‘‘Simplified General Correlation zontal Channels With Flow Boiling,’’ Exp. Therm. Fluid Sci., 11, pp. 372–386.
for Flow Saturated Boiling and Comparisons of Correlations with Data,’’ 关16兴 Shin, J. Y., Kim, M. S., and Ro, S. T., 1997, ‘‘Experimental Study on Forced
Chem. Eng. Res. Des., 65, pp. 148 –156. Convective Boiling Heat Transfer of Pure Refrigerants and Refrigerant Mix-
关7兴 Jung, D. S., McLinden, M., Radermacher, R., and Didion, D., 1989, ‘‘A Study tures in a Horizontal Tube,’’ Int. J. Refrig., 20, pp. 267–275.
of Flow Boiling Heat Transfer with Refrigerant Mixture,’’ Int. J. Heat Mass 关17兴 Camci, C., Kim, K., Hippensteele, S. A., and Poinsatte, P. E., 1993, ‘‘Evalua-
Transf., 9共1兲, pp. 1751–1764. tion of a Hue Capturing Based Transient Liquid Crystal Method for High-
关8兴 Kandlikar, S. G., 1990, ‘‘A General Correlation for Saturated Two-phase Flow Resolution Mapping of Convective Heat Transfer on Curved Surfaces,’’ ASME
Boiling Heat Transfer Inside Horizontal and Vertical Tubes,’’ ASME J. Heat J. Heat Transfer , 115, pp. 311–318.
Transfer, 112, pp. 219–228. 关18兴 Matsumoto, R., Kikkawa, S., and Senda, M., 1997, ‘‘Effect of Pin Fin Ar-
关9兴 Kandlikar, S. G., 1998, ‘‘Boiling Heat Transfer with Binary Mixtures: Part rangement on Endwall Heat Transfer,’’ JSME Int. Journal, Series B, 40共1兲, pp.
II—Flow Boiling in Plain Tubes,’’ ASME J. Heat Transfer, 120, pp. 388 –394. 142–151.
关10兴 Kattan, N., Thome, J. R., and Favrat, D., 1998, ‘‘Flow Boiling in Horizontal 关19兴 Kimura, I., Kuroe, Y., and Ozawa, M., 1993, ‘‘Application of Neural Networks
Tubes: Part 3-Development of a New Heat Transfer Model Based on Flow to Quantitative Flow Visualization,’’ J. Flow Visualization and Image Process-
Pattern,’’ ASME J. Heat Transfer, 120, pp. 156 –165. ing, 1, pp. 261.
关11兴 Ross, H., Radermacher, R., and Di Marzo, M., 1987, ‘‘Horizontal Flow Boil- 关20兴 Huber, M., Gallagher, J., McLinden, M., and Morrison, G., 1996, NIST Ther-
ing of Pure and Mixed Refrigerants,’’ Int. J. Heat Mass Transf., 30共5兲, pp. modynamic Properties of Refrigerants and Refrigerant Mixtures Database
979–992. 共REFPROP兲 Version 5.0, National Institute of Standards and Technology,
关12兴 Jung, D. S., McLinden, M., Radermacher, R., and Didion, D., 1989, ‘‘Hori- U.S.A.
zontal Flow Boiling Heat Transfer Experiments with a Mixture of R22/R114,’’ 关21兴 Dittus, F. W., and Boelter, L. M. K., 1930, University of California (Berkeley)
Int. J. Heat Mass Transf., 32, pp. 131–145. Publications of Engineering, 2, University of California, Berkeley, CA, pp.
关13兴 Yoshida, S., Mori, H., Matsunaga, T., and Ohishi, K., 1991, ‘‘Heat Transfer to 443.
Non-Azeotropic Mixtures of Refrigerants Flowing in a Horizontal Evaporator 关22兴 Thome, J. R., and Shakir, S., 1987, ‘‘A New Correlation for Nucleate Pool
Tube,’’ ASME/JSME Thermal Eng. Proc., 2, pp. 295–300. Boiling of Aqueous Mixtures,’’ AIChE Symp. Ser., 83共257兲, pp. 46 –51.
Evaporation of Water Droplets
Placed on a Heated Horizontal
Orlando E. Ruiz
e-mail: orlando-eគruiz@hp.com
Surface
Hewlett Packard,
Pen Technology & Manufacturing Center, A numerical analysis of the evaporation process of small water droplets with diameters of
Solutions Development Group, 1 mm or less that are gently deposited on a hot isothermal solid surface has been per-
P.O. Box 4048, formed. This study considers the internal fluid motion that occurs as a result of the
Agüadilla, PR 00605 thermocapillary convection in the droplet and it determines the effect of fluid motion on
the heat transfer between the drop and the solid surface. This study is particularly rel-
William Z. Black evant because the internal fluid motion has not been considered in previous numerical and
e-mail: william.black@me.gatech.edu analytical models presented in the literature. To assess the effects of internal fluid motion,
Regents Professor the model results are compared to numerical results provided by a heat conduction model
Georgia Institute of Technology, that neglects the fluid motion. The Navier-Stokes and Thermal Energy equations are
George W. Woodruff School of solved using the Artificial Compressibility Method with Dual Time Stepping. Boundary-
Mechanical Engineering, fitted grids are used to track the changes in the droplet surface shape during the evapo-
Atlanta, GA 30332-0405 ration process. The numerical simulations have demonstrated that the internal fluid mo-
tion provides vastly different temperature distributions in the drop compared to the results
from the heat conduction model that neglects fluid motion. The evolution of the droplet
geometry was simulated from an initial spherical-shaped cap until the contact angle was
close to the receding contact angle. 关DOI: 10.1115/1.1494092兴
Keywords: Computational, Convection, Droplet, Evaporation, Thermocapillary
Introduction condensation process of hemispherical water drops was deter-

mined to be insignificant. These models have considered simpli-
Dropwise evaporative cooling is the process by which a single
fying assumptions regarding the shape of the droplet during the
fluid droplet impacts a heated surface and cools by latent heat
evaporation process, contact angle behavior, and liquid-solid in-
absorption which transforms the liquid into the vapor phase. The
terface thermal conditions.
heat transfer rate during the evaporation process is much greater
The physical processes that take place during the droplet evapo-
than traditional air cooling techniques. Dropwise evaporative
ration are complicated and consist of conjugate heat transfer be-
cooling of a hot surface is widely used in many areas of engineer-
tween the solid and the liquid, convective fluid motion within the
ing. For instance, this technique is applied in metallurgical appli-
droplet, free surface flow, contact line motion, and surface evapo-
cations, nuclear thermal management, electronic industries, and
ration. The heat transfer interaction between the solid and the
fire suppression systems. The microelectronics industry has been a
liquid is complicated and it is based on the fact that as the droplet
major motivating influence on the research in the spray-cooling
evaporates, the contact line travels over the heated surface making
field. The thermal loads projected for future microelectronic tech-
the contact surface vary with time. When droplets are placed on a
nologies are beyond the capabilities of traditional air cooling sys-
tems. As a result, droplet evaporation has received consideration solid metallic surface with a high thermal conductivity, the surface
as a means to provide the high heat flux cooling that is necessary temperature changes are very small during the droplet lifetime
to maintain microelectronic packages at reasonable temperatures. 关1,7兴. For these cases, the contact surface can be treated as an
To understand the thermal interaction of the spray with the isothermal boundary. This simplification allows concentrating on
heated surface, a substantial amount of experimental research has the transport processes in the evaporating droplet without the
been conducted to optimize the parameters of the spray cooling added complexity of the solid-liquid thermal interaction.
process. Some researchers have studied the spray cooling process Convective fluid motion within the droplet can be attributed to
by proposing theoretical and numerical models of the dropwise the interaction between buoyant and thermocapillary convection
evaporation process. These models have been expanded based on cells. During the evaporation process, buoyant convection may
statistical arguments to describe the spray cooling process as the occur as a result of the density stratification in the fluid. The
collective influence of independent drops 关1兴. During the past two thermocapillary convection is the result of an instability due to
decades, researchers have modeled this process by considering surface tension gradients at the droplet free surface. These gradi-
heat transfer by conduction as the dominating mode of thermal ents occur as a result of a non-uniform temperature distribution at
energy transport and they have neglected motion of the fluid in the the free surface. The colder fluid has higher surface tension and it
droplet. Various experimental studies 关2–5兴 have demonstrated pulls the warmer fluid which has a lower surface tension. This
that the droplet evaporation process is a combination of the inter- behavior sets up viscous forces in the fluid at the free surface
action among the coupled fluid dynamics, surface chemistry, and which drag adjacent fluid layers toward the cooler regions of the
heat transfer. Most of the theoretical and numerical dropwise droplet.
evaporation models have neglected the effects of fluid motion In this study the droplet evaporation process is investigated
within the liquid droplets. An exception is the work of Lorenz and from a numerical standpoint. The value of a droplet evaporation
Mikic 关6兴, were the effects of the thermocapillary flow in the model as a starting point for more complex spray cooling models
has been proven consistently by previous researchers 关1,5,8兴. The
difficulty of measuring local heat fluxes, temperature distributions
HEAT TRANSFER. Manuscript received by the Heat Transfer Division March 5, 2001; within the droplet, and the evolution of the droplet geometry has
revision received May 7, 2002. Associate Editor: P. S. Ayyaswamy. created a need to model this process. There are several benefits to
ment and various experimental studies in the literature 关2,3,9兴 and
visualizations studies that we have performed, we have neglected
the gravity effects on the droplet shape in this study. The initial
contact angle depends on many factors such as the surface mate-
rial properties, surface finish, temperature, and contamination of
the surface. For water drops on stainless steel surfaces, contact
angles approaching 90 deg have been measured experimentally by
Qiao and Chandra 关10兴 and Chandra et al. 关11兴. Bernardin et al.
关12兴 reported that for aluminum surfaces, water droplets assumed
a spherical cap shape with an initial static contact angle approach-
ing 90 deg for surface temperatures up to 120°C.
The assumption of an isothermal surface is applicable for sur-
faces with high thermal conductivity, such as aluminum and cop-
Fig. 1 Droplet geometry per. This assumption simplifies the analysis because it permits a
study of the influence of the fluid motion inside the drop on the
evaporation process without the complexity of studying the tran-
this type of model. It can be used to study several parameters such sient interaction between the solid and the liquid. Figure 1 pro-
as initial surface temperature, liquid subcooling, solid material vides a sketch of the geometry that has been considered in the
thermal properties, and fluids of interest in spray cooling. A drop- present model. The contact diameter (D o ), also called wetted di-
let evaporation model also can be used to determine optimum ameter, is kept constant during most of the evaporation process.
droplet parameters, evaporation time, heat transfer rates and tem- The apex dimension 共a兲 and the contact angle 共␪兲 are allowed to
perature distributions, as well as the effect of droplet geometry on vary as the volume of the drop decreases due to evaporation. The
heat transfer rates. droplet has been modeled in a cylindrical coordinate system with
A numerical model that assumes a small diameter liquid droplet the radial coordinate 共r兲 along the horizontal direction and the
gently deposited on a heated substrate was developed during this axial coordinate 共z兲 along the vertical direction. In the absence of
study. The droplet is initially shaped as a spherical cap as shown thermal or mechanical loads in the circumferential direction, the
in Fig. 1. The model considers the evaporation process that takes process is assumed to be axisymmetric without swirl.
place after the droplet impacts the substrate, spreads, and reaches To determine the effects of internal fluid motion on the evapo-
an equilibrium geometry. The fluid motion that takes place in the ration process, a comparison between two numerical models with
droplet due to the thermocapillary convection effects has been and without fluid motion has been performed. The numerical
modeled. This study considers the evaporation process for sub- model that accounts for the fluid motion will be referred from this
strate surface temperatures for which nucleate boiling is sup- point on as the convection model and the numerical model with-
pressed and the contact surface with the solid remains isothermal. out fluid motion will be referred as the conduction model. The
A numerical model of an evaporating droplet presents many goal of this study is to accurately model the thermal convection
challenges. During the process, surface tension gradients along the effects during the process and to understand the impact of convec-
droplet surface initiate thermocapillary flow. This flow develops tive motion on the evaporation rates of small drops. Previous nu-
convection cells below the drop surface with dimensions of the merical models of evaporating drops have neglected the internal
order of the droplet radius. Since the droplet surface is an inter- fluid motion. This study extends the scientific understanding of
face between the liquid and the surrounding gas, free surface flow the evaporation process by modeling the fluid motion and its ef-
occurs. These types of problems tend to be numerically complex fects on the thermal field within the drops.
because the free surface position is not known a priori and must
be determined from the flow field solution. The liquid-gas inter- Convection Model. The heat transfer and fluid motion in the
face terminates at a contact line with the solid. For viscous fluids, droplet are governed by the Navier-Stokes and the Conservation
the proper boundary condition at surfaces where a liquid and a of Energy equations. Since no gradients or motion in the circum-
solid are in contact is the well known no-slip condition. The fluid ferential direction will be considered, the equations can be simpli-
adheres to the solid and has the same velocity of the solid where fied for an axisymmetric coordinate system. The dimensionless
the contact occurs. At contact lines, this boundary condition ap- Navier-Stokes and Conservation of Energy equations for a fluid
pears to break down based on a balance of the forces involved in with constant properties and an axisymmetric domain are the fol-
the contact line motion. As a result, the motion of the contact line lowing:
is still an active area of investigation, both experimentally and 1 ⳵ 共 ru 兲 ⳵ v
theoretically. ⫹ ⫽0 (1)
r ⳵r ⳵z
Another challenge that must be faced in developing a numerical
model of the droplet evaporation process is the issue of time
scales. The motion of the fluid develops on a time scale that is
much less than the time scale of the evaporation process. The
⳵ u 1 ⳵ 共 ru 2 兲 ⳵ 共 u v 兲
⫹
⳵t r ⳵r
⫹
⳵z
⳵P
⫽⫺ ⫺
1 ⳵
⳵ r r Re ⳵ r
r 冉冊
⳵u
⳵r
⫹ 2
u
r Re
difference in these two time scales for water drops could be as
large as six orders of magnitude. From a computational stand-
point, this substantially complicates the analysis because the
⫺
1 ⳵ ⳵u
冉冊
Re ⳵ z ⳵ z
(2)
simulations must be performed on the smaller of the two time

scales. As a result, a very large number of numerical iterations
must be carried out before the evaporation effects have an impact
⳵ v 1 ⳵ 共 ru v 兲 ⳵ 共v 2 兲
⫹
⳵t r ⳵r
⫹
⳵z
⫽⫺
⳵P
⫹
1 ⳵
⳵ z r Re ⳵ r
r 冉冊
⳵v
⳵r
⫹
1 ⳵ ⳵v
Re ⳵ z ⳵ z 冉冊
on the thermal and flow fields. Ra
⫹ T (3)
Re2 Pr
冉冊冉冊
Droplet Evaporation Model
⳵ T 1 ⳵ 共 ruT 兲 ⳵ 共v T 兲 1 ⳵ ⳵T 1 ⳵ ⳵T
Based on experimental observations of Zhang and Yang 关2兴, ⫹ ⫹ ⫽ r ⫹ .
Rymkiewicz and Zapalowicz 关3兴, and Di Marzo and Evans 关9兴, ⳵t r ⳵r ⳵z r Re Pr ⳵ r ⳵r Re Pr ⳵ z ⳵ z
gently deposited water drops assume an initial equilibrium shape (4)
as a spherical cap. The gravity effects are small compared to the The velocity field expressed in axisymmetric cylindrical coor-
surface tension effects on the droplet shape. Based on this argu- dinates has u and v components in the radial 共r兲 and axial 共z兲
directions respectively. The variables P and T correspond to the The local mean curvature term in the normal stress boundary con-
pressure and temperature respectively. Since the evaporation pro- dition can be calculated by taking the divergence of the unit nor-
cess is slow and the fluid velocities are not expected to be large, mal vector at the free surface as follows 关13兴,
the viscous dissipation term that typically appears in the Conser-
vation of Energy equation can be neglected. A body force term in ␬ ⫽⫺ⵜ•nជ . (15)
the axial direction has been used to characterize the natural con-
vective motion of the fluid during the transient evaporation pro- The evaporative mass flux is calculated based on the product of
cess. Since the fluid is assumed incompressible, the buoyant con- the mass transfer coefficient, h m , and the difference between the
vection effects are modeled using the Boussinesq approximation. saturated vapor density at the free surface ( ␳ vap,sat) and the vapor
For small drops, this mode of convection can be insignificant in density in the surrounding environment ( ␳ vap,⬁ ),
comparison to the thermocapillary convection in the drop. How-
ever, this body force term can become relevant in situations in- m ⬙ ⫽h m 共 ␳ vap,sat共 T 兲 ⫺ ␳ vap,⬁ 兲 . (16)
volving large drops in which the thermocapillary effects are less
important. The governing equations were dimensionalized based The saturated vapor density depends only on the local free surface
on the length scale (R o ), velocity scale (U sc ), and temperature temperature. The vapor density on the surrounding environment
difference (⌬T⫽T H ⫺T ⬁ ). The pressure was made dimensionless depends on the species concentration in the far field from the
based on the dynamic pressure scale ( ␳ U 2sc ). The dimensionless liquid drop. For the purpose of this study, the environment sur-
numbers Re, Pr, and Ra result from selecting these scales. rounding the droplet was assumed to be dry air and therefore the
The droplet is initially at a quiescent state, and the fluid tem- concentration of the vapor in the far field was assumed to be zero.
perature is assumed to be at a uniform ambient temperature T ⬁ . The vapor density is a thermodynamic property of the fluid and is
The initial fluid velocity is assumed to be zero. The boundary tabulated as a function of temperature and pressure in many ref-
conditions of the fluid domain are specified at three locations, the erences 关14兴. The mass transfer coefficient is determined based on
contact surface with the heated solid, the free surface, and the axis the analogy between heat and mass transfer 关15兴 using the follow-
of symmetry. The contact surface boundary conditions enforce ing relationship,
no-slip, no fluid penetration through the surface, and the equiva-
h nc
lence of the liquid and solid surface temperature distributions at h m⫽ (17)
the contact surface. These boundary conditions are as follows: ␳ airc p,airLe2/3
u 共 r,0,t 兲 ⫽0 (5) The natural convection heat transfer rate between the droplet sur-
face and the surrounding air is very small because both the surface
v共 r,0,t 兲 ⫽0 (6) area and the temperature difference between the droplet surface
and the surrounding air are small. Given this condition, the Nus-
T 共 r,0,t 兲 ⫽T H . (7) selt number can be approximated by its conduction limit value of
where T H is the temperature of the heated surface. 2.0 关16 –18兴. This simplified boundary condition reduced the com-
At the axis of symmetry, the boundary conditions state that no plexity associated in determining the external flow field and the
fluid penetrates this boundary and zero radial derivatives of tem- external thermal convection in a stagnant surrounding. The droplet
perature and axial fluid velocity are imposed. The symmetry shape varies as it evaporates and the heat transfer coefficient is
boundary conditions are summarized as follows: expected to depend directly on the droplet geometry. This simpli-
fied boundary condition did not affect adversely the numerical
u 共 0,z,t 兲 ⫽0 (8) results for the cases considered in this study. It is expected that for
larger drops, the external convection may be more significant as
⳵v
⳵r
冏共 0,z,t 兲
⫽0 (9)
compared to the present cases due to the larger surface area in
contact with the surrounding fluid. As a result, an appropriate
Nusselt number correlation must be used to account for the drop-
⳵T
⳵r
冏共 0,z,t 兲
⫽0. (10)
let geometry variation as the volume decreases.
During the droplet evaporation process, the liquid surface tem-
perature varies due to the absorption of latent heat, conduction
The free surface boundary conditions are complex since this and convection of heat from the solid surface, and the heat trans-
boundary deforms and moves inward as the liquid evaporates at fer with the surroundings. The surface temperature inhomogeneity
the interface. Conservation of mass at the free surface equates the initiates Marangoni convection due to the surface tension depen-
mass flux of the liquid that moves to the free surface with the dence on temperature. The surface tension usually depends on the
mass flux of liquid that evaporates. A normal stress balance con- scalar fields in the system 共e.g., electrical fields and temperature
siders the local pressure difference across the interface to be equal fields兲 as well as on the concentration of foreign materials on the
to the sum of the surface tension stress and the normal viscous interface. Since the temperature field is the focus of this study, an
stress. A balance between viscous and surface tension forces must equation of state must be considered for the surface tension. The
be enforced in a direction tangent to the free surface to incorpo- most common approach is to characterize this dependence by a
rate the thermocapillary convection effects. Finally, the thermal linear law as follows 关19兴,
boundary condition enforces a balance of conduction with natural
convection and evaporative mass transfer to the environment.
These boundary conditions have been derived based on a rela-
␴ ⫽ ␴ o⫺ 冏冏
d␴
dT
共 T⫺T o 兲 , (18)
tively slow evaporation process in which the thermal conductivity, where ␴ o is the reference surface tension at the reference tempera-
viscosity, and density of the vapor phase are much smaller than ture T o , and 兩 d ␴ /dT 兩 is a constant that depends on the fluid.
the liquid phase properties. These boundary conditions are sum- As the fluid warms, buoyant convection occurs within the drop-
marized as follows, let. During the evaporation process there is an interaction of the
Marangoni and buoyant convection which makes it difficult to
␳ 共 Vជ ⫺Vជ I 兲 •nជ ⫽m ⬙ (11) select a characteristic velocity scale for the problem. Since the
P⫺ P ⬁ ⫽⫺ ␬ ␴ ⫹nជ • 共 S= •nជ 兲 (12) velocities associated with buoyant convection are expected to be
smaller than the fluid velocities due to Marangoni convection, the
ជt • 共 S= •nជ 兲 ⫽ⵜ ␴ •tជ (13) selected velocity scale is based on the dominant mode of convec-
tion. The Marangoni velocity scale is selected based on a scale
⫺kⵜT•nជ ⫽h nc 共 T⫺T ⬁ 兲 ⫹m ⬙ h f g . (14) analysis performed on the tangential stress boundary condition,
Eq. 共13兲. The Marangoni velocity scale represents a balance be- evaporation on the surface. Each of these thermal transport pro-
tween surface tension gradients at the interface and the shear cesses has a time scale associated with it. The thermal convection
stresses generated by them 关20兴 in the fluid occurs due to a combination of the thermocapillary
冏冏
and buoyancy effects. Based on the previous scaling analysis, the
d␴ Marangoni velocity scale was chosen to model the dominant
⌬T
dT mode of convection in the process. A time scale associated with
U sc ⫽ . (19)
␮ this convection mode, called the Marangoni time scale, is defined
as follows:
When this velocity scale is used, the Reynolds number is defined
as follows: R o␮
冏冏
t Ma⫽ . (35)
冏冏
d␴ d␴
⌬TR o ⌬T
dT Ma dT
Re⫽ ⫽ , (20)
␯␮ Pr The thermal diffusion occurs at a much slower rate than the Ma-
rangoni convection and the time scale associated with the thermal
and the Marangoni number is defined as diffusion is the thermal diffusion time scale that is defined as
Ma⫽
d␴
dT
冏冏
⌬TR o
. (21)
follows,
t diff⫽
R 2o
. (36)
␮␣ ␣
In addition to scaling the governing equations, the initial and The evaporation process also occurs at a much slower rate than
boundary conditions must be scaled as well. The scaled initial the Marangoni convection. A time scale for this process can be
conditions for the drop are the following, derived based on a thermal energy balance between conduction
and evaporative mass transfer at the free surface. The evaporation
ũ 共 r̃,z̃,0兲 ⫽0 (22) time scale is defined here as follows,
ṽ共 r̃,z̃,0兲 ⫽0 (23) R 2o ␳ h f g
t evap⫽ . (37)
T̃ 共 r̃,z̃,0兲 ⫽0. (24) k⌬T
The contact surface boundary conditions are scaled as follows: The Marangoni time scale is on the order of 10⫺5 to 10⫺6
second for very small water droplets while the evaporation time
ũ 共 r̃,0,t̃ 兲 ⫽0 (25) scale is larger by five to six orders of magnitude. For a larger
ṽ共 r̃,0,t̃ 兲 ⫽0 (26) temperature difference (⌬T), the evaporation process occurs
faster and the velocity associated with the thermocapillary process
T̃ 共 r̃,0,t̃ 兲 ⫽1. (27) is larger. This vast difference in time scales presents a computa-
tional challenge because the governing equations must be scaled
The axis of symmetry boundary conditions take the following based on the Marangoni time scale to accurately resolve the con-
form: vection process. As a result, the number of computational time
ũ 共 0,z̃,t̃ 兲 ⫽0 (28) steps that must be accumulated before the drop experiences sig-
nificant evaporation is extremely large.
⳵ ṽ
⳵ r̃
冏共 0,z̃,˜t 兲
⫽0 (29) Conduction Model. To determine the importance of the in-
ternal fluid motion on the droplet evaporation process, a heat con-
冏
duction model was formulated. The purpose of this model was to
⳵ T̃ provide information on the temperature distribution within the
⫽0. (30)
⳵ r̃ 共 0,z̃,˜t 兲
drop in the absence of fluid motion. The results were compared
with the solution of the convection model to determine the effects
Finally, the scaled free surface boundary conditions produce the of the fluid motion on the evaporation process. The governing
following dimensionless expressions: equation describing the thermal process in an evaporating droplet
in the conduction model is the Heat Conduction Equation. Assum-
Ṽ n ⫺Ṽ I,n ⫽m̃ ⬙ (31)
ing that the process is axisymmetric and the liquid has constant
P̃⫽⫺
˜␬ 1
冉冊冉
⫺T̃ ⫹
2 ⳵ Ṽ n ជ̃ ⳵ nជ
Re ⳵ n
⫺V •
⳵n
冊 (32)
properties, the Heat Conduction Equation can be written in cylin-
drical coordinates as follows:
冉冊
Re Ca
1 ⳵T 1 ⳵ ⳵T ⳵ 2T
⳵n
⫹
⳵s 冉冊
⳵ Ṽ t ⳵ Ṽ n ជ̃ ⳵ជt ⳵ nជ
⫺V • ⫹
⳵n ⳵s
⫽⫺
⳵ T̃
⳵s
(33)
⫽
␣ ⳵t r ⳵r
r
⳵r
⫹ 2.
⳵z
(38)
Since the conduction model shares the same thermal initial and
⳵ T̃ Ma boundary conditions as the convection model, these equations will
⫺ ⫽Bi T̃⫹ m̃ ⬙ . (34) only be mentioned here. The fluid is initially at uniform ambient
⳵n Ja
temperature T ⬁ . The contact surface is isothermal at temperature
The normal and tangential partial derivatives are ⳵ / ⳵ n and ⳵ / ⳵ s, T H , Eq. 共7兲. The axis of symmetry boundary condition was shown
respectively. The new dimensionless numbers that result from the previously in Eq. 共10兲. Finally, the free surface thermal balance
scaling of these boundary conditions are the Biot number 共Bi兲, was presented in Eq. 共14兲. Temperature, length, and time scales
capillary number 共Ca兲, and the Jacob number 共Ja兲. are chosen as before to nondimensionalize these equations. The
During the droplet evaporation process, various modes of ther- temperature is scaled based on the temperature difference between
mal energy transport are coupled. The fluid motion in the drop the heated plate and the ambient temperature. The chosen length
allows for thermal energy transport by means of a convection scale is the initial contact radius R o . The time scale for the Heat
process. Also, a portion of the thermal energy is transported Conduction Equation is typically based on the diffusion time
through heat diffusion and the droplet surface experiences a tem- scale. However, in this analysis the time scale is selected equal to
perature reduction due to the latent heat absorbed as a result of the Marangoni time scale so that the results can be compared on a
one-to-one basis with the results provided from droplet evapora- ated at the cell centers. Second order accurate forward and back-
tion model that includes fluid motion. Based on these scaling ward differences were used to approximate the partial derivatives
quantities, the governing equation for the conduction model in in the boundary conditions.
dimensionless form is The system of equations was marched iteratively between
冋冉冊册
physical time-steps by using a Line Relaxation approach 关26兴. At
⳵ T̃ 1 1 ⳵ ⳵ T̃ ⳵ 2 T̃ the end of a physical time step, a converged solution was obtained
⫽ r̃ ⫹ 2 . (39)
⳵ t̃ Ma r̃ ⳵ r̃ ⳵ r̃ ⳵ z̃ when the dimensionless discrete divergence was below 10⫺4 and
The initial and boundary conditions are scaled in a similar fashion the dimensionless variables variation between sub-iterations was
as the convection model. below 10⫺8 . The grid was generated at every physical time-step
using an algebraic grid generation approach. Specifically, a trans-
finite interpolation method based on the four-boundary scheme
关27,28兴 was used to generate the boundary-fitted grid at each
Numerical Method physical time-step.
The numerical techniques used to solve both the convection and
conduction models described in the previous section are based on Results
the Finite Difference Method. Boundary-fitted non-uniform grids The numerical simulations were accomplished using a com-
were used to represent the physical domain and to track the drop- puter code developed during the course of this investigation. The
let surface as the domain deforms and shrinks due to evaporation. computer code was validated for numerical accuracy by compari-
A generalized curvilinear coordinate transformation was used to son with the numerical solutions to several benchmark problems
map the physical grid into a stationary uniform grid. The govern- 关21兴. In this study, the effects of three major variables on the
ing equations and boundary conditions are transformed from the evaporation process were considered. These variables are the ini-
physical coordinates r and z to the generalized curvilinear coordi- tial droplet contact diameter D o , initial droplet contact angle ␪ o ,
nates ␰ and ␩ as shown in Fig. 2. As a result of the transformation, and the heated surface temperature T H . The initial droplet diam-
the governing partial differential equations become more complex eter was assigned a value of either 1.0 mm or 0.1 mm. The size
and contain additional terms that account for the geometric trans- was selected because droplets within this range of sizes are ob-
formation and the grid motion. For details of the geometric trans- tained in many spray cooling applications. The initial contact
formation and the resulting partial differential equations, refer to angle was assigned a value of 90 deg or 60 deg. The heated
Ruiz 关21兴. The benefit of the transformation results from the fact surface temperature was set at either 80°C or 100°C. These tem-
that the boundaries become coordinate lines in the computational peratures are low enough so that nucleate boiling does not occur
domain. The boundary conditions are transformed and applied on at the contact surface. The ambient temperature T ⬁ was assumed
these coordinate lines. This approach reduces the difficulty of to be constant with a value equal to 20°C. Table 1 shows these
implementing boundary conditions especially on boundaries that variables and the dimensionless parameters relevant to this prob-
move and deform. In addition, the presence of coordinate lines at lem 共Ma,Ca,Bi,Pr,Re兲. In addition to the eight primary cases 共1–
the boundaries permits moving boundaries such as free surfaces to 8兲, two additional cases 共9 and 10兲 for droplets with similar initial
be located with considerable accuracy. volumes and different contact angles were also considered. Based
The Artificial Compressibility method was used to solve for the on the selection of the primary variables, the maximum Ma-
fluid velocities and temperature distribution in the droplet domain. rangoni number considered was approximately 152,000 and the
This numerical technique was first introduced by Chorin 关22兴 to smallest was approximately 7000. The values for the Ca, Bi, and
obtain steady-state solutions to the incompressible Navier-Stokes Pr numbers do not vary significantly since they are based almost
equations. The method has been used with much success by many entirely on fluid properties. The Reynolds number is equal to the
investigators. The Artificial Compressibility method has also been ratio of the Marangoni and the Prandtl number and its value cov-
extended to obtain time accurate solutions of the incompressible ers a range between approximately 2300 and 68,500.
Navier-Stokes equations by using the Dual-Time Stepping ap- The time scale associated with the thermocapillary convection
proach 关23兴. In this study, the formulation of Rogers and Kwak is at least five orders of magnitude smaller than the evaporation
关24兴 has been extended to couple the Conservation of Energy time scale. The evaporation time scale characterizes the thermal
Equation and simultaneously solve for the temperature and veloc- interaction at the droplet surface and the deformation of the drop-
ity fields. The numerical scheme is implicit and second order ac- let shape. In this study, the difference in time scales and their
curate in time. The convection derivatives were approximated us- significance in the numerical process has been used as an advan-
ing a third order upwind flux-difference splitting approach 关25兴 tage. Based on numerical studies, the evolution of both the tem-
and the viscous derivatives were approximated using second order perature and flow fields is characterized by the Marangoni time
central differences in a collocated grid. Source terms were evalu- scale. Both fields become ‘‘stationary’’ before the deformation of
the free surface can affect them. The deformation process occurs
at a much slower rate that is based on the evaporation time scale.
Table 1 Simulation variables and dimensionless parameters

„ T ⴥ Ä20°C…
Fig. 2 Generalized curvilinear coordinate transformation
Fig. 3 Short time solution: Isotherm and Streamfunction contours „with velocity vectors
superimposed… for a 0.1 mm initial contact diameter water drop on a 100°C heated surface.
Initial contact angleÄ90 deg. Environment: T ⴥ Ä20°C, dry air: „a… 0.5 ms, „b… 1.0 ms, „c… 1.5
ms, and „d… 2.0 ms.
To obtain numerical solutions in a reasonable time, the numerical 10.0 was used for the large diameter drops and it was 1.0 for the
simulations of the convection model were performed in two small drops. These integration time steps allowed the solution to
stages. During the first stage, the governing equations were nu- advance at a faster rate and permitted a more economical solution
merically integrated using a small integration time step to solve for the thermal and flow fields. A long time solution for a droplet
the short time transient evolution of the temperature and flow evaporation case required between 96 to 120 hours of runtime on
fields. A dimensionless integration time step ⌬t of 0.1 was suffi- the SGI Origin2000 parallel platform. The short and long time
ciently accurate for the present computations. This value corre- transient solutions were patched together to obtain a complete
sponds to a dimensional time step of 0.1 times the Marangoni time history of the evaporation process.
scale. This first stage of the numerical simulation is characterized To determine the proper grid resolution required for the simu-
by a heating period in which the drop temperature increases due to lations in this investigation, a grid resolution study was per-
heat transfer with the solid surface. During the same time frame, formed. During the grid resolution study, the droplet evaporation
the flow field develops and affects the heating process and tem- problem was solved using four different grids with increasing lev-
perature field within the drop. The solution evolves until the ther- els of grid refinement. The four grids used had refinement levels
mal and flow field changes are very small. At this point in time, of 61⫻31, 91⫻46, 121⫻61, and 181⫻91 points in the ␰ and ␩
the isotherm and streamfunction contours appear to have reached directions respectively. Each refinement level approximately
their steady state distributions. For the drop sizes selected in this doubled the total number of grid points. Based on a comparison of
study, this process occurs during approximately the first five per- the velocity profiles, temperature profiles, average Nusselt num-
cent of the total evaporation time. The mass loss and volume ber, and apex temperature, the 121⫻61 grid was selected as an
changes that occur during this stage are small. As this nearly acceptable resolution level to perform all the numerical simula-
asymptotic state is approached, the temperature and flow field tions. This resolution level provided reasonable accuracy without
solutions resemble the steady-state solutions obtained by solving an excessive number of grid points. The 181⫻91 grid was not
the problem without droplet mass loss and without free surface used due to the greater storage constraints and due to the exces-
deformation. sive computational time requirements that would be required with
This type of simulation will be referred to as the short time this level of grid refinement.
solution. The number of physical time steps required to complete Thermal and flow solutions are shown below that are based on
the short time solution depends on the Marangoni time scale. the selection of one of the droplet evaporation cases. The case that
Based on these short time simulations, a small amount of the was selected for this comparison consists of a water droplet ini-
droplet mass evaporates during the evolution of the thermal and tially at room conditions (T ⬁ ⫽20°C). The droplet has an initial
flow fields. The numerical simulations were performed using six contact radius of 0.1 mm and an initial contact angle of 90 deg.
processors on an SGI Origin2000 parallel platform. The large The heated surface temperature T H is maintained at 100°C. This
drops (D o ⫽1.0 mm) required on the order of 264 hours of com- corresponds to case 6 in Table 1. The results of the other cases in
putational time while the small drops (D o ⫽0.1 mm) required be- Table 1 are not presented because the results of case 6 show the
tween 120 to 144 hours for these short time solutions. same trends and characteristics and due to space limitations. Iso-
The second stage of the numerical solution which is referred to therm and streamfunction contours have been used to quantify the
as the long time solution is characterized by changes in the drop changes in the thermal and flow fields for both the short and long
geometry and volume due to the evaporation effects. Very small time solutions. Isotherm contours are also shown for the conduc-
changes occur in the temperature and flow fields as these fields tion model based on the same conditions used for the convection
adjust to changes in the drop geometry. During this stage the model.
solution is advanced using a larger integration time step than was
used during the short time solution. The larger integration time Short Time Solution. Figure 3 shows a time sequence of the
step is carefully chosen so that it does not jeopardize the solution isotherm and streamfunction contours throughout the water drop-
accuracy and stability. A dimensionless integration time step of let for the conditions identified as Case 6 in Table 1. These con-
Fig. 4 Long time solution: Isotherm and Streamfunction contours „with velocity vectors su-
perimposed… for a 0.1 mm initial contact diameter water drop on a 100°C heated surface. Initial
contact angleÄ90 deg. Environment: T ⴥ Ä20°C, dry air: „a… 5 ms, „b… 20 ms, „c… 40 ms, and „d…
60 ms.
tours represent the evolution of the thermal and flow fields during gradients at this boundary. The changes in the droplet volume due
the initial transient period described by the short time solution. to the evaporation process were very small during the short time
The short time solution for case 6 has a total duration of 5.23 ms. solution interval.
The isotherm contours shown are for 0.5, 1.0, 1.5, and 2.0 ms The instantaneous contours displayed in this section show the
after the start of the evaporation process. During approximately relevant features that occur during the startup process. The
the initial 1.5 ms of the evaporation process, the violent startup of streamfunction contour lines exhibit a curved pattern near the con-
the thermal and flow fields cause the cold fluid in the drop to be tact line (r⫽R CL ). Also, the streamfunction contour lines are con-
entrained inside a circulation cell. The droplet surface warms very centrated near the droplet surface. This is clear evidence of the
fast due to the strong thermal convection that occurs early in the high velocity gradients that exist near the droplet surface. The
heating process. The isotherms are concentrated near the contact surface tension gradient along the droplet surface results in a tan-
surface (z⫽0) demonstrating the existence of a thermal boundary gential stress that drags the fluid layers next to the surface toward
layer near this boundary. The isotherms meet the axis of symmetry cooler regions. The streamfunction contours decrease in magni-
orthogonally due to the symmetry required at that boundary. As tude away from the center of circulation as can be seen from the
time increases 共⬎2.0 ms兲, the changes in thermal field occur at a increased spacing between contour lines. As time increases, the
slower rate and a nearly steady temperature distribution results. A velocity of the fluid throughout the drop decreases until a nearly
small amount of the total droplet volume has left due to evapora- steady velocity field is established.
tion during this time. For times greater that 2.0 ms, the isotherms
are less concentrated near the contact surface when compared to Long Time Solution. The long time solution results of the
the initial startup process. The temperatures inside the drop are evaporation process presented in this section are a continuation of
well above the initial isothermal temperature that existed at the short time solution results for the same conditions identified in
time⫽0. the previous section 共Case 6兲. For this simulation, the numerical
A circulation cell with a size of the order of the contact radius calculations were continued until a physical time of approximately
of the drop was observed in all the cases considered in this study. 72.1 ms had transpired. The contact angle predicted at this time
Since the model assumes axial symmetry about the vertical axis, was approximately 26.0 deg and the volume remaining in the drop
the circulation cell forms a toroid in three-dimensional space. was 17.6 percent of the original volume. Since the contact angle is
During the startup process, the radial and axial location of the greater than the receding angle which is approximately 10 deg for
center of circulation shifts with time. The motion of the cell center water on a metallic surface 关12兴, the contact line remained at the
decays until the flow field becomes nearly stationary. The circu- same location throughout the entire time of the simulation. The
lation cell pulsates and moves during this initial transient. This convergence of the numerical computations deteriorated beyond
pulsating behavior can be observed using computer animations of 72.1 ms. Beyond this time, the grid became very distorted. Grid
the isotherms and viewing the streamfunction contours as a func- distortion is inevitable because as the contact angle approaches
tion of time. the receding angle, the droplet resembles a thin film. Based on
During the short time solution large temperature and velocity personal experimental observations of small water droplets evapo-
gradients occurred at the contact surface with the solid. The flow rating on heated surfaces, the deformation process at these low
field was characterized by a tangential velocity distribution at the contact angles becomes non-symmetric and the symmetry as-
droplet surface while the fluid traveled from the contact line to the sumed in the model is no longer valid. Rymkiewicz and Zapalow-
droplet apex. The flow field at the axis of symmetry consisted of icz 关3兴 and Chandra et al. 关11兴 have also observed this behavior.
cooler fluid and it resembled an impingement type flow. The tem- It is possible that the circumferential component of the velocity
perature distribution inside the droplet showed very curved iso- is no longer insignificant from this point on. The numerical model
therms demonstrating the strong influence of the flow field on the assumes that the problem is axisymmetric with no circumferential
thermal process. The isotherms were concentrated near the contact motion. Experience with other numerical problems 共e.g., the flow
surface with the solid providing evidence of the large temperature over a backward facing step and the axisymmetric sudden expan-
Fig. 5 Isotherm contours comparison between the numerical solutions with fluid motion and
without fluid motion for a water drop with an initial contact diameter of 0.1 mm, 100°C heated
surface, and initial contact angleÄ90 deg. Environment: T ⴥ Ä20°C, dry air: „a… 10 ms, „b… 20
ms, „c… 40 ms, and „d… 60 ms.
sion flow兲 using the present formulation has shown that the con- motion in the thermal process. The contour lines meet the axis of
vergence deteriorates as the flow field becomes three-dimensional. symmetry orthogonally as expected. The streamfunction contours
Therefore difficulty with the convergence of the numerical com- show that only one cell remains during the time interval of the
putations may be one way of detecting a transition in the flow simulation. The center of the circulation cell moves down as the
field. Though the numerical formulation takes into account the cell size decreases and the droplet surface compresses the circu-
non-orthogonal contributions in the fluxes and on the boundary lation cell against the contact surface.
conditions, there will be some discretization error associated with
the distortion of the grids. Conduction Model. Figure 5 show instantaneous contours of
Figure 4 shows long time instantaneous isotherm and stream- the dimensionless temperature distribution in the drop based on
function contours for this case. The contours are shown at times of the results of the convection and conduction models. The isotherm
5, 20, 40, and 60 ms. These contours show the long time changes contour lines for the conduction model are almost horizontal in
in the droplet geometry and their effects on the temperature dis- contrast with the distorted isotherms that are observed in the con-
tribution in the drop. As the evaporation process occurs, the drop- vection model results. As the droplet evaporates, the contour lines
let volume decreases as demonstrated by the decrease in the drop- adjust to the new shape and the boundary conditions. The iso-
let apex height and the motion of the droplet surface. The position therms intersect the axis of symmetry orthogonally, as expected.
of the isotherms changes to satisfy the energy balance at the When these results are compared with the long time solution 共time
boundaries. No significant variation occurs in the isotherm spac- ⬎5 ms兲 of the convection model, it can be noticed that the con-
ing near the contact surface. The isotherms are very distorted near duction model underpredicts the speed of the evaporation process.
the droplet free surface indicating a strong influence of the fluid The temperature levels inside the drop predicted by the conduc-
tion model are lower in the top half of the drop throughout most
of the process compared to the predictions of the convection
model. The influence of the fluid motion produces a warmer drop-
let surface. Since the evaporative mass flux is based on the local
temperature distribution at the droplet surface, a warmer surface
will result in a larger evaporative mass flux. Therefore, the con-
vection model predicts a faster evaporation rate from the surface
of the drop as demonstrated in Fig. 5.
Droplet Volume. In Figure 6 a comparison of the normalized
droplet volume is made with the experimental data from Crafton
关29兴. The normalized droplet volume is defined as the current
droplet volume divided by the initial droplet volume. The experi-
mental data considers a water droplet with an initial contact diam-
eter of 1.05 mm and initial contact angle of 112 deg placed on an
aluminum surface at 95°C. The surroundings of the droplet were
at 21°C and a relative humidity of 30 percent. The numerical
results are in close agreement with the experimental data. The
Fig. 6 Water droplet normalized volume comparison with ex- comparison demonstrates that the normalized volume decreases in
perimental data of Crafton †29‡ a relatively linear fashion as time elapses.
Conclusions Nomenclature
A numerical study of the evaporation process of small water a ⫽ droplet apex
droplets on hot isothermal solid surfaces has been performed. The Bi ⫽ Biot number, Bi⫽h nc k/R o
present numerical study has taken into account the effects of fluid Ca ⫽ capillary number, Ca⫽ 兩 d ␴ /dT 兩 ⌬T/ ␴ o
motion on the evaporation of small droplets by modeling the ther- cp ⫽ liquid specific heat
mocapillary convection that occurs during the process. Most of Do ⫽ initial contact diameter
the experimental studies in the literature have concentrated on D vap-air ⫽ binary diffusion coefficient of the vapor species into
larger drops and temperatures above saturation for which nucleate air
boiling occurs. No other simulations in the current literature have g ⫽ acceleration of gravity
considered the internal fluid motion and the droplet geometrical
hfg ⫽ latent heat of vaporization
changes that occur during the evaporation process.
hm ⫽ mass transfer coefficient
A computational strategy consisting of modeling the evapora-
tion process in two solution stages called the short and long time h nc ⫽ natural convection heat transfer coefficient
solutions was implemented. The short time solution showed large Ja ⫽ Jacob number, Ja⫽h f g /c p ⌬T
changes in the thermal and flow fields with small changes in the k ⫽ liquid thermal conductivity
geometry due to evaporation. In addition, the short time solution k air ⫽ air thermal conductivity
showed that the temperature and flow fields become stationary Le ⫽ Lewis number, Le⫽ ␣ air /D vap-air
after the initial transient period. The long time solution showed no m ⫽ droplet mass
significant variation in the thermal and flow fields as the geometry m⬙ ⫽ mass flux across the droplet surface
changed due to the evaporation of the water drop. This computa- Ma ⫽ Marangoni number, Ma⫽ 兩 d ␴ /dT 兩 ⌬TR o / ␮␣
tional strategy of dividing the solution in a short time and a long nជ ⫽ unit normal vector to the droplet surface
time solution has been successful because of the asymptotic trend Nu ⫽ Nusselt number, Nu⫽h nc D/k air
of the thermal and flow fields during the short time solution. As a P ⫽ pressure
result, the computational strategy allowed the advance of the long Pr ⫽ Prandtl number, Pr⫽␯/␣
time solution using larger integration time steps without compro- r ⫽ radial coordinate
mising the numerical accuracy and reducing the total solution Ra ⫽ Rayleigh number, Ra⫽g ␤ ⌬TR 3o / ␯ ␣
time.
Re ⫽ Reynolds number, Re⫽UscRo /␯
The simulations using the convection model were terminated
when the contact angle approached the receding angle. The nu- Ro ⫽ initial contact radius
merical convergence deteriorated as the droplet contact angle ap- S= ⫽ viscous stress tensor
proached the receding angle value of 10 deg. There is a possibility T ⫽ temperature
that this convergence deterioration is a result of the numerical t ⫽ time
model being incompatible with the assumption of an axisymmet- t diff ⫽ thermal diffusion time scale
ric flow field and deformation process when the contact line mo- t evap ⫽ evaporation time scale
tion starts. Experimental observation has shown that the evapora- TH ⫽ heated surface temperature
tion process becomes asymmetric as the contact line motion t Ma ⫽ Marangoni time scale
occurs. The numerical convergence deterioration that occurs when ជt ⫽ unit tangential vector to the droplet surface
the contact angle approaches the receding angle seems to indicate ⌬t ⫽ physical time increment
that the axisymmetric assumption is no longer appropriate. Per- ⌬T ⫽ temperature difference scale
haps, a three-dimensional model that takes into account the cir- u, v ⫽ radial and axial velocity components
cumferential fluid motion is better suited to model the contact U sc ⫽ velocity scale
angle transition and the asymmetric deformation process that oc-
Vជ ⫽ velocity vector
curs when the contact line moves.
The conduction model results were compared with the results z ⫽ axial coordinate
of the convection model. The conduction model predicted slower ⬃ ⫽ denotes a dimensionless quantity
rates of change of the droplet volume, height, and contact angle. Greek
The conduction model predicted larger values of the volume,
␣ ⫽ liquid thermal diffusivity
height, and contact angle compared to the convection model at
␣ air ⫽ air thermal diffusivity
corresponding times. Based on these comparisons, it is obvious
that fluid motion plays a very important role in the transport pro- ␧ s v ⫽ convergence parameter for the state vector
cesses during the evaporation process of small water drops in pseudotime variation
horizontal surfaces. The fluid motion significantly affects the ther- ␪ ⫽ contact angle
mal field in the droplet, and the temperature distribution at the ␬ ⫽ local mean curvature
droplet surface determines the rate of mass loss due to evapora- ␮ ⫽ liquid dynamic viscosity
tion. A model without fluid motion underpredicts the rate of ␯ ⫽ liquid kinematic viscosity
change of the droplet mass and overpredicts the evaporation ␰, ␩, ␶ ⫽ computational space coordinates
times. ␳ vap,sat ⫽ saturated density of the vapor at the surface tempera-
The numerical results were in close agreement with experimen- ture
tal data of the normalized volume for gently deposited small water ␳ vap,⬁ ⫽ vapor density in the ambient
droplets. The close agreement with the experimental results sug- ␴ ⫽ surface tension
gests that using the Reynolds Analogy and the conduction limit Subscript
共Nu⫽2.0兲 to determine the mass transfer coefficient at the free
surface had no adverse impact on the droplet volume predictions. ⬁ ⫽ ambient condition
These results demonstrate the applicability of the numerical model air ⫽ air property
to predict the droplet shape evolution as the fluid evaporates. This I ⫽ interface
type of model can be used to predict temperature distributions and n ⫽ normal component
fluid velocities in small drops were non-invasive measuring tech- o ⫽ reference state
niques are unfeasible and very difficult to accomplish. t ⫽ tangential component
References 关14兴 Black, W. Z., and Hartley, J. G., 1996, Thermodynamics, HarperCollins Col-
lege Publishers, New York, NY.
关1兴 Liao, Y., 1992, ‘‘Dropwise Evaporative Cooling of Solid Surfaces,’’ Ph.D. 关15兴 Incropera, F. P., and DeWitt, D. P., 1990, Fundamentals of Heat and Mass
thesis, University of Maryland, College Park, MD. Transfer, John Wiley & Sons, New York.
关2兴 Zhang, N., and Yang, Wen-Jei, 1982, ‘‘Natural Convection in Evaporating
关16兴 Yang, W., 1978, ‘‘Mechanics of Droplet Evaporation on Heated Surfaces,’’
Minute Drops,’’ ASME J. Heat Transfer, 104, pp. 656 – 662.
Lett. Heat Mass Transfer 5, pp. 151–166.
关3兴 Rymkiewicz, J., and Zbigniew, Z., 1993, ‘‘Analysis of the Evaporation Process
关17兴 Kreith, F., and Black, W. Z., 1980, Basic Heat Transfer, Harper & Row, NY, p.
for Water Droplet on Flat Heated Surface,’’ Int. Commun. Heat Mass Transfer,
20, pp. 687– 697. 259.
关4兴 Yang, Wen-Jei, 1975, ‘‘Theory on Vaporization and Combustion of Liquid 关18兴 Mills, A. F., 1999, Basic Heat & Mass Transfer, Prentice Hall Inc., NJ.
Drops of Pure Substances and Binary Mixtures on Heated Surfaces,’’ Institute 关19兴 Carey, V. P., 1992, Liquid-Vapor Phase Change Phenomena—An Introduction
of Space and Aeronautical Science, University of Tokyo, Report No. 535, pp. to the Thermophysics of Vaporization and Condensation Processes in Heat
423– 455. Transfer Equipment, Hemisphere Publishing Corporation, Washington.
关5兴 Bonacina, C., Del Giudice, S., and Comini, G., 1979, ‘‘Dropwise Evapora- 关20兴 Ostrach, S., 1982, ‘‘Low-Gravity Fluid Flows,’’ Annu. Rev. Fluid Mech., 14,
tion,’’ ASME J. Heat Transfer, 101, pp. 441– 446. pp. 313–345.
关6兴 Lorenz, J. J., and Mikic, B. B., 1970, ‘‘The Effect of Thermocapillary Flow on 关21兴 Ruiz, O. E., 2000, ‘‘Numerical Analysis of the Dropwise Evaporation Pro-
Heat Transfer in Dropwise Condensation,’’ ASME J. Heat Transfer, 92, pp. cess,’’ Ph.D. thesis, Georgia Institute of Technology, Atlanta, GA.
46 –52. 关22兴 Chorin, A. J., 1967, ‘‘A Numerical Method for Solving Incompressible Viscous
关7兴 Di Marzo, M., Tartarini, P., Liao, Y., Evans, D., and Baum, H., 1993, ‘‘Evapo- Flow Problems,’’ J. Comput. Phys., 2, pp. 12–26.
rative Cooling Due to a Gently Deposited Droplet,’’ Int. J. Heat Mass Transf., 关23兴 Merkle, C. L., and Athavale, M., 1987, ‘‘Time-Accurate Unsteady Incompress-
36, pp. 4133– 4139. ible Flow Algorithm Based on Artificial Compressibility,’’ AIAA Paper 87-
关8兴 Tio, K.-K., and Sadhal, S. S., 1992, ‘‘Thermal Analysis of Droplet Spray 1137-CP.
Evaporation from a Heated Solid Surface,’’ ASME J. Heat Transfer, 114, pp. 关24兴 Rogers, S. E., Kwak, D., and Kiris, C., 1989, ‘‘Numerical Solution of the
220–226. Incompressible Navier-Stokes Equations for Steady-State and Time-Dependent
关9兴 Di Marzo, M., and Evans, D. D., 1987, ‘‘Evaporation of a Water Droplet Problems,’’ AIAA Paper 89-0463.
Deposited on a High Thermal Conductivity Solid Surface,’’ Proceedings of the
关25兴 Rogers, S. E., and Kwak, D., 1991, ‘‘An Upwind Differencing Scheme for the
24th National Heat Transfer Conference and Exhibition, Heat and Mass
Incompressible Navier-Stokes Equations,’’ Appl. Numer. Math., 8, pp. 43– 64.
Transfer in Fire, Pittsburgh, PA, pp. 11–18.
关26兴 Rogers, S. E., 1995, ‘‘A Comparison of Implicit Schemes for the Incompress-
关10兴 Qiao, Y. M., and Chandra, S., 1995, ‘‘Evaporative Cooling and Enhancement
by Addition of a Surfactant to Water Drops on a Hot Surface,’’ ASME National ible Navier-Stokes Equations with Artificial Compressibility,’’ AIAA Paper
Heat Transfer Conference, Portland, Oregon, 2, pp. 63–71. 95-0567.
关11兴 Chandra, S., Di Marzo, M., Qiao, Y. M., and Tartarini, P., 1996, ‘‘Effect of 关27兴 Vinokur, M., and Lombard, C. K., 1983, ‘‘Algebraic Grid Generation with
Liquid-Solid Contact Angle on Droplet Evaporation,’’ G Ital. Med. Lav Ergon, Corner Singularities,’’ Advances in Grid Generation, ASME FED Vol. 5, K. N.
27, pp. 141–158. Ghia and U. Ghia, eds., New York, pp. 99–106.
关12兴 Bernardin, J. D., Mudawar, I., Walsh, C. B., and Franses, E. I., 1997, ‘‘Contact 关28兴 Shih, T. I-P., Bailey, R. T., Nguyen, H. L., and Roelke, R. J., 1991, ‘‘Algebraic
Angle Temperature Dependence for Water Droplets on Practical Aluminum Grid Generation for Complex Geometries,’’ Int. J. Numer. Methods Fluids, 13,
Surfaces,’’ Int. J. Heat Mass Transf., 40, pp. 1017–1033. pp. 1–31.
关13兴 Aris, R., 1962, Vectors, Tensors, and the Basic Equations of Fluid Mechanics, 关29兴 Crafton, E. F., 2001, ‘‘Measurements of the Evaporation Rates of Heated Liq-
Prentice-Hall, Inc., Englewood Cliffs, NJ. uid Droplets,’’ M.S. thesis, Georgia Institute of Technology, Atlanta, GA.
A Cavity Activation and Bubble
Growth Model of the Leidenfrost
John D. Bernardin
Point
Assoc. Mem. ASME This study presents a new mechanistic model of the Leidenfrost point (LFP); the minimum
liquid/solid interface temperature required to support film boiling on a smooth surface.
Issam Mudawar The model is structured around bubble nucleation, growth, and merging criteria, as well
e-mail: mudawar@ecr.purdue.edu as surface cavity size characterization. It is postulated that for liquid/solid interface
Fellow ASME temperatures at and above the LFP, a sufficient number of cavities (about 20 percent) are
activated and the bubble growth rates are sufficiently fast that a continuous vapor layer is
Boiling and Two-Phase Flow Laboratory, established nearly instantaneously between the liquid and the solid. The model is appli-
School of Mechanical Engineering, cable to both pools of liquid and sessile droplets. The effect of surface cavity distribution
Purdue University, on the LFP predicted by the model is verified for boiling on aluminum, nickel and silver
West Lafayette, IN 47907 surfaces, as well as on a liquid gallium surface. The model exhibits good agreement with
experimental sessile droplet data for water, FC-72, and acetone. While the model was
developed for smooth surfaces on which the roughness asperities are of the same magni-
tude as the cavity radii (0.1–1.0 ␮m), it is capable of predicting the boundary or limiting
Leidenfrost temperature for rougher surfaces with good accuracy.
关DOI: 10.1115/1.1470487兴
Keywords: Boiling, Bubble Growth, Cavities, Heat Transfer
1 Introduction formed, which included data for acetone, benzene, FC-72, and
water on heated aluminum surfaces with various surface finishes.
The competitive demands of industry for products with en-
hanced material properties which can be manufactured more effi- The same study also explored the effects of surface material 共cop-
ciently and with greater cost effectiveness are continually shaping per, nickel, silver, and aluminum兲, liquid subcooling, liquid degas-
and advancing processing techniques. For example, processing of sing, surface roughness, and chemical residue on the Leidenfrost
aluminum alloys has received considerable attention from the au- point. In addition, several Leidenfrost point models developed
tomobile and aerospace industries because of such attributes as over the past four decades were reviewed and assessed. These
high strength-to-weight ratio, corrosion resistance, and recyclabil- models include hypotheses adopted from different disciplines,
ity. However, the replacement of steel components with aluminum such as hydrodynamic instability theory, metastable physics, ther-
alloy counterparts has been restricted, in part, by limited knowl- modynamics, and surface chemistry. Table 6 of Ref. 关2兴 reveals
edge of the heat transfer aspects associated with quenching of these models fail to accurately and consistently predict the
extrusions, castings, forgings, and other heat treated parts. Leidenfrost temperature for sessile droplets.
Quenching involves rapid cooling of a part to control micro- The model proposed and verified in this paper is based on sur-
structural development and hence dictates material properties 关1兴. face cavity size characterization as well as bubble nucleation,
Figure 1 shows a typical temperature-time history of an aluminum growth, and merging criteria. It is consistent with the relationship
part during a quench. The curve is divided into four distinct re- between surface cavities and the boiling phenomena reported in
gimes, each possessing unique heat transfer characteristics. In the early nucleate boiling literature. In these studies, scanning elec-
high temperature, or film boiling regime, the quench proceeds tron microscopy identified micron-sized cavities on macroscopi-
rather slowly as liquid-solid contact is prevented by the formation cally polished surfaces and high-speed photography recorded
of an insulating vapor blanket. The lower temperature boundary of bubble formation speculated to originate from vapor trapped
this regime is referred to as the Leidenfrost point 共LFP兲, below within these cavities 关3–5兴. In conjunction with these early obser-
which partial liquid-solid contact increases cooling rate. As dis- vations, a bubble incipience model to predict the surface superheat
cussed by Bernardin and Mudawar 关1兴, most of the material trans- required to form vapor bubbles from surface cavities was devel-
formations for aluminum alloys occur at temperatures above the oped 关6 –9兴. The focus of continuing investigations was to corre-
LFP, while warping and distortions are caused by thermal stresses late the observed heat flux and superheat characteristics to the
resulting from the large cooling rates at temperatures below the number density of active cavities 关4,5,10,11兴. It has been well
LFP. Consequently, accurate knowledge of the Leidenfrost tem- documented in these studies that as the surface superheat is in-
perature and the parameters which govern its behavior is para- creased, the number of active nucleation sites increases up to
mount to controlling the quenching process and subsequent mate- some maximum point at which bubble coalescence occurs and a
rial properties. It must be emphasized that boiling is an interfacial vapor blanket begins to develop, covering nearly 50 percent of the
phenomenon. Consequently, the Leidenfrost temperature corre- surface at critical heat flux 关4兴.
sponds to that of the liquid-solid interface at the LFP, which may
In the present study, it is hypothesized that as the Leidenfrost
differ significantly from temperatures within the solid.
temperature is approached from the boiling incipience tempera-
In a previous study by the authors 关2兴, a fairly comprehensive
ture, smaller and more numerous surface cavities become acti-
experimental assessment of the Leidenfrost phenomenon was per-
vated, and the growth rate of these bubbles increases appreciably.
For liquid-solid interface temperatures at and above the LFP, a
HEAT TRANSFER. Manuscript received by the Heat Transfer Division June 30, 2001; sufficient number of cavities are activated and the bubble growth
revision received January 7, 2002. Associate Editor: T. Y. Chu. rates are large enough that liquid in immediate vicinity of the
surface is nearly instantaneously converted to vapor upon con-
tact, enabling a continuous insulating vapor layer to form between
the liquid and the solid.
To help explain the rationale of the proposed LFP model, Fig. 2
displays the sessile droplet evaporation time versus wall super-
heat, exhibiting the four distinct boiling regimes indicated in the
cooling curve of Fig. 1. In this example, the interface temperature
refers to that at which the liquid and the solid come in contact.
Included in Fig. 2 are photographs depicting the vapor layer de-
velopment for sessile water droplets approximately 2 ms after
making contact with a polished aluminum surface at four interface
temperatures. At an interface temperature of T i ⫽137°C, corre-
sponding to the transition boiling regime, individual vapor
bubbles occupying roughly 15 percent of the liquid-solid contact
area are visible throughout the liquid film. At 151°C, approaching
the LFP, the bubble density increases significantly, covering
nearly 50 percent of the droplet underside. As the interface tem-
perature increases further, bubble density also increases, signaling
the formation of a continuous vapor layer beneath the droplet. The
LFP corresponds to the minimum liquid-solid interface tempera-
ture required to sustain a continuous vapor layer, as suggested in
Fig. 2 for T i ⫽165°C. At and above the LFP, the vapor layer
beneath the droplet allows surface tension forces in the liquid to
reduce the droplet’s outer diameter noticeably as shown in Fig. 2
for T i ⫽180°C.
To fully appreciate the proposed model, the structure of solid
surfaces and its influence on the boiling process must first be
explored.
Characterization of Surfaces. A typical surface is made up
of many imperfections, including pits, scratches, and bumps. In
addition, there is no definite region on a particular length scale
which can be considered as roughness or waviness. As mentioned
Fig. 1 Temperature-time history of a surface during quenching by Ward 关12兴, a surface generally exhibits self-similarity, meaning
in a bath of liquid that its appearance remains basically the same over a wide range
of magnifications. This behavior is described in Fig. 3共a兲, in
which a hypothetical surface profile is shown at various magnifi-
cations. Surface roughness features appear to repeat themselves as
the magnification is increased again and again. Based on this type
Fig. 2 Sessile droplet evaporation curve and corresponding photographs of water droplets approximately 2
ms after contact with a polished aluminum surface
Fig. 3 Depiction of „a… an actual surface profile exhibiting self-similarity and the correspond-
ing cavity size distribution, „b… sensitivity limitation of a stylus of a surface contact profilome-
ter, and „c… a polished aluminum surface profile „with an arithmetic average surface roughness
of 26 nm… measured with a contact profilometer and the corresponding cavity size distribution
of surface description, it is reasonable to conclude that the surface of 1 to 10 ␮m. Thus, the resulting surface cavity distribution
consists of relatively large craters, which are filled with many measured with such an instrument would erroneously exhibit the
smaller cavities, and so on 关11兴. Consequently, the surface cavity characteristic dome-shape of Fig. 3共c兲 commonly employed in
sizes would be expected to fit an exponential distribution as boiling heat transfer literature. Cavities which serve as bubble
shown in Fig. 3共a兲. nucleation sites 共0.1 to 5 ␮m兲 would not be detected with a sur-
Several techniques, including surface contact profilometry, face contact profilometer 关2,11兴.
scanning electron microscopy, and various optical, electrical, and Scanning electron microscopy images 共SEMS兲 help identify
fluid methods, are available for assessing surface features 关13兴. As and characterize the cavities which serve as nucleation sites. Fig-
mentioned by Ward 关12兴, each of these techniques is bandwidth ure 4 displays surface cavity distributions for a polished alumi-
limited, meaning it can only resolve surface features of a certain num surface determined from SEMS at two different magnifica-
size interval. Figures 3共b兲 and 3共c兲 describe this limitation for a tions. Each distribution is limited by the SEM magnification.
surface contact profilometer, showing how at higher magnifica- However, by combining the distributions from both SEMS, a
tions the surface appears to get smoother as fine surface features complete cavity size distribution covering the range responsible
are no longer detected. This resolution limit is the result of the for bubble nucleation of common fluids is obtained.
physical size of the diamond stylus, typically having a tip radius In the past, a number of investigations were performed to char-
Fig. 4 Cavity size distributions for a polished aluminum surface determined from scanning electron microscopy
images at „a… 1000Ãmagnification, „b… 4800Ãmagnification, and „c… combined magnifications
acterize surface features responsible for nucleate boiling to derive the assumed conical-shaped cavities兲. Later, Wang and Dhir 关18兴
quantitative relationships for nc a , the number of surface cavities determined the relationship between the number of surface cavi-
which are activated 关11,14 –16兴. More recently, Yang and Kim ties per unit area, nc, and those that become activated, nc a , for
关17兴 determined nc a using scanning electron and differential in- water boiling on a polished copper surface. They also confirmed
terference microscope data along with the vapor entrapment cri- Gaertner’s 关19兴 statistical spatial distribution of active cavities.
terion 共contact angle, ␪, must be greater that the cone angle, ␾, of Using a Poisson distribution function, Gaertner derived the fol-
lowing expressions for the statistical distribution, f (d), and aver- heated surface. For a relatively short duration over which rapid
age, d̄, of nearest-neighbor distances, d, between cavities bubble nucleation occurs at high superheats, the contact between
the liquid and solid can be modeled as one-dimensional transient
f 共 d 兲 ⫽2 ␲ nc a d exp共 nc a ␲ d 2 兲 (1) conduction between two semi-infinite bodies. The transient tem-
perature distribution in the liquid, or the available superheat,
0.50
d̄⫽ . (2) T ash(y,t), is given by 关20兴
冉冊
冑nc a
y
The remainder of this paper presents the development of the T ash⫽T i ⫹ 共 T f ⫺T i 兲 erf . (6)
new theoretically-based LFP model. Experimental data and nu- 2 冑␣ f t
merical schemes required to solve the model equations are in- where the interface temperature, T i , is
cluded and discussed. Finally, comparisons are made between the
model predictions and empirical data to demonstrate both the ro- 共 k ␳ c p 兲 1/2
s T s⫹共 k ␳ c p 兲 f T f
1/2
bustness and accuracy of this model. T i⫽ . (7)

共 k ␳ c p 兲 1/2
s ⫹共 k␳c p兲 f
1/2
2 LFP Model Development y is the distance into the liquid measured normal to the liquid-
The methodology used to construct the present LFP model re- solid interface, and T s and T f are, respectively, the surface and
lies upon two important aspects concerning bubble nucleation and liquid temperatures prior to the contact.
its relationship to surface temperature and cavity shape and distri- The minimum condition necessary for bubble nucleation is met
bution. First, based upon earlier bubble nucleation criteria 关6 –9兴, when the available superheat at a distance y from the solid sur-
increasing surface superheat beyond the boiling incipience tem- face, is equal to the required superheat for a hemispherical bubble
perature causes both larger and smaller surface cavities to activate whose radius, r, is equal to y. This condition is represented by
and bubble growth rates to increase. Secondly, for a typical pol- equating the required and available superheats from Eqs. 共5兲 and
ished surface, there is an exponential increase in the number of 共6兲, respectively:
surface cavities with decreasing cavity mouth radius 关17,18兴, as
shown in Fig. 3.
In the present study, the authors postulate that at some large
T sat exp 冉冊
2␴v fg
rh f g
⫽T i ⫹ 共 T f ⫺T i 兲 erf
r
冉冊
2 冑␣ f t
. (8)
liquid-solid interface temperature corresponding to the LFP, a suf-

ficient number of cavities will activate to produce enough vapor to Cavity Size Distribution. Surface cavities and other defects,
separate the liquid from the solid, and hence, induce film boiling. typically on the order of 1 to 10 ␮m, have long been known to be
Discussed below are the various sub-models used to support the highly influential in controlling nucleate boiling by serving as
overall LFP model. In the next section, a solution procedure based nucleation sites. In this study, scanning electron microscopy
upon these sub-models is outlined. 共SEM兲 was utilized to characterize the surface cavity distributions
of macroscopically polished surfaces from which empirical
Bubble Nucleation. The criteria for bubble nucleation from Leidenfrost temperature measurements were made 关2兴. NIH Im-
cavities have been rigorously investigated theoretically and veri- age, an image processing and analysis program for the Macintosh
fied experimentally by many researchers. The significant aspects was utilized in conjunction with digitized SEM images of the
of the bubble nucleation model are outlined below. In the devel- polished surfaces to determine the number and sizes of the surface
opment that follows, it is assumed that the surface cavities are cavities. The program determined the mouth area of each irregular
conical. cavity, and then calculated an equivalent circular cavity mouth
The pressure drop across a spherical bubble interface of radius radius which would provide an equal mouth area.
r is given as From inspection of various SEM images at different magnifica-
tions, it was apparent that the number of cavities per unit area, n,
2␴ having an equivalent mouth radius between r and r⫹⌬r, could be
P g⫺ P f ⫽ . (3)
r fit by the exponential function
A relation for the liquid superheat, ⌬T sat , required to provide n⫽a 1 exp共 ⫺a 2 r 兲 . (9)
the necessary gas pressure, P g , for initiation of bubble growth
can be found by integrating the Clausius-Clapeyron equation Using the scanning electron microscopy images of the various
along the saturation line, surfaces used in this study, the following curve fits were obtained
over a cavity size range of 0.07 to 1.0 ␮m:
冕 P g,sat
P f ,sat
d P⫽ 冕 T sat⫹⌬T sat
␶ sat
hfg
Tv fg
dT. (4) n⫽3.379 exp共 ⫺10.12r 兲共 aluminum兲 (10a)
n⫽4.597 exp共 ⫺12.20r 兲共 nickel兲 (10b)
By substituting Eq. 共3兲 for the pressure difference ( P g,sat
⫺ P f ,sat) and holding the latent heat of vaporization and specific n⫽13.16 exp共 ⫺16.07r 兲共 silver兲 , (10c)
volume difference constant, Eq. 共4兲 can be integrated to give the ⫺2 ⫺1
following expression for the surface superheat temperature re- where the units for n and r are sites.␮m .␮m and ␮m, respec-
quired to initiate the growth of a hemispherical vapor bubble of tively. The curve fits had acceptable least square residuals with a
radius r, worst case value of 0.87.
冉冊
The cumulative number of surface cavities in the radius interval
2␴v fg r min⭐r⭐rmax , is then obtained through integration,
T rsh⫽T sat exp . (5)
rh f g
By assuming constant values for h f g and v f g 共evaluated at the nc⫽ 冕 r max
r min
n 共 r 兲 dr⫽
a1
a2
关 exp共 ⫺a 2 r min兲 ⫺exp共 ⫺a 2 r max兲兴 .
mean temperature (T rsh⫹T sat)/2兲, integrating Eq. 共4兲 produces a (11)
difference of only 6 percent 共for water and a surface superheat
temperature of 190 °C兲 from the results obtained by substituting Bubble Growth. Due to the relatively high superheat and
temperature-dependent properties and performing the more com- short duration over which vapor is created in the film boiling
plicated integration. regime, it is believed the rapid bubble growth will be initially
The superheat available for bubble nucleation is provided by dominated by inertia rather than heat diffusion. For this condition,
the transient heat diffusion following contact of the liquid with the bubble growth is described by the Rayleigh equation 共neglecting
Fig. 5 Temperature dependence of vapor bubble growth for
water as predicted by the numerical solution to the Rayleigh
equation
viscous effects兲 which can be derived from the momentum equa-

tion for incompressible and irrotational flow 关21兴, or from energy
conservation principles 关22兴, incorporating the pressure drop
across a spherical interface, 2 ␴ /R.
RR̈⫹
3 2 1
2
Ṙ ⫽
␳f 冋
共 P g⫺ P ⬁ 兲⫺
2␴
R
, 册 (12)
where Ṙ and R̈ are, respectively, the first and second derivatives

of bubble radius with respect to time, and P ⬁ is the liquid pressure
far from the bubble interface. While no analytical solution to the
Rayleigh equation exists, an asymptotic solution (RⰇR o ) which
neglects surface tension forces, has been obtained 关22,23兴.
However, for early stages of bubble growth, surface tension
forces cannot be neglected. Consequently, Eq. 共12兲 must be solved Fig. 6 „a… Transient maximum cavity activation and bubble ra-
to accurately describe these early stages of bubble growth. In dius and „b… nearest-neighbor cavity distances for 25 percent
solving the differential Rayleigh equation, the following interme- cavity activation at three different times following liquid-solid
contact for water on a polished aluminum surface with an in-
diate step
冉冊冋册
terface temperature of 145°C
1 2␴ 1 ⌬ P R3 ␴R2
d 共 R 3/2Ṙ 兲 ⫽ ⌬ P⫺ R 1/2dt⫽ d ⫺
␳f R R 3/2Ṙ ␳f 3 ␳f tion is several orders of magnitude faster than that of the thermal
(13) boundary layer. Therefore, it is assumed the early stage of bubble
was used in the present study to reduce Eq. 共12兲 to the following growth is described by the solution to the Rayleigh Eq. 共14兲 until
integral the bubble dome reaches the maximum bubble stability point in
the growing thermal boundary layer predicted by Eq. 共8兲, after
冕 R dR which the bubble growth is controlled by this slower diffusion rate
冋册
t⫽ (14)
0 2 2␴ 0.5 of the thermal boundary layer. This two-staged bubble growth
共 P ⫺ P⬁兲⫺ model is fairly consistent with the numerical bubble growth model
3␳ f g ␳fR
of Lee and Merte 关24兴, which predicts a rapid inertia-controlled
Figure 5 displays the numerically predicted temperature depen- bubble growth that converges to a much slower thermally con-
dence for bubble growth for surface temperatures corresponding trolled growth as the bubble expands.
to nucleate, transition, and film boiling of water. As expected, the At interface temperatures well above the boiling point of the
growth rate increases appreciably with increasing surface tem- liquid, the number of active surface cavities and the bubble
perature. growth rates can become significantly large that bubble interfer-
The bubble growth predicted by Eq. 共14兲 is similar to the rapid ence begins to take place. Figure 6 describes this interference for
inertia-controlled growth examined numerically and analytically water in contact with a polished aluminum surface with an inter-
by Lee and Merte 关24兴 and Bankoff and Mikesell 关26兴, respec- face temperature of 145°C. Figure 6共a兲 displays the transient
tively. The bubble growth rates displayed in Fig. 5 are also very maximum stable bubble radius supported by the growing thermal
similar to data for bubble growth in superheated water presented boundary layer as predicted by Eq. 共8兲. Figure 6共b兲 shows the
in the same references. The hemispherical bubble geometry nearest-neighbor distance distribution given by Eq. 共1兲 and the
adopted in the present model for the inertia-controlled growth is transient cumulative cavity density for a polished aluminum sur-
consistent with the description given by Carey 关27兴. face 100, 1000, and 2000 ␮s following liquid-solid contact. In
Fig. 6共b兲, it is assumed that only 25 percent of the surface cavities
Interaction of the Thermal Boundary Layer and the Grow- satisfy the vapor entrapment criterion and serve as nucleation
ing Bubbles. As will be shown in the next section, the bubble sites, an estimate consistent with the findings of Yang and Kim
growth predicted by the numerical solution to the Rayleigh equa- 关17兴 and Wang and Dhir 关18兴.
Fig. 7 Schematic representation of different forms of cavity cancellation: „a… poor vapor entrapment, „b… neighbor bubble
overgrowth, and „c… bubble merging
In comparing Figs. 6共a兲 and 6共b兲, it is apparent that the bubble a polished surface, this radius is typically well within the range of
radius exceeds half of the nearest-neighbor distance for a large cavity radii available on the surface. As time progresses and ther-
percentage of the active cavities, a condition which is necessary mal boundary layer thickens, all cavities within a specific cavity
for two bubbles to interfere with one another. This trend increases radius interval are activated. This interval is given by the two
significantly as time progresses. However, not all cavities will roots of Eq. 共8兲, namely, r min(t) and r max(t), as displayed in Fig.
participate equally in this process. Because of several cancellation 8共b兲, where r max is the radius of the largest activated cavity at a
effects discussed below, only a small fraction of the cavities will given instant, not the largest cavity on the surface. Similarly, r min
activate to form bubbles which grow until they begin to interfere is the smallest activated cavity.
with bubbles from neighboring cavities. Table 1 presents curve fits for r min(t) and r max(t) obtained by
Figure 7 is a schematic representation of different forms of solving Eq. 共8兲 over a 2 ms time interval for a variety of fluids and
surface cavity cancellation which occur before or during the de- liquid-solid interface temperatures. These curve fits were used in
velopment of the vapor layer. Figure 7共a兲 displays the vapor en- the remainder of the LFP model calculations to determine the
trapment mechanism for conical surface cavities. When the liquid Leidenfrost temperature for sessile droplets.
initially makes contact with the solid surface, only those cavities Assuming only a fraction, ␺, of the cavities actively participate
with a cone angle smaller than the advancing contact angle will in the growth of the vapor layer due to the cancellation effects
entrap vapor and serve as bubble nucleation sites 关25兴. Two other described in the previous section, and that bubbles grow from
types of cavity cancellation occur during bubble growth from cavities as hemispheres, the time dependence of the cumulative
nucleation sites. As illustrated in Fig. 7共b兲, a bubble from an ac- number of activated cavities per unit area can be found by inte-
tivated cavity can overgrow a non-activated, vapor entrapped cav- grating the cavity size distribution over the radius limits r min(t)
ity, thus canceling it out as a nucleation site. Two growing bubbles and r max(t):
冕
may collide and merge as depicted in Fig. 7共c兲. In this case, the r max共 t 兲
bubbles, represented as hemispheres for simplicity, may form a nc a 共 t 兲 ⫽ ␺ a 1 exp共 ⫺a 2 r 兲 dr
single larger bubble which extends beyond the stability limit of r min共 t 兲
the thermal boundary layer, causing condensation and bubble
shrinkage to temporarily occur. The net effect is the cancellation a1
of an active bubble source by bubble merging. ⫽␺ 兵 exp共 ⫺a 2 r min共 t 兲兲 ⫺exp共 ⫺a 2 r max共 t 兲兲其 . (15)
a2
It should be noted that the LFP model described in this paper is
applicable to sessile as well as impinging droplets. In addition, Since the inertia-controlled bubble growth rate predicted by Eq.
since the model is constructed around cavity activation and bubble 共14兲 is orders of magnitude greater than the thermal boundary
growth arguments, and not on the expanse of the surrounding layer growth rate, it is assumed all bubbles initiated with r o
liquid, it should be applicable to pool boiling as well. ⬍r max(t) will rapidly grow to r max(t), the maximum stable hemi-
spherical bubble radius supported by the growing thermal bound-
ary layer. A hemispherical bubble will not be stable for sizes be-
3 LFP Model Solution Procedure yond r max(t) as condensation on the leading front of the growing
Upon contact between a sessile droplet and a heated surface, a bubble would significantly reduce its growth rate 关26兴. This is
thermal boundary layer begins to develop in the liquid. At some consistent with bubble incipience model of Hsu 关7兴 and the ex-
time t o , the thermal boundary layer would have grown suffi- perimental results of Clark et al. 关3兴. Consequently, the limiting
ciently large to satisfy the bubble nucleation criterion for conical- condition considered here is that once the bubbles reach the ther-
shaped cavities with a mouth radius r o as shown in Fig. 8共a兲. For mal boundary layer limit of r max(t) they will continue to grow at
Table 1 Active cavity radii equations for sessile droplets of
various liquids in contact with a hot surface for different inter-
face temperatures
tent with the 14.5 percent value for T i ⫽137°C in Fig. 2 deter-
mined with the image analysis software. A value of 0.05 for ␺
indicates that even if 25 percent of the cavities satisfy the vapor
entrapment criterion, as discussed previously and illustrated in
Fig. 6, only 20 percent of these nucleation sites actively partici-
pate in the vapor layer growth; the remaining sites being canceled
out by the effects illustrated in Fig. 7. Based on this comparison,
a value of 0.05 for ␺ was used consistently for all subsequent
calculations presented in this study. Due to the complex shapes of
surface features and the limited means of analyzing these shapes,
Fig. 8 Transient cavity nucleation model including „a… cavity it is extremely difficult to characterize the surface cavities which
nucleation superheat criteria and corresponding cavity size serve as potential nucleation sites. In addition, the contact angle
distribution with transient activation window, and „b… transient
used for the vapor entrapment criterion is highly dependent on the
maximum and minimum active cavity radii for water in contact
with a hot surface with an interface temperature of 165°C spreading velocity of the liquid, surface contamination, as well as
surface roughness 关28兴. Since the present models for surface char-
acterization and bubble nucleation are limited in their degree of
accuracy, a more accurate means of determining the percent of
the same rate as the thermal boundary layer, i.e., ṙ max(t). This actively participating surface cavities, ␺, is currently unavailable
two-stage growth is consistent with the bubble growth findings of and warrants further investigation. Nevertheless, it should be em-
Lee and Merte 关24兴. phasized that while the choice of ␺ will influence the vapor layer
Given this bubble growth model, the time-dependent percent growth rate, the strong temperature-dependence of the latter,
area coverage of the liquid-solid interface by vapor, AB%(t), is ⌬AB%/⌬t, which is used to identify the LFP in the present
then given by model, is still very well preserved.
AB% 共 t 兲 ⫽nc a 共 t 兲 ␲ r 2max共 t 兲 (16) 4 LFP Model Assessment

which, upon substitution of Eq. 共15兲, gives Figure 9共a兲 shows the temperature dependence of the transient
a1 vapor layer growth for a sessile water droplet on a polished alu-
AB% 共 t 兲 ⫽ ␺ 兵 exp共 ⫺a 2 r min共 t 兲兲 minum surface with the cavity distribution given by Eq. 共10a兲.
a2 The time for complete vapor layer development (AB%⫽100) is
shown to rapidly decrease as the interface temperature is in-
⫺exp共 ⫺a 2 r max共 t 兲兲其 ␲ r 2max共 t 兲 . (17)
creased from 145 to 185°C. While the model predicts an eventual
To determine the cavity cancellation parameter, ␺, in Eq. 共17兲, 100 percent vapor layer growth for the interface temperature of
the vapor layer development in Fig. 2 corresponding to 137°C was 145°C, other effects such as bubble departure and liquid motion
compared to vapor layer predictions of the LFP model. A value of which are not accounted for in the model, would interrupt this
0.05 for ␺ resulted in a 15 percent vapor layer coverage, consis- development within a few milliseconds of liquid-solid contact,
Fig. 9 Temperature dependence of the „a… transient vapor
layer coverage and „b… average vapor layer growth rate for a
sessile water droplet on a polished aluminum surface
and hence prevent film boiling from occurring. Figure 9共b兲 pre-
sents the trend observed in Fig. 9共a兲 in a slightly different manner.
Figure 9共b兲 shows that as the interface temperature increases be- Fig. 10 Average vapor layer growth rate for sessile droplets of
„a… water on various polished metallic surfaces and „b… ac-
yond the liquid saturation temperature, the average vapor layer etone, FC-72, and water on polished aluminum
growth rate will increase exponentially. Intuition suggests that at
some minimum interface temperature, the LFP, the average vapor
layer growth rate will become sufficiently high to support film
boiling. To determine the minimum average vapor layer growth
rate required to support film boiling, experimental LFP data for
sessile water droplets on aluminum were employed. The water- layer growth rate versus interface temperature for sessile water
aluminum system was used earlier in the model development to droplets on various metallic surfaces and sessile acetone, FC-72,
determine the percentage of actively participating surface cavities and water droplets on aluminum. Using these plots and an average
and this system was also highly scrutinized in an experimental vapor layer growth rate of 0.05, the LFP was determined for each
study of the LFP 关2兴. Shown in Fig. 9共b兲 is the experimentally of these fluid-solid systems. These LFP model predictions are
determined Leidenfrost temperature of 162°C (T s ⫽170°C) for compared to measured values 关2兴 in Table 2. Excellent agreement
sessile water droplets on aluminum which corresponds to an av- is obtained for all cases except acetone, where the agreement is
erage vapor layer growth rate of 0.05. This value of the average only fair. Even so, these results are quite promising considering
vapor layer growth rate was used throughout the LFP model as- the large differences in the wetting characteristics as well as the
sessment of different liquid-solid systems. thermodynamic and thermal properties of the fluids and solids
This same technique, as described by Carey 关27兴, has been used used in the comparison. In addition, the differences between the
to determine the critical vapor bubble formation rate needed to LFP predictions and present measurements are significantly
sustain homogeneous nucleation within a superheated liquid. In smaller than the majority of previous LFP predictive tools pre-
the homogeneous nucleation superheat limit model, the vapor sented in 关2兴. The difference exhibited for acetone on aluminum
bubble formation rate increases exponentially with increasing liq- may be due to the limitation imposed on the number of actively
uid temperature, much like the vapor blanket growth rate in the participating cavities, accounted for in the parameter ␺ in Eq.
present study. Carey explains how empirical data were used to 共17兲. As discussed earlier, the existing techniques for modeling
determine a critical vapor bubble formation rate, and how this vapor entrapment and bubble nucleation as well as characteriza-
single bubble formation rate was used to determine the homoge- tion of surface cavities are limited and warrant continued study.
neous nucleation superheat limit of several different liquids in- Adopting newer techniques may lead to a more accurate means of
cluding water. determining the number of actively participating cavities and
Figures 10共a兲 and 10共b兲 display, respectively, the average vapor hence further strengthen the present LFP model.
Table 2 Comparison of measured Leidenfrost temperatures ization. After evaluating the model with an extensive experimental
†2‡ and predictions based on the present LFP model for ac- database, the following key conclusions can be drawn about its
etone, FC-72, and water on various polished metallic surfaces validity and use:
1. The number of surface cavities which act as bubble nucle-
ation sites increases exponentially with increasing liquid-
solid interface temperature.
2. Bubble growth rates predicted by the solution to the Raleigh
equation are several orders of magnitude greater than the
growth rate of the thermal boundary layer for conditions
consistent with film boiling of common fluids. Conse-
quently, the bubbles emanating from active surface cavities
grow instantaneously to the maximum allowable radius as
predicted by bubble nucleation theory, and thereafter, grow
at the rate of the diffusing thermal boundary layer.
3. For interface temperatures at and above the LFP, the present
model predicts the number of active sites and bubble growth
rates are large enough that a complete vapor layer is estab-
lished between the liquid and solid almost instantaneously
upon contact.
4. The present model is substantiated by a large experimental
data base for sessile droplets, provided the surface roughness
features are on the same order of magnitude as the cavities
responsible for bubble nucleation. For rougher surfaces, the
model predicts a lower bound for the sessile droplets.
Further Justifications of the LFP Model and Application to
Rough Surfaces. For a perfectly smooth surface which is void
of all surface cavities, the current LFP model predicts that the Acknowledgment
liquid can be heated to an infinitely high temperature and film
The authors gratefully acknowledge the support of the Office of
boiling would never be reached. Realistically, the maximum tem-
Basic Energy Sciences of the U.S. Department of Energy 共Grant
perature that the liquid can be heated to, above which it is instan-
No. DE-FE02-93ER14394.A003兲.
taneously converted to vapor, is referred to as the kinetic or ther-
modynamic superheat limit. Methods to predict this superheat
limit, which is well above the Leidenfrost temperature of the Nomenclature
liquid-solid systems presented earlier in this study, can be found a 1 , a 2 ⫽ coefficients in cavity size distribution
elsewhere 关29兴. AB % ⫽ percent liquid-solid interface area coverage by
Gallium, a liquid metal with a melting point of 29.8°C and vapor
density of 5900 kg.m⫺3, was used in the present study to provide c p ⫽ specific heat at constant pressure
a smooth liquid surface nearly free of defects. Using a thermal d ⫽ nearest-neighbor cavity distance
monitoring system consisting of a temperature controller, car-
d̄ ⫽ average nearest-neighbor cavity distance
tridge heater, and thermocouple, sessile droplet evaporation ex-
f (d) ⫽ nearest-neighbor cavity distance distribution
periments, similar to those described in 关2兴, were performed to
h f g ⫽ latent heat of vaporization
determine the LFP of water on liquid gallium. Results revealed a
k ⫽ thermal conductivity
gallium temperature of 260°C was needed to provide a water/
n ⫽ number of surface cavities per unit area per unit
gallium interface temperature of 222°C corresponding to the LFP.
interval 共sites ␮m⫺2 ␮m⫺1兲
Impurities in the gallium caused by oxidation and contaminant
nc ⫽ cumulative number of surface cavities per unit
metals were speculated to provide a few heterogeneous nucleation
area 共sites ␮m⫺2兲
sites which prevented the water from obtaining its maximum su-
nc a ⫽ cumulative number of active surface cavities per
perheat temperature limit of 273°C predicted using the thermody-
unit area 共sites ␮m⫺2兲
namic homogeneous nucleation model 关30兴, or 310°C according
P ⫽ pressure
to the kinetic homogeneous nucleation model 关27兴. However, the
R ⫽ bubble radius
results do indicate that a dramatic reduction in surface cavities
greatly increases the Leidenfrost temperature, which is in agree- Ṙ ⫽ first derivative of bubble radius with respect to
ment with the present LFP model. time
While this model was developed for polished surfaces, it also R̈ ⫽ second derivative of bubble radius with respect
provides a limiting condition for surfaces possessing roughness to time
features orders of magnitude larger than the cavity radii respon- r ⫽ surface cavity radius
sible for bubble nucleation 共0.1 to 1 ␮m兲. The model effectively r a ⫽ radius of active surface cavity 共␮m兲
predicts a lower limit to the Leidenfrost temperature for sessile T ⫽ temperature
droplets and pools of liquid. Contamination and surface roughness t ⫽ time
will act to increase the Leidenfrost temperature by requiring a v f g ⫽ specific volume difference between vapor and
thicker vapor layer to inhibit liquid-solid contact. This is sup- liquid
ported by experimental data for sessile droplets of different liquids y ⫽ normal distance from solid surface
on surfaces of various roughnesses 关2兴. Greek Symbols
␣ ⫽ thermal diffusivity
⌬AB %/⌬t ⫽ average vapor layer growth rate
5 Conclusions ⌬T sat ⫽ surface superheat, T s ⫺T sat
This study presented a new theoretically based LFP model ␾ ⫽ cavity cone angle
which was constructed around vapor bubble nucleation, growth, ␪ ⫽ contact angle
and interference criteria, along with surface cavity size character- ␳ ⫽ density
␴ ⫽ surface tension 关11兴 Cornwell, K., and Brown, R. D., 1978, ‘‘Boiling Surface Topography,’’ Proc.
␺ ⫽ fraction of actively participating cavities Sixth Int. Heat Transfer Conf., 1, Toronto, Canada, pp. 157–161.
关12兴 Ward, H. C., 1982, ‘‘Profile Description,’’ in Rough Surfaces T. R. Thomas,
Subscripts ed., Longman Group, New York, pp. 72–90.
关13兴 Thomas, T. R., 1982, ‘‘Stylus Instruments,’’ in Rough Surfaces T. R. Thomas,
a ⫽ active ed., Longman Group, New York, pp. 12–70.
ash ⫽ available superheat 关14兴 Brown, W. T., Jr., 1967, ‘‘Study of Flow Surface Boiling,’’ Ph.D. thesis, M.I.T.,
f ⫽ liquid Cambridge, MA.
关15兴 Mikic, B. B., and Rohsenow, W. M., 1969, ‘‘A New Correlation of Pool-
g ⫽ vapor Boiling Data Including the Effect of Heating Surface Characteristics,’’ ASME
i ⫽ liquid-solid interface J. Heat Transfer, 91, pp. 245–250.
leid ⫽ Leidenfrost condition 关16兴 Bier, K., Gorenflo, D., Salem, M., and Tanes, Y., 1978, ‘‘Pool Boiling Heat
max ⫽ maximum Transfer and Size of Active Nucleation Centers for Horizontal Plates With
Different Surface Roughness,’’ Proc. Sixth Int. Heat Trans. Conf., 1, Toronto,
min ⫽ minimum Canada, pp. 151–156.
o ⫽ initial, nucleation 关17兴 Yang, S. R., and Kim, R. H., 1988, ‘‘A Mathematical Model of the Pool
rsh ⫽ required superheat Boiling Nucleation Site Density in Terms of the Surface Characteristics,’’ Int.
s ⫽ surface, solid J. Heat Mass Transf., 31, pp. 1127–1135.
关18兴 Wang, C. H., and Dhir, V. K., 1993, ‘‘Effect of Surface Wettability on Active
sat ⫽ saturation Nucleation Site Density During Pool Boiling of Water on a Vertical Surface,’’
⬁ ⫽ liquid condition far from bubble interface. ASME J. Heat Transfer, 115, pp. 659– 669.
关19兴 Gaertner, R. F., 1963, ‘‘Distribution of Active Sites in the Nucleate Boiling of
Liquids,’’ Chem. Eng. Prog., Symp. Ser., 59, pp. 52– 61.
References 关20兴 Eckert, E. R. G., and Drake, R. M., 1972, Analysis of Heat and Mass Transfer,
关1兴 Bernardin, J. D., and Mudawar, I., 1995, ‘‘Validation of the Quench Factor McGraw-Hill, New York.
Technique in Predicting Hardness in Heat Treatable Aluminum Alloys,’’ Int. J. 关21兴 Panton, R. L., 1984, Incompressible Flow, John Wiley & Sons, New York.
Heat Mass Transf., 38, pp. 863– 873. 关22兴 Mikic, B. B., Rohsenow, W. M., and Griffith, P., 1970, ‘‘On Bubble Growth
关2兴 Bernardin, J. D., and Mudawar, I., 1999, ‘‘The Leidenfrost Point: Experimental Rates,’’ Int. J. Heat Mass Transf., 13, pp. 657– 666.
Study and Assessment of Existing Models,’’ ASME J. Heat Transfer, 121, pp. 关23兴 Van Stralen, S. J. D., Sohal, M. S., Cole, R., and Sluyter, W. M., 1975,
894 –903. ‘‘Bubble Growth Rates in Pure and Binary Systems: Combined Effect of Re-
关3兴 Clark, H. B., Strenge, P. S., and Westwater, J. W., 1959, ‘‘Active Sites for laxation and Evaporation Microlayers,’’ Int. J. Heat Mass Transf., 18, pp. 453–
Nucleate Boiling,’’ Chem. Eng. Prog., Symp. Ser., 55, pp. 103–110. 467.
关4兴 Gaertner, R. F., and Westwater, J. W., 1959, ‘‘Population of Active Sites in 关24兴 Lee, H. S., and Merte, H., 1996, ‘‘Spherical Vapor Bubble Growth in Uni-
Nucleate Boiling Heat Transfer,’’ Chem. Eng. Prog., Symp. Ser., 55, pp. 39– formly Superheated Liquids,’’ Int. J. Heat Mass Transf., 39, pp. 2427–2447.
48. 关25兴 Bankoff, S. G., 1958, ‘‘Entrapment of Gas in the Spreading of a Liquid over a
关5兴 Kurihara, H. M., and Myers, J. E., 1960, ‘‘The Effects of Superheat and Sur- Rough Surface,’’ AIChE J., 4, pp. 24 –26.
face Roughness on Boiling Coefficients,’’ AIChE J., 6, pp. 83–91. 关26兴 Bankoff, S. G., and Mikesell, R. D., 1959, ‘‘Bubble Growth Rates in Highly
关6兴 Bankoff, G. S., 1959, ‘‘The Prediction of Surface Temperatures at Incipient Subcooled Nucleate Boiling,’’ Chem. Eng. Prog., Symp. Ser., 55, pp. 95–102.
Boiling,’’ Chem. Eng. Prog., Symp. Ser., 55, pp. 87–94. 关27兴 Carey, V. P., 1992, Liquid-Vapor Phase-Change Phenomena: An Introduction
关7兴 Hsu, Y. Y., 1962, ‘‘On the Size Range of Active Nucleation Cavities on a to the Thermophysics of Vaporization and Condensation Processes in Heat
Heating Surface,’’ ASME J. Heat Transfer, 84, pp. 207–213. Transfer Equipment, Hemisphere, New York.
关8兴 Han, C. Y., and Griffith, P., 1965, ‘‘The Mechanism of Heat Transfer in Nucle- 关28兴 Bernardin, J. D., Mudawar, I., Walsh, C. B., and Franses, E. I., 1997, ‘‘Contact
ate Pool Boiling-Part I,’’ Int. J. Heat Mass Transf., 8, pp. 887–904. Angle Temperature Dependence for Water Droplets on Practical Aluminum
关9兴 Lorenz, J. J., Mikic, B. B., and Rohsenow, W. M., 1974, ‘‘The Effects of Surfaces,’’ Int. J. Heat Mass Transf., 40, pp. 1017–1034.
Surface Condition on Boiling,’’ Proc. Fifth Int. Heat Transfer Conf., 4, Tokyo, 关29兴 Skripov, V. P., 1974, Metastable Liquids, John Wiley & Sons, New York.
pp. 35–39. 关30兴 Spiegler, P., Hopenfeld, J., Silberberg, M., Bumpus, Jr., C. F., and Norman, A.,
关10兴 Gaertner, R. F., 1965, ‘‘Photographic Study of Nucleate Pool Boiling on a 1963, ‘‘Onset of Stable Film Boiling and the Foam Limit,’’ Int. J. Heat Mass
Horizontal Surface,’’ ASME J. Heat Transfer, 87, pp. 17–29. Transf., 6, pp. 987–994.
Rewetting of an Infinite Slab With
A. K. Satapathy
Uniform Heating Under
e-mail: aksatapathy@rec.ori.nic.in
R. K. Sahoo
Quasi-Steady Conditions
The two-dimensional quasi-steady conduction equation governing conduction controlled
Department of Mechanical Engineering, rewetting of an infinite slab, with one side flooded and the other side subjected to a
Regional Engineering College, constant heat flux, has been solved by Wiener-Hopf technique. The solution yields the
ROURKELA-769 008, quench front temperature as a function of various model parameters such as Peclet
Orissa, India number, Biot number and dimensionless heat flux. Also, the critical (dryout) heat flux is
obtained by setting the Peclet number equal to zero, which gives the minimum heat flux
required to prevent the hot surface being rewetted. 关DOI: 10.1115/1.1484111兴
Keywords: Analytical, Boiling, Heat Transfer, Heat Pipes, Modeling, Nuclear
1 Introduction ary heat flux and the dryout induced by the heat flux is of specific
interest while considering the decay heating of a nuclear fuel 关5兴
The process of quenching of hot surfaces is of practical impor-
or in the design of heat pipes for thermal radiators 关6,7兴. Chan and
tance in nuclear and metallurgical industries. For instance, in the
Zhang 关7兴 observed that the existence of heat flux on the wall
event of a postulated loss-of-coolant accident 共LOCA兲 in water
poses an unsteady state solution for the heat conduction equation,
cooled reactors, the clad surface of the fuel elements may reach
even after the equation is transformed to the Lagrangian coordi-
very high temperature because the stored energy in the fuel cannot nate moving with the quench front. In this respect, they also con-
be removed adequately by the surrounding steam. In order to sidered the rewetting velocity as well as the plate temperature 共at
bring the reactor to a cooled shutdown condition, an emergency far ahead of the quench front兲 to be time variant. In the present
core cooling system is activated to reflood the core. The time paper, however, precursory cooling in the dry region has been
delay in re-establishing the effective cooling may result in a clad- included in the boundary condition in order to consider the quasi-
ding temperature rise, significantly above the saturation tempera- steady state conduction equation. Further, reported literature on
ture. If the cladding temperature rises above the rewetting tem- analytical investigations indicates that Wiener-Hopf solution for
perature, a stable vapor blanket will prevent the immediate return the rewetting model with a boundary heat flux does not exist. In
to liquid-solid contact. Rewetting is the re-establishment of liquid the present analysis, Wiener-Hopf technique has been employed
contact with a hot cladding surface and, thereby, bringing it to an because of its accuracy and computational simplicity. Besides, the
acceptable temperature. Also, quenching phenomenon is of con- advantage of using the Wiener-Hopf technique may be recognized
siderable practical interest in many other applications such as in case of handling discontinuous boundary conditions, where the
steam generators, evaporators, cryogenic systems and metallurgi- singularity due to the discontinuity can be readily resolved by
cal processing. The cooling process during quenching is charac- decomposing an appropriate kernel function in the complex Fou-
terized by the formation of a wet patch on the hot surface, which rier domain.
eventually develops into a steadily moving quench front. As the In the present study, the physical model consists of an infinitely
quench front moves along the hot surface, two regions can be extended vertical slab with one side flooded and the other side
identified: a dry region ahead of the quench front and a wet region subjected to a uniform heat flux. The model assumes constant but
behind the quench front. The upstream end of the solid 共wet different heat transfer coefficients for the wet and dry regions on
region兲 is cooled by convection to the contacting liquid, while the flooded side. The two-dimensional quasi-steady conduction
its downstream end 共dry region兲 is cooled by heat transfer to a equation governing the conduction-controlled rewetting of the in-
mixture of vapor and entrained liquid droplets, called precursory finite slab has been solved by Wiener-Hopf technique. The present
cooling. solution involves the exact decomposition of the kernel function,
The rewetting model for a two-dimensional two-region heat so that the solution may be valid for all range of values of the
transfer with a step change in heat transfer coefficient at the parameters used in the model. The solution has been compared
quench front has been solved for a single slab 关1–3兴 or for a with other analytical solutions and depicted in the graphical form.
composite slab 关4兴. In the single slab model the unwetted side is
considered to be adiabatic, whereas in case of a composite slab a 2 Mathematical Model
three layer composite is considered to simulate the fuel and the
cladding separated by a gas filled gap between them. The solution The two-dimensional transient heat conduction equation for the
method commonly employed is Wiener-Hopf technique. The two- slab is
dimensional rewetting model for a single slab with a uniform heat
flux and precursory cooling has been solved by an approximate
integral method 关5兴. The one-dimensional rewetting model with a
k 冉 ⫹
⳵X2 ⳵Y 2冊
⳵ 2T ⳵ 2T
⫽␳C
⳵T
⳵t
0⬍X⬍ ␦ 0⬍Y ⬍L L→⬁ (1)
uniform heat flux has been solved for a smooth plate 关6兴 and for where L is the length of the slab and ␦ is the thickness of the slab.
both smooth and grooved plates 关7兴, considering the dry region to The density, specific heat and thermal conductivity of the slab
be adiabatic. material are ␳, C, and k respectively. The origin of the coordinate
The analysis of rewetting of a hot surface subjected to a bound- frame is at left-bottom corner of the slab. To convert this transient
equation into a quasi-steady state equation, the following transfor-
Contributed by the Heat Transfer Division for publication in the JOURNAL OF mation is used:
HEAT TRANSFER. Manuscript received by the Heat Transfer Division September 8,
2000; revision received March 29, 2002. Associate Editor: T. Y. Chu. x̄⫽X ȳ⫽Y ⫺ut
analysis, the heat transfer coefficient h 1 is assumed to be constant
over the entire wet region. The coolant temperature is taken to be
equal to its saturation temperature T s . On the dry side of the slab,
the wall is cooled by the surrounding vapor. The heat transfer
coefficient accounting for both convective and radiative cooling
effects on the dry side is assumed equal to h 2 , a constant, which
is smaller than h 1 . The temperature of the surrounding vapor is
assumed equal to T w , which can be interpreted as the initial tem-
perature of the hot surface without a boundary heat flux. This is
justified because the vapor in dry region would be superheated
due to the existence of imposed surface heat flux on the wall. The
rewetting 共quench front兲 temperature is denoted by T 0 .
Following Yao 关5兴, it may be envisaged that the temperature
field is sufficiently flat in the axial direction at infinity. Conse-
quently, the first and second derivatives of temperature in
ȳ-direction can be neglected at far upstream of the quench front
共at ȳ→⫺⬁兲 as well as at far downstream of the quench front 共at
ȳ→⫹⬁兲. The above two assumptions are adequate to prescribe
the temperature at infinity (ȳ→⫾⬁). The far-field boundary con-
ditions then become
q q
T⫽T s ⫹ 共 ␦ ⫺x̄ 兲 ⫹ ȳ→⫺⬁
k h1
(3)
q q
T⫽T w ⫹ 共 ␦ ⫺x̄ 兲 ⫹ ȳ→⫹⬁
k h2
The conventional rewetting models 共without a boundary heat
flux兲 usually assume the vapor temperature in the dry region equal
to its saturation temperature so that it would be used as a sink
temperature. In the presence of a boundary heat flux, however, it
is well justified to assume the vapor temperature equal to the
initial wall temperature because the vapor would be superheated
owing to the existence of the boundary heat flux. The surface
temperature of the slab at far ahead of the quench front 共at x̄
⫽ ␦ , ȳ→⫹⬁兲 can be calculated 共Eq. 共3兲兲 to be equal to (T w
⫹q/h 2 ). In situations when T 0 ⭓T w , the temperature of dry re-
gion wall T( ␦ ,ȳ) will be above T w and, hence, the wall will be
cooled by the vapor. On the other hand, for T 0 ⬍T w , only a finite
part of the dry region wall immediately ahead of the quench front
Fig. 1 „a… Physical domain of infinite slab; and „b… common
strip of analyticity in the complex Fourier plane.
will be less than T w , whereas for the remaining part it will be
more than T w over an infinite length. This implies that the former
part of the dry region wall of a finite length will be heated by the
vapor while the latter part of an infinite length will be cooled by
where u is the constant quench front velocity and x̄, ȳ are normal
the vapor and the overall effect is to cool the dry region wall.
and axial coordinates respectively 共Fig. 1共a兲兲. Experiments have
Thus, on the whole, in both the situations the vapor temperature
shown that, if the slab is long enough compared to the penetration
would behave as the sink temperature. Moreover, the boundary
depth to heat transfer field, the temperature distribution around the
conditions in Eq. 共3兲 suggest that precursory cooling in the dry
heat source/sink soon becomes independent of time. That is, an
region cannot be neglected in the case of existence of boundary
observer stationed at the origin of the moving (x̄,ȳ) coordinate
heat flux on the wall. Equation 共2兲 can be expressed in the follow-
system fails to notice any appreciable change in the temperature
ing dimensionless form
distribution around him as the front moves on. This is identified as
the apparent steady state or quasi-steady state condition. Thus the ⳵ 2␪ ⳵ 2␪ ⳵␪
transformed heat conduction equation in a coordinate system ⫹ ⫹Pe ⫽0 0⬍x⬍1 ⫺⬁⬍y⬍⬁ (4)
⳵x2 ⳵y2 ⳵y
moving with the quench front is
⳵ 2 T ⳵ 2 T ␳ Cu ⳵ T
⫹ ⫹ ⫽0 0⬍x̄⬍ ␦ ⫺⬁⬍ȳ⬍⬁ (2) ⳵␪
⳵ x̄ 2 ⳵ ȳ 2 k ⳵ ȳ ⫹Q⫽0 at x⫽0 ⫺⬁⬍y⬍⬁
⳵x
The above equation is the governing partial differential equation
in quasi-steady state for the slab, in which ⳵ T/ ⳵ t⫽0 in the mov- ⳵␪
⫹B 1 ␪ ⫽0 at x⫽1 y⬍0
ing coordinate system. ⳵x
In conduction-controlled rewetting analysis, it is believed that ⳵␪
conduction of heat along the slab from the dry region to wet ⫹B 2 共 ␪ ⫺1 兲 ⫽0 at x⫽1 y⬎0
region is the dominant mechanism of heat removal ahead of the ⳵x
(5)
quench front, which results in a lowering of the surface tempera- Q
ture immediately downstream of the quench front and causes the ␪ ⫽ ⫹Q 共 1⫺x 兲 at y→⫺⬁
quench front to progress further. Since only axial conduction is B1
considered, the effect of coolant mass flux, coolant inlet subcool- Q
ing and its pressure gradient etc. are not accounted for explicitly, ␪ ⫽1⫹ ⫹Q 共 1⫺x 兲 at y→⫹⬁
B2
but implicitly in terms of wet region heat transfer coefficient,
which is incorporated in the boundary condition. In the present ␪⫽␪0 at x⫽1 y⫽0
The non-dimensional variables used above are in which ␥ ⫽( ␣ 2 ⫹s 2 ) 1/2. The transformed boundary conditions
are
x̄ ȳ T⫺T s h 1␦ h 2␦
x⫽ , y⫽ , ␪⫽ , B 1⫽ , B 2⫽ , ⌽ ⬘ 共 ␣ ,0兲 ⫽0
␦ ␦ T w ⫺T s k k
Pe⫽
␳ Cu ␦
k
, Q⫽
q␦
k 共 T w ⫺T s 兲
and ␪ 0⫽
T 0 ⫺T s
T w ⫺T s
. ⬘ 共 ␣ ,1兲 ⫹B 1 ⌽ ⫺ 共 ␣ ,1兲 ⫽⫺
⌽⫺
i
␣ ⫺is
Q Q
B 1 1⫹ ⫺ 冋
B2 B1 册 (10)
It may be verified that for no heat flux condition with adiabatic ⬘ 共 ␣ ,1兲 ⫹B 2 ⌽ ⫹ 共 ␣ ,1兲 ⫽0
⌽⫹
dry side 共by setting Q⫽0 B 2 ⫽0 and Q/B 2 ⫽0兲, the boundary
conditions in Eq. 共5兲 reduces to that of conventional two-region where prime denotes the transform of x-derivatives of ␸ (x,y).
model 共insulated dry wall and without a heating 关2兴兲. The main The general solution of the second order ordinary differential
objective of the present analysis is to obtain the quench front equation 共Eq. 共9兲兲 is
temperature ␪ 0 in terms of wetside Biot number B 1 , dryside Biot
number B 2 , Peclet number Pe and dimensionless heat flux Q. ⌽ 共 ␣ ,x 兲 ⫽C 1 共 ␣ 兲 cosh ␥ x⫹C 2 共 ␣ 兲 sinh ␥ x (11)
Although Eqs. 共4兲 and 共5兲 have been formulated for the case of Imposing the boundary conditions of Eq. 共10兲 into Eq. 共11兲 yields
冋册
bottom flooding, they are also valid for top flooding.
1⫹ 共 B 1 coth ␥ 兲 / ␥
⌽ ⫹ 共 ␣ ,1兲 ⫹ ⌽ 共 ␣ ,1兲
3 Analytical Solution 1⫹ 共 B 2 coth ␥ 兲 / ␥ ⫺
In order to employ the Wiener-Hopf technique, Eq. 共4兲 is first
transformed with a new variable ␸, defined by ␪ (x,y)⫽1
⫹(Q/B 2 )⫹Q(1⫺x)⫺ ␸ (x,y)e ⫺sy , in which s⫽Pe/2. The gov-
⫽⫺
i
冉
B 1 coth ␥
␣ ⫺is ␥ ⫹B 2 coth ␥ 冊冉 1⫹
Q Q
⫺
B2 B1 冊 (12)
erning equation 共Eq. 共4兲兲 then becomes 3.2 Wiener-Hopf Technique. The technique of Wiener-
⳵ 2␸ ⳵ 2␸ 2 Hopf, which has been fruitfully applied to the class of rewetting
⫹ ⫺s ␸ ⫽0 0⬍x⬍1 ⫺⬁⬍y⬍⬁ (6) problems, uses the strategy of solving a functional equation 共Eq.
⳵x2 ⳵y2
共9兲兲 comprising of two unknown functions 共⌽ ⫹ and ⌽ ⫺ 兲 of com-
The boundary conditions can be written sequentially as plex variable. The crucial step in successful execution of the
Wiener-Hopf technique depends on the factorization of a function,
⳵␸ which is analytic in a strip, into the product of two functions that
⫽0 at x⫽0 ⫺⬁⬍y⬍⬁
⳵x are analytic in the overlapping half-planes. In this context, let
⳵␸
⳵x B2 B1冋
Q Q sy
⫹B 1 ␸ ⫽B 1 1⫹ ⫺ e 册 at x⫽1 y⬍0 K 共 ␣ 兲 ⫽K ⫹ 共 ␣ 兲 K ⫺ 共 ␣ 兲 ⫽
1⫹ 共 B 1 coth ␥ 兲 / ␥
1⫹ 共 B 2 coth ␥ 兲 / ␥
(13)
⳵␸ where the functions K ⫹ ( ␣ ), K ⫺ ( ␣ ) are analytic in the domains

⫹B 2 ␸ ⫽0 at x⫽1 y⬎0 (7) D ⫹ and D ⫺ respectively. Now the kernel function K( ␣ ), in con-
⳵x
nection with Eq. 共12兲, is to be decomposed to K ⫹ ( ␣ ) and K ⫺ ( ␣ )
␸ ⫽ 1⫹ 冋 Q Q sy
⫺
B2 B1
e 册 at y→⫺⬁
in accordance with the Wiener-Hopf technique. This is accom-
plished by rearranging Eq. 共12兲 to obtain
␸ ⫽0 at y→⫹⬁ ⌽ ⫹ 共 ␣ ,1兲
K ⫹共 ␣ 兲
⫺
i Q
⫹
1
␣ ⫺is B 2 1⫺␭冉冊冋 1
⫺
1
K ⫹ 共 ␣ 兲 K ⫹ 共 is 兲册
冉冊冋
3.1 Fourier Transform. Fourier transformation of a partial
differential equation and of its associated boundary conditions
generally results in a less complicated problem in the plane of the
transformed variable. If the solution of this subsidiary problem
⫽⫺
i Q
⫹
1
␣ ⫺is B 2 1⫺␭
K ⫺共 ␣ 兲 ⫺
1
K ⫹ 共 is 兲册
can easily be obtained and inverted, then the transform technique ⫺⌽ ⫺ 共 ␣ ,1兲 K ⫺ 共 ␣ 兲 (14)
is straightforward and supposed to be efficient. In the next step of
the analysis, Fourier transform is used to convert the partial dif- where ␭⫽B 2 /B 1 . In Eq. 共14兲, each side characterizes the same
ferential equation 共Eq. 共6兲兲 to an ordinary differential equation. ‘‘entire function’’, through its representation in the upper and
The Fourier transform is defined by lower halves of the ␣-plane. Since ⌽ ⫹ ( ␣ ,1) and ⌽ ⫺ ( ␣ ,1) tend to
zero at infinity in their half planes of analyticity, while K ⫹ ( ␣ ) and
⌽ 共 ␣ ,x 兲 ⫽⌽ ⫹ 共 ␣ ,x 兲 ⫹⌽ ⫺ 共 ␣ ,x 兲 ⫽ 冕 ⫺⬁
⬁
␸ 共 x,y 兲 e i ␣ y dy (8)
K ⫺ ( ␣ ) remain bounded, it follows that the entire function van-
ishes according to Liouville’s theorem 共关8兴, p. 27兲. Therefore,
equating both sides of the Eq. 共14兲 to zero, ⌽ ⫹ ( ␣ ,1) and
with ⌽ ⫺ ( ␣ ,1) are determined as
⌽ ⫺ 共 ␣ ,x 兲 ⫽ 兰 ⫺⬁ ␸ 共 x,y 兲 e i ␣ y dy,
冉冊冋册
0
i Q 1 K ⫹共 ␣ 兲
⌽ ⫹ 共 ␣ ,1兲 ⫽ ⫹ 1⫺
⌽ ⫹ 共 ␣ ,x 兲 ⫽ 兰 ⬁0 ␸ 共 x,y 兲 e i ␣ y dy. ␣ ⫺is B 2 1⫺␭ K ⫹ 共 is 兲
(15)
The parameter ␣ used above is a complex quantity. The far-field
boundary conditions in Eq. 共7兲 indicate that ␸ (x,y) is of the order
exp(sy) at y→⫺⬁, whereas ␸ (x,y) is of the order exp(⫺sy) at
⌽ ⫺ 共 ␣ ,1兲 ⫽⫺
i Q
⫹
1
␣ ⫺is B 2 1⫺␭ 冉冊冋 1⫺
1
K ⫺ 共 ␣ 兲 K ⫹ 共 is 兲册
y→⫹⬁. Thus, the functions ⌽ ⫹ ( ␣ ,x), ⌽ ⫺ ( ␣ ,x) are analytic in 3.3 Quench Front Temperature. Using the above expres-
the domains D ⫹ and D ⫺ respectively 共关8兴, p. 78兲. The domains sions of ⌽ ⫹ ( ␣ ,1) and ⌽ ⫺ ( ␣ ,1), quench front temperature may
D ⫹ and D ⫺ are defined 共Fig. 1共b兲兲 in the entire complex domain be obtained by inverting the Fourier transform 共Eq. 共8兲兲. Such an
as: D ⫹ :Im(␣)⬎⫺s, D ⫺ :Im(␣)⬍⫹s. Applying the Fourier trans- attempt may become tedious because, in order to carry out the
form, Eq. 共6兲 assumes the form Fourier inversion, it would be necessary to evaluate the residues
of the function ⌽共␣,1兲 in the complex domain. Alternatively, in
d 2⌽ the present paper ␪ 0 has been calculated in a simplified approach
⫺ ␥ 2 ⌽⫽0 (9)
dx 2 关1兴 as follows.
⌽ ⫹ 共 ␣ ,1兲 ⫽ 冕0
⬁
␸ 共 1,y 兲 e i ␣ y dy
⫽
i
␣
␸ 共 1,0兲 ⫺
1
i␣ 冕 0
⬁ ⳵␸ 共 1,y 兲 i ␣ y
⳵y
e dy (16)
In the limit ␣→⬁, the second integral appearing in Eq. 共16兲 van-
ishes since the quantity ⳵␸ / ⳵ y is bounded 关1兴. Then, by virtue of
Eqs. 共15兲–共16兲 and invoking an assumption that K ⫹ ( ␣ ) ap-
proaches unity as ␣→⬁ 共the assumption will be validated later兲,
we obtain
␸ 共 1,0兲 ⫽ lim 关 ⫺i ␣ ⌽ ⫹ 共 ␣ ,1兲兴 ⫽

␣ →⬁
冉 Q
⫹
1
B 2 1⫺␭ 冊冋 1⫺
1
K ⫹ 共 is 兲册
(17)
The quench front temperature then becomes
Fig. 2 Quench front temperature for various heat flux and Pe-
␪ 0 ⫽1⫹Q/B 2 ⫺ ␸ 共 1,0兲 ⫽
1 Q
⫹
1
K ⫹ 共 is 兲 B 2 1⫺␭
⫺
␭
1⫺␭
(18) 冋册 clet number
Now the function K ⫹ (is) may be expressed as an ‘‘infinite Eq. 共23兲, the expression for ␪ 0 reduces to exactly the same as that
product series’’ or as a ‘‘contour integral.’’ While the former leads of Olek 关2兴 which, in turn, substantiates the present solution.
to evaluation of the eigen values of a certain transcendental equa-
tion, the latter leads to an integral expression which is seemingly 3.4 Critical Heat Flux. The quench front temperature at the
more convenient for numerical computation. On applying the critical 共dryout兲 heat flux has been deduced by specifying s⫽0 in
Cauchy residue theorem within the strip, the function ln K(␣) can Eq. 共23兲. Thus, K ⫹ (is) simplifies to
be represented by the following contour integral.
ln K 共 ␣ 兲 ⫽ln K ⫹ 共 ␣ 兲 ⫹ln K ⫺ 共 ␣ 兲冋冕
K ⫹ 共 is 兲 ⫽exp
1
␲ 0
␲ /2
ln共 B 1 /B 2 兲 d⍀ ⫽册冑 B1
B2
(24)
⫽
1
2␲i 冕C⫹
ln K 共 ␰ 兲
␰ ⫺ ␣
d␰⫺
2
1
␲i 冕 ln K 共 ␰ 兲
C⫺ ␰ ⫺ ␣
d ␰ (19)
The quench front temperature at the critical heat flux is finally
determined as
where C ⫹ /C ⫺ is an infinite contour lying inside the strip and Q 冑B 2

␪ 0⫽ ⫹ (25)
passing below/above the point ␣ 共Fig. 1共b兲兲. It may be noted that, 冑B 1 B 2 冑 B 1 ⫹ 冑B 2
due to the asymptotic nature of ln K(␣) function 共the order of
ln K(␣) being 1/␣兲, the contribution of vertical sides of the contour The heat flux Q appearing in Eq. 共25兲 may be regarded as the
to the integral vanishes at Re共␣兲→⫾⬁. Equation 共19兲 can be suc- critical heat flux Q cri , which characterizes the maximum allow-
cinctly written as able heat input to a slab to inhibit the dryout of the coolant.
ln K ⫾ 共 ␣ 兲 ⫽⫾
1
2␲i 冕 C⫾
ln K 共 ␰ 兲
␰⫺␣
d␰ (20) 4 Results and Discussion
Numerical values of the quench front temperature are obtained
from which it follows that K ⫾ ( ␣ )⫽1 as ␣→⬁, as assumed ear- from the expressions in Eqs. 共23兲 and 共25兲, for a practical range of
lier. In order to evaluate the function K ⫹ (is), the contour C ⫹ may model parameters B 1 , B 2 , Pe and Q. For this purpose, the integral
be shifted to the real axis to yield appearing in Eq. 共23兲 has been numerically calculated by Simp-
son’s 1/3 rule with 101 equally spaced base points. Experimental
ln K ⫹ 共 is 兲 ⫽
1
2␲i 冕⫺⬁
⫹⬁ ln K 共 ␰ 兲
␰ ⫺is
d␰ (21)
investigations on quenching 关9兴 reveal the existence of four dis-
tinct heat transfer regimes along the wall, the regimes being de-
marcated by the characteristic hot surface temperature. These four
Further, as the functions ln K(␰), ␰ ln K(␰) exhibit even and odd zones are: forced convection of subcooled liquid, nucleate boiling,
properties respectively, Eq. 共21兲 thereby reduces to wet and dry transition boiling and film boiling. Quench front is
冕
observed to exist in the transition zone. The heat transfer coeffi-
s ⬁ lnK 共 ␰ 兲 cient in the transition zone is shown to be 105 – 106 W/m2 -K and
ln K ⫹ 共 is 兲 ⫽ d␰ (22)
␲ 0 ␰ 2 ⫹s 2 the vapor cooling heat transfer coefficient in the film boiling zone
is in the order of 102 W/m2 -K. In the present analysis the values
For computational purposes it is advantageous to transform ␰, by of h 1 and h 2 are adopted from the experimental results of Barnea
␰ ⫽s tan ⍀, to finally obtain the quench front temperature et al. 关9兴. Hence the values of h 2 are set equal to 10⫺3 h 1 and,
␪ 0⫽
1 Q
⫹ 冋 B1
K ⫹ 共 is 兲 B 2 B 1 ⫺B 2
⫺
B2
B 1 ⫺B 2 册 (23)
therefore, B 2 ⫽10⫺3 B 1 .
The variation of quench front temperature with heat flux and
Peclet number is shown in Fig. 共2兲, for a fixed value of Biot
in which, number. Here ␪ 0 is found to increase with increase in Peclet num-
冋冕再冎册
ber. With fixed material properties and dimensions, Peclet number
1 ␲ /2 1⫹ 共 B 1 coth s sec ⍀ 兲 /s sec ⍀ and Biot number represent the quench front velocity and the heat
K ⫹ 共 is 兲 ⫽exp ln d⍀ .
␲ 0 1⫹ 共 B 2 coth s sec ⍀ 兲 /s sec ⍀ transfer coefficient respectively. For prescribed values of heat flux
and Biot number, ␪ 0 increases with increase in quench front ve-
It is of interest to examine the limiting solution of the above locity. This may be due to the fact that a higher relative velocity
equation for the case that has been investigated by Olek 关2兴, between the slab and the coolant allows less time for sufficient
namely, the rewetting of an infinite slab without any heating or heat transfer to take place, resulting in a higher value of ␪ 0 . The
precursory cooling. By assigning Q⫽0, B 2 ⫽0 and Q/B 2 ⫽0 in above trend also reflects the fact that, for the same rewetting rate,
Fig. 3 Quench front temperature for various heat flux and Biot Fig. 5 Quench front temperature for various wetside Biot
number number and Peclet number without heating and precursory
cooling
an increasing slab thermal diffusivity tends to reduce ␪ 0 whereas results are in good agreement with those in 关5兴 for lower values of
the increasing slab thickness has the opposite effect. Further, for a Biot and Peclet numbers while the deviation becomes more pro-
fixed Peclet number, ␪ 0 increases with increase in Q. Apparently, nounced as Biot and Peclet numbers become large. In Yao’s 关5兴
a higher heat flux causes more heat transfer to the slab and hence analysis, the temperature distribution along the width of the slab
this would increase ␪ 0 . was assumed to be quadratic and with this assumption, the solu-
The dependence of quench front temperature on Biot number tion was obtained for a two-dimensional conduction model. Ap-
and dimensionless heat flux is shown in Fig. 共3兲, where ␪ 0 de- parently, the solution may deviate at higher Biot and Peclet num-
creases with increase in Biot number for a given Pe and Q. A bers due to the above approximation. Finally, the model is reduced
higher Biot number results in a higher heat transfer coefficient. to the conventional model 共by setting Q⫽0, B 2 ⫽0 and Q/B 2
This enhanced heat transfer coefficient may cause to decrease ␪ 0 . ⫽0兲 and illustrated in Fig. 共5兲. As expected, ␪ 0 increases with
The above trends are in obvious accord with the predictions based increase in Peclet number and with decrease in Biot number.
on physical ground. In all cases, ␪ 0 decreases as Biot number The Wiener-Hopf technique yields a solution for the quench
increases, reflecting the fact that a quench front progresses more front temperature 共Eq. 共23兲兲, which is more elegant and accurate
easily when the heat transfer to the coolant is increased. On a than results obtained by other analytical methods. In particular,
similar ground, conversely, one would conclude that an increasing Wiener-Hopf solution is superior to the one by separation of vari-
Q has the opposite effect on the quench front velocity. ables, since it overcomes the accuracy problems due to slow con-
The dependence of quench front temperature on Biot number vergence of the series expansions that stem from discontinuity of
and dimensionless heat flux is shown in Fig. 共4兲, with Pe⫽0. The the surface heat flux at the quench front 关2兴. The technique makes
physical meaning of Pe⫽0 is that the quench front ceases to move use of decomposing a kernel function in the complex Fourier
when Q approaches its critical value. This is the case that the plane so as to resolve the singularity arising out of discontinuous
surface can no longer be wetted. For Q⬎Q cri , the quench front boundary conditions at the quench front, as in the case of a rewet-
will reverse its direction and the wetted surface will be dried. In ting problem. The explicit formula for the quench front tempera-
this case, the slab will be heated by a heat flux that exceeds the ture obtained in the present study is valid for all Biot and Peclet
maximum heat removal capacity by convection and boiling and, numbers. However, the present model is limited to small Peclet
thus, dryout would occur. Further, the present solution has been numbers with regard to heat pipes. This is due to the fact that, in
compared with those of Yao 关5兴 in Fig. 共2兲 and in Fig. 共4兲. The the case of large Peclet numbers, a thermal boundary layer is
formed near the cooling surface of a heat pipe and this has not
been incorporated in the present model. Besides, large Biot num-
bers are usually associated with large Peclet numbers, unless the
internal heating is large. Since the internal heating effect is also
not considered in the model, the present analysis is limited to both
small Biot and Peclet numbers in case of heat pipes.
5 Conclusion
An analytical solution for rewetting of an infinite slab with a
uniform heating has been obtained, employing the Wiener-Hopf
technique. In general, quench front temperature is found to in-
crease with increase in Peclet number and dimensionless heat flux,
and with decrease in Biot number. The boundary conditions in the
present formulation require liquid/vapor temperatures and liquid/
vapor heat transfer coefficients as input parameters, these limita-
tions being inherent in a conduction-controlled rewetting model.
The arbitrariness of the choice of their values may be eliminated if
a conjugate heat transfer model is considered, where the energy
equations of solid, liquid and vapor regions need to be solved
Fig. 4 Quench front temperature at the critical heat flux simultaneously.
Nomenclature Subscripts
B ⫽ Biot number 0 ⫽ quench front
1 ⫽ wet region
C ⫽ specific heat
2 ⫽ dry region
h ⫽ heat transfer coefficient s ⫽ saturation
k ⫽ thermal conductivity w ⫽ initial wall condition
L ⫽ length of the slab
Pe ⫽ Peclet number
References
关1兴 Levine, H., 1982, ‘‘On a Mixed Boundary Value Problem of Diffusion Type,’’
q ⫽ heat flux Appl. Sci. Res., 39, pp. 261–276.
Q ⫽ dimensionless heat flux 关2兴 Olek, S., 1988, ‘‘On the Two-region Rewetting Model with a Step Change in
s ⫽ half of the Peclet number the Heat Transfer Coefficient,’’ Nucl. Eng. Des., 108, pp. 315–322.
关3兴 Tien, C. L., and Yao, L. S., 1975, ‘‘Analysis of Conduction Controlled Rewet-
t ⫽ time ting of a Vertical Surface,’’ ASME J. Heat Transfer, 97, pp. 161–165.
T ⫽ temperature 关4兴 Olek, S., 1994, ‘‘Quenching of a Composite Slab,’’ Int. Commun. Heat Mass
Transfer, 21, pp. 333–344.
u ⫽ quench front velocity 关5兴 Yao, L. S., 1976, ‘‘Rewetting of a Vertical Surface with Internal Heat Genera-
X, Y ⫽ physical coordinates tion,’’ AlChE Symposium Series: Solar and Nuclear Heat Transfer, 73, pp.
46 –50.
x̄, ȳ ⫽ coordinates in quasi-steady state 关6兴 Peng, X. F., and Peterson, G. P., 1992, ‘‘Analysis of Rewetting for Surface
x, y ⫽ dimensionless coordinates in quasi-steady state Tension Induced Flow,’’ ASME J. Heat Transfer, 114, pp. 703–707.
关7兴 Chan, S. H., and Zhang, W., 1994, ‘‘Rewetting Theory and the Dryout Heat
Greek Alphabets Flux of Smooth and Grooved Plates with a Uniform Heating,’’ ASME J. Heat
Transfer, 116, pp. 173–179.
␦ ⫽ thickness of the slab 关8兴 Roos, B. W., 1969, Analytical Functions and Distributions in Physics and
␭ ⫽ ratio of dryside to wetside Biot numbers Engineering, John Wiley and Sons, New York.
关9兴 Barnea, Y., Elias, E., and Shai, I., 1994, ‘‘Flow and Heat Transfer Regimes
␪ ⫽ dimensionless temperature during Quenching of Hot Surfaces,’’ Int. J. Heat Mass Transf., 37, pp. 1441–
␳ ⫽ density 1453.
A Natural Circulation Model
of the Closed Loop, Two-Phase
Thermosyphon for Electronics
S. I. Haider
Cooling
Yogendra K. Joshi
This study presents a model for the two-phase flow and heat transfer in the closed loop,
G. W. Woodruff School of Mechanical two-phase thermosyphon (CLTPT) involving co-current natural circulation. The focus is
Engineering, on CLTPTs for electronics cooling that exhibit complex two-phase flow patterns due to the
Georgia Institute of Technology, closed loop geometry and small tube size. The present model is based on mass, momen-
Atlanta, GA 30332 tum, and energy balances in the evaporator, rising tube, condenser, and the falling tube.
The homogeneous two-phase flow model is used to evaluate the friction pressure drop of
the two-phase flow imposed by the available gravitational head through the loop. The
Wataru Nakayama saturation temperature dictates both the heat source (chip) temperature and the con-
Therm Tech International, denser heat rejection capacity. Thermodynamic constraints are applied to model the satu-
920-7 Higashi Koiso, Oh-Iso Machi, ration temperature, which also depends upon the local heat transfer coefficient and the
Kanagawa 255-0004, Japan two-phase flow patterns inside the condenser. The boiling characteristics of the enhanced
structure are used to predict the chip temperature. The model is compared with experi-
mental data for dielectric working fluid PF-5060 and is in general agreement with the
observed trends. The degradation of condensation heat transfer coefficient due to dimin-
ished vapor convective effects, and the presence of subcooled liquid in the condenser are
expected to cause higher thermal resistance at low heat fluxes. The local condensation
heat transfer coefficient is a major area of uncertainty. 关DOI: 10.1115/1.1482404兴
Keywords: Cooling, Electronics, Heat Transfer, Thermosyphons, Two-Phase
Introduction A modeling methodology based on mass, momentum, and en-

ergy balances is presented, which also utilizes thermodynamic
We report on the modeling of the two-phase flow and heat
constraints in the four major components of the thermosyphon
transfer in the closed loop, two-phase thermosyphon 共CLTPT兲 in- loop. The fluid absorbs heat by boiling inside the evaporator en-
volving co-current natural circulation. Most of the available mod- closure. The resulting two-phase flow rises up from the evaporator
eling literature deals with two-phase thermosyphons with counter- through a vertical tube, bends by 90 deg, and then undergoes
current circulation within a closed, vertical, wickless heat pipe condensation by passing through a naturally cooled fin-tube heat
arrangement. A need was felt to develop a model for design and exchanger. Depending upon the inlet vapor quality, flow rate, satu-
optimization of CLTPTs for microelectronics applications. These ration and ambient temperatures, and the condenser inside/outside
devices allow not only large heat removal capability from discrete heat transfer characteristics, the flow may or may not completely
electronic devices, but also provide considerable flexibility in the condense within the available condenser length. If the condensa-
placement of the condenser and evaporator. tion is not complete by the condenser outlet, the falling tube will
Figure 1 shows a schematic diagram of the experimental setup experience a higher pressure drop due to two-phase flow and will
described by Yuan et al. 关1兴 to which the present model results are yield a smaller gravitational head due to vapor presence. As a
compared. Their CLTPT consists of four major components; the result, the evaporator will receive both vapor and liquid, the entire
evaporator, the rising tube, the condenser, and the falling tube. A thermosyphon would operate at a constant saturation temperature/
micro fabricated three-dimensional boiling enhancement structure pressure, and all heat transfer in the evaporator would be latent.
within the evaporator is attached to its bottom wall. The terms On the other hand, if the condensation is complete, it would be
‘‘rising’’ and ‘‘falling’’ describe the general fluid flow direction followed by a subcooled liquid core toward the condenser outlet.
through the connecting tubes in a natural circulation thermosy- The presence of a subcooled liquid core in the condenser would
phon where condenser is placed at a higher elevation than the deteriorate the condenser’s heat transfer performance, the con-
evaporator. Figure 1 shows four distinct thermodynamic states denser outlet would be at a temperature less than the saturation
along the loop as S1 共evaporator outlet兲, S2 共condenser inlet兲, S3 temperature, the falling tube would experience single-phase pres-
共condenser outlet兲, and S4 共evaporator inlet兲. To simplify the sure drop and would give rise to a maximum gravitational head at
analysis, the evaporator wall thickness is neglected. As a result, the evaporator, and some of the heat addition in the evaporator
the square base of the enhanced boiling structure is in intimate would be sensible.
contact with the heat generating chip surface. This plane termed as After rejecting heat in the condenser, the condensate then flows
the ‘‘wall,’’ is the reference plane with which the various eleva- vertically downward through the falling tube, bends by 90 deg,
tions of the thermosyphon are measured to reflect the respective and enters the evaporator. As no heat losses are considered
gravitational heads. The desired range of the wall temperature is through the connecting tubes, the vapor quality, void fraction, and
typically 80–95°C for microelectronics applications. temperature remain constant along the rising and falling tubes.
The difference between the gravitational heads of the liquid-rich
falling tube and the vapor-rich rising tube causes the natural cir-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 28, culation of the two-phase flow through the loop. The homoge-
2001; revision received April 4, 2002. Associate Editor: F. B. Cheung. neous two-phase flow model is used to evaluate the two-phase
from the initial ambient condition. In the following, we present a
brief overview of the GATPT modeling literature in order to relate
it to the saturation temperature modeling in the present CLTPT
geometry. Modeling saturation temperature in the present CLTPT
is very important because of the level of rise of the vapor pressure
in the operating heat flux range. The successful prediction of the
chip wall temperature is critically dependent on the saturation
temperature.
Harley and Faghri 关3兴 modeled a GATPT arrangement using the
mass, momentum, and energy balances, and solving them for the
well-separated, counter-current liquid and vapor flows due to a
Nusselt type condensation on the wall. The maximum temperature
variation in their thermosyphon is only 6 C in the test range. They
modeled the saturation temperature as the falling condensate film
interface temperature that drives the conjugate heat transfer
through the film and the wall. They solved the vapor momentum
balance and the velocity field by using a guessed pressure field.
Then, they used the Clasius-Clapeyron equation to find the new
saturation temperature corresponding to the new pressure, and ap-
plied the general gas law to find the new vapor density. The cou-
pling between the heat transfer and the flow geometry was used to
determine the correct saturation pressure and temperature fields on
Fig. 1 Schematic of the experimental setup of the co-current,
closed loop, two-phase thermosyphon
the interface. In a similar analysis of Zuo and Gunnerson 关4兴,
operating saturation temperature was specified and not modeled.
El-Genk and Saber 关5兴 studied the effect of initial filling ratio
and vapor temperature on the design and operation of closed two-
friction pressure drop along the loop. An established natural con-
phase thermosyphons of the inside diameter range 15– 45 mm and
vection correlation is used for the condenser air-side plate-fin ge-
the evaporator length range 0.5–2.0 m. They gave special empha-
ometry. Nonetheless, the local in-tube condensation heat transfer
sis to calculating the operating filling ratio by applying mass bal-
coefficient is an area of uncertainty due to its lack of sensitivity to
ances to the individual components of the thermosyphon. They
the vapor convective effects, and the laminar to turbulent Rey-
also calculated the axial distribution of the liquid film thickness
nolds number range encountered during condensation in the
along the condenser, evaporator, and adiabatic sections. A signifi-
present set-up.
cant effect on the thermosyphon performance was found due to
Saturation temperature that couples the evaporator and con-
the pressure dependence of the latent heat of vaporization. How-
denser is identified as a key variable, as it dictates both the heat
ever, in their model the saturation temperature is not modeled as a
source 共chip兲 temperature and the condenser’s capacity to reject
function of heat input, but is rather treated as an independent
heat to the ambient. Thermodynamic constraints are reconciled
parameter set for parametric study. Vincent and Kok 关6兴 simulated
with the thermophysical balances in an attempt to model the sys-
an oil-cooled closed-loop thermosyphon using 375 kg water for a
tem saturation temperature, which strongly depends upon the pre-
1 MW industrial application. They applied a control-volume based
diction of the local condensation heat transfer coefficient and the
energy and momentum balance, but did not deal with the thermo-
void fractions in the system components. To the best of our
dynamics of the thermosyphon. The saturation temperature was
knowledge, this is the first attempt to model the saturation tem-
treated as an independent parameter in their parametric study,
perature in a CLTPT as a function of heat transfer, fluid flow, and
though it varied from 90°C to 250°C.
thermodynamic characteristics of the system. The experimentally
Wang and Ma 关7兴 modeled and conducted experiments on in-
observed boiling characteristics of the enhanced structure reported
clined two-phase thermosyphons. They concluded that when the
by Yuan et al. 关1兴 are used in the model to predict the wall
liquid filling is larger than 10 percent, the liquid filling and vapor
temperature.
pressure will affect the condensation heat transfer in the thermo-
syphon. They concluded the existence of an optimum angle of
Literature Survey
inclination between 30–50 deg from the horizontal. They also
The available modeling literature on two-phase thermosyphons reported other investigations where an optimum angle of inclina-
is dedicated to the closed, gravity assisted, two-phase thermosy- tion was found between 10–50 deg. An improved gravity-assisted
phons 共GATPTs兲. They are vertical, wickless heat pipes whose drainage could reduce the condenser thermal resistance, which is
lower and upper sections, separated by an adiabatic section, serve the dominant part in the present system, and result in a lower wall
as the evaporator and condenser, respectively. The working fluid 共chip兲 temperature. Their work indirectly shows that different
absorbs heat in the evaporator section, and the vapor rises up to saturation temperatures can be realized for the same heat flux
the condenser section where it rejects heat to the ambient by con- input and same initial fill ratio, depending upon the mode of heat
densing on the vertical pipe wall. The liquid then flows downward transfer realized in the condenser.
on the wall under the effect of gravity as a thin film whose thick-
ness is much smaller than the pipe diameter. The counter-current
liquid and vapor streams are well separated. A survey of GATPTs Present Two-Phase Thermosyphon Model
by El-Genk and Saber 关2兴 shows ethanol, acetone, R-11, and A rationally based thermosyphon model was developed by sat-
R-113 as working fluids of interest, wall heat fluxes of 0.99–52.62 isfying the mass, momentum, and energy balances, as well as the
kW/m2, filling ratios of 0.01–0.62, inner diameters of 6 –37 mm, thermodynamic constraints. The model utilized the visual obser-
evaporator section height of 50– 609.6 mm, and vapor temperature vations made by Yuan et al. 关1兴 for the rising tube, falling tube,
of 261–352 K. and the evaporator. A steady-state thermosyphon operation is as-
The typical GATPT length scales are much larger than that in sumed and a one-dimensional analysis is performed. The pressure
the present CLTPT arrangement. Because of their smaller size, drops in the rising and falling tubes are modeled using the homo-
loop geometry, and liquid film thickness of the order of the diam- geneous two-phase flow model, which assumes equal liquid and
eter of the condenser tubing, CLTPTs also face more complex vapor phase velocities. All two-phase flow parameters are cross-
two-phase flow patterns and higher pressure and temperature rises section averaged, with vapor assumed to be an ideal gas in ther-
Fig. 2 Mass, momentum, and energy balance iterations in the thermosyphon model
modynamic equilibrium with the liquid phase. The model neglects quality that would satisfy both iterations, i.e., the overall mass,
pressure drops in the evaporator and the condenser; as well as the momentum and energy balance as well as the thermodynamic
heat losses from the rising and falling tubes. The neglected quan- constraints.
tities are believed to be of the order of magnitude of the uncer-
tainties involved in predicting similar quantities in other parts of Overall Loop Pressure Balance. The general formulation of
the thermosyphon. The model assumes constant liquid/vapor ther- the idealized one-dimensional steady-sate, two-phase flow pres-
mophysical properties with the exception of vapor density that sure gradient in an inclined tube can be found in Carey 关8兴 as
varies with saturation temperature and pressure that are, in turn, follows:
冉冊冉冊冉冊
interdependent.
dp p dA dp
Figure 2 shows a summary of the present two-phase thermosy- ⫺ ⫽⫺ ⫺ ⫹ 关共 1⫺ ␣ 兲 ␳ l ⫹ ␣ ␳ v 兴 g sin ␪
phon model. The figure reflects two distinct iterations coupled dz A dz dz fr
冋册
through two key design parameters; the total two-phase mass flow
rate circulating through the loop and the vapor quality at the 1 d G 2 x 2 A G 2 共 1⫺x 兲 2 A
⫹ ⫹ (1)
evaporator outlet. The first iteration is based on the overall mo- A dz ␳ v ␣ ␳ l 共 1⫺ ␣ 兲
mentum balance. This means that the sum of the friction and
acceleration pressure drops in the entire thermosyphon loop must The above equation accounts for the conservation of mass with
be equal to the available gravitational pressure head—the driving possible phase change; individual momentum transports of both
force behind the natural circulation—which is the liquid-rich fall- liquid and vapor phases during two-phase flow, as well as the
ing tube gravitational head minus the vapor-rich gravitational interfacial shear force balance.
head of the evaporator and the rising tube. The homogeneous Assuming an adiabatic flow through the rising and falling tubes
two-phase flow model was used to evaluate the two-phase friction with constant area of cross-section leads to a constant quality and
pressure drops in the system. The first iteration backs out a total a constant void fraction. This eliminates the acceleration terms,
two-phase mass flow rate whose circulation through the loop and the following equation results:
冉冊冉冊
would satisfy the overall pressure balance along the loop.
The second iteration decomposes the total two-phase flow rate dp dp
⫺ ⫽⫺ ⫹ 关共 1⫺ ␣ 兲 ␳ l ⫹ ␣ ␳ v 兴 g sin ␪ (2)
into the vapor and liquid parts by solving the individual mass and dz dz fr
energy balances on the evaporator and the condenser. This also
requires the knowledge of the air-side natural convection charac- Thus, the overall pressure gradient in the rising and falling tubes
teristics of the condenser. The experimentally observed boiling is a sum of the friction and gravitational pressure gradients. The
characteristics of the enhanced boiling structure are also built in friction pressure gradient term includes the combined effect of
the model that allows us to predict the wall temperature for a both liquid and vapor phases.
given saturation temperature. In the entirety, the model looks for Let H pump be the pumping head 共Pa兲 available to the loop
the values of the total mass flow rate and the evaporator outlet through an external power source. In that case, H pump would act
opposite to the gravitational, frictional, and acceleration pressure Both f t p and f lo are evaluated using the following conventional
losses. Therefore, the following equation can be written: forms for laminar 共Re⬍2,300兲 and turbulent flows, respectively.
N N N
f ⫽16.0 Re⫺1.0 (9)
H pump⫹ 兺
i⫽1
⌬p f r ⫹ 兺
i⫽1
⌬ p gr ⫹ 兺
i⫽1
⌬ p acc⫽0 (3)
f ⫽0.079 Re⫺0.25 (10)
where N is the total number of distinct pressure drop components The Reynolds numbers for the liquid-only and two-phase flows
in the closed loop. We are following a sign convention that H pump are defined as.
is positive; ⌬ p f r is always negative; ⌬p gr is positive in a falling
component and negative in a rising component; and ⌬p acc is posi- Gdh
tive if the flow decelerates and negative if the flow accelerates. Relo ⫽ (11)
␮l
The above equation is a formal way of expressing the fairly es-
tablished methodology for modeling mass flow rate through the Gdh
closed loops, e.g., by Cammarata et al. 关9兴 and Tengblad 关10兴. Ret p ⫽ (12)
As the gravitational pressure change is positive for the falling ␮
¯
tube and negative for the rising tube and the evaporator, Eq. 3 can We used the following expression for the two-phase mean viscos-
be rewritten as: ity proposed by McAdams et al. 关11兴.
N N
兺 ⌬p ⫺ 兺 ⌬p
1 x 1⫺x
H pump⫹ 共 H fall⫺H rise兲 ⫽⫺ fr acc (4) ⫽ ⫹ (13)
i⫽1 i⫽1 ␮
¯ ␮v ␮l
where both H fall and H rise are the magnitudes of the falling and More details and references on the homogeneous flow model can
rising gravitational pressure heads that would act in tandem with be found in Collier 关12兴.
any externally imposed H pump . In the absence of any externally
imposed head H pump , as is the case at present, the difference Modeling the System Saturation Temperature. A unique
between the falling and rising heads would be the sole driver of aspect of the present model is its ability to predict the saturation
the natural circulation through the two-phase thermosyphon. The temperature in the CLTPT. Thermodynamic considerations are im-
two-phase pressure drop is neglected inside the horizontal con- portant as the saturation temperature and pressure vary signifi-
denser, as are fitting pressure drops in the system and entrance/ cantly with the heat flux due to the volumetric constraints of the
exit pressure losses. two-phase thermosyphon.
The vapor density for PF-5060, a perflourocarbon 共Perfluoro-2-
Rising and Falling Tube Pressure Drops. During the ther- methylpentane, C6 F14兲 with a molecular weight 338.04, was com-
mosyphon operation, the rising tube experiences two-phase flow puted by using the general gas law. This helped estimate the total
while the falling tube is filled with two-phase flow or saturated charge, m charge , in the system using the initial fill ratio at the
liquid returning back to the evaporator, depending upon whether ambient saturation condition. We used the following empirical
the condensation is complete. equation for the inter-conversion of p sat and T sat for PF-5060
Homogeneous Two-phase Flow Model. The homogeneous b
two-phase flow model was used to analyze the flow characteristics log10共 p sat兲 ⫽a⫺ (14)
共 T sat⫹273.15兲 ⫺c
in the rising and falling tubes. The model treats the two-phase as
a single phase with mean fluid properties that are functions of the where a⫽4.04938, b⫽1108.035, c⫽56.584 and, p sat and T sat are
corresponding liquid and vapor properties and the quality. This in bar and degree C, respectively. The equation was found from
also means assuming equal liquid and vapor velocities, i.e., no slip NIST 共CAS Registry Number 355-04-4兲, and relates the PF-5060
(S⫽1). The flow through the rising and falling tubes is consid- experimental data on saturation pressure and temperature within
ered adiabatic, so temperature, vapor quality, and void fraction the range, ⫺19.22 C to 177.96 C. It was preferred over the clas-
remain unchanged along the rising and falling tubes. sical Clasius-Clapeyron equation, as it accounts for the dependen-
Considering the two-phase flow as an equivalent single phase cies of i f g and ␳ l on saturation state.
flow, the frictional part of the overall pressure gradient in Eq. 2 is So far, we have no information on either the saturation tempera-
evaluated using the following conventional form: ture or pressure during the thermosyphon operation. We propose
冉冊
the heat balance coupling between the evaporator and the con-
dp 2 f tpG 2
⫺ ⫽ (5) denser as the missing link to the solution of saturation tempera-
dz ¯␳ d h ture. The balance, which would relate the saturation temperature
fr
with the heat flux, is implicit in the present model. Summarily, T sat
where f tp is the effective two-phase friction factor, and the mean is iteratively solved to the value that would be low enough to
two-phase density is defined as a function of liquid and vapor absorb the heat input in the evaporator and high enough to reject
densities, and vapor quality 共x兲 as follows: the same in the condenser, for the imposed flow circulation. In
1 x 1⫺x addition to the condenser heat transfer characteristics, the predic-
⫽ ⫹ (6) tion of T sat also requires the knowledge of the condenser void
¯␳ ␳ v ␳l fraction as a function of the initial system filling ratio, and the
The mass velocity G 共kg/m2/s兲 is given by: operating void fractions of the rising tube, falling tube, and the
evaporator. This way the mass balance gives an estimate of the
ṁ tot ṁ l ⫹ṁ v amount of liquid in the condenser that has to match with the liquid
G⫽ ⫽ (7) film distribution dictated by the condensation characteristics along
A A
the condenser wall. Thus, for a given initial charge in the thermo-
The hypothetical single-phase pressure gradient considering the syphon, m charge , the following overall mass balance needs to be
entire two-phase flow at the liquid density, is evaluated for bench- satisfied in terms of the component volumes and their predicted
marking purposes, as follows: operating void fractions.
⫺ 冉冊
dp
dz lo
⫽
2 f lo G 2
␳ ld h
(8) m charge⫽
4
兺关 ␳ 共 1⫺ ␣ 兲 ⫹ ␳ ␣ 兴 V
i⫽1
l i v i i (15)
The evaporator and condenser void fractions are dictated by the
heat flux input and the respective condenser heat transfer charac-
teristics. The above equation implicitly reconciles the heat transfer
h i 共 x 兲 ⫽h lo 冑冉
1⫹
␳l
␳v 冊
⫺1 x (21)
rates and the void fractions in a way that leads to the saturation
temperature. Intuitively the saturation temperature should be a Using the above h i (x) equation gave the following closed form
function of the condenser heat transfer characteristics, besides the solution of Eq. 16 for the local vapor quality 共x兲 along the con-
system initial filling ratio, and the heat flux. This was corroborated denser length 共z兲 applicable within the condensation region (x c,i
by blowing a fan over the condenser, and noting that the saturation ⭐x⭐x c,o ).
冉冑冉冊冑冉冊冊
temperature dropped by about 5°C around 40 W/cm2 wall heat
flux. ␳l ␳l
Az⫽B 共 x c,i ⫺x 兲 ⫹C 1⫹ ⫺1 x c,i ⫺ 1⫹ ⫺1 x
␳v ␳v
Tube Length Required for Complete Condensation. As the
(22)
in-tube condensation heat transfer coefficient h i (x) varies along
the condenser length due its dependence on vapor quality, a dif- where
ferential analysis was performed to determine the variation of va-
por quality 共x兲 along the condenser length 共z兲. The analysis was
␲ F 共 T lat⫺T amb兲
based on a condensing vapor-liquid mass balance causing a A⫽ (23)
change in quality 共dx兲 along a differential tubing element of length ṁ toti f g
dz; and the rejection of the associated latent heat from the con-
densing fluid to the ambient air through the wall of the element
under the tube inside and outside heat transfer characteristics. The
resulting formulation is as follows. B⫽
ln 冉冊
Do
Di
⫹
1
(24)
␩ h o D o,eff
冉冊
2k
␲ F 共 T sat⫺T amb兲
冋册
dx
冉冊
⫺ṁ toti f g ⫽ (16)
dz Do 2
冉冊
ln C⫽ (25)
1 Di 1 ␳l
⫹ ⫹ h lo D i ⫺1
␳v
h i共 x 兲 D i 2k ␩ h o D o,eff
The rectangular plate fin array attached to the condenser tubing Equation 22 gives the following condenser tubing length required
to enhance the natural convection heat transfer constitutes parallel for complete condensation (x⫽0).
冉冑冉冊冊
vertical channels that are open to the ambient air from all sides.
We used a semi-empirical model from Bar-Cohen and Rohsenow Bx c,i C ␳l
关13兴 for the air-side free convection heat transfer coefficient h o L cc ⫽ ⫹ 1⫹ ⫺1 x c,i ⫺1 (26)
A A ␳v
within the channels. As no more than a few degrees of subcooling
is anticipated, a constant h o is justified for the entire condenser The flow at the condenser outlet may be two-phase, saturated, or
tubing. subcooled depending upon whether the available condenser tubing
The effective outside diameter D o,eff accounts for the effect of length L c is less than, equal to, or greater than L cc , respectively.
the outside surface geometry in terms of the tube outside diameter
D o , the rectangular fin height 共H兲, width 共W兲, spacing 共S兲, and Condenser Outlet Thermodynamic State. Assuming that no
thickness 共t兲, as follows. heat is rejected and no phase change occurs in the rising and
冋冉冊册
falling tubes, the evaporator outlet thermodynamic state is ex-
1 ␲ 2 pected to be the same as the one at the condenser inlet 共T c,i
D o,eff⫽ 2HW⫺ D ⫹ ␲ D o S⫹2t 共 W⫹H 兲
␲ 共 S⫹t 兲 2 o ⫽T c,o ; x c,i ⫽x c,o 兲, and vice versa 共T e,i ⫽T c,o ; x e,i ⫽x c,o 兲. For
(17) steady-state thermosyphon operation, the above assumption also
We used the following correlation from Ananiev et al. 关14兴 for establishes equalities in the latent and sensible heat transfer in the
local in-tube condensation heat transfer coefficient, as required by evaporator and the condenser, i.e, q c,lat⫽q e,lat and q c,sen⫽q e,sen .
Eq. 16 However, there are two possible scenarios for applying the overall
condenser heat balance depending upon whether the condensation
h i 共 x 兲 ⫽h lo 冑 ␳l
␳共 x 兲
(18)
is complete or not. First, we compute the condenser length re-
quired for complete condensation L cc using Eq. 26. If L cc is less
than the available condenser tubing length L c , the remaining con-
denser length is filled with subcooled liquid. On the other hand, if
where the liquid-only heat transfer coefficient h lo is computed L cc turns out to be larger than L c , the entire condenser is filled
from the typical single-phase heat transfer correlations assuming with two-phase flow.
the entire two-phase flow to be liquid. The same liquid-only heat
transfer coefficient would also be applicable to the subcooled re- Complete Condensation. In this scenario, the available con-
gion, if it existed. For turbulent flow, we use. denser tubing length is sufficient for complete condensation, and
the condenser outlet vapor quality is zero (x c,o ⫽0). Here, first,
h lo,turd h the latent heat added in the evaporator is rejected in the conden-
⫽0.021 Re0.8
lo Pr
0.43
(19)
kl sation length of the condenser tubing at a constant saturation tem-
perature while vapor quality decreases. Once the condensation
For laminar flow with a constant heat flux boundary condition, we
process is complete, the sensible heat added in the evaporator is
use.
rejected in the remaining length of the condenser tubing, which is
h lo,lamd h filled with sub-cooled liquid. The condenser outlet temperature
⫽4.36 (20) and the amount of sensible heat rejected by the condenser are
kl
dictated by a balance between the heat transfer characteristics of
Using Eq. 6 for the homogeneous flow mean two-phase density, the subcooled section and the subcooled liquid’s capacity to reject
the local condensation heat transfer coefficient from Ananiev heat, as follows.
et al. 关14兴 is rewritten as a function of the local vapor quality, as The sensible heat, q c,sen , rejected by the sub-cooled liquid sec-
follows. tion of length L sub to the ambient air is as follows
␲ L subF sub共 LM TD sub,c f 兲 Integrating the above expression from the maximum vapor quality
冋冉冊册
q c,sen⫽ (27) x c,i at the condenser inlet through the minimum vapor quality x c,o
Do
ln generally occurring at the condenser outlet, we get the following
冋
1 Di 1 formulation of the total vapor core volume in the condenser.
⫹ ⫹
h lo D i 2k ␩ h sub,o D o,eff
冉冊 ␳l
冦
where the counter-flow log-mean-temperature difference for the B (x c,i ⫺x c,o )
␲ D 2i ␳v 1
冉冊冉冊
subcooled section is defined in terms of the condenser inlet, outlet, V v ,cond⫽ 1⫺
and ambient temperatures as follows. 4A ␳l ␳l
⫺1 ⫺1 (x c,i ⫺x c,o )
冉冊册
共 T c,i ⫺T amb兲 ⫺ 共 T c,o ⫺T amb兲 ␳v ␳v
冉冊冉冉冊冊冉冊
LMTDsub,c f ⫽ (28)
T c,i ⫺T amb ␳l ␳l
ln 1⫹ ⫺1 x c,i C
T c,o ⫺T amb ␳v ␳v
冉冉冊冊冉冊
Therefore, the following specific heat balance also has to hold. ⫻ln ⫹
␳l ␳l
q c,sen⫽ṁ totc p,l 共 T c,i ⫺T c,o 兲 1⫹ ⫺1 x c,o ⫺1
(29) ␳v ␳v
冋冉冑冉
Thus, the amount of sensible heat addition in the evaporator is
冊冑冉冊冊
dictated by the degree of subcooling attained in the condenser.
␳l ␳l
Incomplete Condensation. Here, the available condenser tub- ⫻ 1⫹ ⫺1 x c,i ⫺ 1⫹ ⫺1 x c,o
␳v ␳v
ing length is not sufficient for all the vapor to condense, and the
冉冑冉冊册冧
condenser outlet is two-phase that also suggests a constant tem-
perature throughout the condenser (T c,o ⫽T c,i ). Thus, no sub- 1 1
⫺ ⫺
冊冑冉冊
cooled liquid core exists and no sensible heat rejection is involved
here; and the entire thermosyphon operates at a constant satura- ␳l ␳l
tion temperature/pressure. The condenser outlet vapor quality x c,o 1⫹ ⫺1 x c,o 1⫹ ⫺1 x c,i
␳v ␳v
can be obtained by solving Eq. 22 for the known total condenser
length L c , as follows.
冋冑冉冊
(35)
␳l As the above formulation is developed in terms of inlet and
AL c ⫽B 共 x c,i ⫺x c,o 兲 ⫹C 1⫹ ⫺1 x c,i
␳v outlet vapor qualities, it is also valid for the case of complete
冑冉冊册
condensation with minimum vapor quality x c,o (⫽0) occurring
␳l before the condenser outlet.
⫺ 1⫹ ⫺1 x c,o (30)
␳v
Evaporator Mass and Energy Balances. The following
Condenser Vapor Volume. The model assumes an annular mass balance was applied at the evaporator inlet and outlet.
two-phase flow all along the condensation length. A liquid film of
mean thickness ␦ growing on the tube wall surrounds the vapor ṁ tot ⫽ṁ v ,i ⫹ṁ l,i ⫽ṁ v ,o ⫹ṁ l,o (36)
core that diminishes in size due to condensation. The peripheral The overall evaporator heat transfer was, in general, assumed to
variation in the liquid film thickness due to gravity is considered be divided into latent and sensible parts.
insignificant for volumetric calculations. The local void fraction,
␣, at a general location z is defined in terms of the local mean film q e ⫽q e,lat⫹q e,sen (37)
thickness ␦ ⫽ ␦ (z) and the condenser tube inside diameter D i , as In case of a complete condensation in the condenser, sensible
follows: heat must be added in the evaporator to compensate for the sen-
sible heat rejection from the condenser subcooled liquid core
共 D i ⫺2 ␦ 兲 2 (q e,sen⫽q c,sen⫽0). On the other hand, if the condensation is not
␣⫽ (31)
D 2i complete, there would be no sensible heat rejection from the con-
denser or addition in the evaporator (q e,sen⫽q c,sen⫽0). In the
Using this expression, a differential vapor volume along the con-
evaporator, the sensible heat is added first to the subcooled liquid
densation length turns out to be
returning from the condenser to bring its temperature up to the
␲ 2 saturation level. Once the saturation temperature is reached, fur-
dV v ⫽ D ␣ dz (32) ther heat addition to the fluid is latent and increases its vapor
4 i
quality through boiling on the enhanced surface.
Using the following expression for local homogeneous flow void The following latent heat balance also needs to be satisfied in
fraction the evaporator.
␳l
␳v
x冉冊 q e,lat⫽ṁ toti f g 共 x e,o ⫺x e,i 兲 (38)
冉冊
␣⫽ (33) Evaporator Gravitational Pressure Drop. In the absence of
␳l a reliable model to predict the two-phase friction and acceleration
1⫹ ⫺1 x
␳v pressure drops through the cubic evaporator enclosure, we only
considered the gravitational pressure change in the evaporator.
along with the z⫽z(x) formulation in Eq. 22, Eq. 32 can be Following Tengblad 关10兴, we assumed the evaporation rate to be
冋册
rewritten as: uniform all along the evaporator. So, the well-mixed vapor quality
dV v ⫺ ␲ D 2i
B 冉冊
␳l
␳v
x 冉冊冉冊
C
␳l
␳v
⫺1
␳l
␳v
x
increases linearly along the evaporator height, z, as follows.
冉冊
x o ⫺x i
冉冊冉冉冊冊
⫽ ⫹ x⫽x i ⫹ z (39)
dx 4A ␳l ␳l 3/2 ⌬H
1⫹ ⫺1 x 2 1⫹ ⫺1 x
␳v ␳v where ⌬H is the change in the elevation 共gravitational level兲 from
(34) the inlet to the outlet of the evaporator.
Using the above idealization with the homogeneous flow mean Parametric Study
density given by Eq. 6, the gravitational pressure drop in the
An interesting feature of the present model is its highly coupled
evaporator turns out to be, as follows.
nature, which has created a very iterative computational environ-
⌬ p gr ⫽⫺g 冕 ⌬H
¯␳ dz
ment. For example, to perform the mass and energy balances on
the evaporator, one needs to know whether the condensation is
冋册
o
complete or not and whether subcooled liquid or a two-phase flow
⫺ ␳ l g⌬H 冉冉冊冊
1⫹
␳l
␳v
⫺1 x o
should be expected at the evaporator inlet. Conversely, the same
balances on the condenser require the knowledge of heat addition
冉冊冉冉冊冊
⫽ ln (40) in the evaporator. Besides, the evaporator and condenser balances
␳l ␳l require the mass flow rate that is not known until the pressure
⫺1 共 x o ⫺x i 兲 1⫹ ⫺1 x i
␳v ␳v balances are applied, which in turn require the vapor qualities that
are not known until the evaporator/condenser mass and energy
Even though the above expression is apparently derived for the balances have been applied. We used the commercially available
evaporator, it is equally applicable to the other thermosyphon software Engineering Equation Solver 共EES兲关15兴 to solve this
parts such as the rising and falling tubes. In case of upward flow system of coupled non-linear equations.
through the evaporator or the rising tube, ⌬H is positive that The model predictions are compared with experimental data for
results in a negative ⌬ p gr or a drop in the gravitational pressure dielectric working fluid PF-5060. As shown in Fig. 1, the rectan-
head. For the downward flow through the falling tube, ⌬H is gular loop in the present experimental setup lies in a vertical plane
negative that signifies a positive ⌬p gr or a rise in pressure.
with the exception of the horizontal condenser, which is oriented
Using Eq. 39 with Eq. 33 for local homogeneous void fraction,
perpendicular to the vertical plane. The two-tube-pass condenser
冉冊冋冉
the overall void fraction of the evaporator is modeled as
has 75 cm long aluminum tubing, and is naturally cooled by am-
␳l bient air using an array of aluminum plate fins attached to its 8.5
冕
mm outside-diameter tubing. The 0.16 mm thick rectangular fins
1 xo ␳v 1
冊
¯␣ e ⫽ ␣ dx⫽ 1⫺ are 60 mm high and 27 mm wide with 4.34 mm spacing. An
x o ⫺x i xi ␳l ␳l overall fin efficiency 共␩兲 of 0.70, and a correction factor 共F兲 of
⫺1 ⫺1 共 x o ⫺x i 兲
␳v ␳v
冉冉冉冊冊冊册
0.90 are used for the present computations. The thermosyphon
冉冉冊冊
data were taken by boiling the working fluid PF-5060 over a cubic
␳l enhanced structure confined inside an 8 cm3 evaporator. A 3.125
1⫹ ⫺1 x o
␳v mm inside diameter transparent tygon tubing is used as flexible
⫻ln (41) rising and falling tubes that connect the plexiglass evaporator to
␳l
1⫹ ⫺1 x i the 6.9 mm inside diameter condenser tubing. More details on the
␳v experimental procedure, or the magnified geometry of the con-
Evaporator Enhanced Boiling Surface Characteristics. denser, evaporator and the micro-fabricated enhanced boiling
Experiments by Yuan et al. 关1兴 were conducted with the dielectric structure can be obtained from Yuan et al. 关1兴.
fluid PF-5060. The following relation between the wall heat flux The condenser elevation and the separation between the rising
共W/cm2兲 and the wall superheat 共K兲 was found by curve-fitting and falling tubes are adjustable. The tested condenser elevation is
their observed boiling data for 50% initial fill height. 30 cm above the chip surface. For a 30 cm⫻30 cm arrangement,
the total thermosyphon fluid volume is 44.41 cm3. The volume
q w ⫽1.2263共 T w ⫺T sat兲 1.1578 (42) distribution among the evaporator, the rising tube, the condenser,
and the falling tube is 16.4 percent, 10.1 percent, 63.1 percent,
The above fit corresponds to a maximum spread of ⫾2.5 C in 10.3 percent, respectively. For a 50 percent initial fill height at an
the wall superheat and a maximum spread of ⫾4.5 W/cm2 in the ambient temperature of 23.4 C, the total initial liquid and vapor
wall heat flux, due to the uncertainty in the experimental data. The volumes are 11.86 cm3 and 32.55 cm3, respectively. A 0 percent
above equation obtained with the highly confined boiling en- initial fill height means that the evaporator is completely filled but
hancement structure does not represent a typical boiling curve, there is no liquid in the rising tube; while a 100 percent fill height
and rather suggests a roughly constant evaporator heat transfer means that the rising tube is completely filled up to the end of the
coefficient not very sensitive to the wall superheat. We expect two 90 deg bend but there is no liquid in the horizontal section of the
reasons for this unconventional trend. First, as opposed to the rising tube. The 23.4 C ambient temperature corresponds to a
typical pool boiling curve, every data point corresponds to a dif-
saturation pressure of 0.2703⫻105 Pa compared to an atmospheric
ferent system pressure. Secondly, the above equation may also
reflect any suppression caused by the convective effects of the pressure of 1.01325⫻105 Pa. Using a constant liquid density, ␳ 1
condensate returning to the evaporator. It appears that the dynamic ⫽1680 kg/m3 with the computed vapor density ␳ v ⫽3.71 kg/m3 ,
situation in the evaporator is more similar to flow boiling than the liquid and vapor masses are 19.93⫻10⫺3 kg and 0.12
pool boiling. As the equation is based on actual data, it implicitly ⫻10⫺3 kg, respectively. Thus, the total fluid mass in the system is
accounts for the pressure variation and any boiling suppression 20.05⫻10⫺3 kg. The following figures show various predictions
involved. by the present model over the tested heat flux range of 5.9– 40.2
The wall heat flux, q w , and the wall temperature, T w , in the W/cm2. They demonstrate the capabilities of the model, and pro-
above empirical equation are based on the projected base area of vide the estimate of the quantities of interest.
the cubic enhanced structure. In the present study this also simu- Figure 3 shows that the variation of available gravitational head
lates the footprint area of the heat dissipating chip to which the between 2272 Pa to 3452 Pa, and the total mass flow rate between
enhanced boiling structure would be attached. Using this area, the 2.6⫻10⫺3 kg/s to 3.2⫻10⫺3 kg/s it is able to impose through the
overall heat transfer input to the evaporator, Q w (W), could be loop. Figure 4 shows the predicted condenser inlet/outlet vapor
evaluated. In the present model, Eq. 42 is used to predict the wall qualities, and shows an almost complete condensation throughout
temperature for the input heat flux and the computed T sat . Finally, the test range. As expected, as heat flux increases, the evaporator
for the given ambient temperature, T amb , the overall thermal re- outlet vapor quality increases. Whether the condensation is com-
sistance of the thermosyphon is evaluated as. plete would strongly depend upon the condensation correlation
and the void fraction model used. A less than 1.0 evaporator outlet
T w ⫺T amb
R tot⫽ (43) vapor quality, as suggested by Fig. 4, is in accordance with our
Qw visual observations. We were able to make the visual observation
Fig. 6 Condenser and evaporator overall void fractions
Fig. 3 Available gravitational pressure head and loop mass
flow rate
the possible sensible heat transfer through the sub-cooled liquid
region, as its impact on the prediction of condenser void fraction
as the plexiglass evaporator and the connecting tygon tubings are could far outweigh its numerical significance relative to the latent
transparent. We did experimentally observe a liquid-vapor two- part.
phase plug/slug flow leaving the evaporator and rising through the Figure 6 shows a void fraction variation of 4.0 percent in the
vertical tygon tubing to the condenser throughout the test range. evaporator and 2.3 percent in the condenser over the entire heat
We were able to clearly see significant liquid fraction being pulled flux range tested. As a related observation from the model, the
into the vertical tygon tubing throughout the heat flux test range of overall void fraction of the system does not change significantly.
practical interest. Because of the dynamic boiling activity on the At 23.4 C ambient temperature, the initial void fraction of the
micro-fabricated cubic boiling structure inside the evaporator, the system was 73.29 percent while at 55.2 C saturation temperature,
generated vapor drags some of the saturated liquid inside the it became 73.62 percent. Thus, about 0.45 percent of the initial
evaporator into the vertical tygon tubing. So, the re-circulatory liquid volume was evaporated to increase the vapor density by
two-phase flow involves some liquid that is neither evaporated nor three folds to 11.25 kg/m3. The corresponding change of 0.33
condensed. Thus, our flow visualizations are consistent with the percent in the system void fraction is negligible, with the possi-
predicted evaporator outlet quality to be significantly less than bility that individual void fractions may vary rather significantly
unity that still corresponds to significantly high void fraction due in the evaporator, condenser, and rising and falling tubes.
to the liquid-vapor density difference. As shown in Fig. 6, the model predicts an asymptotic plateau-
Figure 5 shows that all heat transfer is latent throughout the ing of the condenser and evaporator overall void fractions starting
tested range, which again implies that condensation is either just around 25–30 W/cm2. As Figs. 4 and 5 have shown an almost
completed or is incomplete at the condenser outlet. On the con- complete condensation at the condenser outlet throughout the test
trary, if the applied condenser were long enough to condense all range, the falling tube is filled with liquid phase returning to the
generated vapor before the condenser outlet, the condensed liquid evaporator with very little vapor. For an almost constant liquid
would be subcooled by the time it would reach the condenser volume in the system, the overall void fraction of the rising tube
outlet and the evaporator inlet. Then, a part of the heat conducted would also start plateauing around 25–30 W/cm2. Thus, the over-
through the enhanced boiling structure would be used to sensibly all void fraction of each of the four thermosyphon components
heat the returning subcooled liquid up to the saturation tempera- becomes constant above 25–30 W/cm2 heat flux. Further rise of
ture corresponding to the evaporator pressure. We accounted for saturation temperature and pressure, caused by increasing heat
flux, would compress additional vapor in the same vapor volume
in the rising tube, which explains the drop of available gravita-
tional pressure head as shown in Fig. 3. However, an increased
mass flow rate with higher vapor quality is predicted to remove
the increasing heat flux, and is sustained by a rather declining
available pressure head due to the non-linear dependence of the
pressure drop on vapor quality and void fraction that are on the
rise, as suggested by Eq. 5 through 13 of the homogeneous two-
phase flow model.
Figure 7 shows various predicted Reynolds numbers. It can be
seen that the liquid-only Reynolds numbers for the 3.175 mm dia
rising and falling tubes, and the 6.9 mm dia. condenser tube are
within the laminar range, and are fairly insensitive to the heat flux.
Fig. 4 Condenser inlet and outlet vapor qualities However, the Reynolds number of the vapor core is turbulent over
Fig. 5 Latent and sensible contributions to the heat flux Fig. 7 Two-phase flow Reynolds numbers
percent in the condenser over the entire heat flux range tested,
compared to a corresponding 0.33 percent change in the overall
system void fraction.
Currently, there are two major uncertainties in the model. The
first uncertainty is inducted by using a conventional condensation
heat transfer correlation developed for large tubes. With an almost
constant evaporator heat transfer coefficient, the observed trends
in the total thermal resistance are caused by the condensation
thermal resistance. At low heat flux, the condensation heat transfer
coefficient is expected to be poor due to small vapor flow rates. As
heat flux increases, so does the vapor flow rate in the core, which
should drive the condensation heat transfer coefficient up. The
Fig. 8 A comparison between the predicted and observed currently used Ananiev correlation does not account for an order
saturation temperatures of magnitude variation in the convective effects in the vapor core
inside the condenser. More realistic predictions are expected once
better condensation correlations are used that are sensitive to the
the tested heat flux range, and varies by a factor of eight over the vapor convective effects. Thus, a deteriorating condenser heat
tested heat flux range of 5.9– 40.2 W/cm2. The finding suggests a transfer coefficient can explain higher thermal resistances ob-
turbulent vapor core in the middle surrounded by a laminar liquid served at lower heat fluxes. The second uncertainty stems from
film on the condenser wall. This suggests a need to assess the role using a homogeneous flow assumption of equal mean liquid and
of the turbulent vapor convective effects, as well as the no-slip vapor velocities inside the condenser, i.e., slip factor S⫽1. This is
condition between the mean liquid and vapor velocities in the clearly not the case for the condenser, where the vapor velocity
condenser. Contrary to the currently used homogeneous flow as- could considerably change over the tested heat flux range due to
sumption, the mean liquid and vapor velocities may be different the well-separated annular flow. The introduction of a slip factor is
due to separated flow in the condenser, especially, at high heat bound to affect the void fraction prediction throughout the ther-
fluxes. Sudden expansion at the condenser inlet could change the mosyphon.
void fraction without changing the quality. It is concluded that with all basic features of the model success-
Figure 8 compares the observed and predicted saturation tem- fully identified, the key lies in using the two-phase heat transfer
peratures. It is believed that the discrepancy in the predicted and correlation developed for small diameter horizontal tubes and
measured saturation temperatures is caused by a combined effect validated with fluids similar to PF-5060, such as R-113.
of the uncertainties in the condensation heat transfer and the indi-
vidual void fractions. The saturation temperature was measured by
using thermocouples inside the liquid pool of the evaporator. It is Acknowledgment
worth emphasizing that the present model did not introduce any
empirical constant. An underprediction of T sat at low heat fluxes The present research is sponsored through the DARPA
and an over prediction at high heat fluxes suggests that we may be HERETIC Program under contract N00164-99-C-0039, managed
using a higher condensation heat transfer coefficient at low heat by Dr. Dan Radack and Dr. Elias Towe. This work was done when
flux and a lower heat transfer coefficient at higher heat flux. It the authors were at the University of Maryland, College Park.
may be due to the fact that Ananiev correlation does not account
for the factor of eight variation in the vapor core Reynolds num-
ber. The use of liquid-only Reynolds number in annular two-phase Nomenclature
flow tends to hide the convective effects of the vapor core by
lumping the laminar liquid and turbulent vapor flow rates together. cp ⫽ specific heat, J/kg/K
This also shows the importance of correctly predicating the local dh ⫽ hydraulic diameter, m
condensation heat transfer coefficient, which is recognized as a Di ⫽ Tube-inside diameter, m
major area of uncertainty in the present model. It is quite possible Do ⫽ Tube-outside diameter, m
that for the present thermosyphon setup, the condensation heat f ⫽ friction factor
transfer coefficient may not follow the large tube condensation F ⫽ heat exchanger correction factor for the cross flow
correlations due to the laminar condenser operation and surface effect
tension effects. The possibility of existence of subcooled liquid g ⫽ acceleration due to gravity, m/s2
film within the condensation length, i.e., in the laminar liquid G ⫽ two-phase flow mass velocity, kg/m2/s
layer beneath the saturated liquid-vapor interface, cannot be ruled hi ⫽ Condenser inside heat transfer coefficient, W/m2/K
out. This way subcooled liquid could still exist even though there ho ⫽ Condenser outside heat transfer coefficient, W/m2/K
may not be a 100 percent liquid filled subcooled core right before H ⫽ magnitude of the gravitational pressure head, Pa
the condenser outlet. ifg ⫽ latent heat of vaporization, J/kg
k ⫽ thermal conductivity, W/m/K
L ⫽ tube length, m
Conclusions m ⫽ coolant mass, kg
The present paper proposes an analytical approach to modeling ṁ ⫽ mass flow rate, kg/s
the closed loop two-phase thermosyphon 共CLTPT兲 performance. LMTD ⫽ log-mean temperature difference, K
The model identifies the fundamental elements, and has a closure p ⫽ pressure, Pa
in the form of the prediction of the system saturation temperature. Pr ⫽ Prandtl number
The model is compared with experimental data for dielectric q ⫽ heat flux, W/cm2
working fluid PF-5060 and is in general agreement with the ob- Q ⫽ total heat input, W
served trends. The simulation results for the given experimental R ⫽ thermal resistance, K/W
setup and the test range suggest an annular two-phase flow with a Re ⫽ Reynolds number
turbulent vapor core in the middle surrounded by a laminar liquid T ⫽ temperature, degree C
film on the condenser wall. The results predict all heat transfer in V ⫽ volume, m3
the condenser and evaporator to be latent. Simulations show a x ⫽ two-phase flow vapor quality
void fraction variation of 4.0 percent in the evaporator and 2.3 z ⫽ general location along the condenser tubing, m
Greek Letters References
关1兴 Yuan, L., Joshi, Y. K., and Nakayama, W., 2001, ‘‘Effect of Condenser Loca-
␣ ⫽ void fraction tion and Tubing Length on the Performance of a Compact Two-Phase Ther-
␦ ⫽ local condensate liquid film thickness, m mosyphon,’’ Proc. 2001 ASME International Mechanical Engineering Confer-
␩ ⫽ condenser fin efficiency ence and Exposition, Session on Heat Pipes and Multiphase Heat Transfer in
Electronic System, November 11–16, New York, NY.
␮ ⫽ viscosity, kg/m/s 关2兴 El-Genk, M. S., and Saber, H. H., 1998, ‘‘Heat Transfer Correlations for Liq-
␪ ⫽ tube inclination with the horizontal, radian uid Film in the Evaporator of Enclosed, Gravity-Assisted Thermosyphons,’’
ASME J. Heat Transfer, 120, pp. 477– 484.
␳ ⫽ density, kg/m3 关3兴 Harley, C., and Faghri, A., 1994, ‘‘Complete Transient Two-Dimensional
Analysis of Two-Phase Closed Thermosyphons Including the Falling Conden-
Subscript sate Film,’’ ASME J. Heat Transfer, 116, pp. 418 – 426.
关4兴 Zuo, Z. J., and Gunnerson, F. S., 1994, ‘‘Numerical Modeling of the Steady-
acc ⫽ acceleration state Two-Phase Closed Thermosyphon,’’ Int. J. Heat Mass Transf., 37共17兲, pp.
amb ⫽ ambient 2715–2722.
关5兴 El-Genk, M. S., and Saber, H. H., 1999, ‘‘Determination of Operation Enve-
c ⫽ condenser lopes for Closed, Two-phase Thermosyphons,’’ Int. J. Heat Mass Transf., 42,
cc ⫽ complete condensation pp. 889–903.
cf ⫽ counter flow 关6兴 Vincent, C. C. J., and Kok, J. B. W., 1992, ‘‘Investigation of the Overall
Transient Performance of the Industrial Two-Phase Closed Loop Thermosy-
e ⫽ evaporator phon,’’ Int. J. Heat Mass Transf., 35, No. 6, pp. 1419–1426.
fall ⫽ falling tube 关7兴 Wang, J. C. Y., and Ma, Y., 1991, ‘‘Condensation Heat Transfer Inside Vertical
and Inclined Thermosyphons,’’ ASME J. Heat Transfer, 113, pp. 777–780.
fr ⫽ frictional 关8兴 Carey, Van P., 1992, Liquid-Vapor Phase-change Phenomena, Hemisphere
gr ⫽ gravitational Publishing Corporation.
关9兴 Cammarata, G., Fichera, A., Froghieri, M., Misale, M., and Xibilia, M. G.,
i ⫽ inlet 1999, ‘‘A New Modeling Methodology of Natural Circulation Loop for Sta-
l ⫽ liquid bility Analysis,’’ Single and Two-Phase Natural Circulation, Proceedings of
EUROTHERM SEMINAR No. 63, September 6 – 8, 1999, Genoa, Italy.
lat ⫽ latent 关10兴 Tengblad, N., 1996, ‘‘Cooling of Electronic Components with Closed Loop
lo ⫽ liquid-only Two-Phase Thermosiphons,’’ Engineering Licentiate Thesis, The Royal Insti-
o ⫽ outlet tute of Technology, Stockholm, Sweden, pp. 17–50, Chap. 3.
关11兴 McAdams, W. H., Woods, W. K., and Heroman, Jr., L. C., 1942, ‘‘Vaporization
rise ⫽ rising tube Inside Horizontal Tubes—II—Benzene-Oil Mixtures,’’ Trans. ASME, 64, p.
sat ⫽ saturation 193.
关12兴 Collier, J. G., 1981, Convective Boiling and Condensation, Second Edition, pp.
sen ⫽ sensible 30–34. McGraw-Hill Book Company 共UK兲 Ltd.
sub ⫽ subcooling 关13兴 Bar-Cohen, A., and Rohsenow, W. M., 1984, ‘‘Thermally Optimum Spacing of
Vertical Natural Convection Cooled, Parallel Plates,’’ ASME J. Heat Transfer,
tot ⫽ total 106, p. 116.
tp ⫽ two-phase 关14兴 Ananiev, E. P., Boyko, L. D., and Kruzhilin, G. N., 1961, ‘‘Heat Transfer in the
Presence of Steam Condensation in a Horizontal Tube,’’ Proc. 1st Int. Heat
v ⫽ vapor Transfer Conf., part II, p. 290.
w ⫽ wall, i.e, chip surface 关15兴 Engineering Equation Solver 共EES兲, F-Chart Software, www.fChart.com.
Heat Transfer Characteristics
of Melting Ice Spheres Under
Forced and Mixed Convection
Y. L. Hao
An experimental investigation is conducted to characterize melting and heat transfer
Y.-X. Tao during convective melting of an ice sphere in horizontally flowing water. The shape
e-mail: ytao@eng.fiu.edu changes of melting ice spheres with time at different velocities and supply temperatures of
water and at different initial ice temperatures are recorded using a digital video cam-
Department of Mechanical Engineering, corder. From the recorded images, the time variations of local melting rate, local heat
Florida International University, transfer coefficient and local Nusselt number at various angular positions are obtained.
Miami, FL 33199 The effects of water temperature and velocity on these local parameters are analyzed and
the total melting rate and average heat transfer coefficient are determined. Guided by a
shape factor characterizing the particle shape variation, an empirical correlation for
average Nusselt number in convective melting of a solid particle is obtained. The experi-
mental results provide important particle-level information needed to improve the accu-
racy of numerical models for convective melting of a single particle or packed particles.
关DOI: 10.1115/1.1494090兴
Keywords: Heat Transfer, Melting, Mixed Convection
1 Introduction The previous studies mentioned above focus mainly on either

the melting process of a sphere in a pool of liquid or heat transfer
Melting of ice or solid bodies occurs in many important engi-
between fluid and sphere in a flowing liquid. To our knowledge,
neering and environmental applications such as material process-
there has been no complete study on the full melting process of a
ing, space applications and water resource and energy conversion sphere in a flowing liquid, especially where the shape of the solid
and conservation. Tkachev 关1兴 examined the melting of ice changes considerably from that of a sphere. In order to fill this
spheres as well as vertical and horizontal cylinders using photo- gap, Hao and Tao 关12兴 presented the visualization and qualitative
graphic techniques. Merk 关2兴 obtained a boundary layer solution analysis of convective melting process of ice spheres and ice cyl-
and predicted a minimum Nusselt number of the melting of ice inders and examined the influence of water velocity. They further
spheres using the von Karman-Pohlhausen integral method. reported more detailed flow visualization based on the particle
Schenk and Schenkels 关3兴 observed the region of dual flow around image velocimetry 共PIV兲 technique 共Hao and Tao 关13兴兲. In their
an ice sphere produced by freezing a 3-mm-thick ice layer onto a study, the distributions of velocity, streamline, and vorticity
hollow copper sphere of diameter 3.2 cm, and determined the around the ice sphere are obtained for different upstream veloci-
subsequent melting rate from pictures obtained during free- ties and temperatures. The focus is on flow characteristics around
convection melting in water. The results provided an equation for the melting ice sphere and effects of velocity and temperature.
the Nusselt number in terms of the Grashof and Prandtl numbers. The visualization of the melting of a dyed ice sphere is also con-
Vanier and Tien 关4兴 conducted experiments on free-convection ducted to qualitatively investigate the motion of the melt and the
melting of a submerged ice sphere in water, focusing on boundary mixing with mainstream. These results are very helpful in under-
layer motion existing at 4°C. Accelerated melting in the bottom standing the interaction between fluid and melting solid particles,
region of the sphere was due to heavy circulation in the wake of and lay a foundation for this paper to examine the melting and
the boundary layer flow. This resulted in the formation of a scal- heat transfer characteristics of ice particles in flowing water.
lop at the bottom. Eskandari 关5兴 and Eskandari et al. 关6兴 reported
on a series of experiments to study forced convection heat transfer
2 Experimental Setup and Procedure
from a flowing stream of water over an ice sphere, and developed
two relationships to describe the process. Anselmo 关7,8兴 per- The experiments are performed in a horizontal, open channel
formed an extensive theoretical and numerical analysis of the made of Plexiglas. The experimental setup, similar to the one
melting of a solid sphere partially or fully submerged in a pool of presented in Hao and Tao 关13兴, as shown in Fig. 1, consists of a
its own melt. Anselmo accounted for variation in heat transfer closed and forced water circulation loop and a temperature control
coefficient due to change in sphere diameter and flow condition. circulation loop. A MARCH TE-7R-MD pump moves water from
Mukherjee et al. 关9兴 conducted a visualization study on melting of the reservoir and discharges it into the head tank, from which the
an ice sphere in a pool of water using dyed ice. Aziz et al. 关10兴 water flows into the channel. Water flows downstream in the chan-
used the weighing method to obtain experimental data on heat nel past the sphere, which is mounted near the middle of the
transfer between a sphere and forced convective water, and pre- channel, and is then routed back to the reservoir. The discharge
sented an empirical correlation. Mcleod et al. 关11兴 conducted a line has a gate valve and a John C. Ernst Co S100 flow meter to
visualization study of melting of a wax sphere in hot water. They control and measure the flow rate of water. Water temperature is
calculated the melting rate using a simple theoretical analysis controlled by a refrigerated circulator with a measured fluctuation
which estimates melt layer thickness and the heat flux from the of ⫾0.1°C during a typical experiment.
fluid. The dimensions of the test section are 500⫻152⫻216 mm 共L
⫻W⫻H兲, as shown in Fig. 2. A uniform velocity at the inlet of the
test section is attained by the use of a flow straightener made of
HEAT TRANSFER. Manuscript received by the Heat Transfer Division October 24, Duocel aluminum foam. An overflow plate is placed at the down-
2001; revision received April 16, 2002. Associate Editor: C. Amon. stream location to control the water level. Using the gate placed at
Fig. 1 Schematic of the test apparatus
the exit location instead of the overflow plate can also control the
water level. A water level of 130 mm can be achieved in the
present study. Twelve thermocouples are mounted at the inlet, test
section, and exit so that the water temperature can be measured at Fig. 3 Volume element in the ice sphere and volume elements
different locations along the flow direction. A pair of thermo- between t and t ¿ dt
couples is placed in the center of the ice sphere to measure the
central temperature of the ice sphere during the melting process. A
computer with OMEGA DAS-TC/B data acquisition system is
stainless steel. In this study, the diameter of the ice sphere is 36
used to record the thermocouples’ readings during tests.
mm, and the diameter of the thin plastic straw is 2 mm. The insert
The ice sphere is made by freezing filtered tap water in two
length of the straw into the sphere is about 18 mm. A pair of
hemispherical moulds. A thin plastic straw is sandwiched in the
thermocouples is placed in the center of ice sphere through the
sphere to allow the sphere to be fixed on a movable base made of
plastic straw. The Plexiglas walls of the test section allow optical
access for photography and the passage of light. A Sony DCR-
VX1000 digital camcorder is used to continuously record the im-
age from the front side, as shown in Fig. 1 and Fig. 2. A Minolta
RD-175 digital camera is used to intermittently take pictures from
the top. To obtain a clear image of the ice sphere, an ivory white
background plate is vertically placed at 200 mm from the backside
of the test section. Two 45-Watt Philips reflector floodlights are
mounted above the front side of the test section. One floodlight
illuminates the ice sphere from the upper front side, and another
illuminates the background plate.
During a typical test, the pump is activated and a steady, open-
channel flow is established at the desired flow rate and tempera-
ture by adjusting the opening of the gate valve and using the
refrigerated circulator. The digital video camcorder and the com-
puter data acquisition system begin to record video images and
temperatures. The test begins when the sphere, fixed on the mov-
able basis by a thin plastic straw, is placed into the horizontal
water flow, and it ends when any remaining ice breaks away from
the plastic straw and drifts downstream. After the test is finished,
the shape data of the ice particle in the melting process can be
obtained from the digital video images on the computer by using
an image analysis software. The maximum uncertainty of the di-
mensional measurements is ⫾0.2 mm.
As illustrated in Fig. 3共a兲, we consider an ice sphere situated in
Fig. 2 Test section a horizontal flow of water from the right. The local mass melting
rate can be determined in terms of the volume variation of volume difference in the interior of ice sphere is less than 0.02°C. Based
element on the ice surface. As a result of the shape change and on this, we simplify the problem by introducing the following
different local melting rates at different locations on the surface, equation:
the centroid of the particle does not stay at a fixed physical loca-
tion referenced to the ground. To take this into account and also to
find a simple yet accurate way to calculate the local melting rate, dV s dT̄ s
we define a time dependent, geometric center of the ice particle at Q s ⫽ ␳ s c ps 共 T̄ s ⫺T si 兲 ⫹ ␳ s c ps V s , (4)
dt dt
a given time by finding the intersection of the maximum
x-direction linear dimension and maximum y-direction linear di- where T̄ s is the average internal temperature of the ice particle.
mension lines on the center plane image 共recall that the particle is We approximate T̄ s by using the center temperature measured in
symmetrical about the vertical center plane兲. At a given time in- the melting process. T si is the ice temperature at the surface of the
terval from t to t⫹dt, the particle center moves from (x c ,y c ) to ice, which is equal to the melting temperature of ice, T si ⫽T li
(x c ⫹dx c ,y c ⫹dy c ), as illustrated in Fig. 3共b兲. The corresponding ⫽0°C. Equation 共4兲 is applicable to both the spherical and non-
volume element of the ice particle at an angular position, ␪, de- spherical shapes of the ice particle during the melting process.
fined at the moving origin, therefore changes from V( ␪ ) to Also based on the analytical solution mentioned above, we may
V( ␪ )⫹dV( ␪ ). Accordingly, the radius changes from R( ␪ ) to assume that the heat flux conducted into the ice particle is uniform
R( ␪ )⫹dR( ␪ ). This treatment, although arbitrary, allows easy on the particle surface. It then follows that
data reduction and presentation using spherical coordinates with
the origin coinciding with the above-defined geometric center of
ice sphere, as shown in Fig. 3共a兲. Therefore, it follows that the q s 共 ␪ 兲 ⫽q̄ s ⫽Q s /A S . (5)
local melting rate per unit surface area at an angular position ␪ The local heat transfer coefficient can be determined from Eq.
共measured from the horizontal centerline兲, ṁ ⬙ ( ␪ ), can be deter- 共2兲 after the local melting rate and the conduction heat flux are
mined as follows: determined,
冉冊
1
␳ s dV 共 ␪ 兲 dR 共 ␪ 兲 dx c dy c h共 ␪ 兲⫽ 关 ṁ ⬙ 共 ␪ 兲 h ls ⫹q s 共 ␪ 兲兴 . (6)
ṁ ⬙ 共 ␪ 兲 ⫽⫺ ⫽⫺ ␳ s ⫹cos ␪ ⫹sin ␪ . T l⬁ ⫺T li
⌬A s 共 ␪ 兲 dt dt dt dt
(1) The total melting rate over the surface of ice sphere can be
obtained by integrating the local melting rate over the surface of
where ⌬A s ( ␪ ) is the surface area element at ␪. The second and
ice sphere. We can also experimentally determine the total melting
third terms in the right-hand side parenthesis of the above equa-
rate by measuring the volume change of the ice sphere over time
tion results from the aforementioned definition of the moving geo-
based on
metric center 共origin兲 as a result of shape change. The variations
of radius as a function of an angular position, R( ␪ ), and coordi- dV s
nates of the geometric center, x c and y c , with time t can been ṁ⫽⫺ ␳ s . (7)
dt
obtained directly from the digital video images of the ice particle
in the melting process. The average heat transfer coefficient over the surface of the ice
The heat transfer from the liquid to the solid-liquid interface is particle can be determined by
controlled by convection in the present study. The local heat flux ṁh ls ⫹Q s
at an angular position ␪ can be determined by applying an energy h̄⫽ . (8)
balance to the ice surface element A S 共 T l⬁ ⫺T li 兲
Uncertainties were present in various experimental measure-
q 共 ␪ 兲 ⫽h 共 ␪ 兲共 T l⬁ ⫺T li 兲 ⫽ṁ ⬙ 共 ␪ 兲 h ls ⫹q s 共 ␪ 兲 (2) ments. The maximum random uncertainties in temperature, tim-
ing, dimension and coordinates, and flow rate of water mea-
where h( ␪ ) is the local heat transfer coefficient at the angular surements are ⫾0.5°C, ⫾0.5 s, ⫾0.2 mm, and ⫾2.0 percent, re-
position ␪. On the right side of above equation, the first term is the spectively. Based on the method of Kline and McClintock 关15兴,
latent heat absorption rate, and the second term is the heat flux the maximum uncertainties in the shape change, melting rate, heat
conducted into the ice particle to raise the internal temperature. transfer coefficient, Reynolds number, Grashof number, Nusselt
The conduction heat flux will equal zero after the internal tem- number, and Stefan number are estimated to be ⫾2.66 percent,
perature of ice particle increases to the surface temperature that ⫾4.98 percent, ⫾5.16 percent, ⫾4.66 percent, ⫾4.17 percent,
maintains at the melting temperature, 0°C. When the ice shape is ⫾7.82 percent, and ⫾4.0 percent, respectively.
spherical, the total heat transfer rate conducted into the ice is
determined by
3 Results and Discussion
Q s⫽ 冕 0
R
4 ␲ r 2 ␳ s c ps
⳵Ts
⳵t
dr. (3)
3.1 General Characteristics. The typical video pictures of
the ice sphere in the melting process are shown in Fig. 4. The
top-view images obtained from the experiment reveal that the
The transient temperature distribution T s (r,t) in the solid spheri- shape change of ice sphere in the traverse direction to the flow is
cal region has been obtained under the condition that the initial very symmetrical about the vertical center plane, as shown in
temperature of the ice sphere is uniform and the surface tempera- Fig. 4共b兲. Therefore, in the following discussion, we focus our
ture is maintained at the constant temperature 共Gebhart 关14兴兲. The attention to local melting and heat transfer from the side view,
shape of melting ice gradually becomes irregular because the local which shows significant deviation from symmetry in the vertical
melting rate is not uniform over the surface of ice 共Hao and Tao direction.
关12兴兲, so that it is impossible to obtain the analytical solution of Figure 5共a兲 shows the typical shape change of a melting ice
T s (r,t) and also very difficult and complex to obtain the numeri- sphere, viewed from the side, with time for forced convection
cal solution of T s (r,t). To find an alternative solution, we first dominated conditions (Gr0 /Re20⬍1). In Fig. 5共a兲, x stands for the
perform an analytical calculation for a spherical particle under the coordinate in the horizontal direction, and y is in the vertical di-
same temperature condition as in the present study. It is found that rection. Water flows horizontally from the right side to the left
the temperature gradient is nearly zero in the interior of ice sphere side. As will be discussed later, there exists a second upper bound-
except for a very thin layer under the surface, and the temperature ary layer separation point due to the discharge of melt near the
Fig. 4 Typical video pictures of the ice sphere in the melting
process „water flows horizontally from the right side to the left
side…: „a… Side view; „b… top view. Fig. 6 Temperature-time histories at the inlet and the ice cen-
ter: T l 0 Ä26°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0 Ä36 mm,
Gr0 ÕRe20Ä0.319. 1—inlet; 2—center of ice.
rear portion of the surface, as shown in Fig. 5共b兲 where the shapes
of the particle have been plotted in the spherical coordinate sys-
As shown in a preceding paper 共Hao and Tao 关13兴兲, the non-
tem with the center of the particle coinciding with the origin.
uniform flow field around the sphere results in different interac-
Figure 6 shows the history of temperature measured by the ther-
tion among mass, momentum, and energy between ice particle and
mocouples at the inlet of test channel and in the center of ice
flow at different local locations on the ice surface. These differ-
particle for the same case as in Fig. 5. The time variations of total
ences cause the shape of a convectively melting ice sphere to
heat transfer rate, total latent heat of fusion and heat conducted
become irregular in later stages, as shown in Fig. 5. This situation
into the inside of ice particle over the surface of ice particle are
may occur in many practical processes. The model prediction and
shown in Fig. 7.
empirical correlations from the existing studies in open literature
Upon immersion into flowing water with a temperature higher
do not include this type of melting process. To characterize the
than the phase change temperature 共0°C兲, the ice sphere begins to
shape change during the melting process, we show in Fig. 8 the
melt, and its surface temperature remains at the phase change
equivalent spherical diameter based on the same volume and the
temperature. When the initial temperature of ice is lower than
equivalent spherical diameter based on the same surface area. The
0°C, heat is transferred as a result of the temperature difference
difference of these two diameters increases as ice shapes becomes
between ice and water, causing the ice on the surface to melt and
more irregular. We define the shape factor based on the two diam-
the sphere’s internal temperature to increase, as shown in Fig. 6.
eters as follows:
The heat conducted into the ice decreases with the increase in the
internal temperature, as shown in Fig. 7. When the internal tem- d 3V
perature of the ice sphere rises to 0°C, the heat from water to ice ␾⫽ (9)
d 3s
equals the latent heat for ice melting, and the ice temperature
remains constant at 0°C. Finally, the thermocouple probe placed This dimensionless parameter varies between 0 and 1. The
in the center of the ice sphere is exposed to flowing water as the shape factor equals 1 when a particle is spherical. The shape fac-
ice reduces to a critical size. The temperature measured by the tor will be less than 1 when a particle becomes non-spherical,
probe then quickly rises as shown by the final segment of curve 2 which means that an irregularly shaped particle has a larger sur-
in Fig. 6. When the remaining ice breaks away from the plastic face area than a sphere with the same volume. The variation of
straw and drifts downstream, the test ends. shape factor in the melting process is also shown in Fig. 8. It
Fig. 5 Shape change of ice particle in the melting process: T l 0 Ä26°C, v l ⴥ

Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0 Ä36 mm, Gr0 ÕRe20Ä0.319. 1—0 s; 2—20 s; 3—35
s; 4—50 s; 5—65 s; 6—80 s; 7—95 s; 8—110 s; 9—125 s: „a… In Cartesian
coordinates; „b… in spherical coordinates with the center of particle coincid-
ing with the origin.
Fig. 7 Variations of total heat transfer rate, total latent heat for
fusion, and total conducted heat into the inside with time in the
melting process: T l 0 Ä26°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0
Ä36 mm, Gr0 ÕRe20Ä0.319. 1—convective heat transfer from wa-
ter to interface; 2—total latent heat for fusion; 3—heat con-
ducted into the inside of ice.
Fig. 9 Local melting rate ṁ ⬙ „ ␪ … at different time in the melting

process: T l 0 Ä26°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0 Ä36 mm,
indicates that a spherical ice particle gradually changes its shape
Gr0 ÕRe20Ä0.319.
and deviates from initial spherical shape as convective melting of
the particle continues.
3.2 Local Melting Rate and Heat Transfer Coefficient. lasts about 10 to 15 seconds within the range of the present testing
conditions. The variations in the local heat transfer coefficient
3.2.1 Low Gr 0 /Re 20 (⬍1). The local melting rate, local heat includes the effect of heat conducted into the ice when the internal
transfer coefficient, and local Nusselt number in the same case as temperature is lower than the melting temperature during the early
in Figs. 5 to 8 are shown in Figs. 9–11. The test conditions are melting period. The highest value of conducted heat rate is less
maintained at a water velocity of 0.06 m/s, a water temperature of than 20 percent of the total heat rate from the water to the ice, as
26°C, an initial ice temperature of ⫺16°C, an initial ice diameter shown in Fig. 7. For the results shown in Figs. 9 and 10, the trend
of 36 mm, and an initial Reynolds number of 1793. For these test of local heat transfer coefficient is almost the same as that of the
results, the ratio of corresponding Gr/Re2 is less than unity, indi- local melting rate. This means that the conduction effect is gener-
cating an inertia-dominant flow. In the figures, the positive value ally not significant except for a short early effective period. The
of the angular coordinate ␪ stands for the angle measured from the effect of the heat conduction, or the conduction effective time
horizontal centerline to the upper surface and the negative value period, does increase slightly with a decrease in the initial tem-
of ␪ represents the angle to the lower side. perature of ice particle.
The local melting rate and local heat transfer coefficient are The variation of the local Nusselt number with ␪ at different
distributed irregularly at first when the water flow is disturbed time is shown in Fig. 11. The local Nusselt number decreases with
after the ice particle with the stainless steel basis is suddenly
placed in the water flow. The first 5 seconds of the curve shown in
Figs. 9 and 10 represents this irregular distribution. After stable
water flow resumes, the local melting rate and local heat transfer
coefficient distribute more smoothly. The initial transient period
Fig. 8 Variation of diameters and shape factor with time in the

melting process: T l 0 Ä26°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0
Ä36 mm, Gr0 ÕRe20Ä0.319. 1—diameter based on volume of ice, Fig. 10 Local heat transfer coefficient h „ ␪ … at different time in
d V ; 2—diameter based on surface area of ice, d s ; 3—shape the melting process: T l 0 Ä26°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ16°C,
factor, ␾. d 0 Ä36 mm, Gr0 ÕRe20Ä0.319.
Fig. 11 Local Nusselt number Nu„␪… at different time in the melting process: T l 0 Ä26°C, v l ⴥ
Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0 Ä36 mm, Gr0 ÕRe20Ä0.319.
time over almost all the surface except for the regions around boundary layer beyond the first upper separation point. The devel-
␪⬇130 deg and ␪⬇⫺130 deg in the final melting period. This opment of new boundary layer causes the heat transfer to weaken,
trend is contrary to the local melting rate and local heat transfer and the local Nusselt number decreases until it finally reaches the
coefficient, since the effects of volume diameter, which decreases secondary minimum at ␪⬇130 deg, where the new boundary layer
with the melting of ice particle as shown in Fig. 8, are also in- separates 共see Fig. 5共b兲 for the second upper separation point兲. It
cluded. The Reynolds number, which also includes the effect of is worth mentioning that for a non-melting sphere, the separation
size change during the melting process, decreases from the initial point usually appears at about ␪⬇80 deg for the same Re range.
value of 1793 to the final value of 439 in this case. In Fig. 11, Gr0 /Re20⫽0.319; therefore we still see some thermal
After the initial stage, the stable boundary layer is formed over buoyancy effect on the lower surface of the melting particle as the
the most surface of ice sphere except for the region beyond the melt discharges from the surface. The downward motion of the
separation point 共Hao and Tao 关13兴兲. The characteristics of heat melt induced by thermal buoyancy prevents from the reattachment
transfer are strongly influenced by the nature of boundary layer
of boundary layer beyond the lower separation point, as compared
development on the surface. The local Nusselt number decreases
with the formation of new boundary layer beyond the first upper
with increasing 兩␪兩 from the horizontal centerline to ␪⬇110 deg
separation point 共see Fig. 5兲. Therefore, no secondary separation,
and ␪⬇⫺120 deg as a result of boundary layer development along
the ice surface 共Fig. 11兲. The local Nusselt number reaches the or the second minimum of local Nusselt number, appears on the
minimum on the upper and lower surface 共␪⬇110 deg and lower side. Before the lower separation point, the downward free
␪⬇⫺120 deg, respectively兲, due to the separation of the boundary convection also makes the boundary layer on the lower surface
layer there. Beyond the separation point the local Nusselt number thicker than that on the upper side. That can explain why the local
increases with increasing 兩␪兩 and even reaches a slightly higher Nusselt number of lower surface is smaller than that of upper
value than that over the front surface because considerable distur- surface during t⫽30⬃60 s.
bance exists near the rear surface of the ice sphere, where wake Another feature implied in Fig. 11 is that shape changes due to
eddies sweep the surface. The strong local heat transfer results in the melting further influence heat transfer by changing the flow
a relatively high local melting rate over the rear surface in the pattern around the particle. The shape of the ice sphere gradually
earlier stage, as shown in Fig. 9. Therefore, the rear shape of the changes into a scallop shape because the higher heat transfer
ice sphere gradually flattens and eventually becomes completely causes higher melting rate on the front and rear surfaces. In turn,
flat, as shown in Fig. 5. The variation of the local Nusselt number the flow pattern around the ice gradually changes, along with
with ␪ on the upper surface is characterized by two minima before changes in the curvature of the front surface and the maximum
t⫽100 s. This is because secondary separation of the boundary radius in the cross direction of water flow. The most obvious
layer happens on the upper surface 共also see Fig. 5共b兲. The most change on heat transfer is that of the local Nusselt number, which
significant feature of the boundary layer with melting that differs decreases from a maximum to a minimum on the rear surface
from that without melting is that the melt from the surface dis- 共␪⫽180 deg兲 while the rear shape of the ice changes from spheri-
charges into the boundary layer. The strong disturbance enhances cal to flat during the entire melting process. The surface contour
the heat transfer in the region just beyond the separation point, beyond the location of the maximum radius experiences a sudden
causing further melting. The higher density of the melt with re- change when the rear shape of ice becomes flat. As a result the
spect to the mainstream causes the melt on the upper surface to boundary layer cannot continue to develop. Therefore, the separa-
flow downward along the upper surface but quickly mixes back- tion point always appears at the position of the maximum radius
ward into the mainstream 共recall Gr0 /Re20⬍1兲 and forms a new in the traverse direction of water flow when the rear shape of ice
Fig. 12 Peak local Nusselt number Nupeak and the location ␪ peak during the
melting process: 1—T l 0 Ä26°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0 Ä36 mm,
Gr0 ÕRe20Ä0.319, 2—T l 0 Ä16°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ16°C, d 0 Ä36 mm,
Gr0 ÕRe20Ä0.093.
becomes a flat. The strong disturbance beyond the separation surface of ice, resulting in an increase in the local melting rate, as
point can be isolated in a small region so that a high local Nusselt shown in Fig. 13共b兲. As a result the size of the ice in the case of
number remains there and even slightly increases during the final v l⬁ ⫽0.06 m/s is smaller than in the case of v l⬁ ⫽0.04 m/s, as
period of melting. The influence of disturbance is weakening in shown in Fig. 13共a兲. Changes in the local Nusselt number are
the region near ␪⫽180 deg while the rear shape of ice becomes determined by changes in the local heat transfer coefficient and
flat, resulting in a gradual decrease in the local Nusselt number. the ice diameter, based on the equivalent sphere volume. The local
The phenomenon of secondary separation on the upper surface Nusselt number for the case of v l⬁ ⫽0.06 m/s is initially higher
共illustrated by the second upper minimum point on the curve in than in the case of v l⬁ ⫽0.04 m/s, as shown in Fig. 13共c兲. As time
Fig. 11兲 also gradually disappears. The reduced heat transfer on elapses, the local Nusselt number for the higher water velocity
the lower front surface in the earlier period results in a lower becomes smaller. This is due to the fact that the decrease in the
melting rate than that of the upper front side, and as a result the size is faster for the 0.06 m/s case, resulting in a decrease in the
curvature of the lower front surface gradually becomes slightly time-dependent Re 共for example, at t⫽200 s in Fig. 13共c兲兲.
bigger than that of upper front surface in the final period. The Effect of Flow Temperature
shape change also causes the flow over the lower front surface to Figure 14 shows another set of test results for observing the
flow faster than that over the upper front side. Therefore, the local effects of water temperature. The tests are also conducted at the
Nusselt number on the lower front surface gradually becomes water velocity of 0.06 m/s, the initial ice temperature of ⫺16°C,
higher than that on the upper front surface during the final period. and the initial ice diameter of 36 mm. Two water temperatures of
The above discussion may lead us to conclude that during con- 30°C and 16°C are selected, yielding Re0⫽1895 and 1558, respec-
vective melting the melting layer on the ice surface is being mixed tively. The results indicate that the increase in water temperature
with the boundary layer flow resulting from the main flow. The results in an increase in local melting rate and a decrease in melt-
combined convective mass transfer and heat transfer 共which in- ing time, as shown in Fig. 14共a兲 and 共b兲. The local Nusselt number
cludes both forced and free convection兲, along with the particle for high water temperature is smaller than that for lower water
shape change, complicate the distribution of local Nusselt number, temperature at the same time, especially in the later period of the
as shown in Fig. 11, and reveal multiple flow separation points. melting process, as shown in Fig. 14共c兲, because the size de-
Even before flow separation occurs, the combined mass and heat creases rapidly in high water temperature.
transfer play an important role in the local Nusselt number distri-
bution. If we define a peak Nusselt number, Nupeak, the maximum 3.2.2 High Gr 0 /Re 20(⬎6). For a melting process with a high
value of the Nusselt number over the portion of surface where the Gr0 /Re20, the shape variation of the ice sphere is significantly
boundary layer is attached, as shown in Fig. 11, and its corre- different. This is due to the strong influence of thermal buoyancy
sponding location, ␪ peak , we can plot Nupeak as a function of as visualized in Hao and Tao 关13兴. Figure 15 shows the results at
␪ peak , as shown in Fig. 12共a兲. For Gr0 /Re20⫽0.319, it can be seen a low water velocity of 0.01 m/s, water temperature of 21°C,
that the peak Nusselt number decreases with ␪ peak 共from about 15 initial ice temperature of ⫺3°C, initial ice diameter of 36 mm, and
deg to ⫺35 deg兲; i.e., shifting from the upper surface to the lower initial Reynolds number of 279. The initial Grashof number is
surface. This trend corresponds to the time history of the melting 5.43⫻105 , and the initial (Gr0 /Re20) reaches 6.98. The thermally
process as shown in Fig. 12共b兲. Initially, ␪ peak remains at about 15 induced density difference between the cold melt and mainstream
deg upper surface and gradually moves downwards. For the lower causes a strong downward flow in the region behind the ice sphere
Gr0 /Re20(⫽0.093), there exists a period when ␪ peak remains at and a downward extension of the wake. The upper separation
zero degree, similar to the forced flow around a non-melting point moves towards the rear of ice sphere. The effects of separa-
sphere. These complex features need further study and theoretical tion and disturbance caused by separation are relatively weak. The
analysis. melt from the surface moves into the boundary layer and further
Effect of Flow Velocity weakens these effects. Therefore, no significant drop in local melt-
Figure 13 shows the results at a water temperature of 16°C, an ing rate is observed at ␪⬇110 deg and ␪⬇⫺130 deg, as shown in
initial ice temperature of ⫺9°C, an initial ice diameter of 36 mm, Fig. 15共b兲; instead a relatively uniform distribution of local melt-
and a water velocity of 0.06 m/s and 0.04 m/s 共Re0⫽1558 and ing rate over the rear surface of ice sphere is found. This type of
1039, respectively兲, to demonstrate the influence of water velocity distribution of local melting rate does not cause the ice sphere
on the melting process. The variations of radius, local melting become scallop-shaped. The shape instead changes into an ellip-
rate, and local Nusselt number with ␪ is plotted in the time se- soid from the side view, in which the minor axis is in the hori-
quence of 50, 100, 150, and 200 second, respectively, to make zontal direction, and the major axis is in the vertical direction, as
more original experimental data available. The increase in water shown in Fig. 15共a兲. On the contrary, in the cases with higher
velocity enhances the heat transfer between the water and the initial Reynolds numbers the lowest local melting rate appears in
Fig. 13 Compared experimental results at different water velocities: T l 0 Ä16°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ9°C, d 0 Ä36 mm,
Gr0 ÕRe20Ä0.093. - - - - - - T l 0 Ä16°C, v l ⴥ Ä0.04 mÕs, T s 0 ÄÀ9°C, d 0 Ä36 mm, Gr0 ÕRe20Ä0.209. „a… R ; „b… ṁ ⬙ ; „c… Nu.
the regions near the separation points, and the relative higher local ent times is shown in Fig. 15共c兲. In this case, the trend in local
melting rates occurs in the rear portion. This kind of distribution heat transfer coefficient is almost the same as that of local melting
of local melting rate corresponds to the shape change into scallop. rate because the initial ice temperature is slightly lower than 0°C,
Further details in the flow characteristics for high Gr/Re2 cases and the influence of heat conducted into inside of ice particle is
can be found in Hao and Tao 关13兴. relative small. The variation of local Nusselt number with ␪ at
The variation of local heat transfer coefficient with ␪ at differ- different times is shown in Fig. 15共d兲. In general, the local Nusselt
Fig. 14 Compared experimental results at different water temperatures: T l 0 Ä30°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ9°C, d 0
Ä36 mm, Gr0 ÕRe20Ä0.440. - - - - - - T l 0 Ä16°C, v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ9°C, d 0 Ä36 mm, Gr0 ÕRe20Ä0.093. „a… R ; „b… ṁ ⬙ ; „c… Nu.
number decreases as time elapses, contrary to that of local melting 3.3 Total Melting Rate and Average Heat Transfer Coeffi-
rate and local heat transfer coefficient, which increase. This is cient. The variations of total melting rate and average heat
because the local Nusselt number is defined by the time dependent transfer coefficient with time at different water velocities are
equivalent sphere diameter that decreases at a rate much faster shown in Figs. 16 and 17. All six tests are conducted at a supply
than the increased rate of local heat transfer coefficient. water temperature of 16°C, initial ice temperature of ⫺8°C, and
Fig. 15 Experimental results at a slow water velocity: T l 0 Ä21°C, v l ⴥ
Ä0.01 mÕs, T s 0 ÄÀ3°C, d 0 Ä36 mm, Gr0 ÕRe20Ä6.98. „a… Shape change; „b… ṁ ⬙ ;
„c… h ; „d… Nu.
initial ice diameter of 36 mm. The water velocities are varied from process because the decrease in volume and surface area of ice is
0.01 to 0.06 m/s. The increase in the water velocity enhances the more pronounced than the influence of increase in water velocity.
convection heat transfer between water and ice surface resulting The trends of variation of the average heat transfer coefficient
in an increase in the total melting rate and average heat transfer with time are similar in all experimental results with different
coefficient. velocities, as shown in Fig. 17. The average heat transfer coeffi-
The volume and surface area of ice decrease over time. The cient increases with time during the early melting period because
melting rate is the highest when the melting process begins, as the size of the ice decreases as the melting continues. In the final
shown in Fig. 16. The melting rate gradually decreases over time period, the shape of the ice becomes very irregular, and the con-
as the volume and surface area of the particle decrease. The trend vective heat transfer between water and ice surface is weakened,
of the decrease in melting rate with the elapse of time is stronger especially in the rear where the shape becomes flat. This causes a
for higher velocity of water. At a high velocity the melting rate reduction in the average heat transfer coefficient.
may be lower than at lower velocity near the end of the melting As discussed in the above section, free convection caused by
Fig. 17 Variation of average heat transfer coefficient with

Fig. 16 Variation of melting rate of ice particle with time at time at different water velocities: T l 0 Ä16°C, T s 0 ÄÀ8°C, d 0
different water velocities: T l 0 Ä16°C, T s 0 ÄÀ8°C, d 0 Ä36 mm. Ä36 mm.
Fig. 20 Variation of average heat transfer coefficient with time
Fig. 18 Variation of Nusselt number as a function of „GrÕRe2…
at different water temperatures: v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ8°C, d 0
at different water velocities: T l 0 Ä16°C, T s 0 ÄÀ8°C, d 0
Ä36 mm.
Ä36 mm.
the density difference between the cold melt and the mainstream
has an important effect on the heat transfer between the water and
the ice surface and the melting process. Figure 18 shows the Nus-
selt number based on the average heat transfer coefficient and
volume diameter as a function of (Gr/Re2) at different water ve-
locities. For a given velocity, the Nu-Gr/Re2 relation actually
shows a similar dependence on the ice size for all the cases. Fig-
ures 19, 20, and 21 show the variation of melting rate and average
heat transfer coefficient with time and Nusselt number as a func-
tion of (Gr/Re2), respectively, for the melting ice particles under
different supply water temperatures. All the five tests are con-
ducted at the water velocities of 0.06 m/s, the initial ice tempera-
ture of ⫺8°C, and the initial ice diameter of 36 mm. The water
temperatures are varied at 11, 16, 21, 26, and 30°C, respectively.
All those results show a strong correlation among Nu, Gr, Re, Fig. 21 Variation of Nusselt number as a function of „GrÕRe2…
and Ste 共temperature effect兲. To develop an empirical correlation, at different water temperatures: v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ8°C, d 0
we conduct a total of thirty tests with water velocity of 0.01 to Ä36 mm.
0.10 m/s and water temperature of 4 to 30°C. Some of them were
repeated to test the repeatability of the experiments under identi-
cal conditions. The qualitative results of repeating tests agree rea-
sonably well, and no detectable differences in main quantitative
results are found. The regression analysis yields the following
empirical correlation for the average Nusselt number in a melting
process,
h̄d V
Nu⫽ ⫽0.118Re0.431Gr0.0898Pr0.748Ste⫺0.167 (10)
k l0
Fig. 19 Variation of melting rate of ice particle with time at Fig. 22 Correlation and all of results for dimensionless aver-
different water temperatures: v l ⴥ Ä0.06 mÕs, T s 0 ÄÀ8°C, d 0 age heat transfer coefficient in the present study: T l 0 Ä4
Ä36 mm. È30°C, v l ⴥ Ä0.01È0.10 mÕs, T s 0 ÄÀ23È0°C, d 0 Ä36 mm.
which is valid for This correlation considers the irregular shape of ice. A shape fac-
tor that indicates the irregularity of particle shape in melting pro-
80⭐Re⭐3200 0.0016⭐Gr/Re ⭐6.98
2
cess is defined.
7.91⭐Pr⭐12.69 0.05⭐Ste⭐0.39
Acknowledgments
0.007⭐d V ⭐0.036 m 0.65⭐ ␾ ⭐1
The supports from NASA 共Grant No. NAG3-1797兲 and NSF
Each of the properties 共␳ l0 , ␮ l0 , c pl0 , k l0 , and ␤ l0 兲 in the 共Grant No. HR67206268兲 are greatly appreciated.
dimensionless numbers is evaluated at the film temperature, T f
⫽(T l⬁ ⫹T m )/2. The R-squared value for Eq. 共10兲 is 0.9378. The
Nomenclature
comparison of the experimental data with the values calculated by
Eq. 共10兲 is shown in Fig. 22. The deviation for all test data is As ⫽ surface area of ice particle, m2
within ⫾9.4 percent. cp ⫽ specific heat, J/kg K
ds ⫽ surface diameter, m
dV ⫽ volume diameter, m
Summary and Conclusions Gr ⫽ Grashof number, g 0 ␤ l0 ␳ 2l0 d 3V (T l⬁ ⫺T m )/ ␮ 2l0
The melting and heat transfer characteristics of the convective g ⫽ gravitational acceleration, m/s2
melting of an initially spherical ice particle in horizontally flowing h ⫽ local heat transfer coefficient, W/m2 K
water have been studied experimentally to arrive at quantitative h ls ⫽ latent heat of fusion, J/kg
results. Based on the experimental results, the following conclu- k ⫽ thermal conductivity of water, W/m K
sions have been drawn: ṁ ⫽ total melting rate, kg/s
ṁ ⬙ ⫽ local melting rate, kg/m2 s
1 The trends of the local heat transfer coefficient are almost Nu ⫽ local Nusselt number, hd V /k l0
identical to those of the local melting rate on the ice surface. The Pr ⫽ Prandtl number, ␮ l0 c pl0 /k l0
rate of change in local heat transfer coefficient is slightly lower Q ⫽ total heat transfer rate, W
than that of local melting rate in the early-stage of melting pro- q ⫽ heat flux, W/m2
cess. The local heat transfer coefficient includes the effect of heat R ⫽ outside radius, m
conduction into the inside of ice. Although the conduction effect is Re ⫽ Reynolds number, ␳ l0 v l⬁ d V / ␮ l0
generally not significant except for a short early effective period, r ⫽ radius coordinate inside ice, m
the effective time period of heat conduction does increase with the Ste ⫽ Stefan number, c pl0 (T l⬁ ⫺T m )/h ls
decrease in the initial temperature of ice particle. T ⫽ temperature, °C
2 For Gr0 /Re20⬍1 and at a given time during the early melting t ⫽ time, s
period, a typical distribution of local Nusselt number over the V ⫽ volume, m3
front portion of the ice sphere exhibits a peak local Nusselt num- v ⫽ velocity, m/s
ber (Nupeak) at a location ( ␪ peak) starting from the upper surface, x, y ⫽ coordinates, m
then decreases with increasing ␪ from ␪ peak , and reaches minima
at the upper and lower surfaces at an angular location of about 110 Greek Symbols
deg to 130 deg, depending on the flow velocity, and ⫺120 deg ␤ ⫽ coefficient of thermal expansion, K⫺1
near the separation point. The location ␪ peak for Nupeak shifts from ␪ ⫽ angle from the horizontal centerline, degree
the upper surface towards the lower surface as melting proceeds. ␮ ⫽ viscosity, Pa•s
There exist two flow separation points on the upper surface and ␳ ⫽ density, kg/m3
one separation point on the lower surface of a melting particle, ␾ ⫽ shape factor, d 3V /d 3s
resulting from combined mass and heat transfer and particle shape
change. The second separation point on the upper rear surface Subscripts
corresponds to a second minimum value on the local Nusselt num- 0 ⫽ at the inlet; initial; respective value of property in
ber distribution curve. For high Gr0 /Re20 共⬎6兲 cases, thermal dimensionless number
buoyancy effects become more pronounced. The local Nusselt c ⫽ center of particle
number distribution does not show a second separation point on l ⫽ liquid, water
the upper surface of the melting particle. i ⫽ interface
3 Gr0 /Re20 also correlates to the shape variation of the ice m ⫽ phase change
sphere in the melting process. The shape over the rear ice sphere s ⫽ solid, ice
becomes flat and the overall shape of the ice changes into a scal- ⬁ ⫽ mainstream
lop when Gr0 /Re20 is small 共less than 1兲. With a high Gr0 /Re20 Other
(⬎6), the shape over the rear of ice sphere does not become flat
— ⫽ average
and the shape of ice sphere changes into an ellipse from the side
view. Further investigation is needed to define more precisely the
criteria of distinguishing these two different types of shape References
change. 关1兴 Tkachev, A. G., 1953, ‘‘Heat Exchange in Melting and Freezing of Ice,’’ in
4 The increase in water velocity results in an increase in the Problem of Heat Transfer During Change of Phase: A Collection of Articles,
AEC-tr-3405, translated from Russian, State Power Press, pp. 169–178.
local melting rate and the local heat transfer coefficient and a 关2兴 Merk, H. J., 1953, ‘‘The Influence of Melting and Anomalous Expansion on
decrease in the total melting time. The increase in water tempera- the Thermal Convection in Laminar Boundary Layers,’’ Appl. Sci. Res., 4, pp.
ture results in an increase in the local melting rate and the local 435– 452.
heat transfer, which eventually causes the local Nusselt number to 关3兴 Schenk, I., and Schenkels, F. M., 1968, ‘‘Thermal Free Convection from an Ice
Sphere in Water,’’ Appl. Sci. Res., pp. 465– 476.
decrease because of the decrease in the particle size. 关4兴 Vanier, Cr. R., and Tien, C., 1970, ‘‘Free Convection Melting of Ice Spheres,’’
5 The average heat transfer coefficient first increases with time AIChE J., 16, pp. 76 – 82.
and later rapidly decreases. The average heat transfer coefficient 关5兴 Eskandari, V., 1981, ‘‘Forced Convection Heat Transfer from Ice Spheres in
increases with the velocity and temperature of water except near Flowing Water,’’ Master’s thesis, University of Toledo, Toledo, OH.
关6兴 Eskandari, V., Jakubowski, G. S., and Keith, T. G., 1982, ‘‘Heat Transfer from
the end of melting period. Spherical Ice in Flowing Water,’’ Joint AIAA/ASME Fluids, Plasma, Thermo-
6 An empirical correlation that can be used to calculate the physics, and Heat Transfer Conference, St Louis, MO, ASME 82-HT-58, pp.
average heat transfer coefficient in a melting process is obtained. 1–5.
关7兴 Anselmo, A., Prasad, V., and Koziol, J., 1991, ‘‘Melting of a Sphere when 关11兴 McLeod, P., Riley, D. S., and Sparks, R. S. J., 1996, ‘‘Melting of a Sphere in
Dropped in a Pool of Melt with Applications to Partially-Immersed Silicon Hot Fluid,’’ J. Fluid Mech., 327, pp. 393– 409.
Pellets,’’ Heat Transfer in Metals and Containerless Processing and Manufac- 关12兴 Hao, Y. L., and Tao, Y. X., 1999, ‘‘Convective Melting of a Solid Particle
turing, ASME HTD Vol. 162, pp. 75– 82. in a Fluid,’’ Proceedings of the 3rd ASME/JSME Joint Fluids Engineering
关8兴 Anselmo, A., Prasad, V., Koziol, J., and Gupta, K. P., 1993, ‘‘Numerical and Conference, San Francisco, California, July 18 –23, 1999, FEDSM99-7091,
Experimental Study of a Solid-pellet Feed Continuous Czochralski Growth pp. 1– 6.
Process for Silicon Single Crystals,’’ J. Cryst. Growth, 131, pp. 247–264. 关13兴 Hao, Y. L., and Tao, Y. X., 2001, ‘‘Melting of Solid Sphere Under Forced and
关9兴 Mukherjee, M. K., Shih, J., and Prasad, V., 1994, ‘‘A Visualization Study of Mixed Convection: Flow Characteristics,’’ ASME J. Heat Transfer, 123, pp.
Melting of an Ice Sphere in a Pool of Water,’’ 1994 International mechanical 937–950.
Engineering Congress and Exposition, The Winter Annual Meeting, ASME 关14兴 Gebhart, B., 1993, Heat Conduction and Mass Diffusion, McGraw-Hill, Inc.,
94-WA/HT-14, Chicago, Illinois, November 6 –11, 1994. New York, Chap. 4.
关10兴 Aziz, S. A., Janna, W. S., and Jakubowski, G. S., 1995, ‘‘Forced Convection 关15兴 Kline, S. L., and McClintock, F. A., 1953, ‘‘Describing Uncertainties in
Heat Transfer From an Isothermal Melting Ice Sphere Submerged in Flowing Single-Sample Experiments,’’ Mech. Eng. 共Am. Soc. Mech. Eng.兲, 75, pp.
Water,’’ ASME HTD-Vol. 317-1, pp. 213–217. 3– 8.
Directional Spectral Emittance
of a Packed Bed: Influence
Dominique Baillis
e-mail: domino@cethil.insa-lyon.fr of the Temperature Gradient
Jean-François Sacadura in the Medium
Center de Thermique de Lyon (CETHIL),
UMR CNRS 5008, Recently a new experimental set-up for measuring the directional spectral emittance has
Institut National des Sciences Appliquées de been developed. Both sides of packed opaque spheres are simultaneously heated with
Lyon, identical power laser beams (CO2 10.6 ␮m) and the isothermal condition in the medium
20, avenue Albert Einstein 69621, is assumed. In this paper we consider the coupled conduction-radiation equations to
Villeurbanne Cedex, investigate the effect of the temperature nonuniformity on the calculated value of the
France emittance and to verify if the isothermal assumption is valid. Moreover, the sensitivity of
the emittance and the temperature profile to the values of parameters (such as the poros-
ity, the face temperatures, and the particle hemispherical reflectivity, etc.) is studied.
关DOI: 10.1115/1.1466459兴
Keywords: Heat Transfer, High Temperature, Modeling, Packed Beds, Participating

Media, Porous Media, Properties, Radiation
Introduction measuring the directional spectral emittance. Both sides of the

sample are simultaneously heated with identical power laser
Radiative heat transfer through semi-transparent media 共absorb-
beams 共CO2 10.6 ␮m兲 and the isothermal condition in the medium
ing, emitting and scattering media兲 is governed by the Radiative is assumed. Directional spectral emittances have been measured
Transfer Equation 共RTE兲. The RTE is stated in terms of radiation for packed beds of oxidized bronze spherical particles. Experi-
intensity, a scalar that varies with spatial location, direction, and mental results have been compared with emittances calculated by
wavelength, and is coupled through temperature to the energy solving the radiative transfer equation using the discrete ordinates
conservation equation. The RTE is an integro-differential equation method. Radiative properties are predicted using the geometric
with relatively few analytical solutions. A wide variety of approxi- optics theory combined with a correlated scaling factor for large
mate and computational techniques have therefore been developed size parameter proposed by Singh and Kaviany 关5兴.
to solve the RTE. Moreover, experiments to measure intensity In this paper, the coupled conduction-radiation equations are
have been developed. An extensive review of the radiative transfer considered in order to investigate the effect of the temperature
in dispersed media dealing with theoretical prediction and experi- non-uniformity on the calculated value of the emittance, and to
mental studies was carried out by Viskanta and Mengüc 关1兴 and by verify if the isothermal assumption is valid. Moreover, results ob-
Baillis and Sacadura 关2兴. Radiative heat transfer prediction de- tained from uncorrelated and correlated radiative properties are
pends upon accurate material property information in addition to compared. An analysis of sensitivity of parameters such as poros-
accurate RTE modelling techniques, and the properties themselve ity, particle diameter, and surface temperature on the temperature
continue to be an important source of uncertainty. To this end, profile and on emittance results is also studied here.
radiative heat transfer through relatively simple media such as Firstly, the two samples of packed bed of oxidized bronze
packed bed spheres has received much attention. A general review spherical particles studied are described. Then emittance measure-
of radiation heat transfer in packed beds was given by Kaviany ments are briefly recalled. Then the model of coupled conduction-
关3兴. radiation is described. Finally, results obtained from two samples
Kamiuto 关4兴 proposed a correlated-scattering theory to deter- are presented and discussed.
mine radiative properties of a packed bed consisting of relatively
large spheres. Singh and Kaviany 关5兴 showed that the radiative Packed Spheres Systems
properties for a bed of large opaque spheres can be obtained from The participating porous medium considered in this study is a
the independent properties by scaling the optical thickness while mono-dispersion of spherical particles of oxidized bronze. It is
leaving the albedo and the phase function unchanged. These au- assumed that the packing geometry is random. When heated, the
thors also concluded from this study that the results obtained from bronze oxidizes and a black oxide forms on the particle boundary
the correlated scattering theory of Kamiuto do not generally show surface. In order to ensure that oxide growth does not continue
good agreement with the results obtained from the Monte Carlo during measurements, the samples are heated in a furnace at 923
method. Jones et al. 关6兴 measured the spectral directional distribu- K for 72 hours. Two different samples are studied. For each me-
tion of radiation intensity at the exposed boundary of a packed dium, spheres are assumed to have a nominal uniform diameter, as
bed of opaque spheres using a direct radiometric technique. The specified by the supplier. The dimensional characteristics of the
purpose of these measurements was to provide an experimental two samples are given in Table 1.
data base of radiative intensity with which to correlate intensity
field solutions of the radiative equation in packed beds. Recently Emittance Measurement
Lopes et al. 关7兴 have developed a new experimental set up for
Emittance measurements are briefly recalled here, more details
can be found in Lopes et al. 关7兴. The experimental set up 共Fig. 1,
HEAT TRANSFER. Manuscript received by the Heat Transfer Division December 15, Ref. 关8兴兲 was developed to minimize the temperature gradient
2000; revision received October 12, 2001. Associate Editor: R. D. Skocypec. inside the medium. Indeed, both sides of the sample are simulta-
Table 1 Sample data square slab. The medium can be treated as one-dimensional. The
one-dimensional, radiation/conduction energy equation is:
⫺
⳵
⳵y 冉冊
kb
⳵T
⳵y
⫹4 ␲ ␤ ␭ 冕 ⬁
␭⫽0
共 1⫺ ␻ ␭ 兲 I ␭b d␭⫺2 ␲ ␤ ␭ 冕 ⬁
␭⫽0
共1
冕
⫺␻␭兲
1
␮ ⫽⫺1
I ␭ d ␮ d␭⫽0 (1)
The bed thermal conductivity can be obtained by using parallel-

neously heated with a high power laser 共4 kW, CO2 10.6 ␮m兲 series models from the thermal conductivity of the bed particle
which was split into two identical beams using a beam splitter 共bronze兲, the thermal conductivity of the interstitial medium 共air兲
(50 percent⫺50 percent⫾1 percent). The power of both laser being negligible:
beams is measured by a calorimeter. Energy distribution of the
laser beam is homogenized using two copper tubes with a square 1
opening and with side-faces buffed as a mirror. The sample tem- k b ⫽ 共 1⫺ ␦ 兲 k bronze (2)
perature is determined by a pyrometer. Values obtained are given 2
in Table 1. Uncertainties of temperatures have been estimated to
be less than 3.5 percent. The bronze conductivity, k bronze is given by the relation:
The directional spectral emittance can be defined as the ratio of
the spectral fluxes emitted by the sample and by the reference log k bronze⫽C 0 ⫹C 1 log T⫹C 2 共 log T 兲 2 ⫹C 3 共 log T 兲 3 (3)
blackbody. The radiometer Heimann KT4S collects the heat flux with C 0 ⫽⫺37.8999, C 1 ⫽43,3167, C 2 ⫽⫺15,8575, C 3
emitted by the sample and by a reference blackbody at the same ⫽1,9488. Equation 共1兲 is solved with the measured temperature
temperature, within the same spectral range and solid angle 共Fig. boundary conditions (T(0),T( ␶ 0 )).
1兲. The radiometer is mounted on a rotating arm allowing the For a one-dimensional semi-transparent medium with azimuthal
measurement of the radiation emitted by the sample at several symmetry the RTE is written:
angles. Different narrow band infrared filters are used to measure
spectral emitted radiation. The area of the sample surface heated ⳵ I ␭共 ␶ , ␮ 兲
by the laser beams is 12⫻12 mm. Uncertainties of emittance ␮ ⫹I ␭ 共 ␶ , ␮ 兲 ⫽S ␭ 共 ␶ , ␮ ,T 兲 (4)
⳵␶
measurements have been estimated to be less than 11 percent.
where the source term is:
冉冕冊
Theoretical Emittance Model-Combined Conduction-
␻␭ 1
Radiation Heat Transfer S ␭ 共 ␶ , ␮ ,T 兲 ⫽ 共 1⫺ ␻ ␭ 兲 I ␭b 共 T 兲 ⫹ I ␭共 ␶ , ␮ ⬘ 兲 p 共 ␮ ⬘, ␮ 兲 d ␮ ⬘
2 ⫺1
Governing Equations. Lopes et al. 关7兴 have considered the
medium as isothermal. In this paper our aim is to verify the va- On the bed’s surface, the incoming boundary condition inten-
lidity of this hypothesis. It is necessary to solve the energy con- sity is taken as the sum of the blackbody radiation at the tempera-
servation equation to determine the complete temperature field to ture of the surroundings I b␭ (T⫽293 K) and the collimated nor-
be input in the RTE. The oxidized bronze spherical packed bed mally incident radiation I 0 incoming from the power laser at the
used in the experiment is considered as a finite plane parallel wavelength 10.6 ␮m.
Fig. 1 Emittance measurement system „Lopes et al. †11‡…
␭⫽10.6␮ m
再 ␶ ⫽0→
␶ ⫽ ␶ o→
再
I ␭ 共 0,␮ 兲 ⫽I 0 ⫹I b␭ 共 T⫽293K 兲 ␮ 0 ⭐ ␮ ⭐1
I ␭ 共 0,␮ 兲 ⫽I b␭ 共 T⫽293K 兲 0⬍ ␮ ⬍ ␮ 0
再
I ␭ 共 ␶ 0 , ␮ 兲 ⫽I 0 ⫹I b␭ 共 T⫽293K 兲 ⫺1⭐ ␮ ⭐⫺ ␮ 0
I ␭ 共 ␶ 0 , ␮ 兲 ⫽I b␭ 共 T⫽293K 兲 ⫺ ␮ 0 ⬍ ␮ ⬍0
(5.a)
␭⫽10.6␮ m 再 ␶ ⫽0→I ␭ 共 0,␮ 兲 ⫽I b␭ 共 T⫽293K 兲 0⭐ ␮ ⭐1

␶ ⫽ ␶ o →I ␭ 共 ␶ 0 , ␮ 兲 ⫽I b␭ 共 T⫽293K 兲 ⫺1⭐ ␮ ⭐0
(5.b)
I 0 ⫽ P/ 共 2 ␲ 共 1⫺cos ␪ 0 兲 S 兲 (6)
with P, the power laser received by the sample P⫽625 W,

␪ 0 , the divergence half angle of the incident collimated beam
␪ 0 ⫽2.5 deg,
S, the sample surface S⫽1.44 10⫺4 m2 . Numerical calculations
have shown that blackbody radiation at the temperature of the
surroundings incoming on the sample face can be neglected.
The directional spectral emittance of the medium is defined as:
I ␭ 共 0,␮ 兲
␧ ␭ 共 0,␮ 兲 ⫽ ␮ ⬍0 (7)
I ␭b 共 T 共 0 兲兲
Solution Method. The discrete ordinates method 共DOM兲 as-
sociated with the control volume method is used to solve the
coupled radiation conduction equations 共Doermann 关9兴兲. A
quadrature over 24 directions is adopted in the DOM. The spheri-
cal space is discretized into 12 directions for the positive range of
␮ and 12 other symmetric directions for the negative ␮. This
quadrature is precise and adapted to our case. Indeed the incident
radiative flux on the sample being collimated, a first zone in which Fig. 2 Particle hemispherical reflectivity for Samples A and B
is the collimated beam is considered ( ␮ 0 ⭐ ␮ ⭐1). After this zone, „Lopes et al. †7‡…
in the interval (0⭐ ␮ ⭐ ␮ 0 ) a combination of two Gaussian
quadrature with 11 directions is used 共Nicolau 关10兴兲. A linear
scheme 共diamond兲共Doermann 关9兴兲 is employed to evaluate the lated by the Monte Carlo method and by independent theory. The
radiative intensity in the control volume. The spectral integrations scaling factor scales the extinction coefficient, leaving the phase
in Eq. 共1兲 are computed by solving Eq. 共4兲 for constant spectral function and the albedo unchanged:
band intensities between 1.3 ␮m and 16.3 ␮m. 153 bands are
considered. The general solution algorithm is iterative. An initial ␤ c ⫽ ␥␤ u (11)
temperature-linearized form is adopted. The RTE is solved itera- The values of ␥ for f v ⬍0.7 can be obtained from the following
tively using a marching algorithm starting from the boundary polynomial expression 共Singh and Kaviany 关5兴兲:
faces. The spatial domain is discretised in control volume. A vari-
able discretization of Tchebycheff with a grid concentration near ␥ ⫽1⫹1.84f v ⫺3.15f 2v ⫹7.2 f 3v (12)
boundaries is used. The dimension of the volume j is given by:
The particle spectral hemispherical reflectivity ( ␳ ␭ ) has been
⌬y j ⫽1/2兵 cos关(j⫺1)␲/ny兴⫺cos关 j␲/ny兴其 with ny the number of
determined for the two samples studied by Lopes et al. 关7兴 from
control volumes. A grid sensitivity study indicated that 90 control
an identification method 共Gauss method of linearization兲 applied
volumes are sufficient to achieve a converged solution for both
to bidirectional spectral reflectance data obtained from an experi-
samples.
mental device using a Fourier transform infrared spectrometer.
Radiative Properties. The spherical particles have a size pa- The ␳ ␭ values obtained for the two samples are shown in Fig. 2
rameter x(x⫽ ␲ d/␭) larger than 59, wavelengths being less than
15.8 ␮m and particles diameter being larger than 300 ␮m. When Results-Influence of Parameters
the particle size parameter x is much larger than unity, the geo- Different cases carried out for the two different packed beds of
metrical optics theory can be used to predict radiative properties. oxidized bronze spherical particles are studied. All cases pre-
If a packed bed consists of large diffuse spheres, then the inde- sented show numerical results, except case 1, which shows experi-
pendent uncorrelated radiative properties become 共Brewster 关11兴兲: mental results obtained by Lopes et al. 关7兴. The reference case
共case 3兲 is the solution of the coupled conduction-radiation heat
p 共 ␮ 兲 ⫽ 共 8/3␲ 兲共冑1⫺ ␮ 2 ⫺ ␮ cos⫺1 ␮ 兲 (8) transfer model described before, the data are those of Table 1, the
bed conductivity (k b ) is given by Eqs. 共2 to 3兲, radiative proper-
␻ u⫽ ␳ ␭ (9) ties are correlated 共Eqs. 8 to 12兲, and the emittance is defined by
Eq. 共7兲. For other cases when not specified, the data are the same
␤ u ⫽1.5f v /d (10) as those of reference case.
Case 1: experimental results
To take into account the dependent effects due to the overlap- Case 2: isothermal medium
ping particles for a bed of large opaque spheres, Singh and Kavi- Case 3: reference case
any 关5兴 proposed scaling factors so that the independent radiative Case 4: in the emittance calculation 共Eq. 7兲, the blackbody
properties can be scaled to give the dependent properties of the emission is I ␭b (T( ␶ 0 /2)) instead of I ␭b (T(0))
particulate media. The scaling factor ␥ proposed by Singh and Case 5: radiative coefficients are uncorrelated 共Eqs. 8, 9, 10兲
Kaviany 关5兴 is evaluated by finding the ratio of the slopes calcu- Case 6: porosity ␦ ⫽ ␦ 共Table 1兲 ⫾5 percent ␦ 共Table 1兲
Fig. 3 „a… Sample A-temperature profile for cases 3 and 5; and
„b… Sample B-temperature profile for cases 3 and 5.
Case 7: particle diameter d⫽d 共Table 1兲 ⫾5 percent d 共Table

1兲
Case 8: thickness of medium L⫽L 共Table 1兲 ⫾5 percent L
共Table 1兲
Case 9: particle hemispherical reflectivity ␳ ␭ ⫽ ␳ ␭ 共Fig. 2兲 ⫾5
percent ␳ ␭ 共Fig. 2兲
Case 10: bed thermal conductivity k b ⫽k b 共Eq. 2兲 ⫾5 percent Fig. 4 „a… Sample A-relative deviation between the emittance
k b 共Eq. 2兲 of case 3 and one of cases i Ä2,4,5 for ␭Ä3 ␮ m; and „b…
Case 11: T(0)⫽T(0) 共Table 1兲 ⫾5 percent T(0) 共Table 1兲, Sample B-relative deviation between the emittance of case 3
T( ␶ 0 )⫽T( ␶ 0 ) 共Table 1兲 ⫾5 percent T( ␶ 0 ) 共Table 1兲 and one of cases i Ä2,4,5 for ␭Ä3 ␮ m.
Case 12: incident laser power P⫽625 W⫾5 percent 625 W
Emittance obtained from different cases are compared espe-
cially for the wavelength, 3 ␮m, for which high intensity values For Sample B the relative deviation between emittance of the
are measured. case 3 and that of the case 2 is less than 7.7 percent 共Fig. 4共b兲兲; for
Sample A these relative deviations are smaller, they are less than 4
Temperature Gradient and Its Influence on Emittance Re- percent 共Fig. 4共a兲兲. As is expected, reference case 3 emittances are
sults. To study the influence of taking into account the coupled closest to experimental case 1 emittances for Sample B 共Fig.
conduction-radiation heat transfer, the reference case 3 is com- 5共b兲兲. For Sample A, the emittances of isothermal case 2 are
pared to the isothermal case 2. In Fig. 3, normalized temperatures closer to the experimental case 1 emittances 共Fig. 5共a兲兲. This fact
(T/T(0)) of case 3 are shown. The normalized temperature of can be explained by experimental uncertainties. Indeed they have
isothermal case 2 being T/T(0)⫽1, the relative deviation between been estimated to be on the order of 11 percent 共Lopes et al. 关7兴兲
the temperatures normalized of case 3 and case 2 is less than 3 and deviations between experimental case 1 emittances and those
percent for Sample B and less than 1.5 percent for Sample A obtained from case 2 or 3 remains less than this value.
共Figs. 3共a, b兲兲. The temperature gradient is more important for The agreement between theoretical and experimental emit-
Sample B than for Sample A due to the fact that Sample B is tances confirms the validity of the value of the bed conductivity
thicker than Sample A, and Sample B extinction coefficient ( ␤ obtained from the simple parallel-series model. Larger values of
⫽5968.8 m⫺1 ) is smaller than the Sample A extinction coefficient bed conductivity could also be valid, in which case the tempera-
( ␤ ⫽10765.4 m⫺1 ) 共extinction coefficients being calculated from ture gradient would decrease and emittance results would be
the radiative properties model 共Eqs. 10, 11兲兲. nearer to case 2 emittances. On the contrary for smaller values of
percent 共Fig. 4共a兲兲. As a result, the weak temperature gradient has
a significant influence on emittance calculation, especially for
Sample B.
Comparison Between Results Obtained From Correlated

and Uncorrelated Radiative Properties. When the radiative
properties are uncorrelated 共case 5兲, the extinction coefficient is
2.6 times smaller than in the case of correlated radiative properties
共due to the gamma factor, ␥ ⫽2.6兲. As a result, the temperature
inside the medium and therefore emittance are higher for case 5
than for case 3. The maximum relative deviation between the
temperature of case 5 and the isothermal case 2 (T/T(0)⫽1) is
7.5 percent for Sample B and 4.5 percent for Sample A 共Figs. 3共a,
b兲兲. It can be noted that radiative properties, when uncorrelated,
have a significant influence on the temperature gradient in the
medium.
A maximum deviation on emittance of 13 percent is observed
between cases 3 and 5 in the direction ␪ ⫽0deg for Sample B, and
8.5 percent for Sample A 共Figs. 4共a, b兲兲. Moreover, it can be
observed that emittances for case 5 are very far from the experi-
mental ones 共Fig. 5兲. Therefore, the use of the Singh and Kaviany
correlation seems to be appropriate and necessary.
Sensitivity of Emittance and Medium Temperature to Tem-

perature Boundary Conditions and Thermophysical Proper-
ties. The computed temperatures and intensities are functions of
the temperature boundary conditions and of the thermophysical
properties.
The sensitivities of temperature to each parameter are functions
of the spatial position. The relative variation of normalized tem-
perature ⌬T ⬘ /T ⬘ are calculated for parameter variation ⌬x i /x i
⫽⫾5 percent 共cases 6 . . . 12兲. A same value of parameter varia-
tion is used for each parameter in order to compare parameter
sensitivities. The value of ⫾5 percent represents the maximum
parameter uncertainty. The sensitivities of medium temperature to
hemispherical reflectivity and sample thickness are smallest, and
they are greatest for the porosity and boundaries temperature
共Figs. 6, 7兲. Sensitivities are less than 0.35 percent for Sample B
and less than 0.14 percent for Sample A. As a result, uncertainties
on data should not induce a significant variation on the tempera-
ture gradient.
The sensitivities of the emittance to each parameter are func-
tions of wavelength and direction. Figures 8, 9 show the sensitivi-
ties of emittance (⌬␧ ␭ /␧ ␭ ) as a function of the angle ␪ and for
␭⫽3 ␮ m. Except for the reflectivity sensitivity that is maximum
for the direction ␪ ⫽82deg, other parameter sensitivities are maxi-
mum for the normal direction to the sample, ␪ ⫽0deg. It can be
observed that sensitivity to sample thickness is nearly zero, this is
due to the fact that sample thickness is sufficiently large. The
sensitivities are the largest for the following parameters: porosity
共␦兲, boundary temperatures (T(0),T( ␶ 0 )) and hemispherical re-
flectivity ( ␳ ␭ ). For Sample A, the hemispherical reflectivities sen-
sitivity is larger than the porosity and the faces temperatures sen-
sitivities, while for Sample B, these sensitivities are on the same
order. This is due to the fact that the extinction coefficient ␤ is
higher for the Sample A resulting in a smaller sensitivity to the
boundary temperatures and porosity. The emittance sensitivities
Fig. 5 „a… Sample A emittance polar distribution for ␭Ä3 ␮ m; remains weak; they are less than 1.8 percent. Hence uncertainties
and „b… Sample B emittance polar distribution for ␭Ä3 ␮ m. on data do not induce significant variation on the emittance cal-
culation; emittances remain very near to those of case 3.
bed conductivity the gradient temperature would increase and the-

oretical emittance would increase resulting in theoretical emit- Conclusions
tance further from experimental ones. An experimental setup for measuring the directional spectral
To better study the influence of the temperature gradient inside emittance has been recently developed. The isothermal condition
the medium, case 4 is considered, where the blackbody intensity is in the medium has traditionally been assumed. In this paper, the
calculated with the temperature at spatial position ␶ 0 /2 instead of coupled conduction-radiation equations are considered in order to
boundary temperature. The relative deviation between the emit- investigate the effect of temperature non-uniformity on the calcu-
tance of reference case 3, and that of case 4 is 14.5 percent for lated value of the emittance and to analyze the sensitivities of
Sample B 共Fig. 4共b兲兲. For Sample A the relative deviation is 7.5 temperature gradient and emittance to parameters.
Fig. 7 Sample B-relative deviation between the normalized
temperature profiles T ⬘ of case 3 and cases i Ä6 to 12: „a…
Fig. 6 Sample A-relative deviation between the normalized ⌬ x i Õ x i Ä¿5 percent; and „b… ⌬ x i Õ x i ÄÀ5 percent.
temperature profiles T ⬘ of case 3 and cases i Ä6 to 12: „a…
⌬ x i Õ x i Ä¿5 percent; and „b… ⌬ x i Õ x i ÄÀ5 percent.
ties, the temperature gradient could be larger. Simulations are
strongly recommended to verify the validity of experimental re-
sults based on the isothermal assumption.
Comparison between emittance results obtained from correlated
The largest temperature gradient is observed for case B, it re- and uncorrelated radiative properties confirms that the use of the
mains less than 3 percent and induces a variation on emittance of Singh and Kaviany correlation is suitable and necessary. If radia-
7.7 percent in the direction near 0 deg. Weak gradient tempera- tive properties are not correlated, the deviation between theoreti-
tures are observed for samples considered in this study. The tem- cal and experimental emittance is significant.
perature gradient inside the medium is dependent on the thickness, The sensitivity of temperature gradient and emittance to varia-
on the bed conductivity and on the extinction coefficient of the tion in parameters such as porosity, particle dimension, sample
sample. For samples with other radiative and conductive proper- thickness, particle hemispherical reflectivity, bed thermal conduc-
Fig. 8 Sample A-relative deviation between the emittance of Fig. 9 Sample B-relative deviation between the emittance of
case 3 and one of cases i Ä6 to 12 for ␭Ä3 ␮ m: „a… ⌬ x i Õ x i case 3 and one of cases i Ä6 to 12 for ␭Ä3 ␮ m: „a… ⌬ x i Õ x i
Ä¿5 percent; and „b… ⌬ x i Õ x i ÄÀ5 percent. Ä¿5 percent; and „b… ⌬ x i Õ x i ÄÀ5 percent.
tivity, face temperature and laser power remain weak. Uncertain-

ties of these parameters do not induce significant uncertainties on
emittance calculation. The more sensitive parameters for emit- Greek Symbols
tance are the porosity, the surface temperatures and the hemi-
spherical reflectivity. Overall, the identification of porosity and
␤ ⫽ extinction coefficient
hemispherical reflectivity, or alternatively ␤ and ␻ ␭ , at different ␦ ⫽ porosity
temperatures and different wavelengths from emittance measure- ␪ ⫽ angle of direction of propagation as measured from
ments will be difficult due to the weak sensitivities. the normal of the sample
␧ ⫽ emittance
Nomenclature ␥ ⫽ scaling factor
␮ ⫽ cos␪
d ⫽ 2r, particle diameter ␳ ⫽ particle hemispherical reflectivity
F ⫽ quadratic difference ␶ ⫽ optical coordinate
fv ⫽ volume fraction ␶ 0 ⫽ optical thickness of the sample
I ⫽ intensity ⍀ ⫽ solid angle
I ␭b ⫽ spectral blackbody intensity ␻ ⫽ albedo
kb ⫽ bed thermal conductivity
L ⫽ thickness of a participating medium Subscripts
N ⫽ number of directions in discrete ordinates method b ⫽ blackbody
共DOM兲 c ⫽ correlated
P ⫽ laser power received by the sample e ⫽ experimental
p ⫽ phase function t ⫽ theoretical
R ⫽ reflectance u ⫽ uncorrelated
S ⫽ term source ␭ ⫽ wavelength
T ⫽ absolute temperature
T⬘ ⫽ normalized temperature (T ⬘ ⫽T/T( ␶ ⫽0)) Abbreviations
U ⫽ uncertainties DOM ⫽ Discrete Ordinates Method
x ⫽ ⫽2 ␲ r/␭, size parameter RTE ⫽ Radiative Transfer Equation
References 关7兴 Lopes, R., Moura, L. M., Baillis, D., and Sacadura, J. F., 2001, ‘‘Directional
Spectral Emittance of a Packed Bed: Correlation Between Theoretical Predic-
关1兴 Viskanta, R., and Mengüç, M. P., 1989, ‘‘Radiative Transfer in Dispersed tion and Experimental Data,’’ J. Heat Transfer, 123共2兲, pp. 205– 416.
Media,’’ ASME Appl. Mech. Rev., 42共9兲, pp. 241–259. 关8兴 Lopes, R., Moura, L. M., Delmas, A., and Sacadura, J.-F., 1998, ‘‘Directional
关2兴 Baillis, D., and Sacadura, J. F., 2000, ‘‘Thermal Radiation Properties of Dis-
Spectral Emittance of Ceramic Material: Theoretical Prediction Compared to
persed Media: Theoretical Prediction and Experimental Characterization,’’ J.
Experimental Data,’’ 7th AIAA/ASME Joint Thermophysics and Heat Transfer
Quant. Spectrosc. Radiat. Transf., 67, pp. 327–363.
Conference, Albuquerque, New Mexico.
关3兴 Kaviany, M., 1995, Principles of Heat Transfert in Porous Media, 2nd ed.,
Springer-Verlag. 关9兴 Doermann, D., 1995, ‘‘Modélisation des transferts thermiques dans des maté-
关4兴 Kamiuto, K., 1990, ‘‘Correlated Radiative Transfer in Packed-Sphere Sys- riaux semi-transparents de type mousse à pores ouverts et prédiction des pro-
tems,’’ J. Quant. Spectrosc. Radiat. Transf., 43共1兲, pp. 39– 43. priétés radiatives,’’ Ph.D. thesis, n° 95 ISAL. 0010, INSA de Lyon, France.
关5兴 Singh, B. P., and Kaviany, M., 1992, ‘‘Modeling Radiative Heat Transfer in 关10兴 Nicolau, V. P., 1994, ‘‘Identification des propriétés radiatives des matériaux
Packed Beds,’’ Int. J. Heat Mass Transf. 35共6兲, pp. 1397–1405. semi-transparents diffusants,’’ Ph.D. thesis, n° 94 ISAL. 0001, INSA de Lyon,
关6兴 Jones, P. D., McLeod, D. G., and Dorai-Raj, D. E., 1996, ‘‘Correlation of France.
Measured and Computed Radiation Intensity Exiting a Packed Bed,’’ J. Heat 关11兴 Brewster, M. Q., 1992, Thermal Radiative Transfer and Properties, John
Transfer, 118, pp. 94 –102. Wiley and Sons, New York, Inc., pp. 301–336.
R-22 and Zeotropic R-22/R-142b
Mixture Condensation in Microfin,
F. J. Smit High-Fin, and Twisted Tape Insert
Ph.D. student
Rand Afrikaans University,
Johannesburg, South Africa
Tubes
J. P. Meyer Using mixtures of the zeotropic refrigerant mixture R-22/R-142b, a series of experiments
e-mail: JPMeyer@postino.up.ac.za
was performed to determine the sectional and average heat transfer coefficients. Experi-
Mem. ASME
ments were also conducted to compare three different heat transfer enhancement methods
Professor
to that of smooth tubes. They were microfins, twisted tapes, and high fins. Measurements
University of Pretoria,
at different mass fluxes were obtained at six refrigerant mass fractions from 100 percent
Department of Mechanical and
R-22 up to a 50 percent/50 percent mixture of R-22/R-142b. All condensation measure-
Aeronautical Engineering,
ments were conducted at an isobaric inlet pressure of 2.43 MPa. This pressure corre-
Pretoria, 0002, South Africa
sponds to a saturation temperature of 60°C for R-22. The measurements were taken in
9.53 mm outer diameter smooth tubes and microfin tubes with lengths of 1603 mm. The
heat transfer coefficients were determined with the Log Mean Temperature Difference
equations. It was found that microfins were more suitable as an enhancement method than
twisted tubes or high fins. Also, that the heat transfer coefficients and pressure drops
decrease as the mass fraction of R-142b increases. 关DOI: 10.1115/1.1484394兴
Keywords: Condensation, Enhancement, Finned Surfaces, Refrigeration, Tubes
Introduction usually manufactured in industrialized countries, HCFCs are still

being used for new equipment under development such as hot-
Microfin tubes perform outstanding in enhancing heat transfer
water heat pump heaters. Especially R-22 is very popular and it
and are widely used to save energy in the air-conditioning and
seems at this stage as if it will be used up to the year 2040, if
refrigeration industries. For this reason, its heat transfer character- allowed by legislation and if the cost stays reasonable.
istics have been extensively studied over the past 20 years. De- The refrigerant R-22 in hot-water heat pumps is especially used
tailed literature reviews are presented by Bergles 关1,2兴 and Thome in countries with a mild climate in winter, that have no natural gas
关3兴. According to Thome the heat transfer enhancement of micro- and where electrical heating is usually used for the heating of
fin tubes 共relevant to condensation兲, can be explained by: water. Heating of water with heat pumps is very energy-efficient.
1. the increase in wetted surface area per unit length which Savings of approximately 67 percent can be realised, compared to
usually ranges from 1.4 to 1.9, depending on the number of heating with direct electrical resistance heaters 关4,5兴. Hot-water
fins, their height and shape, and the helix angle; heat pumps are vapor compression cycles, which use water-cooled
2. an increase in liquid phase convective heat transfer due to condensers for the heating of hot water. The most widely used
the fins; refrigerant for hot-water heat pumps is R-22 with which a maxi-
3. the increase in wetting around the circumference of the tube mum hot-water temperature of 60°C to 65°C is possible with
at low mass fluxes due to capillary forces, which maintain approximately the same condensing temperatures. This is possible
the flow in the completely wetted wall regime rather than the by making use of the refrigerant’s superheat, which is approxi-
partly dry wall regime of stratified flow; mately at a temperature of 120°C at the compressor outlet. The
4. an effect of swirl on mist flow at low qualities that increases maximum condensation temperature is limited by the maximum
condensing pressure to control the amount of wear in the
turbulence and forces entrained liquid droplets to the wall.
compressor bearings, the load on the bearings, and to keep the
In spite of the number of published studies, there is a huge lubrication oil from decomposing at higher compressor discharge
demand for further experimental research in order to test new temperatures.
microfin tubes and refrigerants. Manufacturers of air-conditioning Although hot-water temperatures of 60°C to 65°C are adequate
and refrigeration equipment that use microfin tubes in their prod- for domestic use, they are low when compared to temperatures
ucts usually offer different refrigerants depending on where the that can be produced by fossil fuel and direct electric resistance
manufacturer and customer are located. Manufacturers in Europe systems. This limits the potential applications of hot-water heat
tend to offer R-134a for air-conditioning and natural refrigerants pumps. Smit and Meyer 关6兴 as well as Johannsen 关7兴 showed
such as ammonia and propane for refrigeration and process cool- analytically that a zeotropic mixture of R-22 and R-142b
ing. Japanese manufacturers offer R-134a, R-404A, and R-407C. (CClF2 CH3 ) could be used to obtain higher temperatures. A hot-
In Thailand and China they offer R-22 (CHClF2 ) units. Manufac- water outlet temperature of 120°C is possible if only R-142b is
turers in the USA offer many refrigerants, and although many used. The disadvantage of using only R-142b is that its heating
options are available, R-22 remains extremely popular. capacity is 15% lower and its heating COP 共coefficient of perfor-
In nonindustrialized countries where air-conditioning, heat mance兲 is 7% lower when compared to R-22. Furthermore, it is
pump and refrigeration equipment is being developed, that is not flammable, but the flammability is decreased by adding R-22. A
mixture of 60% R-22 with 40% R-142b to 80% R-22 with 20%
R-142b by mass, is usually recommended. With these mixtures
HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 21, the heating capacities are about the same as when just R-22 is
2000; revision received April 5, 2002. Associate Editor: T. Y. Chu. used, but the COP is increased while hot-water temperatures of
80°C 共80%/20% R-22/R-142b兲 to 90°C 共60%/40% R-22/R-142b兲 The test condenser consisted of eight separate coaxial double
can be achieved. Mixtures of R-22 with R-142b form zeotropic tube condensers in series labeled A, B, C, up to N, where N⫽8, as
mixtures with glides of 7°C and 5°C for 60% and 80% R-22 shown in Fig. 1. The inner tube of each smooth tube test section
respectively. was a hard-drawn refrigeration copper tube with an inner diameter
Literature searches by Smit 关8兴, Kebonte 关9兴 and Bukasa 关10兴 of 8.11 mm and an outer diameter of 9.53 mm 共3/8 inch兲. The
showed that apparently no detailed literature on heat transfer co- microfin tube test sections had the same outside diameter but the
efficients for microfin tubes at the recommended mass fractions of inner diameter was 8.92 mm. It also had 60 helically ribbed mi-
R-22 with R-142b at a dew point temperature of 60°C or more has crofins at a helix angle of 18 deg with an area ratio of 1.59 共fin
been published. In addition, it seems as if no detailed data on area divided by inner diameter area without fins兲. The fins had a
condensation exist for the mixture of R-22/R-142b. Meyer et al. triangular profile with a rounded tip. The fin height was 0.203 mm
关11兴 published on average condensation coefficients at this high and the included angle of the fins was 51 deg. A profile of the
condensing temperature but in the annulus of coiled tube-in-tube microfin tube is shown schematically in Fig. 2. The twisted tape
heat exchangers. Shizuya et al. 关12兴 published on sectional heat test section consisted of a 9.53 mm outer diameter smooth tube
transfer coefficients for smooth tubes, but only for a mass fraction with a 0.3 mm thick plate helically twisted to a twist ratio of 3.
ratio of 50/50. The twist ratio is the axial pitch for a 180° turn of the tape divided
The first objective of this work was to determine the sectional by the tube diameter. The width of the twisted tape was 7.9 mm.
and average heat transfer coefficients of the zeotropic mixture The high fin 共or internally finned兲 test section was a 9.52 mm
R-22 with R-142b for microfin tubes at the following mass frac- outer diameter tube with 6 fins with a height of 0.6 mm, thickness
tions: 90%/10%, 80%/20%, 70%/30%, 60%/40% and 50%/50%. of 0.2 mm, area ratio of 1.26 and a spiral angle of 18 deg. The
The heat transfer coefficients were to be determined at an inlet cross sectional shape of the fins was rectangular. The inner tube
pressure of 2.43 MPa. At this pressure the dew point temperature diameter measured at the roots of the fins was 8.87 mm. The
for condensation will vary between 60°C 共100% R-22兲 and profile of the high fin is shown schematically in Fig. 3.
85.6°C 共50% R-22兲. During these measurements, the average Spacers were used halfway in each test section to keep the inner
pressure drops were also to be determined. The second objective
tubes from bending. The heat transfer length of each section was
was to compare three different heat transfer enhancement meth-
1603 mm and the distance between pressure drop measuring
ods, namely: microfins, twisted tapes and high fins to that of
points 1900 mm. The outer tube for all the test sections was also
smooth tubes.
a hard-drawn copper tube with an inner diameter of 17.27 mm and
an outer diameter of 19.05 mm 共3/4 inch兲. The thermal conduc-
Test Facility tivity of all tubes was 339 W/m°C. Sight-glasses were installed
A test facility was specifically constructed to measure in-tube between all the test sections to observe the refrigerant quality
condensation of pure refrigerants and refrigerant mixtures. The visually. All test sections were well insulated with 13 mm of ar-
characteristics of R-22 and the R-22/R-142b mixtures considered, moflex inside a 50 mm glass wool box to prevent heat leakage
as calculated from the REFPROP 关13兴 database, are given in Table effectively. A bypass line was connected parallel to the test con-
1. The overall test facility is shown in Fig. 1. It was a vapor denser to control the refrigerant mass flow through the test sec-
compression refrigeration and/or heat pump system. The compres- tions. A water-cooled after-condenser was used to ensure that only
sor was a hermetically sealed, reciprocating type with a nominal liquid refrigerant enters a coriolis mass flowmeter with an error of
cooling capacity of 10 kW. An oil separator was connected paral- ⫾0.1%. The sight-glasses before and after the coriolis flowmeter
lel to the compressor with a bypass line. By manually controlling were to ensure that only liquid flow through it. A filter drier fol-
the flow through the bypass line and through the oil separator the lowed it, and also a hand-controlled expansion valve for control-
oil mass fraction in the refrigerant could be controlled. Refrigerant ling the evaporating temperature. A water-heated evaporator was
liquid samples were taken downstream of the after-condenser and used as well as a suction accumulator on the low-pressure side to
analyzed for oil mass fraction according to the ANSI/ASHRAE complete the refrigerant loop.
41.4 关14兴 standard. For this study, full use was made of the oil Two main water loops were used, one flowing through the
separator and only results of oil mass fractions less than 0.1% condensing side and one flowing through the evaporating side. On
were used. the condensing side the water was kept constant at a temperature
Table 1 Thermodynamic properties at different mass fractions of the R-22ÕR142b mixture at a pressure of
2.43 MPa
Fig. 1 Schematic of test facility
of 10°C to 85°C 共depending on the experiments conducted兲 in a mass fluxes, comparisons were possible at approximately the
1000-liter insulated storage tank connected to a 15 kW chiller. The same quality values. A similar flow loop was used on the evapo-
water flow rate through the test sections could be controlled with rating side, also with an insulated 1,000 liter storage tank but
a hand-controlled valve. The flow rate of the water through the connected to a 20 kW electric resistance heater. This water was
test sections was measured with a coriolis mass flowmeter with an also kept constant at a temperature of 8°C to 30°C, depending on
error of ⫾0.2%. The control of the water inlet temperature made the experiments conducted. By increasing or decreasing the tem-
it possible to have more control over the average refrigerant perature of the water through the evaporator, the refrigerant den-
quality in a specific test section. The measuring could be done on sity at the compressor inlet and thus refrigerant mass flow could
more equally spaced quality values. For different refrigerants and also be changed. The water temperatures in both loops could be
thermostatically controlled at a constant temperature with an error
of ⫾1°C. As the storage capacities of the two tanks were rela-
Fig. 2 Profile of the microfin tube „9.53 mm outer diameter… Fig. 3 Profile of the high fin tube „9.52 mm outer diameter…
tively large, the source and sink temperatures were very stable Quality. In the first test section, Q w was equated to the re-
which helped in obtaining quick steady state conditions during frigerant enthalpy change to obtain the outlet enthalpy 共which is
experiments. also the inlet enthalpy for the next test section兲 as it was assumed
Temperatures were measured with resistance temperature de- that no heat losses occured through the insulation, so that
vices 共RTDs兲 mounted longitudinally on the outside of the tubes.
The RTDs were strapped firmly in position with tape and were Q̇ w
also insulated on the outside to prevent heat losses to the environ- h out⫽ ⫹h in (5)
ment. The RTDs were calibrated to measure temperature differ- Ṁ r
ences with errors less than ⫾0.1°C. Temperatures were measured
where h in is the refrigerant inlet enthalpy for the first or second
at the locations shown in Fig. 1. At each of these locations, three
test-section obtained from a refrigerant database 关19兴 using its
RTDs were located at the top, sides and bottom of the tube to take
inlet pressure and temperature as the inlet enthalpy was usually in
care of any circumferential temperature variation. The average
the superheat region. Usually the outlet of the second test section
temperature of the three values was used, with the side RTD
was in the two-phase region. Then the outlet enthalpy of the sec-
weighted double, as the average temperature measurement. Abso-
ond test section was used in Eq. 5 in the place of h in . The inlet
lute pressures on the high-pressure side were measured with a
and outlet enthalpies of each test section as well as the enthalpies
160-mm dial pressure gauge with a range of 0 to 2500 kPa. The
of saturated vapor and liquid were used to determine the inlet and
gauge was calibrated to an error of ⫾5 kPa. Pressure differences
outlet qualities. Hence, the sectional vapor quality is
were measured with a differential pressure transducer with a range
of 0 to 60 kPa, calibrated to an error of ⫾0.05% of full scale. For h⫺h f
small pressure drops, another differential pressure transducer, con- x⫽ (6)
nected parallel to the first, was used with a range of 0 to 10 kPa. h g ⫺h f
It was also calibrated to an error of ⫾0.05% of the reading. The average between the inlet and outlet qualities of each test
section was used to determine the average refrigerant quality of
the test section. For the mixtures, REFPROP 关19兴 was used to
Data Reduction calculate the enthalpy change of the particular mixture along the
dew point temperature curve in the above calculations.
Annulus Heat Transfer Coefficient. Before the test con-
Pressure Drop Experiments. To limit the scope of this study
denser was connected to the experimental set-up shown in Fig. 1,
only the sectional pressure drops for 100% R-22 in smooth and
the annulus heat transfer coefficient of each section was deter-
enhanced tubes were measured. Only average pressure differences
mined individually in a water-to-water configuration. The annulus
were measured for refrigerant mixtures from 90%/10% R-22/R-
heat transfer coefficient was determined by using the modified
142b to 50%/50% R-22/R-142b.
Wilson plot technique 关15兴. For each test section at least 20 data
points were used, all with an energy balance of less than ⫾2% Experimental Uncertainty. A propagation of error analysis
共the average heat transfer between the inside heat transfer and 关16兴 was performed to determine the uncertainty in the measured
outside heat transfer was used as the reference兲. Before the test sectional heat transfer coefficients. Using this method, the uncer-
condenser was connected into the experimental set-up it was dried tainty for the sectional heat transfer coefficients was found to
with high flows of nitrogen and evacuated several times to a pres- range from a low of ⫾4% at the highest heat flux to a high of
sure of less than 6 Pa. Once connected and charged with R-22, it ⫾20% at the lowest heat flux. Uncertainties in the average heat
was operated for a period of 14 days during which time the re- transfer coefficients ranged from ⫾7% to ⫾13% over the mass
frigerant charge and filter drier were changed twice. This was to flux range reported. To ensure low experimental uncertainties the
eliminate possible water moisture in the inner tube from the Wil- mass flow rates of the water in the annulus were adjusted to en-
son plot experiments before the experiments in this study were sure that the relative values of the heat transfer coefficients deter-
conducted. mined from the Wilson plot method do not influence the heat
transfer coefficients in the inner tube negatively.
Heat Transfer Coefficient. Assuming no fouling, the ‘‘sec-
tional’’ heat transfer coefficient 共which is the regionally average
for each section兲 was obtained from:
1 Experimental Results
冉冊
h i⫽ (1)
1 A o ln共 R 0 /R i 兲 1 A i Verification of the Experimental Procedure and R-22 Re-
⫺ ⫺ sults. The heat transfer coefficients and pressure drops were
Uo 2 ␲ kL ho Ao
verified to be accurate by first taking measurements with R-22 in
The sectional heat transfer coefficients were integrated along the both smooth and microfin tubes and comparing the results with
total length of the condenser using the trapezoid rule to obtain the literature, as shown in Figs. 4 –7. The measurements were taken at
integrated average heat transfer coefficient. The overall heat trans- saturation temperatures of 40°C or 50°C, whichever was the high-
fer coefficient was calculated from the sensible heat gain of the est from measurements published in literature and which were
water and the logarithmic mean temperature difference as follows thoroughly tested against the results of other studies. In Fig. 4 the
measured sectional heat transfer coefficients were compared to
Q̇⫽U o A o ⌬T LMTD (2) those of Eckels and Tesene 关17兴 at a mass flux of 400 kg/m2s and
where a saturation temperature of 50°C. The measurements of Eckels
and Tesene were also for a smooth tube with an inner diameter of
Q̇⫽Ṁ w c pw 共 T out⫺T in兲 (3) 8.11 mm and the same microfin size and configuration were used
as in this study. For smooth tubes, the sectional heat transfer co-
The logarithmic temperature difference was calculated by
efficients of the present study underpredicted the sectional heat
共 T r,in⫺T w,out兲 ⫺ 共 T r,out⫺T w,in兲 transfer coefficients by approximately 9% on average, with a
⌬T LMTD⫽ (4) maximum error of approximately 36% at a quality of approxi-
ln关共 T r,in⫺T w,out兲 / 共 T r,out⫺T win兲兴
mately 83%, when compared to those of Eckels and Tesene. For
Refrigerant temperatures in Eq. 4 were from actual measurements microfin tubes, the comparison was better than for smooth tubes.
and the first superheated section was ignored as well as consecu- The sectional heat transfer coefficients of the present study also
tive sections if the refrigerant entered as superheated. underpredicted the sectional heat transfer coefficients by approxi-
Fig. 4 Comparison between literature and measured heat
transfer coefficients of R-22 condensing in smooth and micro- Fig. 6 Comparison between literature and measured pressure
fin tubes at a mass flux of 400 kgÕm2.s and a saturation tem- gradients of R-22 condensing in smooth and microfin tubes at
perature of 50°C a mass flux of 250 kgÕm2.s and a saturation temperature of 50°C
overpredicted by approximately 2% on average and the maximum

mately 4% on average, with a maximum error of approximately error is 3% at a mass flux of approximately 100 kg/m2.s
18% at a quality of 83%, when compared to those of Eckels and The pressure drop per unit length at a saturation temperature of
Tesene. 50°C for smooth and microfin tubes is shown in Fig. 6. The pres-
Figure 5 shows the average heat transfer coefficient for R-22 at sure drop per unit length is calculated by dividing the total pres-
a saturation temperature of 40°C. Results were obtained over a sure drop by the length between pressure taps. The active heat
mass flux range of 89 kg/m2.s to 645 kg/m2.s for smooth tubes and transfer length in each of the test sections is 1,604 mm, while
from 100 kg/m2.s to 650 kg/m2.s for microfin tubes. The inlet pressure tap to pressure tap length is 1,900 mm. The pressure
qualities were 85% to 90% and the average exiting quality was gradients were measured between an inlet quality of approxi-
10% 共also, for most of the other experiments in the following mately 75% and an outlet quality of approximately 30%. All the
section when heat transfer enhancement and refrigerant mass frac- results of the present study are within 10% of the measurements
tions were considered兲. Also shown on the plot are results ob- of Eckels and Tesene 关18兴.
tained by Eckels and Tesene 关17兴, with similar test tubes, also at a The pressure drops during condensation at a saturation tempera-
saturation temperature of 40°C. For smooth tubes, the present ture of 40°C for the smooth and microfin tubes compared to mea-
study overpredicted the average heat transfer coefficients by ap- surements of Eckels and Tesene 关18兴 are given in Fig. 7. The
proximately 6% on average, with a maximum error of 18% at a pressure drops were measured with an approximate inlet quality
mass flux of 89 kg/m2.s. For microfin tubes, the agreement is of 90% and an outlet quality of 10% 共also, for most of the other
better than for smooth tubes. The heat transfer coefficients are experiments in the following section when heat transfer enhance-
Fig. 5 Comparison between literature and measured average Fig. 7 Comparison between literature and measured pressure
heat transfer coefficients of R-22 as function of mass flux in drops of R-22 as function of mass flux in smooth and microfin
smooth and microfin tubes at a saturation temperature of 40°C tubes at a saturation temperature of 40°C
ment and refrigerant mass fractions are considered兲 over a mass ever, at high mass fluxes of 500 kg/m2.s and higher, the difference
flux range of 88 kg/m2.s to 654 kg/m2.s. For both tubes, it was in heat transfer coefficients of high fins and microfins is not sig-
estimated that the results of the present study are all within 10% nificant.
of the measurements of Eckels and Tesene. The pressure drops for the different types of enhancement
methods are shown in Fig. 9. On average the pressure drops of
Enhancement Methods. Measurements were conducted for a microfin tubes are 38% higher than those of smooth tubes. High
wide range of mass fractions, ranging from 100% R-22 to a 50%/ fin tubes increase the pressure drop by 81% in comparison to
50% mixture 共on a mass basis兲 of R-22 with R142b. The results smooth tubes and twisted tapes increase the pressure drop by
shown in Fig. 8 and 9 are for a mass fraction of 70%/30% as it 148%. The microfin tube is therefore the best performer of the
was very representative of all the mass fractions considered. Mea- enhancement methods considered, if taking into consideration its
surements were conducted between mass fluxes of 135 kg/m2.s huge increase in heat transfer coefficient and moderate increase in
to 774 kg/m2.s. An inlet pressure of 2.43 MPa was used for all pressure drop. This result was expected and corresponds with lots
experiments. At this pressure the dew point temperature was of work on the advantages of microfin tubes as a heat transfer
76.13°C and the bubble point temperature was 70.53°C. enhancement method as is also shown in a recent review article by
From Fig. 8 follows that the average heat transfer coefficient Liebenberg et al. 关19兴. Therefore, only the results of the microfin
increases with the use of twisted tapes, high fins and microfins tubes are given and discussed further on in this paper.
compared to the heat transfer coefficient of a smooth tube. The
heat transfer coefficients are increased by approximately 46%, Mass Fraction of R-22ÕR-142b. In Figs. 10–12 the sectional
87% and 113% on average compared to those of smooth tubes, heat transfer coefficients are given for smooth and microfin tubes.
using respectively twisted tapes, high fins and microfins. How- The coefficients are for different refrigerant mass fractions at mass
fluxes of 100 kg/m2.s 共Fig. 10兲, 300 kg/m2.s 共Fig. 11兲, and 600
Fig. 8 Average heat transfer coefficients for R-22ÕR-142b

„mass fraction of 70 percentÕ30 percent… during condensation
at a dew point temperature of 76.13°C for different heat transfer Fig. 10 Sectional heat transfer coefficients in smooth and mi-
enhancement methods and a smooth tube crofin tubes at different mass fractions of R-22ÕR-142b at a
mass flux of 100 kgÕm2.s
Fig. 9 Average pressure drop for R-22ÕR-142b „mass fraction

of 70 percentÕ30 percent… during condensation at a dew point Fig. 11 Sectional heat transfer coefficients in smooth and mi-
temperature of 76.13°C for different heat transfer enhancement crofin tubes at different mass fractions of R-22ÕR-142b at a
methods and a smooth tube mass flux of 300 kgÕm2.s
Fig. 12 Sectional heat transfer coefficients in smooth and mi- Fig. 14 Heat transfer enhancement factor at a mass flux of 300
crofin tubes at different mass fractions of R-22ÕR-142b at a kgÕm2.s for different refrigerant mass fractions
mass flux of 600 kgÕm2.s
where h m is the corrected heat transfer coefficient for a zeotropic

kg/m2.s 共Fig. 12兲. Only the results for mass fractions of 90%/10%, mixture, referred to the temperature difference in saturation which
70%/30%, and 50%/50% of R-22/R-142b are given. Measure- varies between the dew point and bubble point temperatures. h is
ments were also taken for mass fractions of 100%, 80%/20% and the heat transfer coefficient of a pure fluid and h g is the heat
60%/40% but are not given as the results in Figs. 10–12 are transfer coefficient of the vapor phase flowing alone in an inter-
sufficient for discussion purposes. All the measurements were nally enhanced tube calculated by a correlation given by Ravigu-
conducted at an inlet pressure of 2.43 MPa. Therefore the dew rurajan and Bergles 关23兴. The specific heat of the gas phase is c pg
point temperatures were 65.72°C 共90%/10%兲, 76.13°C 共70%/ and the temperature glide is ⌬T, while the enthalpy change during
30%兲 and 85.64°C 共50%/50%兲. The corresponding temperature isobaric condensation of the mixture is h f g . The term on the right-
glides were 2.49°C, 5.60°C, and 6.54°C. hand side in the square brackets of Eq. 7, represents mass transfer
All the results show that the sectional heat transfer coefficients thermal resistance. This term increases by 188% as the mass frac-
decrease as the mass fraction of R-142b increases. For smooth tion of the R-22/R-142b mixture decreases from 90% to 50%, and
tubes the heat transfer coefficient decreases by approximately is therefore the reason why the heat transfer coefficients decreased
30% on average as the R-22 mass fraction decreases from 90% to as the mass fraction of R-142b was increased.
50%. For microfin tubes the decrease is very similar at 27%. Ac- It can also be deduced from the results in Figs. 10–12 that the
cording to Cavallini et al. 关20兴 this decrease is due to an increase sectional heat transfer coefficients increase as the mass flux in-
in mass transfer thermal resistance. This resistance can be ac- creases. Also, that the heat transfer coefficients of the microfin
counted for by the corrections of Silver 关21兴, and Bell and Ghaly tubes are higher than those of smooth tubes at the same mass
关22兴, which can be approximated to fluxes as can be expected 共and as also determined in the previous
h m ⬇ 关 1/h⫹xc pg 共 ⌬T/h f g /h g 兲兴 ⫺1 (7) section兲.
Fig. 13 Heat transfer enhancement factor at a mass flux of 100 Fig. 15 Heat transfer enhancement factor at a mass flux of 600
kgÕm2.s for different refrigerant mass fractions kgÕm2.s for different refrigerant mass fractions
Fig. 16 Average heat transfer coefficients in the microfin tube Fig. 18 Average heat transfer enhancement factor as function
as function of mass flux for different refrigerant mass fractions of mass flux for different refrigerant mass fractions
By fitting polynomials in the data in Figs. 10–12 and dividing condensing liquid at the tube surface is negligible in comparison
the heat transfer coefficients of the microfin tubes by the heat to the inertia forces. The advantage of microfins therefore de-
transfer coefficients of the smooth tubes, the heat transfer en- creases as the mass flux increases.
hancement factors were obtained at mass fluxes of 100 kg/m2.s In Fig. 16 the average heat transfer coefficients in the microfin
共Fig. 13兲, 300 kg/m2.s 共Fig. 14兲 and 600 kg/m2.s 共Fig. 15兲. The tube are given as function of mass flux for different refrigerant
results are shown in Figs. 13–15 for three different mass fractions mass fractions. The heat transfer coefficients increase approxi-
of R-22/R-142b. mately linearly with mass flux. In general, the heat transfer co-
At low mass fluxes of approximately 100 kg/m2 共Fig. 13兲 all the efficient of R-22 is the highest and decreases as the mass fraction
heat transfer enhancement factors vary between 1.8 and 3.6. At a of R-142b is increased. At mass fluxes of approximately 100
mass flux of 300 kg/m2.s 共Fig. 14兲 the heat transfer enhancement kg/m2.s the heat transfer coefficient of the 50%/50% mixture of
factors vary between 1.7 and 2.9, and at a mass flux of 600 R-22/R-142b is approximately 20% lower than that of 100%
kg/m2.s 共Fig. 15兲, they vary between 1.5 and 1.6. The highest heat R-22. At higher mass fluxes of 600 kg/m2.s the differences de-
transfer enhancement factors therefore occur at lower mass fluxes. crease to only 5%.
The reason being that the microfin tube ensures that the flow The influence of refrigerant mass fraction on pressure drop for
regime stays annular for a longer period before it changes to a the microfin tube is much more significant than on the average
stratified/wavy flow regime. The microfins therefore prevent the heat transfer coefficient, as can be deduced from Fig. 17. The
liquid to accumulate at the bottom of the tube due to gravity for a pressure drop increases almost quadratic with mass flux. This ob-
longer period. As the mass flux increases, the flow regime will servation corresponds with that of Shao and Granryd 关24兴 which
stay annular for even a longer period as can be deduced from Figs. suggests ⌬p⫽constant.G 2 . The pressure drop for R-22 is the
14 and 15. The reason is that the influence of gravity forces on the highest and decreases significantly as the mass fraction of R-142b
Fig. 17 Pressure drop in the microfin tube as function of mass Fig. 19 Pressure drop enhancement factor as function of
flux for different refrigerant mass fractions mass flux for different refrigerant mass fractions
is increased. On average, the pressure drop decreases with ap- Acknowledgments
proximately 50% if the mass fraction of R-142b is increased
Support for this project was provided by the Rand Afrikaans
to 50%.
University, NRF and Eskom.
The heat transfer enhancement factor in microfin tubes com-
pared to that in smooth tubes is considered in Fig. 18 as function
of mass flux. The maximum heat transfer enhancement occurs at Nomenclature
low mass fluxes of 100 kg/m2.s to 300 kg/m2.s. For mass fluxes of
U ⫽ overall heat transfer coefficient
80 kg/m2.s to 300 kg/m2.s the heat transfer enhancement factors
vary between approximately 2 and 2.5. At higher mass fluxes of Subscripts
more than 300 kg/m2.s they decrease to approximately 1.5. In c ⫽ critical
general, the heat transfer enhancement factor increases as the exp ⫽ experimental
mass fraction of R-22 is decreased. According to Shao and g ⫽ gas
Granryd 关24兴, the non-ideal properties of mixtures are mainly re- i ⫽ inner or in-tube
sponsible for heat transfer degradation which includes the diffu- in ⫽ inlet of test section
sion resistance due to concentration difference, the local non- LMTD ⫽ logarithmic mean temperature difference
equilibrium state between the phases, enhanced also by the slip in m ⫽ corrected
the liquid-vapor interface, etc. Again, this effect is more dominant o ⫽ outer or annulus side
at lower mass fluxes than at higher mass fluxes. out ⫽ outlet of test section
The pressure drop enhancement factor 共Fig. 19兲 increases with r ⫽ refrigerant
mass fluxes from approximately 1.1 at 100 kg/m2.s to 1.27 at 600 w ⫽ water
kg/m2.s. This increase is small in comparison to the large increase
in mass flux. It is also shown that the pressure drop enhancement
factor increases as the mass fraction of R-142b increases. References
关1兴 Bergles, A. E., 1999, ‘‘Enhanced Heat Transfer: Endless Frontier, or Mature
and Routine?,’’ Enhanced Heat Transfer, 6, pp. 79–99.
关2兴 Bergles, A. E., Jensen, M. K., and Shome, B., 1996, ‘‘The Literature on En-
hancement of Convective Heat and Mass Transfer,’’ Enhanced Heat Transfer,
4, pp. 1– 6.
关3兴 Thome, J. R., 1996, ‘‘Boiling of New Refrigerants: A State-of-the-Art Re-
Conclusions view,’’ Int. J. Refrig., 19共7兲, pp. 435– 457.
The performance of different refrigerant mixtures, varying in 关4兴 Meyer, J. P., and Greyvenstein, G. P., 1991, ‘‘Hot Water for Homes in South
Africa,’’ Energy—The International Journal, 16共7兲, pp. 1039–1044.
mass fraction from 100% R-22 with increases in steps of 10% of 关5兴 Meyer, J. P., and Greyvenstein, G. P., 1992, ‘‘Hot Water for Large Residen-
R-142b, up to a 50%/50% mixture of R-22/R-142b was deter- tial Units, Hospitals and Laundries with Heat Pumps in South Africa: A Tech-
mined for 9.53 mm microfin tubes. Average and sectional heat noeconomic Analysis,’’ Energy, Conversion, and Management, 33, pp. 135–
transfer coefficients were measured over a range of mass fluxes, 143.
关6兴 Smit, F. J., and Meyer, J. P., 1998, ‘‘Investigation of the Potential Effect of
qualities and dew point temperatures from 60°C to 85.64°C. For Zeotropic Refrigerant Mixture on Performance of a Hot-Water Heat Pump,’’
all these tests the inlet pressure was kept constant at 2.43 MPa. ASHRAE Trans., 104, 共Part 1A兲, pp. 387–394.
Measurements were also taken on three different heat transfer 关7兴 Johannsen, A. F. B., 1992, ‘‘Potential of Non-Azeotropic Refrigerant Mixtures
enhancement methods, namely microfins, twisted tubes, and high for Water-Heating Heat Pumps in South Africa,’’ Department of Mineral and
Energy Affairs, Report no. ED 8807, Pretoria, South Africa.
fins. The heat transfer enhancements were compared to those of a 关8兴 Smit, F. J., 1996, ‘‘The Influence of a Non-Azeotropic Refrigerant Mixture on
smooth tube. the Performance of a Hot-Water Heat Pump,’’ M. Eng. dissertation, Rand
When heat transfer enhancement methods were investigated, it Afrikaans University, Johannesburg, South Africa.
关9兴 Kebonte, S. A., 1999, ‘‘Condensation Heat Transfer and Pressure Drop Coef-
was found that microfin tubes are more suitable for heat transfer ficients of R22/R142b in a Water Cooled Helically Coiled Tube-in-Tube Heat
enhancement of R-22/R-142b mixtures, than twisted tapes or high Exchanger,’’ M. Eng. dissertation, Rand Afrikaans University, Johannesburg,
fins. The heat transfer coefficients were increased by approxi- South Africa.
mately 46%, 87%, and 113% on average compared to those of 关10兴 Bukasa, J. M., 1999, ‘‘Average Boiling Heat Transfer and Pressure Drop Co-
efficients of R22/R142b in a Helically Coiled Water Heated Tube-in-Tube Heat
smooth tubes, using respectively twisted tapes, high fins and mi- Exchanger,’’ M.Eng. dissertation, Rand Afrikaans University, Johannesburg,
crofins. However, at high mass fluxes of 500 kg/m2.s and higher, South Africa.
the difference in heat transfer coefficients of high fins and micro- 关11兴 Meyer, J. P., Bukasa, J. M., and Kebonte, S. A., 2000, ‘‘Average Boiling and
fins is not significant. Condensation Heat Transfer Coefficients of the Zeotropic Refrigerant Mixture
R22/R142b in a Coaxial Tube-in-Tube Heat Exchanger,’’ ASME J. Heat Trans-
Comparing pressure drops showed that on average the pressure fer, 122, pp. 186 –188.
drops of microfin tubes were 38% higher than those of smooth 关12兴 Shizuyo, M., Itoh, M., and Hijrkata, K., 1995, ‘‘Condensation of Nonazeotro-
tubes. High fin tubes increased the pressure drop by 81% in com- pic Binary Refrigerant Mixtures Including R-22 as a More Volatile Component
parison to smooth tubes and twisted tapes that increased the pres- Inside a Horizontal Tube,’’ ASME J. Heat Transfer, 117, pp. 538 –543.
关13兴 NIST, 1998, ‘‘NIST Thermodynamic and Transport Properties of Refrigerants
sure drop by 148%. The microfin tube was therefore the best and Refrigerant Mixtures Database,’’ 共REFPROP Ver. 6.01兲, National Institute
performer of the enhancement methods considered, if taking into of Standards and Technology, Gaithersburg, MD.
consideration its huge increase in heat transfer coefficient and 关14兴 ANSI/ASHRAE 1996, ASHRAE STANDARD 41.4, ‘‘Standard Method
moderate increase in pressure drop. for Measurement of Proportion of Lubricant in Liquid Refrigerant,’’ American
Society of Heating, Refrigerating and Air-Conditioning Engineers, Atlanta,
From the measurements on microfin tubes at different mass GA.
fractions of R-22/R-142b, it was found that: 关15兴 Briggs, D., and Young, E., 1969, ‘‘Modified Wilson Plot Technique for Ob-
taining Heat Transfer Correlations for Shell and Tube Heat Exchangers,’’
1. the sectional heat transfer coefficients decrease as the mass Chem. Eng. Prog., Symp. Ser., 65共2兲, pp. 35– 45.
fractions of R-142 are increased; 关16兴 Kline, S., and McClintock, F., 1953, ‘‘Describing Uncertainties in Single-
2. the highest heat transfer enhancements of microfin tubes Sample Experiments,’’ Mech. Eng. 共Am. Soc. Mech. Eng.兲, 75, pp. 3– 8.
关17兴 Eckels, S. J., and Tesene, B. A., 1999, ‘‘A Comparison of R-22, R-134a,
compared to smooth tubes are at low mass fluxes and the R-410a, and R-407c Condensation Performance in Smooth and Enhanced
benefit of microfins decreases as the mass flux increases; Tubes: Part I, Heat Transfer,’’ ASHRAE Trans., 105, Pt. 2
3. the average heat transfer coefficients decrease as the mass 关18兴 Eckels, S. J., and Tesene, B. A., 1999, ‘‘A Comparison of R-22, R-134a,
fraction of R-142b is increased; R-410a, and R-407c Condensation Performance in Smooth and Enhanced
Tubes: Part II, Heat Transfer,’’ ASHRAE Trans., 105, Pt. 2.
4. the average pressure drop is strongly influenced by the re- 关19兴 Liebenberg, L, Bergles, A. E., and Meyer, J. P., 2000, ‘‘A Review of Refrig-
frigerant mass fraction and that the pressure drop decreases erant Condensation in Horizontal Micro-fin Tubes,’’ Proceedings of the ASME
as the mass fraction of R-142b is increased. Advanced Energy Systems Division, International Mechanical Engineering
Congress and Exhibition, November 5–10, Orlando, Florida, AES-Vol. 40, pp. Method for Multicomponent/Partial Condenser,’’ AIChE Symp. Ser., 69, pp.
155–168. 72–79.
关20兴 Cavallini, A., Del Col, D., Doretti, L., Longo, G. A., and Rosetto, L., 2000, 关23兴 Ravigururajan, T. S., and Bergles, A. E., 1985, ‘‘General Correlations for Pres-
‘‘Heat Transfer and Pressure Drop During Condensation of Refrigerants Inside sure Drop and Heat Transfer for Single Phase Turbulent Flow in Internally
Horizontal Enhanced Tubes,’’ Int. J. Refrig., 23共4兲, pp. 4 –25. Ribbed Tubes,’’ ASME HTD, 52, pp. 9–20.
关21兴 Silver, L., 1947, ‘‘Gas Cooling with Aqueous Condensation,’’ Trans. Inst. 关24兴 Shao, D. W., and Granryd, E., 1998, ‘‘Experimental and Theoretical Study on
Chem. Eng., 25, pp. 30– 42. Flow Condensation With Non-Azeotropic Refrigerant Mixtures of R32/
关22兴 Bell, K. J., and Ghaly, M. A., 1973, ‘‘An Approximate Generalized Design R134a,’’ Int. J. Refrig., 12共3兲, pp. 230–246.
Effectiveness Correlations for
Heat and Moisture Transfer
Processes in an Enthalpy
L. Z. Zhang
Exchanger With Membrane Cores
e-mail: BeLzhang@polyu.edu.hk
The performance correlations for the effectiveness of heat and moisture transfer processes
J. L. Niu in an enthalpy exchanger with membrane cores are presented. The physical phenomena
relevant to the heat and moisture transfer in these devices have been used to develop
Department of Building Services Engineering,
a novel set of correlations based on the relevant dimensionless parameters. The total
The Hong Kong Polytechnic University,
enthalpy effectiveness can be calculated from sensible effectiveness, latent effectiveness,
Kowloon, Hong Kong
and the ratio of latent to sensible energy differences across the unit. Studies show that
the sensible effectiveness is a function of NTU, the number of transfer units for heat; while
the latent effectiveness is a function of NTUL, the number of transfer units for moisture.
The relations between NTUL and NTU are derived and studied with the proper sepa-
ration of moisture resistance for membranes. This newly developed dimensionless param-
eter, NTUL, is to summarize the sorption characteristics of membrane material,
the exchanger configurations, as well as the operating conditions. A number of experi-
mental results on an enthalpy exchanger with novel hydrophilic membrane cores has been
used to valid these correlations. 关DOI: 10.1115/1.1469524兴
Keywords: Dehumidification, Heat Transfer, Moisture, Membrane, Exchanger
1 Introduction conductivity of desiccants, the regeneration of desiccants is very

slow and energy consuming, which may in return sacrifice the
Mechanical ventilation with heat recovery is often considered
efficiency in energy recovery. Furthermore, cycling beds are bulky
as one of the key elements of a low energy residential building in and the maintenance of rotating wheels is problematic.
various climates 关1兴. In the past, researches were only focused on Membrane-based enthalpy exchanger is another alternative for
the recovery of sensible heat 关2– 4兴 by neglecting the treatment of enthalpy recovery. This concept is just like an air-to-air sensible
humidity for ventilation air. These systems usually employ tradi- heat exchanger. But in place of traditional metal heat exchange
tional heat exchangers such as fixed plates, sensible heat exchange plates, novel hydrophilic membranes are used, through which both
wheels, heat pipes, and coil run-around loop heat exchangers, of heat and moisture are transferred simultaneously. Since the per-
which the performance is easy to predict. However, in recent meation of moisture through the membrane is in a continuous
years, increasing attention has been paid to energy recovery, or manner, no regeneration is required. Furthermore, it is very easy
enthalpy recovery, in which both the sensible and the latent heat to construct and implement, therefore it is more promising. It
are recovered. The latent load constitutes a large fraction of the should be noted that membranes have been used in air dehumidi-
total thermal load for an air-conditioned building, and thus it is fication for a long time, see Pan et al. 关9兴; Asaeda and Du, 关10兴;
more significant to recover the latent heat, as well as the sensible Wang et al. 关11兴; Cha, et al. 关12兴; to name but a few, and some
heat. energy recovery exchangers have been patented by a couple of
The present techniques for enthalpy recovery rely on the alter- companies. However, the theoretical studies of hydrophilic
nate sorption and regeneration of desiccant materials, either in the membrane-based enthalpy recovery exchangers are still scarce
form of cycling packed beds or rotary wheels. San 关5兴 investigated 关13,14兴.
the heat and mass transfer in a set of two-dimensional cross-flow To optimize the membrane system, the effectiveness of en-
regenerative beds. The effects of operating conditions and outdoor thalpy exchange under various conditions should be calculated.
climate conditions on the enthalpy effectiveness are numerically Due to the complex coupling between heat and moisture transfer,
studied. Simonson and Besant 关6兴 proposed a numerical model for there is no simple design methodology available for the
the heat and moisture transfer in energy wheels during sorption, membrane-based enthalpy exchanger. Finite difference simula-
condensation, and frosting conditions, and the sensitivity of con- tions, which have been used to study the sensitivity of perfor-
densation and frosting to wheel speed and desiccant type are stud- mance to outside conditions 关13,15兴, is time consuming and in-
ied. Simonson and Besant 关7,8兴 also investigated the performances convenient for engineers to estimate and select efficient
of energy wheels both numerically and experimentally in detail. membrane enthalpy recovery devices that provide greater life
The fluctuations of sensible and latent effectiveness with various cycle cost savings. Therefore it is imperative to develop some
operating temperatures and humidities are discussed in their re- simple correlations that could predict the sensible and latent ef-
search. fectiveness by summarizing the couplings between the perfor-
These studies are very interesting and they are certainly helpful mance and the sorption characteristics of membrane material, and
to improve the system efficiency. However, due to the low thermal the operating conditions. This is the objective of this research.
2 Basic Equations
HEAT TRANSFER. Manuscript received by the Heat Transfer Division July 1, 2001; A cross-flow enthalpy exchanger with membrane cores is
revision received November 6, 2001. Associate Editor: C. T. Avedisian. shown in Fig. 1. Two air streams- the supply and the exhaust flow
the number of channels. The subscript ‘‘s’’ means ‘‘in supply
side’’. Previous studies found that the temperature difference be-
tween the two sides of membrane is very small due to the small
thickness of membrane 关13兴. So it is reasonable to assume that
T ms ⫽T me ⫽0.5(T s ⫹T e ).
The effect of axial fluid conduction is totally negligible for Pe
共Peclet number兲⫽100, and is quite small even for Pe⫽10. In prac-
tical applications axial conduction is frequently of considerable
significance for laminar flow of liquid metals, which have very
low Prandtl numbers. For gases axial conduction can be of impor-
Fig. 1 Schematic of a cross-flow enthalpy exchanger with
tance only at extremely low Reynolds numbers 关16兴. Generally
membrane cores
speaking, assumption 共2兲 is valid when the Peclet number is big-
ger than 2. The present values of Peclet number in this case are
from 11 to 60.
in thin, parallel, alternating membrane layers, in order to transfer From assumptions 共1兲 and 共2兲, it is seen that heat and moisture
heat and moisture from one air stream to the other. In air condi- transfer is one-dimensional and along the flow direction. How-
tioning, the supply is usually the outside fresh air and the exhaust ever, due to the cross-flow arrangements, the temperature and hu-
is the room air that needs to be discharged to the outside. The midity distributions in the air streams are two-dimensional.
enthalpy exchanger is just like a traditional plate-type heat recu- The above dimensionless parameters are the commonly recog-
perator. The only difference is that hydrophilic membranes are nized Number of Transfer Units. They give an insight into the
used in place of metal plates. The governing dimensionless equa- characteristics of heat and moisture exchange between fluids and
tions for simultaneous heat and moisture transfer in enthalpy ex- surfaces.
changers, based on the assumptions listed in Table 1, are as fol- Moisture flow rate through the membrane:
lows:
Supply ␪ ms ⫺ ␪ me
ṁ w ⫽D wm (5)
⳵Ts ␦
⫽2NTUs 共 T ms ⫺T s 兲 (1) where ␪ ms , ␪ me are moisture uptake in membrane at two surfaces
⳵x*
共kg.kg⫺1兲, ␦ is the membrane thickness, and D wm is the water
⳵␻ s diffusivity in membrane 共kgm⫺1s⫺1兲.
⫽2NTULs共 ␻ ms ⫺ ␻ s 兲 (2)
⳵x* The equilibrium between the membrane and moisture at its sur-
face can be expressed with a general sorption curve as
Exhaust
w max
⳵Te ␪⫽ (6)
⫽2NTUc 共 T me ⫺T e 兲 (3) 1⫺C⫹C/ ␾
⳵y*
where w max represents the maximum moisture content of the
⳵␻ e membrane material 共i.e., moisture uptake when ␾⫽100 percent兲
⫽2NTULe共 ␻ me ⫺ ␻ e 兲 (4)
⳵y* and C determines the shape of the curve and the type of sorption.
The parameters of ␪, ␾, and ␻ can be correlated by ideal gas
where
state equation and psychrometric relations.
x y The convective heat transfer coefficients are obtained from
x *⫽ , y *⫽ , Nusselt correlations 关17兴 and the mass transfer in boundary layers
xF yF
is often described by Sherwood correlations. By using the well-
n s h s x F y F h s A tot n e h e x F y F h e A tot known Chilton-Colburn Analogy 关18兴
NTUs ⫽ ⫽ , NTUe ⫽ ⫽ ,
ṁ s c ps ṁ s c ps ṁ c c pe ṁ c c pe Sh⫽Nu•Le⫺1/3 (7)
k s A tot k e A tot We have
NTULs⫽ , NTULe⫽ ,
ṁ s ṁ e
h
where T ms and T me are the temperature of membrane in supply k⫽ Le⫺1/3 (8)
c pa
side and exhaust side, respectively, ␻ is moisture uptake in air
streams, x and y are coordinates, h is convective heat transfer For ventilation air and vapor mixture, which is always near
coefficient, k is convective mass transfer coefficient, ṁ is mass atmospheric states, the Lewis number, Le, varies in the range of
flow rate of dry air, x F and y F are lengths of flow channels, n is 1.19 to 1.22, see Ref. 关17兴.
Analogous to the heat transfer effectiveness commonly used in
heat exchanger analysis, the concept of effectiveness can be ap-
Table 1 Assumptions used in governing equations plied to the heat and moisture transfer processes in a membrane
based enthalpy exchanger. For a constant specific heat and heat of
phase change, the effectiveness is defined as
Sensible effectiveness
共 ṁc pa 兲 s 共 T si ⫺T so 兲
␧ s⫽ (9)
共 ṁc pa 兲 min共 T si ⫺T ei 兲
Latent effectiveness
共 ṁc pa 兲 s 共 ␻ si ⫺ ␻ so 兲
␧ L⫽ (10)
共 ṁc pa 兲 min共 ␻ si ⫺ ␻ ei 兲
Enthalpy transfer effectiveness, i.e., total energy transfer effec-
tiveness
Fig. 2 Schematic of the experimental setup
ṁ s 共 H si ⫺H so 兲 the uncertainties are less than 5 percent because most of the tem-
␧ tot⫽ (11) perature differences between the two inlets are bigger than 1.5°C
ṁ min共 H si ⫺H ei 兲
and/or the humidity differences are greater than 3.4 g/kg. Before
where H is the specific enthalpy of air, and it is calculated by 关7兴 the test, tracer gas is used to detect the cross-over of air streams.
H⫽c pa T⫹ ␻ 共 2501⫹1.86T 兲 (12) The membranes used are a novel hydrophilic polymer mem-
brane developed and supplied by one of our vendors. It is a modi-
where T is in °C. fication of poly共vinylchloride兲共PVC兲 membrane by cross-linking
The third term in Eq. 共12兲 usually has a less than 3 percent it with a selective coating and a 0.02 ␮m pore diameter porous
effect, thus it can be neglected. Then the enthalpy effectiveness base material. The final membrane thickness is 20 ␮m and the
can be further simplified as
composite material has an average degree of polymerization of
␧ s ⫹␧ L H * 480. For the applications of such membranes in heat and moisture
␧ tot⫽ (13) transfer, the influences of the microstructure on the performance
1⫹H *
are reflected by the sorption curve of the material 共including shape
where and sorption potential兲 and the diffusivity of water in membrane.
2501共 ␻ si ⫺ ␻ ei 兲 ⌬␻ The sorption curve for the membrane material is measured in a
H *⫽ ⬇2501 (14) constant temperature and humidity chamber and is shown in Fig.
c pa 共 T si ⫺T ei 兲 ⌬T
3. As can be seen, it is a third-type sorption curve that is more
where H * is essentially a ratio of latent to sensible energy differ- sensitive to humidity with increasing humidities. For this material,
ences between the inlets of two air streams flowing through the C⫽2.5, w max⫽0.23 kg/kg. The effective diffusivity is found to be
enthalpy exchanger. H * can in theory vary from ⫺⬁ to ⫹⬁, but 2.16⫻10⫺8 kgm⫺1 s⫺1 from the product manual.
varies typically from ⫺6 to ⫹6 for enthalpy recovery in HVAC The configurations of the membrane exchanger are: width, 0.5
applications. From above equation, it is clear that the total en- m; length, 0.5 m; number of membranes for each side, 15; height
thalpy effectiveness is not a simple algebraic average of sensible of flow channels, 5 mm. The exchanger is in cross-flow arrange-
and latent effectiveness. When H * ⫽1, ␧ tot⫽(␧s⫹␧L)/2. As H * ments, for easy of sealing and construction.
→⬁, ␧ tot→␧L ; as H * →0, ␧ tot→␧s ; as H * →⫺1, ␧ tot→⫾⬁.
3 Experimental
Enthalpy exchange experiments were performed on a cross-
flow membrane exchanger with the apparatus shown schemati-
cally in Fig. 2. The supply air flows from a compressed air bottle
and the exhaust air is supplied from a large humidity and tempera-
ture chamber. For the supply air, humidity is adjusted to the de-
sired point by humidifying air through a bubbler immersed in a
bottle of water and subsequently mixing it with a dry air stream.
The temperature is controlled to the desired point by a hot water
bath. The experimental effectiveness is obtained by measuring the
temperatures and humidities at the inlets and outlets of two air
streams. Temperature is measured by platinum resistance and hu-
midity is measured by chilled-mirror dew point meter. Air mass
flow rate is measured and controlled at two values: 0.01 kg/s and
0.05 kg/s. The uncertainties of measurement are: 0.2°C for tem-
perature; 2 percent for humidity; and 5–10 percent for air flow
rate. The maximum uncertainties for the tested sensible effective-
ness are: 16.8 percent for air mass flow rate of 0.01 kg/s and 12.2
percent for 0.05 kg/s flow rate respectively. The maximum uncer-
tainties for the latent effectiveness are 9.8 percent for air mass
flow rate of 0.01 kg/s and 5.2 percent for 0.05 kg/s respectively. A
large fraction of the uncertainties came from the uncertainties in Fig. 3 Sorption curve for the membrane material, C Ä2.5,
air-flow-rate measurement. However, for most of the test results, w maxÄ0.22
4 Development of Effectiveness Correlations or
A total number of transfer units is used to reflect the sensible ṁ W ⫽U L 共 ␻ s ⫺ ␻ e 兲
heat transfer in an exchanger. For the membrane exchanger that
has equal area on both sides, the total number of transfer units for where
sensible heat is
1 1
A totU ␥ tot⫽U L⫺1 ⫽ ⫹ ␥ m⫹ (26)
NTU⫽ ks ke
共 ṁc pa 兲 min
and
where U is the total heat transfer coefficient. Its general form is
␦
冋册
106
冉冊
⫺1
1 ␦ 1 ␥ m⫽ (27)
U⫽ ⫹ ⫹ (15) D wm 共 ⫺5294/T 兲
⳵␪
hs ␭ he e
⳵␾ ms
The term in the middle is the thermal resistance of membrane,
which value is around 0.005 m2K/kW. Other two terms are con- It is indicated that ␥ tot , the total moisture transfer resistance,
vective thermal resistance. Their values are in the order of 40 has an expression similar to thermal resistance where the first and
m2K/kW, or 8000 times larger than membrane resistance. There- the third term are the convective resistance on the supply side and
fore, membrane resistance for heat transfer can be neglected. exhaust side respectively. The middle term, ␥ m , is the moisture
The sensible effectiveness is a function of two dimensionless diffusive resistance in membrane. U L can be called the total mois-
parameters, NTU and R 1 ⫽(ṁc pa ) min /(ṁcpa)max , the ratio of mini- ture transfer coefficient for the device.
mum to maximum heat capacity rate of two air streams. For un- The differentiation of Eq. 共6兲 gives
mixed cross flow, it can be expressed as 关17兴
⳵␪
冋册
w maxC
exp共 ⫺NTU0.78R 1 兲 ⫺1 ⫽ (28)
␧ s ⫽1⫺exp (16) ⳵ ␾ 共 1⫺C⫹C/ ␾ 兲 2 ␾ 2
NTU⫺0.22R 1
The Eq. 共27兲 can be further simplified as
This empirical equation is fairly accurate except at the extremes of
the variables. ␦
A form similar to Eq. 共16兲 for latent effectiveness may be de- ␥ m⫽ ␺ (29)
D wm
rived if the moisture resistance for membrane could be clarified
and estimated.
The moisture flow rate through the membrane can also be ex-
pressed as
␺⫽
106 共 1⫺C⫹C/ ␾ 兲 2 ␾ 2
e 共 5294/T 兲 w maxC
冏 ms
(30)
ṁ W ⫽k s 共 ␻ s ⫺ ␻ ms 兲 ⫽k e 共 ␻ me ⫺ ␻ e 兲 (17) where the coefficient of diffusive resistance for membrane, ␺, is

co-determined by the operating conditions and the slope of sorp-
␪ me ⫽ ␪ ms ⫹
⳵␪
⳵␾
冏 ms
⌬ ␾ ⫽ ␪ ms ⫹
⳵␪
⳵␾
冏 ms
共 ␾ me ⫺ ␾ ms 兲 (18)
tion curves of membrane material.
Similar to the definition of total number of transfer units for
heat, the total number of transfer units for moisture can be written
Substituting Eq. 共18兲 into Eq. 共5兲, we have as
ṁ W ⫽
D wm ⳵ ␪
␦ ⳵␾ 冉冊 ms
共 ␾ ms ⫺ ␾ me 兲 (19) NTUL ⫽
A totU L
共 ṁ 兲 min
(31)
Using Clapeyron equation to represent the saturation vapor pres- The comparison of total transfer units for moisture and sensible
sure and assuming a standard atmospheric pressure of 101325 Pa heat, assuming equal specific heats for two air streams, gives
gives the relation between humidity and relative humidity as
NTUL U L c pa
␾ e 5294/共 T⫹273.15兲 ␤⫽ ⫽ (32)
⫽ ⫺1.61␾ (20) NTU U
␻ 106
Substituting Eqs. 共15兲, 共26兲, 共27兲, into Eq. 共32兲 suggests that
where the second term on the right side of the equation will gen-
erally have less than a 5 percent effect, thus it can be neglected.
Thus the relation between ␾ and ␻ is expressed by 冉冊 1⫹
hs
he
⫹
␦hs
␭
冉冊
␤⫽ (33)
e 5294/共 T⫹273.15兲 hs ␦
␾⫽ ␻ (21) 1⫹ ⫹ ␺k
106 he D wm s
Substituting Eq. 共21兲 into Eq. 共19兲, we have where, ( ␦ h s /␭) is several orders smaller than other terms, mainly
ṁ W ⫽
D wm ⳵ ␪
␦ ⳵␾ 冉冊 e 5294/共 T⫹273.15兲
106
共 ␻ ms ⫺ ␻ me 兲 (22)
due to the small thickness of membranes, so it can be neglected.
The above equation can be further simplified into
冉冊
ms
hs
From Eq. 共17兲, two equations can be deduced 1⫹
he
冉冊
␻ ms ⫽ ␻ s ⫺ṁ w /k s (23) ␤⫽ (34)
hs ␥m
1⫹ ⫹
␻ me ⫽ ␻ c ⫹ṁ w /k c (24) he 1/k s
Substituting above two equations into Eq. 共22兲 to eliminate ␻ ms In most cases, the enthalpy recovery is implemented with bal-
and ␻ me gives anced flows, i.e., the two air streams have the same flow rates.
Therefore, the convective heat transfer coefficients would have
1
ṁ W ⫽ 共␻ ⫺␻e兲 (25) the same value on both sides of membrane result from similar
␥ tot s fluid fields. Consequently, Eq. 共34兲 can be written in
1
␤⫽ (35)
1⫹ ␣
where
␥m
␣⫽ (36)
␥c
2
␥ c⫽ (37)
ks
where ␥ c is the convective moisture transfer resistance, and ␣ is
the ratio of diffusive resistance to convective resistance for mem-
brane. As can be seen, the total number of transfer units for mois-
ture can be estimated from the total number of transfer units for
sensible heat, and ratio of diffusive to convective moisture resis-
tance. As ␣→⬁, NTUL →0, no moisture can be permeated through
the membrane. In this case, the ‘‘membrane’’ is like a metal plate.
On the other hand, as ␣→0, NTUL →NTU, ⑀ L ⫽ ⑀ s . If ␣⫽1,
NTUL ⫽NTU/2. Under the current operating conditions, the val-
ues of ␣ vary from 2 to 10 共see Fig. 6兲, which implies that mem-
brane resistance for moisture transfer cannot be neglected.
Fig. 4 Calculated and experimental sensible effectiveness for
Similar to the deduction of Eq. 共16兲 for sensible heat transfer, a cross-flow membrane enthalpy exchanger
the correlation for latent effectiveness can be written as
␧ L ⫽1⫺exp 冋 exp共 ⫺NTUL0.78R 2 兲 ⫺1

NTUL⫺0.22R 2
册 (38)
NTUL ⫽ ␤ •NTU (39)
R 2 ⫽ṁ min /ṁ max (40)
A more detailed set up of the analogy between Eqs. 共38兲 and

共16兲 can be seen in the appendix. The latent effectiveness corre-
lations for other flow arrangements, such as concurrent flow and
counter flow, can also be derived from those corresponding corre-
lations for sensible effectiveness, using the definition of Eq. 共35兲.
The value of relative humidity of membrane in supply side, which
is determined by latent effectiveness 共permeation rate兲, needs to
be known before the calculation of diffusive resistance for mem-
brane ␥ m and the diffusive to convective ratio ␣. Iterations are
performed to obtain a converged solution for ␾ ms .

To demonstrate the suitability of the correlations in predicting Fig. 5 Calculated and experimental latent effectiveness for a
the effectiveness, sensible and latent effectiveness are calculated cross-flow membrane enthalpy exchanger
with the proposed correlations and compared with experimental
results, as shown in Figs. 4 and 5 respectively. To make more
extensive comparisons, simulation results with a finite difference
model proposed by Zhang and Jiang 关13兴 are also plotted in the
two figures. The uncertainty for the finite difference results is less
than 1 percent with a grid size of 2 mm.
From Figs. 4 and 5, it is obvious that both the sensible and the
latent effectiveness are properly represented by the correlation
from the present study. The largest discrepancies between the pre-
dictions by the correlation and the experimental data result from
the cases with the smallest air flow rate, where the uncertainties of
the experimental data are the biggest. The average errors between
the predicted and experimental results are 7.3 percent and 8.6
percent for sensible and latent effectiveness respectively.
For a given exchanger, the sensible effectiveness is a fixed
value at the specified flow rates. However the latent effectiveness
will be affected by the two important dimensionless factors pro-
posed in this study: the ratio of diffusive to convective resistance
共␣兲, and the ratio of total number of transfer units of moisture to
that of sensible 共␤兲. The values of ␣ and ␤ are in turn affected by Fig. 6 Variations of ␣ and ␤ with increasing inlet relative hu-
the membrane material types and operating inlet conditions. The midity for first-type „ C Ä0.1… membrane material
Fig. 10 Latent effectiveness for three types of membranes,
NTUÄ4.2
Fig. 7 Variations of ␣ and ␤ with increasing inlet relative hu-
midity for linear type „ C Ä1.0… membrane material
15 at 90 percent RH, while for third-type membrane, ␣ decreases

from 8.6 at 10 percent RH to 0.4 at 90 percent RH. The bigger the
␣, the larger the ␤ and the latent effectiveness.
Figure 9 demonstrates the variations of NTUL , when NTU is
kept constant, with different inlet humidities. The value of NTUL
decreases and increases for first-type and third-type material, with
increasing inlet humidity respectively. The trends of resulting la-
tent effectiveness are the same as those of NTUL , which can be
deduced from Eq. 共31兲, see Fig. 10. For the linear type material,
the NTUL and the latent effectiveness will not change with the
outside conditions. The number of transfer units for moisture
would keep at 0.45 times of that for sensible heat for this material.
The above discussions suggest that an enthalpy exchanger with
linear type membrane cores always performs better than those
with other membrane cores. For example, to obtain a latent effec-
tiveness of 0.6 under an inlet humidity of 50 percent, NTUL
should be at least 2.0, which means that the minimum values of
NTU required for the exchanger are: 4.44 with linear type; 8 with
Fig. 8 Variations of ␣ and ␤ with increasing inlet relative hu- third-type; and 11.1 with first-type membrane. A smaller NTU
midity for third-type „ C Ä10… membrane material usually makes the enthalpy exchanger more compact and cheaper.
6 Conclusions
Enthalpy recovery with new hydrophilic membranes has poten-
tially extensive uses in energy efficient buildings. To evaluate the
system performance, the effectiveness correlations are proposed.
By separating the moisture resistance through membranes and
building up an analogy between the number of transfer units for
moisture and that for sensible heat, the latent effectiveness corre-
lation is written in a form very similar to the empirical correlation
for sensible effectiveness that can be easily found in many litera-
tures. A comparison with established experimental results shows
that the correlations correctly predict the influences of the design
variables on the performance of the membrane system. The stud-
ies also find that the ratio of diffusive to convective resistance 共␣兲
determines the number of transfer units for moisture, with a fixed
value of the number of transfer units for sensible heat. The ratio ␣
reflects the couplings between the moisture transfer and the sorp-
tion characteristics of membrane material and the operating con-
Fig. 9 Variations of NTUL with relative humidity for different ditions. For a given number of transfer units for sensible heat,
membranes, NTUÄ4.2 enthalpy exchangers with linear type membrane cores would have
the highest transfer units for moisture, which in return results in
the highest latent effectiveness.
variations of ␣ and ␤ with increasing inlet relative humidities of
the supply air are plotted in Figs. 6 – 8, for three kinds of mem-
brane material that are most often used. Acknowledgment
The value of ␣ and ␤ with linear type membrane does not This research is funded by the Research Grant Council of the
change with inlet relative humidity. However, for first-type mem- Hong Kong SAR Government and the postdoctoral fellowship of
brane material, ␣ increases from around 0.5 at 10 percent RH to the Hong Kong Polytechnic University.
Nomenclature 共 ṁc pa 兲 s ⳵ T s
dq⫽⫺ dxdy (A3)
A ⫽ transfer area 共m2兲 yF ⳵x
C ⫽ constant in sorption curve
c p ⫽ specific heat 共kJkg⫺1K⫺1兲共 ṁc pa 兲 e ⳵ T e
dq⫽ dxdy (A4)
D wm ⫽ diffusivity of water in membrane 共kgm⫺1s⫺1兲 xF ⳵y
H ⫽ specific enthalpy 共kJ/kg兲
H * ⫽ ratio of latent to sensible energy differences be- ṁ s ⳵␻ s
dṁ W ⫽⫺ dxdy (A5)
tween the inlets of two air streams yF ⳵x
h ⫽ convective heat transfer coefficient 共kWm⫺2K⫺1兲
k ⫽ convective mass transfer coefficient 共kgm⫺2s⫺1兲 ṁ e ⳵␻ e
dṁ W ⫽ dxdy (A6)
ṁ ⫽ mass flow rate of air streams 共kg/s兲 xF ⳵y
ṁ W ⫽ mass flow rate of moisture flow 共kgm⫺2s⫺1兲
NTU ⫽ number of transfer units Combining Eqs. 共A1兲 and 共A3兲 and then Eqs. 共A1兲 and 共A4兲 gives
Nu ⫽ Nusselt number Uy F ⳵Ts
n ⫽ number of channels 共 T s ⫺T e 兲 ⫽⫺ (A7)
R ⫽ ratio for heat/mass capacity 共 ṁc pa 兲 s ⳵x
Sh ⫽ Sherwood number Ux F ⳵Te
T ⫽ temperature 共°C兲共 T s ⫺T e 兲 ⫽ (A8)
U ⫽ total heat transfer coefficient 共kWm⫺2K⫺1兲共 ṁc pa 兲 e ⳵y
U L ⫽ total mass transfer coefficient 共kgm⫺2s⫺1兲 Similarly, combining Eqs. 共A2兲 and 共A5兲 and then Eqs. 共A2兲 and
w max ⫽ maximum water uptake of desiccant 共kgkg⫺1兲共A6兲 gives
x, y ⫽ coordinates 共m兲
x * , y * ⫽ nondimensional coordinates ULy F ⳵␻ s
x F ⫽ length of supply channel 共m兲共 ␻ s ⫺ ␻ e 兲 ⫽⫺ (A9)
ṁ s ⳵x
y F ⫽ length of exhaust channel 共m兲
U Lx F ⳵␻ e
Greek Letters 共 ␻ s⫺ ␻ e 兲⫽ (A10)
ṁ e ⳵y
␺ ⫽ coefficient of moisture diffusive resistance in mem-
brane 共CMDR兲 Differentiating Eqs. 共A7兲 and 共A8兲 with respect to y and x and
␭ ⫽ thermal conductivity of membrane 共kWm⫺1K⫺1兲 taking their sum gives
␪ ⫽ moisture uptake in membrane 共kgkg⫺1兲
␧ ⫽ effectiveness Ux F ⳵ 共 T s ⫺T e 兲 Uy F ⳵ 共 T s ⫺T e 兲 ⳵ 2 共 T s ⫺T e 兲
⫹ ⫽⫺
␥ ⫽ resistance 共m2K/k W for thermal and m2s/kg for 共 ṁc pa 兲 e ⳵x 共 ṁc pa 兲 s ⳵y ⳵x⳵y
moisture兲 (A11)
␾ ⫽ relative humidity Similarly, differentiating Eqs. 共A9兲 and 共A10兲 with respect to y and
␦ ⫽ thickness of membrane 共m兲 x and taking their sum gives
␻ ⫽ moisture content 共kg moisture/kg dry air兲
␣ ⫽ ratio of diffusive to convective moisture resistance U Lx F ⳵共 ␻ s⫺ ␻ e 兲 U Ly F ⳵共 ␻ s⫺ ␻ e 兲 ⳵ 2共 ␻ s⫺ ␻ e 兲
for membrane ⫹ ⫽⫺
ṁ e ⳵x ṁ s ⳵y ⳵x⳵y
␤ ⫽ ratio of total number of transfer units for moisture to (A12)
that for sensible heat
Let dimensionless variables
Subscripts
a ⫽ air T s ⫺T e ␻ s⫺ ␻ e x y
␪ 1⫽ , ␪ 2⫽ , x *⫽ , y *⫽
c ⫽ convective T si ⫺T ei ␻ si ⫺ ␻ ei xF yF
e ⫽ exhaust
i ⫽ inlet and substituting in Eqs. 共A11兲, 共A12兲,
L ⫽ latent, moisture Ux F y F ⳵ ␪ 1 Ux F y F ⳵ ␪ 1 ⳵ 2␪ 1
m ⫽ membrane, diffusive ⫹ ⫽⫺ (A13)
s ⫽ supply, sensible 共 ṁc pa 兲 e ⳵ x * 共 ṁc pa 兲 s ⳵ y * ⳵ x *⳵ y *
o ⫽ outlet U Lx Fy F ⳵ ␪ 2 U Lx Fy F ⳵ ␪ 2 ⳵ 2␪ 2
tot ⫽ total ⫹ ⫽⫺ (A14)
w ⫽ water ṁ e ⳵ x * ṁ s ⳵ y * ⳵ x *⳵ y *
with
Appendix
Ux F y F Ux F y F
Deduction of effectiveness correlations NTUa ⫽ , NTUb ⫽ ,
共 ṁc pa 兲 e 共 ṁc pa 兲 s
Considering a cross-flow membrane exchanger with only one
flow channel. At any point in the exchanger a heat and mass U Lx Fy F U Lx Fy F
balance for an infinitely small volume dxdy can be written from NTULa ⫽ , NTULb ⫽
ṁ e ṁ s
Eq. 共15兲 and Eq. 共25兲 as
Eqs. 共A13兲 and 共A14兲 become
dq⫽U 共 T s ⫺T e 兲 dxdy (A1)
⳵␪1 ⳵␪1 ⳵ 2␪ 1
dṁ W ⫽U L 共 ␻ s ⫺ ␻ e 兲 dxdy (A2) NTUa ⫹NTUb ⫹ ⫽0 (A15)
⳵x* ⳵ y * ⳵ x *⳵ y *
Equation 共A1兲 is a basic heat transfer equation, and Eq. 共A2兲 has
been widely employed as a mass permeation model through a ⳵␪2 ⳵␪2 ⳵ 2␪ 2
membrane in chemical industry 关19兴. NTULa ⫹NTULb ⫹ ⫽0 (A16)
⳵x* ⳵ y * ⳵ x *⳵ y *
Across the elements x F and y F units in length the energy and
moisture balances yield Initial condition: ␪ 1 (0,0)⫽ ␪ 2 (0,0)⫽1
Mason 关20兴 obtained a solution for Eq. 共A15兲 in the form of an Similarly, for moisture effectiveness, we have
冋册
infinite series by employing the Laplace transformation as follow-
⬁ n
ing: 1 共 NTUL 兲 m
兺兺
兺冋册
⬁ ␧ L⫽ 1⫺e ⫺NTUL
共 NTUa 兲共 NTUb 兲 x * y * n R 2 共 NTUL 兲 n⫽0 m⫽0 m!
␪ 1 共 x * ,y * 兲 ⫽e ⫺ 共 NTUa 兲 y * ⫹ 共 NTUb 兲 x *
再冎
n⫽0 共 n! 兲 2 n
关 R 2 共 NTUL 兲兴 m
(A17) ⫻ 1⫺e ⫺R 2 共 NTUL 兲兺 m!
(A27)
Since Eqs. 共A15兲 and 共A16兲 are the same form of differential m⫽0
equations. They are identical if NTUa is replaced by NTULa and

NTUb by NTULb . Therefore, the solution to Eq. 共A16兲 can be We can see at this step that the moisture effectiveness has the
written as same form of expression with sensible effectiveness. The only
兺冋册
⬁
differences are that NTU is replaced by NTUL and R 1 is replaced
共 NTULa 兲共 NTULb 兲 x * y * n by R 2 .
␪2共x*,y*兲⫽e ⫺ 共 NTULa 兲 y * ⫹ 共 NTULb 兲 x * For heat transfer, Eq. 共A26兲 is too complicated, since it has
n⫽0 共 n! 兲 2
(A18) infinite series. Therefore, Kays and London 关21兴 used following
empirical equation to represent the sensible effectiveness as
The overall heat transferred in the exchanger is the integral of
Eq. 共A17兲
Q⫽Ux F y F 共 T si ⫺T ei 兲冕冕 1 1
␪ 1 共 x * ,y * 兲 dx * dy * (A19)
␧ s ⫽1⫺exp 冋 exp共 ⫺R 1 共 NTU0.78兲 ⫺1
R 1 NTU⫺0.22 册 (A28)
0 0
Similarly, Eq. 共27兲, which is similar to Eq. 共26兲, could be approxi-
The overall moisture transferred in the exchanger is the integral mated with
of Eq. 共A18兲
M w ⫽U L x F y F 共 ␻ si ⫺ ␻ ei 兲冕冕 0
1
0
1
␪ 2 共 x * ,y * 兲 dx * dy * (A20) ␧ L ⫽1⫺exp 冋 exp共 ⫺R 2 NTUL0.78兲 ⫺1
R 2 NTUL⫺0.22
册 (A29)
If we define
Ux F y F U Lx Fy F References
NTU⫽ , NTUL ⫽ 关1兴 Dorer, V., and Breer, D., 1998, ‘‘Residential Mechanical Ventilation Systems:
共 pa 兲 min
ṁc 共 ṁ 兲 min Performance Criteria and Evaluations,’’ Energy Build., 27共3兲, pp. 247–255.
关2兴 Dhital, P., Besant, R. W., and Schoenau, G. J., 1995, ‘‘Integrating Run-Around
共 ṁc pa 兲 min ṁ min
R 1⫽ , R 2⫽ Heat Exchanger Systems Into the Design of Large Office Buildings,’’
共 ṁc pa 兲 max ṁ max ASHRAE J., 101共2兲, pp. 979–991.
冕冕
关3兴 Johnson, A. B., Besant, R. W., and Schoenau, G. J., 1995, ‘‘Design of Multi-
1 1 Coil Run-Around Heat Exchanger Systems for Ventilation Air Heating and
⍀ 1⫽ ␪ 1 共 x * ,y * 兲 dx * dy * , Cooling,’’ ASHRAE Trans., 101共2兲, pp. 967–978.
0 0 关4兴 Manz, H., Huber, H., Schalin, A., Weber, A., Ferrazzini, M., and Studer, M.,
冕冕 1 1
2000, ‘‘Performance of Single Room Ventilation Units With Recuperative or
Regenerative Heat Recovery,’’ Energy Build., 31共1兲, pp. 37– 47.
⍀ 2⫽ ␪ 2 共 x * ,y * 兲 dx * dy * 关5兴 San, J. Y., 1993, ‘‘Heat and Mass Transfer in a Two-Dimensional Cross-Flow
0 0 Regenerator With a Solid Conduction Effect,’’ Int. J. Heat Mass Transf., 36共3兲,
pp. 633– 643.
Q MW 关6兴 Simonson, C. J., and Besant, R. W., 1998, ‘‘Heat and Moisture Transfer in
␧ s⫽ , ␧ L⫽
共 ṁc pa 兲 min共 T si ⫺T ei 兲 ṁ min共 ␻ si ⫺ ␻ ei 兲 Energy Wheels During Sorption, Condensation, and Frosting Conditions,’’
ASME J. Heat Transfer, 120, pp. 699–708.
then 关7兴 Simonson, C. J., and Besant, R. W., 1999, ‘‘Energy Wheel Effectiveness: Part
I—Development of Dimensionless Groups,’’ Int. J. Heat Mass Transf., 42, pp.
␧ s ⫽NTU⍀ 1 (A21) 2161–2170.
␧ L ⫽NTUL ⍀ 2 (A22) 关8兴 Simonson, C. J., and Besant, R. W., 1999b, ‘‘Energy Wheel Effectiveness: Part
II—Correlations,’’ Int. J. Heat Mass Transf., 42, pp. 2171–2185.
⬁ 关9兴 Pan, C. Y., Jensen, C. D., Bielech, C., and Habgood, H. W., 1978, ‘‘Permeation
1
⍀ 1⫽
共 NTUa 兲共 NTUb 兲 n⫽0 兺 f 共 NTUa 兲 f 共 NTUb 兲 (A23) of Water Vapor Through Cellulose Triacetate Membranes in Hollow Fiber
Form,’’ J. Appl. Polym. Sci., 22, pp. 2307–2323.
关10兴 Asaeda, M., and Du, L. D., 1986, ‘‘Separation of Alcohol/Water Gaseous
⬁ Mixtures by Thin Ceramic Membrane,’’ J. Chem. Eng. Jpn., 19, pp. 72–77.
1
⍀ 2⫽
共 NTULa 兲共 NTULb 兲 n⫽0 兺
f 共 NTULa 兲 f 共 NTULb 兲 (A24) 关11兴 Wang, K. L., Mccray, S. H., Newbold, D. D., and Cusseler, E. L., 1992,
‘‘Hollow Fiber Air Drying,’’ J. Membr. Sci., 72, pp. 231–244.
关12兴 Cha, J. S., Li, R., and Sirkar, K. K., 1996, ‘‘Removal of Water Vapor and
where VOCs From Nitrogen in a Hydrophilic Hollow Fiber Gel Membrane Perme-
n ator,’’ J. Membr. Sci., 119, pp. 139–153.
zm
兺
⫺z 关13兴 Zhang, L. Z., and Jiang, Y., 1999, ‘‘Heat and Mass Transfer in a Membrane-
f 共 z 兲 ⫽1⫺e (A25) Based Energy Recovery Ventilator,’’ J. Membr. Sci., 163, pp. 29–38.
m⫽0 m! 关14兴 Zhang, Y. P., Jiang, Y., Zhang, L. Z., Deng, Y. C., and Jin, Z. F., 2000, ‘‘Analy-
sis of Thermal Performance and Energy Saving Effect of Membrane Based
There are two cases: Heat Recovery Ventilator,’’ Energy-The International Journal, 25共6兲, 515–527.
If (ṁc pa ) e ⫽(ṁc pa ) min , then NTUa ⫽NTU, and NTUb 关15兴 Zhang, L. Z., Jiang, Y., and Zhang, Y. P., 2000, ‘‘Membrane-Based Humidity
⫽R 1 NTU Pump: Performance and Limitations,’’ J. Membr. Sci., 171, pp. 207–216.
关16兴 Kays, W. M., and Crawford, M. E., 1993, Convective Heat and Mass Transfer,
If (ṁc pa ) e ⫽(ṁc pa ) max , then NTUa ⫽R 1 NTU, and NTUb McGraw-Hill New York, pp. 432– 435.
⫽NTU 关17兴 Incropera, F. P., and Dewitt, D. P., 1990, Fundamentals of Heat and Mass
In both cases, Eq. 共A21兲 can be replaced by the relationship Transfer, 3rd ed., Wiley, New York, pp. 416 – 420, 556 –557.
冋册
⬁ n 关18兴 Taylor, R., and Krishna, R., 1993, Multicomponent Mass Transfer, John Wiley
1 共 NTU兲 m
兺兺
& Sons, New York, pp. 278 –279.
␧ s⫽ 1⫺e ⫺NTU 关19兴 Scovazzo, P., Burgos, J., Hoehn, A., and Todd, P., 1998, ‘‘Hydrophilic
R 1 共 NTU兲 n⫽0 m⫽0 m! Membrane-Based Humidity Control,’’ J. Membr. Sci., 149, pp. 69– 81.
再冎
n
关20兴 Kern, D. Q., and Kraus, A. D., 1972, Extended Surface Heat Transfer,
关 R 1 共 NTU兲兴 m
兺
McGraw-Hill, New York, pp. 617– 621.
⫻ 1⫺e ⫺R 1 共 NTU兲 (A26) 关21兴 Kraus, A. D., Aziz, A., and Welty, J., 2001, Extended Surface Heat Transfer,
m⫽0 m! John Wiley & Sons, New York, pp. 471.
On the Potential for
Homogeneous Nucleation of Salt
From Aqueous Solution in a
Kenneth A. Smith
e-mail: kas@mit.edu
Natural Convection Boundary
Department of Chemical Engineering,
Massachusetts Institute of Technology,
77 Massachusetts Avenue, Room 66-540,
Layer
Cambridge, MA 02139
Recent studies have examined the rate of salt deposition by natural convection on a
1 cylinder heated above the solubility temperature corresponding to the concentration of
Marc Hodes salt in the surrounding solution at conditions typical of the Supercritical Water Oxidation
(SCWO) process (Hodes et al. [1,2], Hodes [3]). The total deposition rate of salt on the
cylinder is the sum of the rate of deposition at the salt layer-solution interface (SLSI)
77 Massachusetts Avenue,
formed on the cylinder and that within the porous salt layer. The rate of deposition at the
Cambridge, MA 02139
SLSI cannot be computed without determining whether or not salt nucleates homoge-
neously in the adjacent (natural convection) boundary layer. A methodology to determine
Peter Griffith whether or not homogeneous nucleation in the boundary layer is possible is presented
e-mail: pgrif@mit.edu
here. Temperature and concentration profiles in the boundary layer are computed under
the assumption that homogeneous nucleation does not occur. If, under this assumption,
supersaturation does not occur, homogeneous nucleation is impossible. If supersaturation
77 Massachusetts Avenue, Room 7-044,
is present, homogeneous nucleation may or may not occur depending on the amount of
Cambridge, MA 02139
metastability the solution can tolerate. It is shown that the Lewis number is the critical
solution property in determining whether or not homogeneous nucleation is possible and
a simple formula is developed to predict the Lewis number below which homogeneous
nucleation is impossible for a given solubility boundary and set of operating conditions.
Finally, the theory is shown to be consistent with experimental observations for which
homogeneous nucleation is absent or present. 关DOI: 10.1115/1.1494089兴
Keywords: Crystal Growth, Cylinder, Heat Transfer, Mass Transfer, Natural Convection,
Phase Change
Introduction rate of salt on the cylinder is the sum of the deposition rate at the
salt layer-solution interface 共SLSI兲 formed on the cylinder plus
Supercritical Water Oxidation 共SCWO兲, oxidation in water at
that within the porous salt layer. The dominant transport mode for
temperatures and pressures exceeding the critical temperature
deposition at the SLSI is 共laminar兲 natural convection 关3兴. Depo-
共374°C兲 and critical pressure 共221 bar兲 of pure water, is an effec-
sition rates at the SLSI have been computed by Hodes et al. 关2,3兴
tive technology for the remediation of hazardous organic wastes. under the assumption that dissolved salt is transported to it by
Reviews of SCWO technology have been provided by Tester et al. molecular diffusion 共with advection兲 and that it subsequently
关4兴 and, more recently, by Gloyna and Li 关5兴. Widespread com- nucleates heterogeneously. A methodology to determine whether
mercialization of SCWO continues to be hindered by scale or not homogeneous nucleation in the boundary layer is possible
buildup 共fouling兲 by ‘‘sticky’’ salts. 共See, for example, Shaw and is presented here and applied to the experimental conditions.
Dahmen 关6兴.兲 Various salts are commonly present in SCWO feeds Molecular diffusion coefficients for salts in supercritical water
or produced during neutralization of halogen-containing wastes. 共SCW兲 are necessary to calculate deposition rates in SCWO sys-
These salts precipitate from solution due to the low dielectric tems and compute the solution Lewis numbers needed to apply the
constant of water at SCWO conditions. Precipitated salts can form results developed below. Butenhoff et al. 关8兴 measured the mo-
agglomerates and coat internal surfaces and, if left uncontrolled, lecular diffusion coefficient of sodium nitrate in SCW. It ranged,
plug SCWO reactors or transport lines. for example, from 1.06* 10⫺8 to 2.17* 10⫺8 m2 /sec as pressure
Recently, rates of deposition from aqueous salt solutions to a was increased from 271.6 to 600.0 bar at 400°C in a 1 molal
heated cylinder at near-supercritical conditions have been mea- aqueous sodium nitrate solution. These are the only data available
sured by Hodes et al. 关1,3兴 in the 共visually accessible兲 test cell for molecular diffusion coefficients of salts in SCW. Lamb et al.
described by Hurst et al. 关7兴. The solutions were preheated to 关9兴 measured the self diffusion coefficient in SCW over the 400 to
about 5°C below the solubility temperature corresponding to the 700°C temperature range and 199 to 1459 bar pressure range.
concentration of salt in the bulk solution surrounding the cylinder Molecular diffusion coefficients in gases and liquids are normally
and the temperature of the cylinder was increased beyond this on the order of 10⫺5 and 10⫺9 m2 /s, respectively 关10兴; and those
solubility temperature to drive deposition. The total deposition measured by Butenhoff et al. 关8兴 and Lamb et al. 关9兴 in SCW lie
between those for gases and liquids. Butenhoff et al. 关8兴 also mea-
1
Current address is: Bell Laboratories, Lucent Technologies, 700 Mountain Ave., sured the thermal diffusivity of the sodium nitrate-water solutions.
Rm. 1C-433A, Murray Hill, NJ 07974, Hodes@lucent.com
Lewis numbers were between about 5 and 7 throughout the tem-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division October 20, perature 共400–500°C兲, pressure 共270–1000 bar兲 and concentra-
2000; revision received April 18, 2002. Associate Editor: J. Georgiadis. tion 共0.25–3.0 molal兲 space investigated by Butenhoff et al. 关8兴.
one phase fluid region from a solid-fluid region. The phase bound-
ary shown is linear, but the physics elucidated here also apply to a
phase boundary with curvature. Throughout this paper, saturation
conditions are assumed to exist at the SLSI. Trajectories of
temperature-concentration (T,C) states connecting the bulk and
SLSI conditions corresponding to solution Lewis numbers of 0, 1,
and ⬁ are also shown in Fig. 1. Note that Fig. 1 does not corre-
spond to a constant heat flux boundary condition at the SLSI
because the temperature difference between the SLSI and bulk
solution is then a function of the thermophysical properties of the
fluid, i.e., Lewis number, etc. This figure represents fixed tempera-
tures at the SLSI and bulk conditions.
The ratio of the thermal boundary layer thickness to the con-
centration boundary layer thickness scales as a 共positive兲 frac-
tional power of Lewis number. Consequently, as the Lewis num-
ber approaches zero, the thermal boundary layer thickness
becomes vanishingly small relative to the concentration boundary
layer thickness and salt 共solute兲 concentration decreases from its
bulk value to its SLSI value at the bulk temperature. Analogously,
temperature increases from its bulk value to its SLSI value at the
SLSI concentration. The Le→⬁ trajectory in Fig. 1 is analogous
Fig. 1 Loci of temperature-concentration states connecting
to that for Le→0. The rectangle bounded by the Le→0 and Le→⬁
bulk and SLSI conditions for selected Lewis numbers. Below trajectories envelopes the physically plausible (T,C) states which
the solubility line is a one phase fluid region and above it a two may exist between the bulk and SLSI conditions. When the Lewis
phase, solid-fluid region. number equals one, dimensionless salt concentration equals di-
mensionless temperature throughout the boundary layer and a
straight line connects the bulk and SLSI conditions. 共Rigorously,
this assumes that the non-dimensional boundary conditions for the
Armellini et al. 关11兴 performed shock-crystallization experi-
energy and species equations are the same.兲 If homogeneous
ments to simulate the rapid precipitation of salts at typical SCWO
nucleation within the boundary layer is to be possible, the locus of
conditions. Measured particle diameters for sodium sulfate and
(T,C) states within the boundary layer must cross the solubility
sodium chloride precipitates typically ranged from 1 to 20 mi-
crons and 5 to 100 microns, respectively, for the flow conditions line. Thus, if the phase boundary is linear and Le⭐1, homoge-
investigated by Armellini et al. 关11兴. Brownian diffusion coeffi- neous nucleation is not possible.
cients of particles in this size range may be calculated from the If nucleation does not occur in the boundary layer, the salt
Stokes-Einstein equation and are several orders of magnitude concentration corresponding to a given temperature must increase
smaller than the corresponding molecular diffusion coefficients. with solution Lewis number because of the reduced thickness of
Moreover, the laminar natural convection mass transfer coefficient the concentration boundary layer relative to the thermal boundary
(h m ) for transport of salt from the bulk solution to the SLSI scales layer. Moreover, beyond some critical Lewis number (Lec ), salt
as the mass diffusion coefficient to the 3/4 power. Thus, if mo- concentration will exceed the solubility concentration at the cor-
lecular clusters of salt nucleate homogeneously in the boundary responding temperature; and supersaturation and/or homogeneous
layer and grow to Brownian-sized particles which diffuse to the nucleation must occur. As the Lewis number is increased beyond
SLSI, the deposition rate will be substantially slower than if salt its critical value, the amount of potential supersaturation increases
molecules diffuse to the SLSI and then nucleate heterogeneously. and, therefore, homogeneous nucleation becomes more likely. In
Even when both particles and molecules diffuse to the SLSI, one the limit as Le→⬁, all of the flow field surrounding the heated
expects that the deposition rate will be slower than for pure mo- cylinder is at the bulk concentration; hence, the level of potential
lecular diffusion because formation and growth of Brownian par- supersaturation is at a maximum throughout the boundary layer.
ticles depletes the driving force for molecular diffusion. At sufficiently high levels of supersaturation, a metastable limit is
The potential for homogeneous nucleation of salt in the natural surpassed and salt will precipitate from solution; however, meta-
convection boundary layer around the SLSI is investigated here. A stable limits for solutes of interest are currently unavailable.
qualitative relationship between the Lewis number of the aqueous If constant power is supplied to the cylinder and salt is allowed
salt solution, the phase boundary curve, and the potential for ho- to accumulate on it, the critical Lewis number will increase with
mogeneous nucleation is developed first. Then quantitative criteria time because the temperature of the SLSI decreases as its area
are developed in order to predict whether or not supersaturation increases. Figure 2 shows three successive loci of (T,C) states
and/or homogeneous nucleation will occur within the boundary connecting a bulk condition to three respective SLSI conditions.
layer. The criteria are then applied to the conditions in the afore- The loci were computed from the analysis presented below and
mentioned deposition experiments. Next, characteristics of aque- correspond to a solution for which Le⫽100, Pr⫽0.72, and Ñ 共de-
ous salt solutions conducive to supersaturation and/or homoge- fined below兲 is of order 1. 共The Prandtl number of 共pure兲 water at
neous nucleation within the boundary layer are identified and a pressure of 250 bar is plotted in Fig. 3 by using thermophysical
specific solutions with such characteristics are discussed. Finally, property data from the NIST Steam Tables 关12兴. For temperatures
the results of a nucleation experiment using lithium carbonate are ranging from 200°C to 700°C it is between 0.79 and 2.0, except in
presented. the interval from 376°C to 405°C where it increases to about 8 at
385°C.兲 At the highest SLSI temperature shown in Fig. 2, the
Lewis number of the solution exceeds the critical Lewis number
Qualitative Relationship Between Lewis Number and
and supersaturation and/or homogeneous nucleation occurs in the
Nucleation Mechanism boundary layer. As the deposition process continues, a condition is
Figure 1 illustrates the relationship between the Lewis number reached such that the solution Lewis number equals the critical
of the aqueous salt solution, the phase boundary and the nucle- Lewis number. Finally, at the lowest SLSI temperature, the solu-
ation mechanism in the boundary layer. It shows a bulk solution tion Lewis number is below the critical value and saturation con-
condition, SLSI condition and a phase boundary which separates a ditions exist only at the SLSI.
Fig. 2 Loci of „ T , C … states connecting a bulk condition to Fig. 4 Hypothetical set of conditions for which supersatura-
three successive SLSI conditions. Loci correspond to LeÄ100, tion andÕor homogeneous nucleation occurs in the boundary
PrÄ0.72, and Ñ ÄO†1‡ layer, but dT Õ dC 円 i Ë dT Õ dC 円 sat, i
As steady-state conditions are approached, the SLSI concentra- Quantitative Nucleation Criteria
tion approaches the bulk concentration and the critical Lewis A sufficient criterion for supersaturation and/or homogeneous
number approaches infinity. Since the Lewis number of any solu- nucleation to occur in the boundary layer is that dT/dC in the
tion is finite, eventually salt must be transported to the SLSI ex- solution at the SLSI exceeds dT/dC of the phase boundary at the
clusively by molecular diffusion. It is noted that, even if super- SLSI, i.e.,
saturation is not present in the natural convective boundary layer
formed adjacent to the nominal SLSI, nucleation is not necessarily
confined to this nominal interface. Indeed, analyses suggest that a
significant amount of heterogeneous nucleation and deposition oc-
dT
⬎冏冏
dT
dC i dC sat,i
(1)
curs within the porous salt layer formed around the heated cylin- Although this criterion is necessary for supersaturation and/or ho-
der 共关2,3兴兲. The relationship between the solution Lewis number, mogeneous nucleation to occur for most systems, exceptions are
the phase boundary, and the nucleation mechanism also applies to possible. A phase boundary and trajectory of (T,C) states 共at Pr
the analogous problem for which solubility increases with tem-
⫽0.72, Le⫽100, Ñ⫽O 关1兴兲 connecting bulk and SLSI conditions
perature and solute is deposited on a cooled cylinder as in the
for which supersaturation and/or homogeneous nucleation must
problem of fog formation.
occur, but for which Eq. 1 is not satisfied, is shown in Fig. 4. In
the usual case for which Eq. 1 applies, it follows from the defini-
tions of the Nusselt and Sherwood numbers that the criterion for
supersaturation and/or homogeneous nucleation to occur is:
Nux C i ⫺C B dT
⬍
Shx T i ⫺T B dC
冏 sat,i
(2)
In this paper, the SLSI is modeled as a constant temperature

vertical flat plate adjacent to a double-diffusive natural convection
boundary layer. Since saturation conditions are assumed at the
SLSI, its concentration is also constant. The boundary conditions
in the bulk solution are constant temperature and concentration
and it follows that the ratio Nux /Shx is a function of Le, Pr, and Ñ
关13兴. The dimensionless buoyancy parameter, Ñ, is, for a parcel of
fluid adjacent to the plate, simply the ratio of the concentration
and temperature induced buoyancy forces, i.e.,
␤ * 共 ␳ A,i ⫺ ␳ A,⬁ 兲
Ñ⫽ (3)
␤ 共 T i ⫺T ⬁ 兲
where ␳ A is salt concentration in kg of salt per m3 of solution. The
volumetric coefficient of thermal expansion 共␤兲 and the species
expansion coefficient 共␤*兲 are defined by
Fig. 3 Prandtl number of pure water versus temperature at a

pressure of 250 bar
␤ ⫽⫺
1 ⳵␳
␳ ⳵T
冏 P,C
(4)
␤ * ⫽⫺
1 ⳵␳
␳ ⳵␳ A
冏 P,T
(5) ␾ ␳A⫽
␳ A ⫺ ␳ A,⬁
␳ A,i ⫺ ␳ A,⬁
(16)
and a convenient form of the Grashof number is Since, for Le→⬁ and Pr⫽O共1兲, the mass transfer boundary layer
is thin when compared with either the momentum or the thermal
g ␤ 共 T i ⫺T ⬁ 兲 x 3 boundary layer thicknesses, the function f ( ␩ ) in Eq. 13 may be
Grx ⫽ (6) expanded in a Taylor series in which only the first nonzero term is
␯2
retained for the ‘‘inner’’ problem, i.e., within the mass transfer
For Leⲏ1, the ratio Shx /Nux scales as approximately the Lewis boundary layer. In terms of a new spatial variable ␨ ⫽ ␩ Sc1/3, the
number to the 1/3 power and, thus, an appropriate expression for similarity formulation for the ‘‘inner’’ problem becomes
it is
d3 f 1
Nux ⍀ ⫺ 关 f ⬙ 共 0 兲兴 2 Sc⫺5/3␨ 2 ⫹Sc⫺1 关 ␾ T ⫹Ñ ␾ ␳ A 兴 ⫽0 (17)
⫽ 1/3 , (7) d␨3 2
Shx Le
d 2 ␾ T 3 ⫺1 d␾T
where ⍀ is a weak function of Le, Pr, and Ñ. The critical Lewis ⫹ Le f ⬙ 共 0 兲 ␨ 2 ⫽0 (18)
d␨2 2 d␨
number beyond which supersaturation and/or homogeneous nucle-
ation occurs becomes: d 2␾ ␳A 3 d ␾ ␳A
冋冏册
2 ⫹ f ⬙共 0 兲 ␨ 2 ⫽0 (19)
⍀ 共 T i ⫺T B 兲 3 d␨ 2 d␨
Lec ⫽ (8)
dT Equation 17 shows that d 3 f /d ␨ 3 is of order Sc⫺1 for Ñ
共 C i ⫺C B 兲 ⱗO(1). It follows that the next term in the Taylor expansion for
dC sat,i
f 共in terms of ␩兲 is smaller by a factor of Sc1/3 and is therefore
When the solution Lewis number equals 1, ⍀ equals 1 for all Pr negligible in the limit as Sc→⬁. Equation 19 can therefore be
and Ñ because the analogy between heat and mass transfer ap- solved for ␾ ␳ A in terms of f ⬙ (0). Moreover, f ⬙ (0) can be ob-
plies. At Pr⫽0.7 and Ñ⫽0.5, which are characteristic of the afore- tained directly from the solution for the ‘‘outer’’ problem for
mentioned deposition experiments 关3兴, ⍀ equals 0.78 at Le⫽14.29 which ␨Ⰷ1 and ␩⫽O 关1兴. In this region, ␾ ␳ A ⬵0 so Eqs. 11 and
based on the results of Gebhart and Pera 关13兴. 共⍀ equals 0.86 for 12 become:
Pr⫽7.0, Ñ⫽0.5 and Le⫽14.29, i.e., it is in the same range even f ⵮ ⫹3 f f ⬙ ⫺2 f ⬘ 2 ⫹ ␾ T ⫽0 (20)
for the maximum Prandtl number of SCW at a pressure of 250
bar.兲 Based on the foregoing values, it is reasonable to assume that ␾ T⬙ ⫹3Prf ␾ T⬘ ⫽0 (21)
⍀ equals approximately 0.9 for 1ⱗLeⱗ15, 1ⱗPrⱗ10 and Ñ subject to the boundary conditions on f and ␾ T in Eq. 14. These
⯝0.5. However, for the deposition experiments, the critical value equations are the conventional ones which apply in the absence of
of the Lewis number may be very much greater than 15 as shown mass transfer and the tabulation of Gebhart 关14兴 provides values
below. Thus, an expression for ⍀ has been developed for the case for f ⬙ (0) as a function of Pr.
in which Le→⬁. To obtain the Sherwood number, Eq. 19 may be integrated once
Following Gebhart and Pera 关13兴, it is convenient to introduce a to give
similarity variable, ␩
␩⫽
y Grx
x 4 冉冊 1/4
(9)
d ␾ ␳A
d␨
1
2 冋
⫽K exp ⫺ Scf ⬙ 共 0 兲 ␨ 3 册 (22)
and then integrated again. Application of the boundary conditions

and to express the stream function in terms of it to the result of the second integration gives
冋冉冊册
⌿ 共 x,y 兲 ⫽ 4 ␯
Grx
4
1/4
f 共␩兲 (10) K 冕冋
⬁
0
exp ⫺
1
2 册
f ⬙ 共 0 兲 ␨ 3 d ␨ ⫽⫺1 (23)
The momentum, energy and species equations then become2 or

f ⵮ ⫹3 f f ⬙ ⫺2 f ⬘ ⫹ ␾ T ⫹Ñ ␾ ␳ A ⫽0
2
␾ T⬙ ⫹3Prf ␾ T⬘ ⫽0
(11)
(12)
K⫽
⌫ 共 1/3兲 2 冋册
⫺3 f ⬙ 共 0 兲 1/3
(24)
where the gamma function (⌫(z)) is defined as 共Abramowitz and

␾ ⬙␳ A ⫹3Scf ␾ ␳⬘ A ⫽0 (13) Stegun 关15兴兲
subject to
f 共 0 兲 ⫽ f ⬘ 共 0 兲 ⫽ f ⬘ 共 ⬁ 兲 ⫽1⫺ ␾ T 共 0 兲 ⫽1⫺ ␾ ␳ A 共 0 兲 ⫽ ␾ T 共 ⬁ 兲
⌫共 z 兲⫽ 冕0
⬁
t z⫺1 e ⫺t dt (25)
⫽ ␾ ␳ A 共 ⬁ 兲 ⫽0 (14) Finally,
where dimensionless temperature and concentration are defined as

T⫺T ⬁
h m ⫽⫺DAB
⳵ ␾ ␳A
⳵␨
冏 y⫽0
⳵␨
⳵y
(26)
␾ T⫽ (15)
T i ⫺T ⬁ or
Shx ⫽0.62849关 f ⬙ 共 0 兲兴 1/3Sc1/3Gr1/4
x (27)
2
At the conditions in the deposition experiments, scale analysis shows that the
Boussinesq approximation applies, the flow is quasi-steady, and the normal velocity The effect of Prandtl number on the Sherwood number is em-
at the SLSI is negligible 关3兴. All of these assumptions are made by Gebhart and Pera bedded in the constant f ⬙ (0). Gebhart and Pera 关13兴 provide
关13兴 to arrive at Eqs. 11 through 14. It is noted that the analysis by Gebhart and Pera
关13兴 was slightly more general. The approach adopted here corresponds to Gebhart f ⬙ (0) for Pr⫽0.7 and 7.0 as 0.67890 and 0.45069 respectively.
and Pera’s result for n⫽0, P⫽1, and Q⫽0. The resulting Sherwood numbers are
Table 1 Summary of ⍀ values and their ranges of applicability Table 2 Prediction of critical Lewis numbers at the beginning
of the deposition experiments and after 6 and 12 minutes of run
time. Relevant temperatures and concentrations at the begin-
ning of the experiments are also provided.
Shx ⫽0.536Le1/3Ra1/4
x for Pr⫽0.7, Le→⬁, Ñ⫽O关 1 兴
(28)
Shx ⫽0.567Le1/3Ra1/4
x for Pr⫽7.0, Le→⬁, Ñ⫽O关 1 兴 Application of Theory to Experiments
(29)
Based on the results of the preceding analyses, ⍀ is set equal to
Sherwood numbers at other Prandtl numbers may be computed 0.72 to provide a conservative estimate of the minimum critical
using the tabulation of f ⬙ (0) as a function of Prandtl number Lewis numbers in the deposition experiments run by Hodes et al.
available in Gebhart et al. 关14兴. When the Lewis number equals 关1兴. The calculations are conservative in that ⍀ is at least 0.72 in
one, the heat and mass transfer analogy applies and Shx ⫽Nux . At accordance with Table 1 and the critical Lewis number is propor-
Le⫽1 and Pr⫽Sc⫽0.7, for example, Shx ⫽Nux ⫽0.386Ra1/4 tional to ⍀ 3 . Based on the results of the Butenhoff et al. 关8兴 study
x 共Ge-
bhart and Pera 关13兴兲 and thus the constant in Eq. 28 共0.536兲 is 39 discussed above, Lewis numbers of aqueous sodium sulfate and
percent too high as a consequence of the Taylor series approxima- potassium sulfate solutions are likely about 5 at the experimental
tion for f ( ␩ ) in the ‘‘inner’’ region. Actually, a 39 percent error is conditions. Critical Lewis numbers at the beginning of the sodium
surprisingly small considering that Sc is only 0.7. Moreover, the sulfate and potassium sulfate deposition experiments and 6 and 12
results of the Gebhart and Pera 关13兴 study at, for example, Pr⫽7 minutes into each run are provided in Table 2. There is a modest
and Sc⫽500, may be used to validate Eq. 27 for high Sc numbers. cross flow of aqueous salt solution past the heated cylinder and
the ‘‘inlet’’ concentration of salt (C in ) is that of the solution en-
When Eq. 27 is used, the constant which multiplies Gr1/4 x is found tering the test cell. The values of C i , C B , and T i have been
to be high by 1.9 percent, low by 0.3 percent, and low by 3.8 extracted from the modeling results of Hodes 关3兴 and T B has been
percent for Ñ equal to 0.5, 1, and 2 respectively relative to the measured by Hodes et al. 关1兴. At the beginning of the sodium
results of Gebhart and Pera 关13兴 for Pr⫽7 and Sc⫽500. Values of sulfate deposition experiments, the critical Lewis number ranges
⍀ calculated from the foregoing results have been summarized in from 4.8 to 9.7. It is unlikely that the Lewis numbers of the
Table 1. The most striking observation which emerges from the aqueous sodium sulfate solutions substantially exceed 4.8, if at all.
comparison is the simple fact that ⍀ does not vary much. After 12 minutes of run time have elapsed, the critical Lewis
Given the value of K, Eq. 22 can also be integrated to provide numbers range from 18.0 to 30.5. It is highly unlikely that the
the solute concentration profile in the boundary layer. The result is Lewis numbers of the aqueous sodium sulfate solutions are as
冉冊
high as 18.0. Few, if any, particles were observed in the boundary
1 f ⬙ 共 0 兲 Sc␩ 3 layer formed around the heated cylinder during the sodium sulfate
␾ ␳ A ⫽1⫺ P , for Le→⬁ and Ñ⫽O关 1 兴 deposition experiments 共Hodes 关3兴兲 which is consistent with the
3 2
(30) critical Lewis numbers in Table 2. At the beginning of the potas-
sium sulfate deposition experiments, the critical Lewis numbers
where P(a,x) is the incomplete gamma function defined by range from 28 to 132. It is extremely unlikely that the Lewis
Abramowitz and Stegun 关15兴. Loci of temperature-concentration numbers of the aqueous potassium sulfate solutions exceed 28.
states in the boundary layer may be computed from this result, Thus the fact that particles were not observed during the potas-
together with those of Ostrach 关16兴 and others for ␾ T (T). Al- sium sulfate deposition experiments is consistent with the theory.
though the asymptotic expressions for Sherwood number 共Eq. 27兲 The critical Lewis numbers as a function of run time corre-
and ␾ ␳ A 共Eq. 30兲 were developed in the limit as Le→⬁, they also sponding to the experiments in which the sodium sulfate and po-
apply at physically realistic Le numbers. The form of the species tassium sulfate concentrations in the inlet stream were 4 wt per-
equation given by Eq. 19 could, following Lèvêque 关17兴, have cent are plotted in Fig. 5. After only thirty minutes the critical
been obtained by assuming a linear velocity profile in the concen- Lewis number has increased by factors of about 6 and 250 at
tration boundary layer. This ‘‘Lèvêque approach’’ suggests a pro- conditions corresponding to the sodium sulfate and potassium sul-
cedure by which the requirement that Sc1/3 be much greater than fate runs, respectively. The increase in the critical Lewis number
one may be made more quantitative. In particular, it is apparent for the potassium sulfate conditions over the same time interval is
that the method cannot be appropriate if the concentration bound- substantially larger because the thermophysical properties of the
ary layer extends beyond the point at which the velocity is a solutions are significantly different.3 Even if homogeneous nucle-
maximum. Thus, a rough criterion may be obtained by using Eq. ation in the boundary layer were to occur at the beginning of a
30 to determine the thickness of the concentration boundary layer sodium sulfate experiment, it is evident from Fig. 5 that it would
cease long before steady state conditions were reached. Homoge-
共by determining the value of Scf ⬙ (0) ␩ 3 at which the dimension-
neous nucleation could dramatically increase the time required to
less concentration is 0.01兲 and comparing the result with the value
establish steady state conditions, however. The very low Brown-
of ␩ at which the velocity is a maximum. This procedure requires
that Sc exceed 8 for Pr⫽0.7 and 25 for Pr⫽7.0. In each case, the 3
Both experiments were run at a pressure of 250 bar which is substantially above
value is rather modest. This is consistent with the observation the critical pressure of pure water 共221 bar兲. However, the temperature of the potas-
above that the Sherwood number expression given by Eq. 28 is sium sulfate solution around the heated cylinder is very close to the critical tempera-
‘‘surprisingly’’ accurate even at Pr⫽Sc⫽0.7. ture of pure water 共374°C兲 as per Table 2.
Fig. 5 Critical Lewis number as a function of time for the so- Fig. 7 Trajectories of „ T , C … states connecting bulk and SLSI
dium sulfate and potassium sulfate deposition experiments in conditions at the beginning of the potassium sulfate deposition
which the concentration of salt in the inlet stream was 4 wt experiments for which C in equals 4 wt percent. Unlabeled tra-
percent jectories correspond to LeÄ7.2, 15, 25, and 50 respectively be-
tween the LeÄ1 and LeÄ100 trajectories.
ian diffusion coefficients of particles would substantially retard

deposition until enough salt accrued on the heated cylinder for the
nucleation mechanism to be predominantly heterogeneous.
dT
冏⫽
⍀ T i ⫺T B
dC i Le1/3 C i ⫺C B
(31)
Trajectories of temperature-composition states which connect where ⍀ is set equal to 0.72.

the bulk solution condition to the SLSI condition when the inlet The four partial trajectories between Le⫽1 and Le⫽100 in Fig.
concentration of sodium sulfate or potassium sulfate in the inlet 6 correspond to Le⫽7.2, 15, 25, and 50 respectively. The critical
stream was 4 wt percent are plotted in Figs. 6 and 7 respectively. Lewis number at the beginning of the 共4 wt percent inlet concen-
The bulk and SLSI conditions shown correspond to the beginning tration兲 sodium sulfate deposition experiments equals 7.2 as indi-
of the experiments. When the Lewis number of the solution is 100 cated in the figure. It is likely that the Lewis number of the aque-
or larger, we assume that Eq. 30 applies and plot complete trajec- ous sodium sulfate solution is about equal to this critical Lewis
tories 共for Pr⫽0.72兲; otherwise, partial trajectories are plotted by number and this is consistent with few, and often zero, particles
expressing Eq. 7 as: being observed in the sodium sulfate deposition experiments. The
three partial trajectories shown in Fig. 7 for the potassium sulfate
experiments between Le⫽1 and Le⫽100 correspond to Le⫽15,
50.6, and 75 respectively. The critical Lewis number is 50.6. The
actual Lewis number of the aqueous potassium sulfate solution is
likely to be much less than this value and, thus, the fact that
particles were not observed during the potassium sulfate deposi-
tion runs is consistent with the nucleation theory. It is noted that
even at a Lewis number equal to 1000, less than 2°C of 共possible兲
supersaturation is predicted and therefore homogeneous nucle-
ation may not occur.
Candidates for Homogeneous Nucleation

Since high Lewis number aqueous salt solutions are conducive
to supersaturation and/or homogeneous nucleation in the bound-
ary layer, large molecular diameter/mass salts are prime candi-
dates for homogeneous nucleation. The uranyl fluoride-water
(UO2 F2 -H2 O) system, for example, is one worthy of consider-
ation. Uranyl fluoride is a rather large molecule with a molecular
mass of 308.02 gm/gmol as compared to 142.04 and 174.27 gm/
gmol for sodium sulfate and potassium sulfate respectively and all
three salts precipitate from solution at similar temperatures. Thus,
it is likely that the Lewis numbers of aqueous uranyl fluoride
solutions exceed those of aqueous sodium sulfate and potassium
Fig. 6 Trajectories of „ T , C … states connecting bulk and SLSI
sulfate solutions, making supersaturation and/or homogeneous
conditions at the beginning of the sodium sulfate deposition nucleation more likely as well. Marshall et al. 关18兴 determined the
experiments for which C in equals 4 wt percent. Unlabeled tra- temperature-composition equilibrium diagram for the
jectories correspond to LeÄ15, 50.6, and 75, respectively, be- UO2 F2 -H2 O system and Hodes 关3兴 outlines how nucleation ex-
tween the LeÄ1 and LeÄ100 trajectories. periments could be performed.
Fig. 8 Idealized phase boundary conducive to homogeneous Fig. 9 Lanthanum selenate solubility data „Friend †21‡… and
nucleation in the boundary layer and trajectories connecting trajectories connecting bulk and SLSI conditions at LeÄ1, Le
bulk and SLSI conditions for LeÄ1 and Le\ⴥ Ä60, and Le\ⴥ
Generally, Lewis numbers of binary gas mixtures are about two pected to become supersaturated and/or nucleate homogeneously
orders of magnitude smaller than those of binary liquid solutions. in the boundary layer. The measured bulk solution and cylinder
For example, Lewis numbers of mixtures of air with various spe- temperatures have an uncertainty of ⫾4.2°C and the uncertainty
cies are of order 1, while those of ordinary aqueous solutions are of the cell pressure measurement is ⫾0.15 MPa 共关3兴兲.
of order 100 共关13兴兲. Hence, salts which precipitate from aqueous An aqueous lithium carbonate 共1 wt percent兲 solution preheated
solutions at liquid-like densities are prime candidates to homoge- to 53°C was circulated through the flow system described by
neously nucleate in the boundary layer. The solubility of many Hurst et al. 关7兴 in which the test cell was visually accessible and
salts in water decreases substantially over temperature ranges be- illuminated by white light. 共The pressure in the experiment was
low 150°C, i.e., at liquid-like conditions. Some examples include 250 bar.兲 Power supplied to the heated cylinder within the bulk
sodium orthophosphate 关 Na3 PO4 兴 , magnesium sulfate 关 MgSO4 兴 , solution increased the measured surface temperature to about
lanthanum selenate 关 La2 (SeO4 ) 3 兴 and lithium carbonate 71°C and the SLSI condition which would be achieved if deposi-
关 Li2 CO3 兴共Linke 关19兴; Stephen and Stephen 关20兴兲. The latter two tion were to occur in the absence of homogeneous nucleation is
salts are subsequently discussed in some detail. In addition to the labeled ‘‘Theoretical Initial SLSI Condition’’ in Fig. 10. A trajec-
Lewis number of the aqueous salt solution, the phase boundary tory corresponding to Le⫽60 and Pr⫽2 which connects the initial
can also dramatically affect the probability that salt will homoge- bulk and theoretical initial SLSI conditions is also shown. 共Since
neously nucleate in the boundary layer. The shape of the hypo- lithium carbonate precipitates from solution at liquid-like densi-
thetical phase boundary shown in Fig. 8, for example, is ex- ties, a Lewis number of at least 60 is a reasonable expectation.兲 At
tremely conducive to homogeneous nucleation. Interestingly, the
shape of this idealized phase boundary is identical to that of the
Le→0 trajectory connecting bulk and SLSI conditions shown in
Fig. 1.
The solubility of lanthanum selenate in water was determined
in a study by Friend 关21兴 and is shown in Fig. 9. 共Pressures were
not reported.兲 The phase boundary mimics the idealized phase
boundary in Fig. 8 to a significant degree. Moreover, lanthanum
selenate is a very large molecule with a molecular mass of 706.69
gm/gmol. Trajectories which connect a bulk condition to a SLSI
condition corresponding to Le⫽1 and Le⫽60 are also shown in
Fig. 9. To compute the Le⫽60 trajectory, the solution Prandtl
number was set equal to 2 which is that of pure water in the
relevant temperature range. 共Thus Sc⫽120 and Eq. 30 is valid.兲
Even at a Lewis number of one, supersaturation and/or homoge-
neous nucleation is expected to occur. Given the very large size of
the lanthanum selenate molecule, a Lewis number of 60 or higher
is quite likely. At a Lewis number of 60, 20°C of supersaturation
would occur at the proposed conditions in the absence of homo-
geneous nucleation in the boundary layer. Thus lanthanum sel-
enate is a prime candidate to nucleate homogeneously in the
boundary layer formed around the SLSI.
Linke 关19兴 compiled the solubility data for lithium carbonate in
water shown in Fig. 10 from sources dating back to the late 19th Fig. 10 Lithium carbonate solubility data „Linke †19‡… and tra-
century. Based on these data, a nucleation experiment was con- jectory connecting bulk and SLSI conditions at LeÄ60. Solubil-
ducted in an effort to validate, at least for one salt, the theory ity boundary separates one phase fluid region from two phase,
developed in this paper at conditions where a salt would be ex- solid-fluid region.
Le⫽60, about 10°C of supersaturation is predicted to occur in the ␩ ⫽ dimensionless similarity parameter 共Eq. 9兲
absence of homogeneous nucleation. However, deposition and/or ␯ ⫽ kinematic viscosity 关m2/s兴
nucleation were not observed at these conditions during one hour ⍀ ⫽ Nux Le1/3/Shx
of elapsed time. Metastability within the aqueous lithium carbon- ␳ ⫽ density 关kg/m3兴
ate solution may explain why deposition and/or nucleation were ␳A ⫽ solute 共salt兲 concentration 关kg A/m3兴
not observed at these conditions. Thus, in an effort to overcome ␾ ␳A ⫽ dimensionless salt concentration
metastability in case it were present, the bulk temperature of the ␾T ⫽ dimensionless temperature
solution in the cell was gradually ramped upward at the rate ⌿ ⫽ stream function 关m2/s兴
of approximately 1°C/min. At a bulk temperature of about 99°C ␨ ⫽ spatial variable ( ␩ Sc1/3)
and heated cylinder surface temperature of about 117°C, deposi-
tion began. Moreover, bright bursts of light were observed around Subscripts
the heated cylinder and especially in the plume above it at a A ⫽ salt component of solution
frequency of about 0.5–3 hertz throughout the deposition process B ⫽ bulk
which was allowed to continue for about one hour. Very probably, c ⫽ critical
the bright bursts of light were reflected from lithium carbonate i ⫽ salt layer-solution interface 共SLSI兲
particles which homogeneously nucleated in the boundary layer. in ⫽ inlet to test cell
Nothing resembling these bright bursts of light was observed o ⫽ beginning of experiment
during the sodium sulfate and potassium sulfate deposition P ⫽ pressure
experiments. sat ⫽ saturation
References
关1兴 Hodes, M., Smith, K. A., Hurst, W. S., Bowers, Jr., W., Griffith, P., and Sako,
Conclusions K., 2002, ‘‘Solubilities and Deposition Rates in Aqueous Sulfate Solutions at
Elevated Temperatures and Pressure,’’ submitted to Int. J. Heat Mass Transfer.
A model has been developed to predict whether or not homo- 关2兴 Hodes, M., Smith, K. A., and Griffith, P., 2002, ‘‘A Natural Convection Model
geneous nucleation and/or supersaturation would occur in a natu- for the Rate of Salt Deposition from Near-Supercritical, Aqueous, Salt Solu-
tions,’’ submitted to J. Heat Transfer.
ral convection boundary layer around a cylinder heated beyond 关3兴 Hodes, M., 1998, ‘‘Measurements and Modeling of Deposition Rates from
the solubility temperature corresponding to the concentration of Near-Supercritical, Aqueous, Sodium Sulfate and Potassium Sulfate Solutions
salt in the surrounding aqueous salt solution. It was shown that the to a Heated Cylinder,’’ Ph.D. thesis, Massachusetts Institute of Technology,
Lewis number is the critical property in determining whether or Cambridge, MA.
关4兴 Tester, J. W., Holgate, H. R., Armellini, F. J., Webley, P. A., Killilea, W. R.,
not homogeneous nucleation is possible and a simple formula was Hong, G. T., and Barner, H. E., 1991, ‘‘Supercritical Water Oxidation Tech-
developed to predict the Lewis number below which homoge- nology: A Review of Process Development and Fundamental Research,’’ ACS
neous nucleation is not possible for a given solubility boundary Symp. Ser., 518.
and set of operating conditions. The model was applied to the 关5兴 Gloyna, E. F., and Li, L., 1998, ‘‘Supercritical Water Oxidation for Wastewater
and Sludge Remediation,’’ Encyclopedia of Environmental Analysis and Re-
conditions for which sodium sulfate and potassium sulfate depo- mediation, R. A. Meyers, ed., John Wiley and Sons, Inc., pp. 4780– 4797.
sition experiments were run by Hodes et al. 关1兴 and yielded con- 关6兴 Shaw, R. W., and Dahmen, N., ‘‘Destruction of Toxic Organic Materials Using
sistent results, i.e., homogeneous nucleation was predicted to be Super-Critical Water Oxidation: Current State of the Technology,’’ Supercriti-
absent. Attributes of aqueous salt solutions which increase the cal Fluids: Fundamentals and Applications, E. Kiran, P. G. Debenedetti and C.
J. Peters, eds., Kluwer Academic Publishers, Dordrecht, The Netherlands.
likelihood of homogeneous nucleation were identified. A lithium 关7兴 Hurst, W. A., Hodes, M. S., Bowers, Jr., W., Bean, V. E., Maslar, J. E., Smith,
carbonate deposition experiment was performed at conditions for K. A., and Griffith, P., 2002, ‘‘Optical Flow Cell and Apparatus for Solubility,
which homogeneous nucleation was expected and evidence Salt Deposition and Raman Spectroscopic Studies in Aqueous Solutions near
strongly suggests that it occurred. the Water Critical Point,’’ J. Supercrit. Fluids, 22共2兲, pp. 157–166.
关8兴 Butenhoff, T. J., Goemans, M., and Buelow, S. J., 1996, ‘‘Mass Diffusion
Coefficients and Thermal Diffusivity in Concentrated NaNO3 Solutions,’’ J.
Phys. Chem., 100, pp. 5982–5992.
关9兴 Lamb, W. J., Hoffman, G. A., and Jonas, J., 1981, ‘‘Self-Diffusion in Com-
Nomenclature pressed Super-Critical Water,’’ J. Chem. Phys., 74共12兲, pp. 6875– 6880.
关10兴 Fogler, H. S., 1992, Elements of Chemical Reaction Engineering, Prentice
C ⫽ salt concentration 关wt percent兴 Hall.
DAB ⫽ mass diffusion coefficient 关m2/s兴关11兴 Armellini, F. J., Tester, J. W., and Hong, G. T., 1994, ‘‘Precipitation of Sodium
f ⫽ dimensionless function of ␩ defined by stream function Chloride and Sodium Sulfate in Water from Sub- to Supercritical Conditions:
150 to 550°C, 100 to 300 bar,’’ J. Supercrit. Fluids, 7, pp. 147–158.
g ⫽ acceleration due to gravity 关m/s2兴关12兴 Harvey, A. H., Peskin, A. P., and Klein, S. A., 2000, ‘‘NIST/ASME Steam
Gr ⫽ Grashof number Properties: Version 2.2,’’ U.S. Department of Commerce.
hm ⫽ mass transfer coefficient 关m/s兴关13兴 Gebhart, B., and Pera, L., 1971, ‘‘The Nature of Vertical Natural Convection
Le ⫽ Lewis number Flows Resulting from the Combined Buoyancy Effects of Thermal and Mass
Diffusion,’’ Int. J. Heat Mass Transf., 14, pp. 2025–2050.
Ñ ⫽ dimensionless buoyancy parameter 关14兴 Gebhart, B., Jaluria, Y., Mahajan, R., and Sammakia, B., 1988, Buoyancy-
Nu ⫽ Nusselt number Induced Flows and Transport, Hemisphere Publishing Corporation.
Pr ⫽ Prandtl number 关15兴 Abramowitz, M., and Stegun, I. A., 1965, Handbook of Mathematical Func-
tions, Dover Publications.
Ra ⫽ Rayleigh number 关16兴 Ostrach, S., 1953, ‘‘An Analysis of Laminar Free-Convection Flow and Heat
Sc ⫽ Schmidt number Transfer about a Flat Plate Parallel to the Direction of the Generating Body
Sh ⫽ Sherwood number Force,’’ Technical Report 1111, NACA.
T ⫽ temperature 关°C兴关17兴 Lèvêque, J., 1928, Annales des Mines, 12共13兲, pp. 305 and 381.
关18兴 Marshall, W. L., Gill, J. S., and Secoy, C. H., 1954, ‘‘Phase Equilibria of
u ⫽ x-direction velocity 关m/s兴 Uranium Trioxide and Aqueous Hydrofluoric Acid in Stochiometric Concen-
v ⫽ y-direction velocity 关m/s兴 trations,’’ J. Am. Chem. Soc., p. 4279.
x ⫽ coordinate direction parallel to gravity 关19兴 Linke, W., 1958, Solubilities of Inorganic and Metal-Organic Compounds: A
y ⫽ coordinate direction perpendicular to gravity Compilation of Solubility Data from the Periodical Literature, D. Van Nos-
trand Company Inc.
Greek Symbols 关20兴 Stephen, H., and Stephen, T., 1963, Solubilities of Inorganic and Organic
Compounds: Volume 1-Binary Systems, Macmillan.
␤ ⫽ volumetric coefficient of thermal expansion 关1/°C兴关21兴 Friend, J. A. N., 1932, ‘‘The Selenates of Lanthanum and Their Solubilities in
␤* ⫽ species expansion coefficient 关m3/kg A兴 Water,’’ J. Chem. Soc., p. 1597.
Thermal Aspects in the
Continuous Chemical Vapor
Hoseon Yoo
Deposition of Silicon
Soongsil University, This paper deals with the continuous chemical vapor deposition of silicon in a horizontal
Seoul 156-743, Korea cold wall reactor, paying special attention to a moving susceptor. A two-dimensional
numerical model, which accounts for variable properties, thermal diffusion, radiative heat
Yogesh Jaluria exchange among surfaces, and conjugate heat transfer between the gas and susceptor, is
Fellow ASME developed and validated. Scale analysis for the susceptor energy balance enables the
e-mail: jaluria@jove.rutgers.edu identification of the characteristic parameters and the prediction of their qualitative ef-
Department of Mechanical fects before carrying out a detailed analysis. The results from the scale analysis are found
and Aerospace Engineering, to be consistent with the numerical predictions. The results show that the present continu-
Rutgers, The State University of New Jersey, ous system is characterized by two newly defined parameters: conductance ratio and
New Brunswick, NJ 08854 susceptor parameter. A pair of performance curves that relate the deposition efficiency to
each parameter successfully provide the susceptor-related design conditions. It is also
revealed that there exists an optimum length of heating zone that maximizes the deposition
efficiency. 关DOI: 10.1115/1.1482084兴
Keywords: Conjugate, Heat Transfer, Mass Transfer, Radiation, Thin Films, Vapor
Deposition
Introduction works have addressed experimental conditions, film properties,

conceptual designs, or equipment development, but have lacked
Chemical vapor deposition 共CVD兲 is an important method for
detailed descriptions of processes and transport phenomena occur-
manufacturing thin films. Since thin films prepared by CVD have
ring in the reactor, that are needed for appropriate understanding
many attractive features, such as large area coverage, good control
and modeling of continuous systems.
of film structure and composition, and conformal deposition, the
Recently, Chiu and Jaluria 关10,11兴 have proposed two types of
technology is being employed in a variety of areas. Engineering
continuous CVD systems based on a horizontal reactor, modeled
applications include the fabrication of microelectronic circuits,
deposition processes, and assessed their feasibility as alternatives
wear- and corrosion-resistant coatings, recording media, solar
to batch processing. One of these systems, where a finite thickness
cells, and optical devices. In view of ever increasing demands on
susceptor moves along the reactor bottom wall in a straight chan-
the deposition performance, one of the most important issues in
nel horizontal reactor 关11兴, is somewhat simpler and more flexible.
the future CVD technology is the productivity. The requirement of
Such a layout allows for uninterrupted operation, precluding the
higher throughputs necessitates new or refined designs in CVD
concern with downtime. Moreover, film uniformity is not of pri-
systems.
mary concern since the susceptor moving at a constant speed,
From the viewpoint of processing, most of the previous studies
under steady state, experiences the same amount of deposition
on CVD have dealt with batch systems, where a limited number of
over the entire reactor length. This system resembles practically
substrates or wafers placed on a stationary or rotating susceptor
used injector-based atmospheric CVD reactor 关12,13兴 in that they
are processed at one time. A number of existing reactors that have
commonly adopt a moving susceptor, on which wafers are con-
evolved to meet the requirements, especially deposition unifor-
tinuously and compactly placed, to achieve uniform deposition
mity, arising from various applications belong to the same cat-
and high throughput. Although some basic aspects have already
egory. Research efforts for modeling transport phenomena occur-
been presented, the continuous system needs to be revisited for
ring in these reactors and chemical reactions involved in
deposition processes have been reviewed by Jensen et al. 关1兴, better understanding of its nature and for obtaining the perfor-
Kleijn 关2兴, and Mahajan 关3兴. The productivity of batch processing mance characteristics applicable to design. In addition, the effects
is essentially limited by downtime associated with product load- of conjugate heat transfer and susceptor motion need to be quan-
ing, reactor startup, and shutdown. tified.
One of the ways to improve production throughput in CVD Motivated by such necessity, the present study considers the
systems is to adopt continuous processing. The feasibility and same continuous system as the previous work 关11兴. The prospect
utility of continuous processing have been exploited in a number of its adoption in the practical process seems to be encouraging
of industrial applications. Yamaji et al. 关4兴 reported that their con- since it is simpler in layout than the injector-based reactor men-
tinuous CVD siliconizing line succeeded in the manufacture of a tioned earlier. We investigate transport phenomena and chemical
silicon-coated steel sheet. Stevenson and Matthews 关5兴 suggested reactions accompanying the deposition process while paying spe-
design criteria for continuous processing of a plasma-assisted cial attention to the moving susceptor. The numerical model en-
physical vapor deposition 共PVD兲 equipment through cost analysis. compasses variable properties, thermal diffusion, radiative ex-
Onabe et al. 关6兴 prepared superconducting tapes using a continu- change among surfaces, and conjugate heat transfer between the
ous CVD technique. In addition, continuous CVD processing has gas phase and moving solid susceptor. A useful feature of this
been actively applied to coatings of glass 关7兴 and fiber 关8,9兴. These work lies in the use of scale analysis, as complementary to the
numerical simulation, in identifying the characteristic parameters
and analyzing the predicted results. In order to represent the per-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 17, formance of continuous CVD processing, a deposition efficiency
2001; revision received March 19, 2002. Associate Editor: C. Amon. is defined and interpreted. The characteristics and importance of
enough, we can employ a two-dimensional model. In describing
two-dimensional steady-state transport phenomena occurring in
the reactor, some commonly accepted assumptions have been in-
troduced 关1–3,14兴. The flow is taken as laminar with negligible
viscous dissipation effects. Note that the present operating condi-
tions correspond to the Reynolds number of 32 when based on the
inlet conditions and channel height. The gases are regarded as
ideal and non-participating for thermal radiation. Since silane is
highly diluted in a single carrier gas, the dilute mixture approxi-
mation can be invoked. It is a known fact that variable property
and Soret effects are important in cold wall horizontal CVD reac-
Fig. 1 Schematic of the present continuous chemical vapor tors 关18兴. Despite its negligible effect, the buoyancy term is re-
deposition system tained for the completeness of modeling.
In order to place the main focus on thermal characteristics as-
sociated with the susceptor motion, a simple, lumped overall sur-
newly defined parameters, as well as of performance curves ex- face reaction is adopted, precluding complicated gas phase reac-
pressed as a function of these parameters, are discussed in detail. tions. Although many studies aimed at developing CVD chemistry
The optimization of the process in terms of the heat supplied to models for silicon deposition from silane-hydrogen have been car-
the system is also addressed. ried out, there is no established model that is commonly accepted
in this area and many uncertainties are involved in using the dif-
Analysis ferent models 关14,18兴. In conjunction with the present susceptor
arrangement, two additional issues need to be considered. One is
Physical System. The continuous CVD system dealt with in that radiative loss from the susceptor surface is expected to be a
this study is depicted in Fig. 1. It is a horizontal reactor, where a dominant heat transport mode when a prescribed heat flux is im-
finite thickness susceptor is allowed to move along the bottom posed from the bottom. The other is that conjugate heat exchange
wall. The moving finite thickness susceptor models the combina- between the susceptor and gas phase may affect the surface tem-
tion of wafers 共or substrates兲 and a thin conveyor belt carrying perature at which chemical reaction takes place. For a proper pre-
compactly placed wafers on it. Continuous deposition processing diction of deposition processes, both of these should be incorpo-
is made possible by such susceptor motion. Silicon deposition rated in the model.
from silane 共SiH4) diluted in hydrogen 共H2) carrier gas is chosen, Owing to the dilute mixture approach, the governing equations
not only because there have been extensive studies on its chemis- and boundary conditions are substantially simplified. A single
try 关2,3,14 –16兴, but also because experimental data for model concentration equation is enough to describe mass transfer be-
validation are readily available 关17兴. cause silane is the only gas species involved in the reaction. Since
Gases flow in and out of the reactor at the atmospheric pressure general versions of the formulation are available elsewhere
P 0 . Note here that continuously moving susceptors are particu- 关1–3,14兴, specific equations applicable to the present system are
larly suited for atmospheric reactors due to easier sealing 关12,13兴. listed in the following:
The susceptor moves at a constant speed u s in the direction of
bulk gas flow. The bottom wall consists of three zones: entrance, ⳵共 ␳u 兲 ⳵共 ␳v 兲
⫹ ⫽0 (1)
heating, and outlet, the first and last of which are thermally insu- ⳵x ⳵y
冉冊冉冊
lated. A uniform heat flux is supplied to the susceptor through the
heating zone, enabling silane to react and deposit silicon onto the ⳵共 ␳u␾ 兲 ⳵共 ␳v␾ 兲 ⳵ ⳵␾ ⳵ ⳵␾
⫹ ⫽ ⌫ ⫹ ⌫ ⫹S ␾ (2)
surface. The upper reactor wall is composed of silica and mirror- ⳵x ⳵y ⳵x ⳵x ⳵y ⳵y
like coating on the outer surface, being radiatively opaque. In
where the general dependent variable ␾, diffusion coefficient ⌫,
order to prevent unwanted deposition, the upper wall is water
and source terms S ␾ are defined as
cooled to the ambient temperature. Due to conduction and heat
conveyed by the susceptor motion, chemical vapor deposition can ␾ ⫽u, v ,T, and ␻ (3)
take place all along the reactor length, which makes the length of
heating zone an important variable. For a given total heat input, as ⌫⫽ ␮ , ␮ ,k/c p , and ␳ D (4)
冉冊冉冊冋冉冊册
an example, changes in the length of the heating zone give rise to
different temperature profiles, thereby altering the overall growth ⳵P ⳵ ⳵u ⳵ ⳵v 2 ⳵ ⳵u ⳵v
S u ⫽⫺ ⫹ ␮ ⫹ ␮ ⫺ ␮ ⫹
rates. ⳵x ⳵x ⳵x ⳵y ⳵x 3 ⳵x ⳵x ⳵y
The present study focuses primarily on the moving susceptor. (5)
冉冊冉冊
All operating conditions and geometric factors, except those rel-
evant to the susceptor, are fixed at typical values 关11,17,18兴. The ⳵P ⳵ ⳵u ⳵ ⳵v
S v ⫽⫺ ⫹ ␮ ⫹ ␮
reactor channel is taken as 0.02 m high by 0.7 m long (H⫻L). ⳵y ⳵x ⳵y ⳵y ⳵y
冋冉冊册
Gas enters the reactor at the average velocity (u avg) of 0.175 m/s
under P 0 ⫽1 atm and T 0 ⫽300 K. The cold wall temperature (T 0 ) 2 ⳵ ⳵u ⳵v
⫺ ␮ ⫹ ⫺␳g (6)
is also set at 300 K. The partial pressure of silane ( P SiH4) at the 3 ⳵y ⳵x ⳵y
冉冊
inlet is taken as 124.1 Pa, from which the inlet mass fraction ␻ 0 is
k ⳵T ⳵cp ⳵T ⳵cp
calculated. A total heat input (Q t ) of 30 kW is chosen so that the S T⫽ ⫹ ⫹S r (7)
susceptor surface temperature attains the experimental conditions c 2p ⳵ x ⳵ x ⳵y ⳵y
冉冊冉冊
in the presence of radiation heat exchange 关17兴. Finally, the
lengths of entrance (L e ) and heating zones (L h ) are taken as 0.1 ⳵ ⳵ ln T ⳵ ⳵ ln T
S ␻⫽ DT ⫹ DT (8)
and 0.3 m, respectively, unless specified otherwise. ⳵x ⳵x ⳵y ⳵y
Mathematical Model. In horizontal CVD reactors two- While the flow and mass transfer equations are solved in the
dimensional modeling is valid if the channel width-to-height 共as- gas phase only (H s ⭐y⭐H s ⫹H), the energy equation applies to
pect兲 ratio is large and forced convection is dominant 关19兴. For the the combined susceptor-gas region (0⭐y⭐H s ⫹H) to incorporate
system under consideration, buoyancy effects were earlier shown conjugate heat transfer. The boundary conditions are specified so
to be negligible.关11兴 Assuming that the present reactor is wide as to cope with such situations. At the inlet, the flow is assumed to
be fully developed, giving rise to a parabolic profile. The suscep- r P SiH4
tor as well as the gas enter the reactor at the ambient temperature. R⫽ 关 mol⫺Si/m2s兴 (15)
The flow, temperature, and concentration are regarded as fully 1⫹A 1 P H2 ⫹A 2 P SiH4
developed at the exit, so that zero gradient conditions are applied where the rate constant is
there. At the susceptor surface, the gas obeys the dilute mixture
approximation, and a species balance on silane yields the mass r⫽1.25⫻109 e ⫺18500/T 关 mol⫺Si/atm m2s兴 (16)
transfer boundary condition. The upper wall is kept cold, and is
In most of stationary reactors, the deposition characteristics
regarded as impermeable. Those conditions are expressed as fol-
have been represented by the local film growth rate
lows:
G⫽ 共 M Si / ␳ Si兲 R (17)
u⫽ 共 6u avg⫺3u s 兲共 ỹ⫺ỹ 2 兲 ⫹u s 共 1⫺ỹ 兲 ; v ⫽0; T⫽T 0 ;
This quantity needs to be modified to characterize the perfor-
␻⫽␻0 at x⫽0 (9) mance in continuous processing. A dimensionless average growth
rate defined as
⳵u ⳵T ⳵␻
⫽0; v ⫽0; ⫽0; ⫽0 at x⫽L (10) ␳ SiḠL
⳵x ⳵x ⳵x ␩⫽ (18)
␳ 0 u avg␻ 0 H 共 M Si /M SiH4兲
⳵␻ ⳵ ln T
u⫽u s ; v ⫽0; ␳ D ⫹D T ⫽M SiH4R at y⫽H s can take the place of G. This parameter indicates the ratio of the
⳵y ⳵y
total deposition rate of silicon along the susceptor to the supply
(11a)
rate at the inlet. From its physical meaning, ␩ can be termed
⳵T deposition efficiency. The species conservation of silicon 共or si-
⫺k s ⫽q w in L e ⬍x⬍L e ⫹L h lane兲 allows us to evaluate it alternatively as 关17,23兴.
⳵y
H ⫹H
兰 Hs ␳ u ␻ 兩 x⫽L dy
⳵T s
and ⫽0 elsewhere at y⫽0 (11b) ␩ ⫽1⫺ (19)
⳵y ␳ 0 u avg␻ 0 H
Since the deposition efficiency includes reactor inlet conditions as
⳵␻ ⳵ lnT
u⫽0; v ⫽0; T⫽T 0 ; ␳ D ⫹D T ⫽0 at u⫽H s ⫹H well as the film growth rate, it can be used as a general perfor-
⳵y ⳵y mance index in CVD systems where film uniformity is not a ma-
(12) jor concern.
where ỹ⫽(y⫺H s )/H and q w ⫽Q t /L h .
Although several sophisticated models for radiation heat trans- Scale Analysis
fer in a horizontal CVD reactor have been proposed, e.g., 关20,21兴, It would be useful, especially in a complicated system, if we
a fairly simple one that retains only the basic feature is employed can identify relevant parameters and infer their qualitative effects
here. Assuming that the inlet and outlet openings are black sur- prior to the detailed study. To this end, a scale analysis for the
faces at the ambient temperature, the reactor can be treated as an susceptor energy balance is carried out. This analysis is based on
enclosure. The susceptor surface needs to be subdivided due to the the assumption that the deposition performance depends primarily
nonuniform temperature, whereas each of the other isothermal on the susceptor temperature, being independent of the species
surfaces is taken as a single area element. Let N be the number of concentration. Further assuming constant properties, the energy
subdivisions in the x-direction, then N⫹3 surface elements consti- equation for the moving susceptor can be rewritten as
冉冊
tute the enclosure. Since the gases are nonparticipating, the net-
radiation method 关22兴 for an enclosure can be used to obtain the ⳵Ts ⳵ 2T s ⳵ 2T s
net radiative loss from each surface, for which both the upper 共 ␳ c p 兲su s ⫽k s ⫹ (20)
⳵x ⳵x2 ⳵y2
reactor wall and the susceptor surface are assumed to be opaque
and gray. The energy balance for surface n in terms of radiosity is where the subscript s is used to denote the susceptor. Equation
共20兲 is scaled as
N⫹3
1⫺␧ n
q o,n ⫹
␧n 兺
j⫽1
F n⫺ j 共 q o,n ⫺q o, j 兲 ⫽ ␴ T 4n (13) Pe
共 ⌬T s 兲 x 共 ⌬T s 兲 x 共 ⌬T s 兲 y
L2
⬃
L2
⫹
H 2s
(21)
Once the radiosity q o,n is known, the net radiative loss q r,n is This relation has two limiting cases depending on the magnitude
calculated by of Pe. From the inlet and outlet boundary conditions, the horizon-
tal temperature difference can be estimated as (⌬T s ) x ⬃T s,max
1⫺␧ n ⫺T0. According to the basic assumptions mentioned here, it is
q r,n ⫽ 共 ␴ T 4n ⫺q o,n 兲 (14)
␧n deduced that
This loss is incorporated into the energy equation via the source k s,ref共 T s,max⫺T 0 兲
term as S r ⫽⫺q r ⫻共radiating area/volume兲 in every susceptor sub- ␩ ⬃ ␪ s⫽ (22)
q wL
division located along the surface. Of course, S r ⫽0 elsewhere.
where the subscript ref designates the reference conditions. the
Deposition Model. In the present system, there are large scale of vertical temperature difference can be expressed in terms
temperature variations along the susceptor surface due to uneven of the supplied heat flux s (⌬T s ) y ⬃q w H s /k s . Since the analysis
heating, motion, and inlet conditions. This means that the deposi- procedure is straightforward, only the final results are presented
tion process cannot be simply treated as diffusion-controlled. Al- here, i.e.,
though gas phase reactions were excluded, the silane concentra-
tion can be influenced by the surface boundary conditions. In view ␩ ⬃K ⫺1 for PeⰆ1 (23)
of these factors, it is necessary to adopt an overall surface reaction
that depends on the reactant concentration as well as the surface
␩ ⬃S ⫺1 for PeⰇ1 (24)
temperature. The mechanism proposed by Claasen et al. 关15兴 and
refined later 关16兴 seems to meet the need. Then, the reaction rate is New dimensionless parameters K and S are defined, respectively,
expressed as as
k sH s
K⫽ (25)
k s,refH s,ref
共 ␳ c p 兲su sH s
S⫽ (26)
k s,ref
In view of the fact that Eq. 共25兲 is associated with the longitu-
dinal conduction capability of a susceptor, K is termed conduc-
tance ratio. Although S physically represents the dimensionless
heat capacity rate of a moving susceptor, we simply call it sus-
ceptor parameter. Since the two parameters are defined in terms
of the susceptor thickness, material properties, and speed only, the
qualitative relations, Eqs. 共23兲 and 共24兲, remain valid regardless of
the change in heat transfer between the susceptor and gas. The
effect of conjugate heat transfer would appear through the quan-
titative dependence of ␩ on K and S.
Although the preceding results need confirmation, it is worth
noting their implications briefly at this stage. First of all, the
present continuous system may be characterized by the two pa-
rameters. The susceptor Peclet number, which appears in the scal- Fig. 2 Comparison of the predicted film growth rate with ex-
ing procedure, is irrelevant to the system performance. Next, in isting experimental data †17‡ and numerical results †11,18‡
each limiting case of Pe the deposition efficiency is expressed in
terms of a single parameter. Such a feature may be of practical
importance in data reduction. Finally, differently from Eq. 共23兲, Numerical Method. The governing equations and boundary
the deposition efficiency should approach a finite value as K ap- conditions are discretized on a nonuniform staggered grid system
proaches zero since ␩⬍1 by definition. using a finite volume method. In correcting the pressure field, an
Properties. Using the dilute mixture approximation, it is approach similar to the SIMPLE algorithm 关28兴 has been em-
easy to evaluate both thermophysical and transport properties of ployed. As noted earlier, the same energy equation covers the
gases. Since silane is present in small fractions, properties of the combined susceptor-gas region along with individual property val-
mixture are taken as those of the hydrogen carrier gas. The prop- ues, yielding the conjugate temperature field. Since the species
erties are allowed to vary with temperature as 关11,18,24,25兴 conservation equation is decoupled from the others under the di-
lute mixture approximation, it is solved after the flow and tem-
␳ ⫽ P 0 M H2 / 共 RT 兲 (27) perature fields have converged.
A typical grid system consists of 450 streamwise and 40 trans-
c p ⫽1.44⫻104 ⫺2.61⫻10⫺1 T⫹8.67⫻10⫺4 T 2 (28) verse nodes in the gas phase. The number of transverse nodes in
the susceptor material depends on the thickness considered. For a
␮ ⫽ ␮ 0 共 T/T 0 兲 0.648 (29) typical case of H s ⫽2 mm, 10 nodes are used. The grid size was
k⫽k 0 共 T/T 0 兲 0.691
(30) chosen by appropriate grid refinement so that the converged solu-
tions are independent of the grid. It is also confirmed that the
⫺6 ⫺1
where ␮ 0 ⫽8.96⫻10 kg/ms and k 0 ⫽1.83⫻10 W/mk. The numerical results are independent of other user-specified variables
power-law dependence is also used for the binary diffusion coef- such as the initial guess and convergence criteria.
ficient of SiH4 –H2 mixture
D⫽D 0 共 T/T 0 兲 1.70 (31)
A large number of simulations have been performed to investi-
where D 0 ⫽6.24⫻10⫺5 m2/s. For a dilute mixture, the thermal gate the influences of susceptor-related factors. The simulated
diffusion coefficient can be expressed as cases are divided into four groups according to the susceptor ma-
terial and thickness. Case A just refers to the reference conditions.
D T ⫽ ␳ Da ␻ (32)
Each of Cases B, C, and D designates the change in k s , H s , and
The factor a is estimated using the Holstein’s approximation 关26兴 both of k s and H s , respectively, while the other variables remain
and curve-fitted Lennard-Jones parameters 关27兴. at the reference values.
The density of silicon is taken as ␳ Si⫽2330 kg/m3. In response Model Validation. In order to validate the present numerical
to the result of scale analysis, a set of reference conditions for the model, two sets of the predicted local growth rate of silicon film
susceptor are chosen based on the previous work 关11兴 as: H s,ref are compared in Fig. 2 with the well-known experimental data of
⫽2 mm, ( ␳ c p ) s,ref⫽1.631⫻106 J/m3K, and k s,ref⫽141.2 W/mK. Eversteyn et al. 关17兴. For comparative discussions, the plot also
Noting that the actual susceptor is a composite element 共of con- includes two representative numerical results 关11,18兴 that com-
veyor belt and silicon wafer兲, its properties do not necessarily monly employed two-dimensional models. A stationary, isother-
correspond to a specific material. The susceptor properties may mal susceptor at 1323 K is used for this comparison to replicate
have wide spectra and vary diversely with temperature depending the experimental conditions. The susceptor in Fig. 2 corresponds
on the material combination. Their effects on the deposition pro- to the heating zone in Fig. 1. Hence, x ⬘ ⫽x⫺L e . One of the
cess can be assessed by a parametric study as is done in this work. present predictions, the solid curve, is referred to as original be-
Those values listed here were taken for silicon at the ambient cause it is obtained from the model without any change. The other
temperature as an approximation. Other numerical data used here is a reference case for later discussion.
include well-known universal constants and molecular masses, Except near the leading edge, the present simulation favorably
which can be found elsewhere. In view of the previous work 关24兴, 共within 20 percent bound兲 agrees with the experimental data. The
emissivities of the upper wall and susceptor surfaces are taken as leading edge discrepancy seems to originate not from improper
0.7 and 0.6, respectively. For the same reason as mentioned ear- modeling but from mismatch in the boundary conditions between
lier, the temperature dependence of emissivity is not taken into the prediction and the experiment. A steep drop in the growth rate
account. always appears in numerical predictions employing an isothermal
Fig. 4 A typical result of the conjugate heat transfer analysis:
Fig. 3 Effect of the susceptor surface emissivity on the tem-
heat transport modes for a susceptor segment
perature distribution
susceptor because silane is depleted sharply from the inlet condi- is needed to maintain a prescribed susceptor temperature. In all
tion to a very low level there. The other numerical results confirm the subsequent simulations, the susceptor surface emissivity is set
this aspect. In the experiment, the surface temperature at the lead- at the previously noted value of 0.6.
ing and trailing edges would be lower than specified, deviating The role of radiation can also be understood in Fig. 4, where a
from the isothermal condition. A gradual change at the leading typical result of the conjugate heat transfer analysis is depicted.
edge and inflective decrease at the trailing edge in the experimen- This is the energy balance for a susceptor segment shown in the
tal growth rate support this arguement. inset of Fig. 4. Four curves represent net conduction (Q k,x⫹⌬x
Regardless of models, all the numerical predictions show the ⫺Q k,x ), convection (Q c ), net radiation (Q r ), and heat input
same qualitative trend over the susceptor length. In terms of quan- (Q w ), respectively. The heat input is q w ⌬x for a segment on the
titative agreement with the experiment 关17兴, the original curve of heating zone and 0 elsewhere. It is confirmed that radiation is the
the present prediction appears to be inferior to the result of Ma- dominant heat transfer mode along the susceptor in the heating
hajan and Wei 关18兴, whereas superior to that of Chiu and Jaluria zone. Two discontinuities in the net conduction curve, which ap-
关11兴. This observation, however, needs some explanations. Ma- pear at the borders of the heating zone, correspond to each of
hajan and Wei 关18兴 used a larger diffusion coefficient than Eq. inflexion points in the temperature distribution 共see Fig. 3兲. Heat
共31兲 and other studies 关2,3,24兴, and assumed the diffusion- conduction through a thin susceptor contributes to smoothen the
controlled deposition. Both of these effects are known to increase horizontal temperature variation. Although convection from the
the local growth rate. Using their diffusion coefficient in the susceptor surface to the gas is much smaller than radiation, it may
present model indeed yielded considerably higher growth rates, affect the deposition performance when gas phase reactions are
which are depicted as the reference case in Fig. 2. It is also known included in the model.
关11兴 that the diffusion-controlled assumption leads to faster film Effect of Conductance Ratio. The scale analysis yielded a
growth than the present reaction kinetics, Eq. 共15兲. Note that the relation between the deposition efficiency and conductance ratio
infinite reaction rate is just a limiting case. These facts imply that for small Peclet numbers. Its validity and utility is examined for
under proper conditions, the predictions might deviate farther S⫽0 because a stationary susceptor best satisfies the limiting con-
from the experiment than seen here. Meanwhile, Chiu and Jaluria dition of PeⰆ1. The local growth rates for two cases with the
关11兴 neglected thermal diffusion that acts as obstructing film depo- same value of K, but obtained from different combinations of k s
sition in silane-hydrogen system. The inclusion of thermal diffu- and H s , are compared in Fig. 5. For each of K⫽1/4 and 4, Case
sion in their model may produce lower growth rates than the plot- B and C yield an identical curve despite the physical difference
ted result. It is deduced from this discussion that the present between them. Such indiscernible agreements appear in all other
model is capable of resolving the basic characteristics of silicon
results, which confirm that the conductance ratio is indeed a char-
deposition occurring in a cold wall horizontal CVD reactor within
acteristic parameter. On the other hand, the difference in pattern
a reasonable tolerance.
among cures that correspond to three values of K indicates that the
Radiation and Conjugate Heat Transfer. In view of their conductance ratio may affect the overall deposition behavior.
contributions, radiation and conjugate heat transfer in the present Figure 6 delineates the stationary deposition efficiency calcu-
continuous system are addressed first. Results are obtained for the lated over a wide range of K and for as many combinations of k s
case of a stationary susceptor. Figure 3 demonstrates the effect of and H s as possible. Cases B, C, and D with the same K are
emissivity on the temperature distribution along the susceptor sur- reduced to a single value of ␩ st , within a bound of numerical
face. The shaded areas in Fig. 3 and plots presented from now on truncation. Although the ␩ st curve does not fit the scale relation,
indicate the part of susceptor that overlies the heating zone. Over Eq. 共23兲, the two results show a similar trend in that they decrease
a wide band of the susceptor surface emissivity, the curves show a toward zero as K increases. At small values of K, however, the
similar pattern, but temperature differences among them along the curve deviates far from the relation, as noted before. With de-
heating zone are quite large. Noting that the surface temperature creasing the conductance ratio, the curve approaches asymptoti-
directly affects the deposition rates, the large differences suffice to cally to the deposition efficiency for K→0. Noting that that varia-
illustrate the importance of radiation. The radiation model can tion of ␩ st is steep in a narrow band of K 共e.g., 0.1⬍ ␩ st ⬍0.5 for
also be used in reactor design to calculate the total heat input that 1⬍K⬍10兲, the overall deposition performance may depend
Fig. 5 Spatial variations of the film growth rate for different Fig. 7 Dependence of the deposition efficiency on the suscep-
values and combinations of conductance ratio tor speed for different values and combinations of conduc-
tance ratio
strongly on the selection of susceptor. From Figs. 5 and 6, along

with Eq. 共23兲, it is evident that the conductance ratio is the only Eqs. 共24兲 and 共26兲, which enable us to deduce that the deposition
parameter representing a stationary system. efficiency can be characterized by the susceptor parameter only.
Although not shown here, the susceptor temperature over the When the speed u s is converted into the susceptor parameter S
heating zone, which primarily affects the deposition rate, de- according to the definition given in Eq. 共26兲, the efficiency curves
creases with increasing K, leading to the pattern of Fig. 6. The with the same value of K nearly coincide with each other. The
␩ st ⫺K plot can be used either to estimate the deposition effi- curves for different values of K disagree with one another due to
ciency for a given susceptor or to select the susceptor material and the difference in ␩ st . This problem can be resolved by normaliz-
thickness that meet a prescribed performance criterion. ing ␩ by ␩ st . Figure 8 depicts the reduced result, i.e., the relation
between the normalized efficiency and susceptor parameter, for
Susceptor Speed. In the present continuous system, the sus- which the data are taken from Fig. 7. Although each set of data
ceptor speed is of fundamental importance. The deposition effi- differs a little from one another, the deviations seem to be small
ciency is readily expected to decrease as the speed increases. In enough to call the result a single curve. Additional calculations for
order to quantify the effect of susceptor speed, ␩ is plotted as a the wide range of K 共as in Fig. 6兲 and various combinations of
function of u s in Fig. 7 for the same values of K and combinations ( ␳ c p ) s u s yield the same curve. A decrease in ␩ / ␩ st over the re-
of k s and H s as in Fig. 5. Regardless of the value of K, all curves gion of large S, where PeⰆ1, qualitatively agrees with the scale
show a similar pattern. As u s increases, ␩ is at first constant, then relation, Eq. 共24兲. From an engineering viewpoint, this curve
decreases gradually, and finally decreases sharply, approaching along with Fig. 6 can be used for susceptor design. Under the
smoothly to zero. In the region of large u s , on the other hand, the present conditions, the deposition efficiency is readily obtained
behavior of ␩ is affected by both the value of K and the combi- for any combinations of susceptor material, thickness, and speed
nation of properties used to obtain a given value. The relation through a simple procedure. Once K and S are known, ␩ st and
between ␩ and u s is independent of k s , but dependent on H s . ␩ / ␩ st can be read in Figs. 6 and 8, respectively. Multiplying them
Such dependence is consistent with the results of scale analysis, produces the required ␩. Therefore, with these plots, we can avoid
Fig. 6 Dependence of the stationary deposition efficiency on Fig. 8 Dependence of the normalized deposition efficiency on
conductance ratio susceptor parameter
Fig. 9 Effect of susceptor parameter on the temperature dis- Fig. 11 Deposition efficiency as a function of heating zone
tribution along the susceptor surface length at four selected values of susceptor parameter
design conditions that result in low efficiencies. In this sense, the tive loss from the surface. Although the temperature is still high,
two curves comprise performance curves for the present continu- the average deposition rate becomes low because of a short depo-
ous system. sition length. When L h is long, the susceptor temperature becomes
The effect of susceptor speed on the local behavior is briefly low, thereby heat losses being reduced. Such moderate tempera-
reported here. Figure 9 presents the temperature distributions ture is ineffective for chemical reaction, decreasing the deposition
along the susceptor surface for five values of S. With increasing S efficiency. The optimum length L h,opt shortens as K increases.
共or equivalent u s 兲, a high temperature band narrows, the maxi- Conduction through the susceptor is enhanced with increasing K,
mum value lowers and shifts downstream toward the end of heat- so that intensive heating is more effective for raising the surface
ing zone, and the profile eventually becomes linear. Heat con- temperature high enough for deposition. If conduction is sup-
veyed by the susceptor movement is responsible for those pressed with decreasing K, extensive heating yields a longer ef-
changes, which in turn leads to the trends seen in Fig. 8. It is fective deposition length. A short L h along with small K leads to a
unlikely that a value of S higher than 0.1 will be adopted in the large radiative loss. This is why ␩ st for a small K decreases more
design of practical systems. sharply from the maximum on the short L h side than on the large
Length of Heating Zone. For a given total heat input, the one.
length of heating zone directly affects the temperature profile due Another result is the dependence of ␩ on L h at represent-
to q w ⫽Q t /L h . The effect on the deposition performance for rep- ative values of S for K⫽1, as depicted in Fig. 11. The overall
resentative values of K and S is investigated. Figure 10 illustrates picture looks similar to Fig. 10. Each curve shows an increasing-
the dependence of ␩ st on L h at four selected values of K for S decreasing pattern, encompassing the local maximum point. The
⫽0. For each K, there exists an optimum length of heating zone optimum length of the heating zone also shortens as S increases.
that maximizes the deposition efficiency. The reason is clear when The influence of S on L h,opt , however, is caused by a different
we consider two limiting cases: L h →0 and L h →L. When L h is mechanism from that of K. With increasing S, the width of the
short, the susceptor temperature on the heating zone tends to rise susceptor at high temperature narrows and the surface temperature
high due to intensive heating, which in turn induces a large radia- decreases 共see Fig. 9兲. This means that more intensive heating is
needed to raise the temperature high enough for deposition. It is
worth noting that the variation band of L h,opt in both of Figs. 10
and 11 is relatively narrow. The value L h ⫽0.3 m used so far is
close to L h,opt for a wide range of K and S. Because of the convex
shape of the efficiency curve, a single value of ␩, other than at the
maximum, corresponds to two different lengths: one is shorter,
and the other longer than L h,opt . The shorter one is preferable
because a smaller reactor is desirable. Other factors, such as mi-
crostructure of deposited film, conveyor drive system, and wafer
loading, should be taken into account in the actual design.
In summary, the dependence of deposition performance on the
length of heating zone stems from how the supplied heat is dis-
tributed along the susceptor and transported to the surroundings.
Since the parameters K and S affect the distribution and transport,
the efficiency curve depends on both of them. It is obvious that
the radiation and conjugate heat transfer play key roles in quanti-
fying such effects in deposition characteristics.
Conclusions
A systematic study has been carried out to investigate the per-
formance characteristics of continuous chemical vapor deposition
Fig. 10 Stationary deposition efficiency as a function of heat- of silicon in a horizontal cold wall reactor, in which the main
ing zone length at four selected values of conductance ratio focus is placed on a moving finite thickness susceptor. The nu-
merical model used for simulation accounts for temperature- T ⫽ temperature
dependent properties, thermal diffusion, radiation among surfaces, u,v ⫽ velocity components in x, y-directions
and conjugate heat transfer between the susceptor and gas. The u avg ⫽ average gas velocity at the inlet
validity and capability of the model have been confirmed by a x, y ⫽ coordinates
reasonably good agreement between the predicted result and Greek Letters
available experimental data, and a fine resolution of the coupled
heat transfer. Scale analysis for the moving susceptor yield two ␧ ⫽ emissivity
characteristic parameters and a pair of qualitative relations, which ␾ ⫽ stands for u, v, T, and ␻
prove to be useful for understanding the deposition behavior and ␩ ⫽ deposition efficiency
analyzing the predicted data. ␮ ⫽ viscosity
When the susceptor is motionless, the stationary deposition ef- ␳ ⫽ density
ficiency is determined by the conductance ratio only. The effi- ␴ ⫽ Stefan-Boltzmann constant
ciency curve monotonically decreases that for the thin susceptor ␪ ⫽ dimensionless temperature
limit to zero as the conductance ratio increases. When the suscep- ␻ ⫽ mass fraction of silane
tor is in motion, the deposition efficiency, normalized by the value Subscripts
under stationary state conditions, is a function of the susceptor
parameter only. With increasing the susceptor parameter 共or 0 ⫽ inlet or reference state
speed兲, the normalized efficiency curve remains almost unity and c ⫽ convection
then decreases sharply toward zero. This variation demonstrates k ⫽ conduction
the feasibility of continuous processing in that the susceptor can r ⫽ radiation
move up to considerable speed without sacrificing deposition ef- ref ⫽ reference conditions
ficiency. Since these two curves represent the performance char- ss ⫽ susceptor surface
acteristics of a continuous system, they can be used to determine st ⫽ stationary susceptor
the susceptor material, thickness, or moving speed in the design of t ⫽ total
practical systems. Finally, an optimum length of heating zone that w ⫽ bottom wall
maximizes the deposition efficiency always exists under a fixed
total heat input and depends on the characteristic parameters. Ad- References
justing the length of the heating zone, along with the use of the 关1兴 Jensen, K. F., Einset, E. O., and Fotiadis, D. I., 1991, ‘‘Flow Phenomena in
performance curves, can play an important role in improving the Chemical Vapor Deposition of Thin Films,’’ Annu. Rev. Fluid Mech., 23, pp.
deposition efficiency of continuous CVD systems. 197–232.
关2兴 Kleijn, C. R., 1995, ‘‘Chemical Vapor Deposition Processes,’’ Computational
Modeling in Semiconductor Processing, M. Meyyappan, ed., Artech House,
Acknowledgments Boston, pp. 97–229.
关3兴 Mahajan, R. L., 1996, ‘‘Transport Phenomena in Chemical Vapor-Deposition
This work was conducted during the first author’s 共H.Y.兲 sab- Systems,’’ Adv. Heat Transfer, 28, pp. 339– 425.
batical leave. He would like to acknowledge the financial support 关4兴 Yamaji, T., Abe, M., Takada, Y., Okada, K., and Hiratani, T., 1994, ‘‘Magnetic
provided by the Korea Research Foundation under Grant Number Properties and Workability of 6.5% Silicon Steel Sheet Manufactured in Con-
tinuous CVD Siliconizing Line,’’ J. Magn. Magn. Mater., 133, pp. 187–189.
KRF-2000-EA0005 and the visiting position furnished by Rutgers 关5兴 Stevenson, P., and Matthews, A., 1995, ‘‘PVD Equipment Design: Concepts
University. The partial support from NSF under Grant Number for Increased Production Throughput,’’ Surf. Coat. Technol., 74–75, pp. 770–
DMI-96-33194 is also acknowledged. 780.
关6兴 Onabe, K., Kohno, O., Nagaya, S., Shimonosono, T., Iijima, Y., Sadakata, N.,
and Saito, T., 1996, ‘‘Structure and Property of YBa2Cu3Ox Tapes Formed on
Nomenclature Metallic Substrate by CVD Technique,’’ Mater. Trans., JIM, 37, pp. 893– 897.
关7兴 Gordon, R., 1997, ‘‘Chemical Vapor Deposition of Coatings on Glass,’’ J.
A1 ⫽ constant in Eq. 共15兲, A 1 ⫽1.75⫻103 atm⫺1 Non-Cryst. Solids, 218, pp. 81–91.
A2 ⫽ constant in Eq. 共15兲, A 2 ⫽4.00⫻104 atm⫺1 关8兴 Bansal, N., and Dickerson, R. M., 1997, ‘‘Tensile Strength and Microstructural
a ⫽ thermal diffusion factor Characterization of HPZ Ceramic Fiber,’’ Mat. Sci. Technol., A222, pp. 149–
157.
cp ⫽ specific heat 关9兴 Lackey, W. J., Vaidyaraman, S., Beckloff, B. N., Moss, III, T. S., and Lewis, J.
D ⫽ binary diffusion coefficient S., 1998, ‘‘Mass Transfer and Kinetics of the Chemical Vapor Deposition of
DT ⫽ thermal diffusion coefficient SiC onto Fiber,’’ J. Mater. Res., 13, pp. 2251–2261.
关10兴 Chiu, W. K. S., and Jaluria, Y., 1999, ‘‘Effect of Buoyancy, Susceptor Motion,
F ⫽ configuration factor and Conjugate Transport in Chemical Vapor Deposition Systems,’’ ASME
G ⫽ local film growth rate Heat Transfer, 121, pp. 757–761.
Ḡ ⫽ 1
average growth rate, L 兰 L0 G(x)dx 关11兴 Chiu, W. K. S., and Jaluria, Y., 2000, ‘‘Continuous Chemical Vapor Deposition
Processing with a Moving Finite Thickness Susceptor,’’ J. Mater. Res., 15, pp.
g ⫽ magnitude of gravitational acceleration 317–328.
H ⫽ height of reactor 关12兴 Dobkin, D. M., 2000, ‘‘Injector-based Atmospheric Pressure Reactors,’’ http://
Hs ⫽ susceptor thickness www.batnet.com/enigmatics/semiconductor_processing/CVD_Fundamentals/
K ⫽ conductance ratio reactors/APCVD_ovvu.html.
关13兴 Silicon Valley Group, Inc., 2000, Annual Report, Securities Exch. Comm.,
k ⫽ thermal conductivity http://www.asml.com/investors/pdf/annual_report_svg00.pdf, p. 7.
L ⫽ reacator length 关14兴 Kleijn, C. R., 2000, ‘‘Computational Modeling of Transport Phenomena and
Le , Lh , Lo⫽ length of entrance, heating, and outlet zones Detailed Chemistry in Chemical Vapor Deposition—A Benchmark Solution,’’
M ⫽ molecular mass Thin Solid Films, 365, pp. 294 –306.
关15兴 Claasen, W. A. P., Bloem, J., Valkenburg, W. G. J. N., and van den Brekel, C.
N ⫽ number of control volumes in the x-direction H. J., 1982, ‘‘The Deposition of Silicon from Silane in a Low-Pressure, Hot-
P ⫽ pressure Wall System,’’ J. Cryst. Growth, 57, pp. 259–266.
Pi ⫽ partial pressure of species i 关16兴 Jensen, K. F., and Graves, D. B., 1983, ‘‘Modeling and Analysis of Low
Pe ⫽ Peclet number, ( ␳ c p ) s u s L/k s Pressure CVD Reactors,’’ J. Electrochem. Soc., 120, pp. 1950–1957.
关17兴 Eversteyn, F. C., Severin, P. J. W., van den Brekel, C. H. J., and Peek, H. L.,
Q ⫽ heat transfer rate 1970, ‘‘A Stagnant Layer Model for the Epitaxial Growth of Silicon from
q ⫽ heat flux Silane on a Horizontal Reactor,’’ J. Electrochem. Soc., 117, pp. 925–931.
qo ⫽ radiosity 关18兴 Mahajan, R. L., and Wei, C., 1991, ‘‘Buoyancy, Soret, Dufour, and Variable
R ⫽ universal gas constant Property Effects in Silicon Epitaxy,’’ ASME Heat Transfer, 113, pp. 688 – 695.
关19兴 Quazzani, J., Chiu, K.-C.,and Rosenberger, F., 1988, ‘‘On the 2D Modeling of
R ⫽ chemical reaction rate Horizontal CVD Reactors and its Limitations,’’ J. Cryst. Growth, 91, pp. 497–
r ⫽ reaction rate constant 508.
S ⫽ susceptor parameter or source term 关20兴 Chinoy, P. B., Kaminski, D. A., and Ghandhi, S. K., 1991, ‘‘Effects of Thermal
Radiation on Momentum, Heat, and Mass Transfer in a Horizontal Chemical Phenomena in Horizontal Chemical Vapor Deposition Reactors,’’ Chem. Eng.
Vapor Deposition Reactor,’’ Numer. Heat Transfer, Part A, 19, pp. 85–100. Sci., 46, pp. 321–334.
关21兴 Kadinski, L., Makarov, Yu N., Schaefer, M., Vasil’ev, M. G., and Yuferev, V. 关25兴 Moffat, H. K., and Jensen, K. F., 1988, ‘‘Three-Dimensional Flow Effects in
S., 1995, ‘‘Development of Advanced Mathematical Models for Numerical Silicon CVD in Horizontal Reactors,’’ J. Electrochem. Soc., 135, pp. 459– 471.
Calculations of Radiative Heat Transfer in Metalorganic Chemical Vapor 关26兴 Holstein, W. L., 1988, ‘‘Thermal Diffusion in Metal-Organic Chemical Vapor
Deposition Reactors,’’ J. Cryst. Growth, 146, pp. 209–213. Deposition,’’ J. Electrochem. Soc., 135, pp. 1788 –1793.
关22兴 Siegel, R., and Howell, J. R., 1992, Thermal Radiation Heat Transfer, 3rd ed.,
关27兴 Neufeld, P., Janzen, A. R., and Aziz, R. A., 1972, ‘‘Empirical Equations to
Hemisphere Publishing Corp., Washington, D.C.
关23兴 Tsai, H. C., Greif, R., and Joh, S., 1995, ‘‘A Study of Thermophoretic Trans- Calculate 16 of the Transport Collision Integrals ⍀ (l,s) * for the Lennard-Jones
port in a Reacting Flow with Application to External Chemical Vapor Depo- 共12-6兲 Potential,’’ J. Chem. Phys., 57, pp. 1100–1102.
sition Processes,’’ Int. J. Heat Mass Transf., 38, pp. 1901–1910. 关28兴 Patankar, S. V., 1980, Numerical Heat Transfer and Fluid Flow, Hemisphere
关24兴 Kleijn, C. R., and Hoogendoorn, C. J., 1991, ‘‘A Study of 2- and 3-D Transport Publishing Corp., New York.
Direct Chill Casting of Aluminum
Christopher J. Vreeman
Capstone Turbine Corporation,
21211 Nordhoff Street,
Alloys: Modeling and Experiments
Chatsworth, CA 91311
J. David Schloz
on Industrial Scale Ingots
Wagstaff Engineering, A continuum mixture model of the direct chill casting process is compared to experimental
3910 North Flora Road, results from industrial scale aluminum billets. The model, which includes the transport of
Spokane, WA 99216 free-floating solid particles, can simulate the effect of a grain refiner on macrosegregation
and fluid flow. It is applied to an Al-6 wt% Cu alloy and the effect of grain refiner on
Matthew John M. Krane macrosegregation, sump profile, and temperature fields are presented. Two 45 cm diam-
Member, ASME eter billets were cast under production conditions with and without grain refiner. Tem-
Assistant Professor perature and composition measurements and sump profiles are compared to the numerical
School of Materials Engineering, results. The comparison shows some agreement for the grain refined case. It is believed
Purdue University, that an incorrect assumption about the actual grain structure prevents good agreement in
West Lafayette, IN 47907 the non-grain refined billet. 关DOI: 10.1115/1.1482089兴
Keywords: Heat Transfer, Manufacturing, Mass Transfer, Processing, Solidification
Introduction positive titanium segregation at the centerline 关2兴, which is present

in grain refining particles that serve as nucleation sites for free-
Direct chill casting is a semi-continuous process in which mol-
floating dendrites. Duplex microstructures accompanied by the
ten aluminum enters the top of a water-cooled mold, is cooled,
negative centerline segregation have also been observed by Chu
and forms a solid ingot drawn out below 共Fig. 1共a兲兲. The molten
and Jacoby 关3兴 and Dorward and Beerntsen 关4兴, and negative cen-
aluminum begins to solidify due to the cooling effect of the mold
terline segregation with positive segregation of titanium was ob-
wall. As the ingot is withdrawn from the bottom of the mold,
served in ingots cast by Gariepy and Caron 关5兴.
water jets impinge on the surface, directly cooling the billet and
forming a solid layer around a liquid melt, commonly referred to To model the transport phenomena which control the segrega-
as the sump. tion in DC casting, single domain models have been used that
A particular defect that occurs in a wide range of casting pro- implicitly couple phenomena in the melt, mushy zone, and solid
cesses, including the direct chill 共DC兲 casting of aluminum alloys, regions of a solidifying ingot. Flood et al. 关6兴 were the first to
is macrosegregation, the uncontrolled redistribution of alloying apply a single domain model for the transport phenomena in DC
elements on the scale of the ingot during solidification. Macroseg- casting to predict macrosegregation in Al-4.5 wt% Cu billets.
regation in DC cast ingots can produce nonuniform mechanical They assumed that the mushy zone consisted only of a rigid,
properties 关1兴, which affects the behavior of the metal during permeable dendritic matrix with the solid moving at the casting
downstream forming and heat treat operations. speed and predicted positive segregation of copper at the center of
A commonly observed, surface-to-surface distribution of alloy- the billet. Reddy and Beckermann 关7兴 included the effect of free-
ing elements at a transverse cross-section of a DC cast ingot re- floating dendrites of varying diameter by employing a modified
veals distinct regions of positive 共solute-rich兲 and negative version of the two-phase model developed by Ni and Beckermann
共solute-depleted兲 segregation 关2兴. Enrichment near the ingot sur- 关8兴 to simulate the DC casting of two Al-4.5 wt% Cu billets. In
face is attributed to a combination of shrinkage induced flow of one case, the solid phase was assumed to form a rigid structure
solute-rich liquid toward the mold wall, where solidification rates moving at the casting speed. In the other case, the transport of
are highest, and exudation induced by local remelting of the ingot solute-depleted, free-floating dendrites was modeled assuming the
surface as the solid shell pulls away from the mold. Once a struc- solid phase formed a rigid structure at a solid volume fraction of
turally sound ingot surface has formed and solidification rates 0.637. In the simulation with no free-floating solid, they predicted
have decreased, more normal segregation behavior is observed, subsurface enrichment, followed by a narrow depleted region just
with the solute composition increasing with decreasing radius within the surface. Positive segregation occurred at the centerline
from the solute-depleted region near the ingot surface. While this and was attributed to the buoyancy driven flow of copper-rich
trend may be expected to continue, ingots produced by DC casting liquid to the bottom of the sump. In the simulation with free-
often exhibit a significant decrease in solute composition near the floating dendrites, they predicted significant negative segregation
centerline. at the centerline, which they attributed to the advection of copper-
Experimental studies have provided some insight into the depleted dendrites to the bottom of the sump. Reddy and Becker-
mechanisms that control macrosegregation in DC cast ingots, es- mann were the first to include the transport of free-floating den-
pecially at the centerline. Yu and Granger 关2兴 investigated nega- drites into a fully coupled, single domain, macrosegregation
tive segregation in Al-Cu-Mg slabs and concluded that a duplex model of the DC casting process and to predict the resulting nega-
microstructure and accompanying negative segregation at the cen- tive segregation at the centerline. More recently, the same authors
ter of DC cast ingots resulted from the settling of solute-depleted, 关9兴 conducted further DC casting simulations of Al-4.5 wt% Cu
free-floating dendrites that formed near the top of the sump and billets to examine the effect of the grain density and interdendritic
were subsequently swept to the bottom of the sump by convection fluid flow on macrosegregation in billets devoid of free-floating
currents in the melt. This theory was also supported by significant dendrites. They demonstrated that, if the mushy zone consists of
a rigid dendritic structure without free-floating dendrites, its per-
meability significantly influences macrosegregation in DC cast
HEAT TRANSFER. Manuscript received by the Heat Transfer Division September 12, billets.
2001; revision received March 26, 2002. Associate Editor: C. Amon. Vreeman et al. 关10兴 developed a fully coupled, single-domain,
Numerical Modeling
The numerical model is that of Vreeman et al. 关10兴, which
simulates solidification in systems with two-phase 共solid-liquid兲
regions consisting of a slurry of free-floating dendrites above a
rigid, dendritic matrix saturated with interdendritic liquid. Mixture
transport equations for mass, energy, and species conservation
关12兴 are applied in all regions of the DC cast billet as it solidifies,
while the application of separate and distinct momentum equa-
tions depends on the local solid morphology 关10兴. The liquid
metal velocities calculated in previous papers under very similar
conditions were low enough to justify the assumption that the flow
can be treated as laminar. In the liquid and slurry regions, for
which the solid volume fraction is less than the designated pack-
ing fraction (g s ⬍g s,p ), modified versions of slurry momentum
equations developed by Ni and Incropera 关13兴 are employed
共Table 1兲. In regions consisting of a rigid, solid structure and
interdendritic liquid (g s ⭓g s,p ), momentum equations developed
by Bennon and Incropera 关12兴 and Prescott et al. 关14兴 are em-
ployed to account for interdendritic fluid flow in a translating solid
matrix.
The model used here to predict solid particle motion relative to
the liquid flow is much simpler than some found in the literature.
Wang and Beckermann 关15兴 have developed a detailed multiphase
model to account for the presence of free-floating equiaxed den-
drites and their effect on composition and grain structure, linking
the model to the macroscopic equations governing thermosolutal
convection and heat and mass transfer. Calculations have been
made, some of which were compared to experiments in salt-water
solutions 关16,17兴.
The parameter that defines the relative extent of the slurry and
rigid mushy zone regions is the packing fraction, g s,p , at which
the free-floating dendrites coalesce to form a rigid dendritic struc-
ture. In general, packing fractions are not explicitly known and
vary significantly from system to system. In addition to the pack-
ing fraction, a characteristic diameter of free-floating dendrites, d,
used to calculate solid and liquid phase velocity differences, is not
explicitly known 关10兴. In this study, d⫽75 ␮ m was used. Lumped
Fig. 1 „a… Schematic of generic DC casting process; „b… into this single, unchanging parameter is the real particle size
AirSlip™ mold cross-section distribution which varies in time and position during solidifica-
tion. A rigorous determination of dendrite size variations requires
implementation of an additional transport equation and a nucle-
ation model to account for the advection and ‘‘generation’’
共through nucleation and/or fragmentation兲 of dendrites, respec-
mixture model of the DC casting process which accounted for the tively. An example of such a model is found in 关7兴.
transport of free-floating dendrites. Separate and distinct mixture To model the solid-liquid interactions in the rigid mushy zone,
the permeability, K, is assumed to be isotropic and is evaluated
momentum equations were employed to account for the different
momentum transfer mechanisms in two-phase regions character- using the Blake-Kozeny expression, K⫽ ␬ o (1⫺g s ) 3 /g 2s . Evalua-
ized by a slurry of free-floating dendrites and a rigid, permeable tion of the permeability coefficient, ␬ o for an equiaxed mushy
dendritic matrix saturated with interdendritic liquid. In a compan- zone relies on empirical data for flow through granular materials
关18兴. In this study, a permeability coefficient of ␬ o ⫽3.75
ion paper 关11兴, the model was used to predict macrosegregation
distributions in Al-4.5 wt% Cu and Al-6.0 wt% Mg billets and the ⫻10⫺11 m2 is used. This value is based on a dendritic arm spac-
ing of 75 ␮m, which is in the range of measured values found in
predicted surface-to-centerline distribution of macrosegregation
DC cast billets. The thermophysical properties used in the study
was found to be consistent with observations in DC cast ingots
are shown in Table 2 and are based on values found in 关19兴.
discussed above. Negative segregation at the centerline increased
The model is closed by assuming thermodynamic equilibrium
with an increase in the packing fraction at which free-floating on the scale of individual control volumes and using the Al-Cu
dendrites are presumed to coalesce into a rigid dendritic structure. phase diagram, with constant equilibrium partition coefficients
Likewise, negative segregation at the centerline and positive seg- 共Table 2兲 and thermodynamic relations developed by Bennon
regation in the enriched region increased with an increase in the and Incropera 关20兴 to account for both primary and eutectic
characteristic diameter of the free-floating dendrites. solidification.
The objective of this study is to make a direct comparison of In order to obtain a steady-state solution, the energy equation
Vreeman’s model to data taken from an industrial scale direct chill was solved by itself first, with the resulting solid fraction and
process. Two billets of an Al-6wt%Cu alloy were cast in the temperature distribution used as initial conditions for further cal-
foundry at Wagstaff, Inc., one with and one without the use of a culations. Recognizing that converged solutions could not be ob-
grain refiner to generate free-floating solid particles in the sump. tained without the time-dependent terms in the conservation equa-
Measured composition profile, temperature history, and sump pro- tions, transient simulations of the full model were marched
file results for the grain refined and non-grain refined billets are forward in time with convergence rigorously enforced at each
compared to the numerical results with and without free-floating time step. While a steady state solution was obtained for the billet
solids. without free-floating dendrites (g s,p ⫽0), attempts to obtain
Table 1 Mixture model transport equations †10‡
steady state numerical solutions with free-floating dendrites discussion of the numerical techniques used is found in 关22兴. The
(g s, p ⬎0) were unsuccessful because of the unstable nature of the solution was carried out as described further in 关11兴, with the
interaction of the fluid flow and the solid in the region of g s,p 共see exception of the thermal boundary condition. Instead of using the
关11兴 for details兲. As the simulations proceeded forward in time, a correlations found in 关11兴, the temperature of the surface of the
quasi-steady state was reached where transient variations in the billet was determined by the measurements described below.
dependent variables reached a minimum range.
The numerical solution of the conservation equations 共Table 1兲
is implemented through the use of a control volume formulation Experimental Methods
and the SIMPLER algorithm 关21兴. The procedures outlined by To understand the behavior of these billets during an actual
Bennon and Incropera 关20兴 are used with upwind differencing to casting process and to evaluate the performance of Vreeman’s
discretize all terms. Modifications are made to the differencing model, two 0.45 m 共18 in兲 diameter Al-6wt%Cu billets were cast
procedure of the so-called advection-like source terms in the spe- at Wagstaff’s foundry in Spokane, Washington, using the Wagstaff
cies equation. For these terms, the discretization procedure was Airslip™ process 共see Fig. 1共b兲兲. The alloy was induction melted
modified to insure that mass is conserved when both solid and and degassed with 100 percent argon, then transferred from the
liquid simultaneously cross control volume interfaces. A complete tilting furnace via a heated refractory trough to the mold table.
Table 2 Thermophysical properties and phase diagram data †11‡
The AirSlip™ mold technology is a ‘‘hot-top’’ design with a
metal inlet diameter of 0.305 m 共12 in兲. A unique feature of this
technology is the use of a porous graphite ‘‘casting ring’’ that is
fed with both dry air and a synthetic lubricant. This design pro-
vides minimal surface macrosegregation by reducing the degree of
chill obtained within the mold bore. A specially designed cooling
water array provides the necessary upward conductance and high
heat transfer rate to solidify both shell and bulk of the billet.
The procedure for casting was to transfer the alloy by tilting the
induction hearth, allowing the metal to fill a cavity comprised of
the casting mold and a steel starter block. Following a prescribed
hold time, the starter block was slowly withdrawn from the mold.
The casting speed was 1 mm/s. Heat transfer for the initial 0.02 m
共0.8 in.兲 of cast length is obtained by conduction from the molten
metal through the starter block to the casting water. After this
period the water then impinges on the billet surface and provides
a much higher rate of cooling.
Sump depths were measured during the start phase by briefly
inserting an aluminum rod into the liquid through the hot top and
marking the level of the free surface of the Al alloy. When the
sump depth became constant, the solidification was judged to be
steady state and a thermocouple rake was inserted into the sump.
Three thermocouples were mounted on this rake and positioned
near the centerline, the mid-radius, and the surface of the billet.
First, a thin guide wire was introduced to provide a support for the
rake. The thermocouple rake was then introduced and allowed to
stabilize at the casting temperature. Following temperature stabi-
lization the rake was ‘‘captured’’ by locating the rake to the upper-
and-outermost area of the mold cavity and clamping the rake to
the guide wire. The rake then moved at the casting speed and
recorded the three vertical temperature profiles as it descended.
Due to the inherent inaccuracy when placing thermocouples, the
position of each thermocouple was established by post-cast sec-
tioning of the billet and machining to the thermocouple tips. These Fig. 2 Mixture copper composition, streamlines, and solid
final positions were found to be at r⫽10 mm, 106 mm, and 220 fraction for billet with grain refiner, using experimental bound-
mm, the last of these being only a few millimeters from the sur- ary conditions
face. The uncertainty in the thermocouple measurements was es-
timated to be ⫾2°C.
Grain refiners are usually added to DC cast alloys in order to
produce a more uniform grain structure in the billet. Grain refine- measurements were made using X-ray fluorescence on 20
ment is normally accomplished by inoculation with ‘‘seed’’ grains samples, taken from each billet at 10 different radial positions. At
and growth rate restriction. The former effect is provided by ad- each of the radial positions, the two samples measured were from
dition of either TiB2 , Tix Cy , or Al3 Ti, while the latter is provided different axial locations in the region of the billet apparently at
steady-state.
by adding solute elements such as titanium that provide constitu-
tional undercooling. Relative uniformity in grain size and equi-
axed shape significantly reduces the anisotropy in the mechanical
properties due to columnar grain growth that occurs without a Results and Discussion
grain refiner. The less oriented grain structure increases the ease Predictions were made of the steady state transport phenomena
with which these billets are rolled or extruded after casting. Two in the casting of the two ingots described above. The numerical
ingots were cast for this study, identical in every respect except results for streamlines, composition, and fraction solid for the
the addition of a grain refinement process to one of them and a grain refined case are shown in Fig. 2. The streamlines show the
slight variation in composition 共grain refined billet⫽6.1 wt% Cu, flow being drawn from the constricted inlet almost horizontally
nongrain refined billet⫽5.9 wt% Cu兲. The grain refiner was a towards the location of jet impingement on the surface. Along
combination of 0.02wt% Ti 共added to restrain growth rate兲 and 6 with the enrichment of the liquid with copper as solidification
ppm B 共added for TiB2 inoculation兲 as Al-3% Ti-1%B rod. This proceeds, the high rate of heat extraction under the jets is the
alloy was introduced by continuously feeding a rod into the engine which drives the buoyancy induced flow in the sump. The
trough. flow accelerates as it approaches and passes the chill, and is
Sump profiles were obtained by two separate methods. In the turned back into the billet by the formation of the rigid solid
case of grain-refined billet, this was accomplished by sudden ad- matrix. The flow then races down the interface of the rigid solid
dition of a molten Al-Si mixture to the hot top of the caster during and the slurry region. As the fluid runs down towards the center-
steady state casting. For the nongrain-refined billet, the Al-Si mix- line, it is continuously entrained into the rigid mushy zone.
ture was replaced by molten Al-3 percent Ti-1 percentB grain In these calculations, the packing fraction is set to g s, P ⫽0.3,
refiner rod. These mixtures were chosen over more traditional which is based on observations in salt solutions and the results of
methods 共e.g., zinc兲 due to the minimal disturbance they cause in numerical studies found in 关11兴 and 关23兴. These works also dem-
the solidification profile. The macrograph in Fig. 4 shows an ex- onstrate the sensitivity of the simulation results to the choice of
ample of an Al-Si sump in a grain-refined billet. The sample was packing fraction. A slurry of solid particles is formed and carried
prepared by slicing the billet in half, polishing the exposed sur- along with the flow down the rigid interface. Because the particles
face, and etching it with a solution of 10 percent HNO3 , 10 per- are formed at temperatures just below the liquidus, they are sig-
cent HCl, and 0.1 percent HF in deionized water. Composition nificantly depleted in copper. The movement of these particles
Fig. 4 Sump profiles: „a… Grain refined billet; and „b… Compari-
son of predictions and experiments.
colder liquid rushing underneath it, but most of the heat extracted
from the ingot is taken from regions closer to the surface. The
effect of this region is seen in the temperature contours in Fig.
3共b兲. Some correspondence can be found between the end of the
shallow gradients in Fig. 3共a兲 and the predicted beginning of the
rigid mushy zone where the fluid quickly assumes the casting
speed. Both Figs. 2 and 3共a兲 present some evidence for double-
diffusive flow cells in the sump, each with relatively uniform
composition and temperature.
An etched macrograph of the grain refined billet is compared to
predictions in Fig. 4. In Fig. 4共a兲, lines representing the solidus
interface (g S ⫽1.0) and the transition between the rigid, packed
Fig. 3 Results with grain refiner: „a… Comparison of predicted solid and the slurry (g S ⫽g S, P ) are seen. These lines are repro-
and measured temperature profiles, and „b… Calculated tem-
duced in Fig. 4共b兲 and compared to calculations. The total sump
perature and fraction solid contours.
depth is only overpredicted by approximately 5 percent, while the
actual rigid mushy zone thickness at the surface is twice the cal-
culated value near the surface. The general shape of the sump is
from where they form 共near the outer radius兲 to where they finally correct, but the profiles are steeper near the center and shallower
join the rigid mushy zone 共closer to the centerline兲 results in a near the surface than predicted. This result suggests that the rigid
region in the center which is deficient in copper. mushy zone is thicker than predicted and, given the agreements in
A comparison of vertical temperature profiles for experiments the temperature profiles, perhaps the packing fraction is slightly
and simulations is found in Fig. 3共a兲. At all three radial positions, lower than assumed (g S ⫽0.20⫺0.25?). Near the surface, it is
the temperature decreases in the melt due to conduction towards possible that the packing fraction was even lower, due to the more
the chill until the liquidus temperature is reached. At that point, horizontal surface, which would slow the flow and allow more
solid begins to nucleate on the grain refiner and a solid-liquid time for particles to settle.
slurry forms. Near the surface, the large heat extraction rate so- The radial composition profile is shown in Fig. 5. Near the
lidifies the alloy fast enough that the particles pack quickly and a centerline, the measured composition was up to 10 percent below
rigid structure forms. The other two profiles behave somewhat the nominal value for the alloy, while numerical results predicted
differently, with a sudden shallowing of the temperature gradient less than half that level of segregation there. The study of the
around the liquidus temperature. This shallow gradient begins at effect of packing fraction on centerline segregation found in 关11兴
roughly 0.06 m from the inlet and continues to approximately 0.12 suggests that this lower composition at the centerline could be an
m and 0.22 m for the midradius and centerline profiles, respec- indication that the actual packing fraction was picked to be too
tively. At those points the temperature gradient became markedly low a value. However, the results for the sump temperature pro-
steeper. The reason for this behavior is found in the flow patterns files discussed above suggest otherwise. The calculated values in-
described above. The liquid metal is chilled by the impinging jets creased as the radius increased until a peak was reached around
and much of the metal runs down the interface at which the solid r⫽0.15 m. The experiments also increased, but reached a plateau
packs until it is entrained into the rigid mushy zone. This flow around r⫽0.07 m. Figure 4 shows that the observed sump profiles
pattern leaves a triangular zone of weak recirculation in the center are less steep than in the simulation near the billet surface, sug-
of the sump around which most of the fluid flows. This ‘‘dead gesting a greater tendency for solid particles to settle there than
zone’’ does contain some solid particles, as it is cooled by the was predicted. As the radius decreased, the situation was reversed
Fig. 5 Comparison of measured and calculated composition
Fig. 7 Comparison of measured and calculated composition
profiles for grain refined billet
profiles for billet with no grain refiner
共predicted sump profiles were less steep than in the real billet兲, so
copper-depleted particles are more likely to end up at the bottom
of the sump at the centerline. It was noted above that the diameter would be about 0.1 wt percent in Fig. 5, which would not account
of the free-floating solid particles was assumed to be fixed at 75 for the discrepancy between the predictions and the measure-
␮m, which was picked based on typical grain sizes in the actual ments. It should also be noted that there are some measured com-
ingot. Using only this average leads to some uncertainty in the
position variations at given radial positions. These data, taken at
results. It should be noted that, in Vreeman and Incropera 关11兴, the
different axial locations, show that the composition did vary along
difference in the level of centerline macrosegregation with a varia-
tion of diameter between 25 and 100 ␮m was found to be approxi- the axis of the billet and call into question the assumption that this
mately 2 percent of the nominal composition. Such a difference process truly reaches a steady-state in the flow and composition
fields.
To understand how well the model simulates the transport phe-
nomena in a non-grain refined process, another numerical case
was run with g S, P ⫽0.0, simulating purely columnar growth. The
results were compared to experimental data from a billet cast in
the same manner as above except the absence of grain refiner.
Figure 6 shows the predicted composition field and streamline and
fraction solid profiles. With g S, P ⫽0.0, the model assumes that all
of the solid is rigid and moving at the casting velocity. The flow
from the inlet to the chill is similar to the previous case, but,
because of the much lower packing fraction, the temperature of
the flow moving down along the interface of rigid mushy zone is
much higher 共above the liquidus temperature兲 than in the grain
refined case. Also, with no Cu depleted solid being swept down
towards the centerline by the buoyancy driven flow, only Cu en-
riched liquid reaches the center. This effect accounts for the pre-
dicted positive segregation at the centerline.
While the grain refined case showed very good agreement be-
tween experiments and modeling, the case with no grain refiner
did not. The radial composition profile in Fig. 7 shows a marked
difference between predicted and measured macrosegregation.
While the predictions suggested a steady increase in composition
as the radius becomes smaller, the experiments show a pattern
more reminiscent of the profile in the presence of grain refiner
共except at the centerline, where there is an increase in copper
content兲. It is well known that the grain structure of these types of
large billets undergo a columnar to equiaxed transition around the
midradius. This change could be brought about by a slurry of solid
particles in the sump generated by fragmentation of the columnar
dendrites. This mechanism of particle generation would produce a
much smaller slurry region and a much lower packing fraction
than the grain refined billet. However, the assumption that all the
solid is rigid and moves at the casting speed is not valid in the
presence of such a particle cloud. At this point, attributing the
Fig. 6 Mixture copper composition, streamlines, and solid discrepancy between the measured and predicted macrosegrega-
fraction for billet with no grain refiner, using experimental tion to a columnar to equiaxed transition is speculative and more
boundary conditions data are needed to confirm this conclusion.
Conclusions References
The continuum mixture model of Vreeman et al. 关10兴 has been 关1兴 Finn, T. L., Chu, M. G., and Bennon, W. D., 1992, ‘‘The Influence of Mushy
Region Microstructure on Macrosegregation in Direct Chill Cast Aluminum-
used to simulate the casting of industrial scale, 45 cm 共18 in.兲 Copper Round Ingots,’’ in Micro/Macro Scale Phenomena in Solidification,
diameter billets of Al-6wt percentCu using Wagstaff’s Airslip™ ASME HTD-Vol. 218, C. Beckermann, et al., eds., ASME, New York, pp.
direct chill casting process. In the case with the addition of an 17–26.
Al-Ti-B grain refiner, some agreement was found between Vree- 关2兴 Yu, H., and Granger, D. A., 1986, ‘‘Macrosegregation in Aluminum Alloy
Ingot Cast by the Semicontinuous Direct Chill 共DC兲 Method,’’ in Aluminum
man’s model and the measured vertical temperature profiles, sump Alloys-Their Physical and Mechanical Properties, EMAS, United Kingdom,
profiles, and macrosegregation in the steady state region of the pp. 17–29.
casting process. The validity of some of the assumed model pa- 关3兴 Chu, M. G., and Jacoby, J. E., 1990, ‘‘Macrosegregation Characteristics of
rameters was discussed. It was surmised that the actual packing Commercial Size Aluminum Alloy Ingot Cast by the Direct Chill Method,’’ in
Light Metals 1990, C. M. Bickert, ed., TMS, pp. 925–930.
fraction is less that the 30 percent solid which was used in the 关4兴 Dorward, R. C., and Beerntsen, D. J., 1990, ‘‘Effects of Casting Practice on
calculations, although it must be recognized that it is likely that Macrosegregation and Microstructure of 2024 Alloy Billet,’’ in Light Metals
the packing fraction in the actual billet will vary with position and 1990, C. M. Bickert, ed., TMS, pp. 919–924.
casting parameters. Without the grain refiner, the model and data 关5兴 Gariepy, B., and Caron, Y., 1991, ‘‘Investigation in the Effects of Casting
Parameters on the Extent of Centerline Macrosegregation in DC Cast Sheet
do not agree as well, possibly because the grain structure in the Ingots,’’ in Light Metals 1991, E. L. Rooy, ed., TMS, pp. 961–971.
real billet undergoes a columnar to equiaxed transition which is 关6兴 Flood, S. C., Katgerman, L., and Voller, V. R., 1991, ‘‘The Calculation of
not accounted for in the model. Macrosegregation and Heat and Fluid Flows in the D.C. Casting of Aluminum
Alloys,’’ in Modeling of Casting, Welding and Advanced Solidification Pro-
cesses V, M. Rappaz, et al., eds., TMS, pp. 683– 690.
关7兴 Reddy, A. V., and Beckermann, C., 1995, ‘‘Simulation of the Effects of Ther-
mosolutal Convection, Shrinkage Induced Flow, and Solid Transport on Mac-
Acknowledgments rosegregation and Equiaxed Grain Size Distribution in a DC Continuous Cast
Al-Cu Round Ingot,’’ in Materials Processing in the Computer Age II, V. R.
The authors wish to thank Mr. Robert Wagstaff of Wagstaff Voller, et al., eds., pp. 89–102.
Engineering for funding the experimental portion of this research. 关8兴 Ni, J., and Beckermann, C., 1991, ‘‘A Volume-Averaged Two-Phase Model for
Transport Phenomena During Solidification,’’ Metall. Mater. Trans. B, 22B,
pp. 349–361.
关9兴 Reddy, A. V., and Beckermann, C., 1997, ‘‘Modeling of Macrosegregation Due
to Thermosolutal Convection and Contraction-Driven Flow in Direct Chill
Continuous Casting of an Al-Cu Round Ingot,’’ Metall. Mater. Trans. B, 28B,
Nomenclature pp. 479– 489.
关10兴 Vreeman, C. J., Krane, M. J. M., and Incropera, F. P., 2000, ‘‘The Effect of
c ⫽ specific heat 关J/kg-K兴 Free-Floating Dendrites and Convection on Macrosegregation in Direct Chill
C Cu ⫽ wt percent of copper Cast Aluminum Alloys—I: Model Development,’’ Int. J. Heat Mass Transf.,
d ⫽ characteristic dendrite diameter 关␮m兴 43, pp. 677– 686.
关11兴 Vreeman, C. J., and Incropera, F. P., 2000, ‘‘The Effect of Free-Floating Den-
D ⫽ mass diffusion coefficient 关m2/s兴 drites and Convection on Macrosegregation in Direct Chill Cast Aluminum
f ⫽ mass fraction Alloys—II: Predictions for Al-Cu and Al-Mg alloys,’’ Int. J. Heat Mass
g ⫽ volume fraction; gravitational acceleration 关m/s2兴 Transf., 43, pp. 687–704.
h ⫽ enthalpy 关J/kg兴关12兴 Bennon, W. D., and Incropera, F. P., 1987, ‘‘A Continuum Model for Momen-
tum, Heat and Species Transport in Binary Solid-Liquid Phase Change
k ⫽ thermal conductivity 关W/m-K兴 Systems—I: Model Formulation,’’ Int. J. Heat Mass Transf., 30, pp. 2161–
kp ⫽ equilibrium partition coefficient 2170.
K ⫽ permeability 关m2兴关13兴 Ni, J., and Incropera, F. P., 1995, ‘‘Extension of the Continuum Model for
P ⫽ reduced pressure 关N/m2兴 Transport Phenomena Occurring During Metal Alloy Solidification—I: The
Conservation Equations,’’ Int. J. Heat Mass Transf., 38, pp. 1271–1284.
r, z ⫽ axisymmetric coordinates 关m兴关14兴 Prescott, P. J., Incropera, F. P., and Bennon, W. D., 1991, ‘‘Modeling of Den-
t ⫽ time 关s兴 dritic Solidification Systems: Reassessment of the Continuum Momentum
T ⫽ temperature 关K兴 Equation,’’ Int. J. Heat Mass Transf., 34, pp. 2351–2358.
u ⫽ axial velocity component 关m/s兴关15兴 Wang, C. Y., and Beckermann, C., 1996, ‘‘Equiaxed Dendritic Solidification
With Convection: Part I. Multiscale/Multiphase Modeling,’’ Metall. Mater.
v ⫽ radial velocity component 关m/s兴 Trans. A, 27A, pp. 2754 –2764.
Vជ ⫽ velocity vector 关m/s兴关16兴 Wang, C. Y., and Beckermann, C., 1996, ‘‘Equiaxed Dendritic Solidification
With Convection: Part II. Numerical Simulations for an Al-4 wt pct Cu alloy,’’
Greek Symbols Metall. Mater. Trans. A, 27A, pp. 2765–2783.
关17兴 Beckermann, C., and Wang, C. Y., 1996, ‘‘Equiaxed Dendritic Solidification
␤ ⫽ contraction ratio With Convection: Part III. Comparisons With NH4Cl-H2O Experiments,’’
␤ s ⫽ solutal expansion coefficient Metall. Mater. Trans. A, 27A, pp. 2784 –2795.
␤ T ⫽ thermal expansion coefficient 关l/K兴关18兴 Ocansey, P., Bhat, M. S., and Poirier, D. R., 1994, ‘‘Permeability for Liquid
␮ ⫽ dynamic viscosity 关kg/s-m兴 Flow in the Mushy Zones of Equiaxed Castings,’’ in Light Metals 1994, U.
␳ ⫽ density 关kg/m3兴 Mannweiler, ed., TMS, pp. 807– 812.
关19兴 Smithell’s Metals Reference Handbook, 1992, 7th ed., E. A. Barnes and G. B.
Subscripts Brook, eds., Butterworth-Heinemann, Ltd., Oxford, pp. 14.1–14.14.
关20兴 Bennon, W. D., and Incropera, F. P., 1988, ‘‘Numerical Analysis of Binary
i ⫽ mold inlet condition Solid-Liquid Phase Change Using a Continuum Model,’’ Numer. Heat Trans-
l ⫽ liquid fer, 13, pp. 277–296.
关21兴 Patankar, S., 1980, Numerical Heat Transfer and Fluid Flow, Hemisphere,
liq ⫽ liquidus New York, pp. 113–134.
m ⫽ mixture 关22兴 Vreeman, C. J., and Incropera, F. P., 1999, ‘‘Numerical Discretization of Spe-
o ⫽ reference, nominal value cies Equation Source Terms in Binary Mixture Models of Solidification and
p ⫽ packed, rigid structure Their Impact on Macrosegregation in Semi-Continuous, Direct Chill Casting
Systems,’’ Numer. Heat Transfer, Part B, 36共1兲, pp. 1–14.
s ⫽ solid 关23兴 Vreeman, C. J., 1997, ‘‘Modeling Macrosegregation in Direct Chill Cast Alu-
S ⫽ solutal minum Alloys,’’ M.S. thesis, School of Mechanical Engineering, Purdue Uni-
T ⫽ thermal versity, West Lafayette, IN.
Thermal and Electrical Energy
Transport and Conversion in
T. S. Fisher
e-mail: tsfisher@purdue.edu
Nanoscale Electron Field
Purdue University,
School of Mechanical Engineering, Emission Processes
1288 Mechanical Engineering Building,
West Lafayette, IN 47907-1288 This paper considers the theory of electron field emission from nanoscale emitters with
particular focus on thermal and electrical energy transport. The foundational theory of
D. G. Walker field emission is explored, and a model is presented that accounts explicitly for the energy
Vanderbilt University band curvature produced by nanoscale tip emitters. The results indicate that the inclusion
Department of Mechanical Engineering, of band curvature strongly influences the energetic distribution of electrons for emitter
Box 1592, Station B, radii less than 50 nm. The energy exchange process between emitted and replacement
Nashville, TN 37235 electrons is shown to allow high local energy transfer rates that can be exploited in direct
thermal-to-electrical energy conversion processes. The dependence of energy conversion
rates on material and operational parameters is demonstrated. Throughout the paper,
opportunities for further research involving nanoscale heat transfer, materials develop-
ment, and modeling are highlighted. 关DOI: 10.1115/1.1494091兴
Keywords: Electronics, Energy Conversion, Heat Transfer, Microscale, Nanoscale
1 Introduction the subject of scientific inquiry 关16兴. Suggested applications for

field emission devices include vacuum field-effect transistors, di-
Prolific advances in microscale and nanoscale technologies
odes and triodes, ion sources, electron guns, flat-panel displays,
have generated renewed scientific interest in direct thermal-to-
scanning microscopes, and many others 关13兴.
electrical energy conversion. These advances have provided re-
Field emission devices are formed by placing an emitting cath-
searchers with new tools to explore physical phenomena that were
ode in close proximity to an anode, separated by vacuum 共typi-
previously inaccessible. For example, the controlled fabrication of
one- and two-dimensional nanoscale structures has fostered sig- cally less than 10⫺5 torr兲. A gate electrode may also be included to
nificant recent improvements in the efficiency and capacity of modulate emission and to provide a more precise spatial separa-
thermoelectric materials and devices 关1,2兴. Further, the develop- tion from the cathode. Figure 1 shows a tip configuration. The
ment of thin-film heterostructures has brought thermionic energy cathode, anode and base electrode materials can be semiconductor
conversion, originally a vacuum-based technology, to the solid 共with narrow or wide band gaps兲 or metals. The gate is typically
state 关3,4兴. The phenomenon of electron field emission is, in some metallic, and the dielectric between the cathode and gate is
respects, similar to thermoelectric and thermionic transport. How- commonly SiO2 . Field emission occurs when a sufficiently strong
ever, the thermodynamics of field emission have received less local electric field 共greater than 0.5 V/Å兲 exists at the location of
attention, although several recent theoretical papers on direct re- emission. Small-scale features within a low average electric
frigeration suggest that field emission devices could operate re- field can produce a high local field. Cathode tip structures are
markably well 关5–10兴. In the present work, we consider the foun- designed to provide this field enhancement. Some field enhance-
dations of and prospects for direct energy conversion by electron ment also occurs in film cathodes due to highly localized areas of
field emission and show that outstanding performance is possible emission 关13兴.
with further progress in fabrication technology, modeling, and Most prior work has focused on device operation near room
engineering. temperature. In fact, the predominant theoretical foundation of
The emission of electrons from an electrically active surface via field emission—basic Fowler-Nordheim theory—neglects the ef-
quantum tunneling into a vacuum is commonly referred to as field fects of temperature on emission characteristics. In contrast, con-
emission. Fowler and Nordheim 关11兴 provided the first theoretical sideration of the thermodynamics of field emission requires the
treatment of field emission from planar metal surfaces. Later stud- inclusion of temperature effects. Fleming and Henderson 关17兴
ies showed that emission could be substantially improved by geo- conducted early experiments on the energy distribution of field-
metric electric field enhancement provided by elongated struc- emitted electrons. Their work prompted an exchange of letters
tures, such as pyramidal or conical tips. Spindt was the first to with Nottingham 关18,19兴, who predicted that field emission could
show such enhancement using molybdenum cones 关12兴. Since produce a heating or cooling effect on the cathode, depending on
then, the emission properties of a wide variety of metallic and the field strength. Subsequent studies by Good and Muller 关20兴,
semiconductor materials and structures have been studied in detail Dyke and Dolan 关21兴, and Gomer 关22兴 provided a more detailed
关13,14兴. The emission properties of carbon-based/diamond mate- treatment of the thermodynamics of field emission.
rials, in particular, are very robust 关15兴. Because they do not re- After these early studies, relatively little work occurred on the
quire heating, carbon-based and other field emitters are often thermodynamics of field emission for several decades. However,
termed cold cathodes. However, the physical reasons behind the studies by Swanson et al. 关23兴, Engle and Cutler 关24兴, and Berg-
outstanding emission properties of carbon-based materials remain eret et al. 关25兴 considered measurements of the so-called Notting-
ham effect 关23,25兴 and the thermodynamics of electron emission
and replacement processes 关24兴. In the past decade, Cutler, Misk-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division May 2, 2001; ovsky, and co-workers have studied various facets of thermody-
revision received November 5, 2001. Associate Editor: A. Majumdar. namics in field emission processes, particularly from diamond
Fig. 1 Nanotips for electron field emission: „a… Schematic of an energy conversion system including nanotip
electron emitters „cathode…, gate electrode, and anode; and „b… Example of a polycrystalline diamond nanotip
emitter surrounded by a monolithic gate. The tip radius is È10 nm.
films, including: cathode heating mechanisms and energy ex- of the planar field emission model by multiplying the applied field
change processes in sharp metallic emitters 关26兴, average electron F by a field enhancement factor ␤ 关13–15兴 such that the local field
energies of emission and replacement 关27,28兴, effects of doping at the tip becomes ␤ F where ␤ is greater than 1.
关29兴, ballistic electron transport 关30,31兴, and internal field emis- The field enhancement factor ␤ depends strongly on small-scale
sion 关32–34兴. This body of work has significantly advanced the structure. For tip emitters, electric field theory indicates that ␤ is
theoretical understanding of energetic processes in field emission. proportional to the inverse of the tip radius R 关35兴. Thus, the tip
Yet, the complexity of the physical processes involved in field radius should be as small as possible to achieve low-voltage emis-
emission, particularly from polycrystalline materials such as sion. However, the use of the field enhancement factor may not be
chemical vapor deposited 共CVD兲 diamond, has prevented a thor- appropriate for energy conversion studies. In effect, the field en-
ough understanding of experimentally observed phenomena. Such hancement factor ␤ linearizes the highly non-linear electric field
understanding will be essential if the promise of energy conver- near the emission surface, as illustrated in Fig. 2. In the figure,
sion by electron field emission is to be realized. both the actual and linearized fields produce the same emission
In the present work, we review the foundational theory of current by assumption. The linear field is necessary to arrive at the
nanoscale field emission with an emphasis on energy transport analytic current-field relation described below as the basic
processes and modeling. The simplifying assumptions that are Fowler-Nordheim relation 关see Eq. 共7兲兴. However, the lineariza-
central to traditional field emission theory are presented, and their tion does not satisfy the anode-side boundary condition on the
applicability to thermodynamic modeling is considered in detail. electric potential and generally underestimates the emission flux
A more accurate model, which explicitly accounts for emitter-size of high-energy electrons 共above the Fermi energy in Fig. 2兲 and
effects and temperature dependence, is presented for thermody- overestimates the emission flux at energies below the Fermi level.
namic calculations. This model is exercised over relevant param-
eter ranges, and its utility in interpreting experimental data is
demonstrated. This work does not focus on specific energy con-
version devices and systems, but rather, provides new insights into
energy transport mechanisms in electron field emission from
nanoscale emitters. Throughout the work, we identify opportuni-
ties for further improvements in modeling and experimental meth-
ods and highlight applications to heat transfer and energy conver-
sion that will be important to the future development of nanoscale
thermodynamic field emission devices.
2 Thermodynamics of Field Emission

2.1 The Field Emission Process. Field emission occurs
when electrons tunnel through a sufficiently narrow potential bar-
rier created by the application of a voltage bias between a cathode
共negative bias兲 and an anode or gate electrode 共positive bias兲.
Fowler and Nordheim presented the first theory of field emission
from a planar metal surface over seventy years ago 关11兴. Applica-
tion of a voltage between planar electrodes creates a constant
electric field F⫽V/L 共modified slightly by the image charge as
discussed later兲, where V is the applied voltage and L is the dis-
tance between the electrodes. Emission from small, non-planar Fig. 2 Electron potential profile near a tip emitter. Solid line
tips can substantially reduce the required field to produce emis- represents actual potential field. Dashed line represents ap-
sion by enhancing the local field near the emission site. This en- proximate, linearized field. Both fields produce the same emis-
hancement has traditionally been described within the framework sion current.
For rectangular barriers, the probability of tunneling is in- fields; however, this approximation is retained here for the pur-
versely proportional to the exponential of the local potential bar- pose of obtaining results without the need to solve Schrödinger’s
rier width 关36兴. Thus, the shape of the potential profile 共which equation numerically for each geometric and field condition. The
depends on the tip geometry兲 has a profound effect on the energy effect of nanoscale features on the emission current is included
distribution of emitted electrons. Nanoscale tips produce ex- through the energy profile V(x) in the transmission coefficient of
tremely large curvature in the vacuum potential near the emission Eq. 共4兲. The supply function, which represents the number of elec-
site. Tip radii of 5 nm have been reported 关15兴. The electrode trons whose x-direction energy 共W兲 incident on the emission sur-
spacing must be small to provide efficient emission and to mini- face falls within dW per second per unit area, can be formulated
mize space-charge effects, which can degrade performance 关37兴. analytically as 关20兴
冋冕册
The combination of small tip radius R and electrode spacing can
produce high emission with low applied bias. Thus, these small- 4␲ ⬁
N 共 W 兲 dW⫽ f 共 W, ␳ 兲 ␳ d ␳ dW
scale features are essential in producing efficient, high-capacity h3 0
再冋册冎
thermal-electrical energy conversion.
In the present work, we employ a model that is more physically 4 ␲ mkT ⫺ 共 W⫺ ␨ 兲
accurate than the basic Fowler-Nordheim approach. The present ⫽ 3 ln 1⫹exp dW (5)
h kT
model allows for curvature of the potential field 共such as shown in
Fig. 2兲 and thus provides more accuracy in simulating the thermo- where h is Planck’s constant, ␳ is the radial electron momentum
dynamics of field emission. Other more detailed models have been 共perpendicular to the emission direction兲, k is Boltzmann’s con-
proposed 关38 – 40兴, but these have typically focused on electronic stant, T is the emitter temperature, ␨ is the Fermi level 共see Fig. 2兲,
applications and thus simplified the thermal considerations that and f (W, ␳ ) is the Fermi-Dirac function:
are central to the present work. The one-dimensional model is
1
冉冊
based on the potential field in a spherically symmetric ball-shell f 共 W, ␳ 兲 ⫽ . (6)
system 关关9,22兴 pp. 37–38兴 and takes the form: W⫺ ␨ ␳2
⫹ ⫹1
冋册
exp
R kT 2mkT
V 共 x 兲 ⫽⫺qF 共 L⫹R 兲 1⫺ ⫹G 共 x 兲 (1)
x⫹R Several simplifying assumptions have been invoked historically
to reduce the integral in Eq. 共3兲 to an analytic form. As described
where q is the magnitude of electron charge, F is the applied above, the use of the field enhancement factor ␤ linearizes the first
共global兲 field, L is the separation distance between electrodes term of Eq. 共1兲 as ⫺q ␤ Fx. Also, the temperature dependence is
共taken as L⫽4 ␮ m for all cases considered later兲, R is the char- eliminated by neglecting the spreading of the Fermi-Dirac func-
acteristic radius of the emitter, x is the distance from the emitter, tion with increasing temperature. With these assumptions and
and G(x) is the image potential experienced by emitting electron. other integration approximations 关20兴, the current density integral
This image potential for a spherical emitter tip can be expressed can be expressed analytically as:
as 关39兴:
G共 x 兲⫽
⫺q 2 K⫺1 R
4 ␲ ␧ 0 K⫹1 共 2R⫹x 兲 2x
(2) J⫽
1.5⫻10⫺6 ␤ 2 F 2
␾
10.4
exp 1/2 exp
␾
冉冊冉
⫺6.44⫻107 ␾ 3/2
␤F 冊 (7)
where ␧ 0 is permittivity of vacuum, and K is the emitter’s dielec- where J is the average current density 共A/cm2兲, F is the average
tric constant. We note that this image potential neglects possible applied electric field 共V/cm兲, and ␾ is an effective work function
effects of surface plasmons on the tunneling electrons 关41兴. Sub- 共eV兲. The foregoing equation is the basic Fowler-Nordheim rela-
sequent results will show that the emitter radius R strongly influ- tion for field emission current. Note that temperature effects are
ences the electrical and thermodynamic characteristics of field absent. However, most experimental data for room-temperature
emission. The first term of Eq. 共1兲 represents the contribution due emission exhibit reasonable agreement with the Fowler-Nordheim
to the applied field F and equals zero at the emission site (x theory.
⫽0). Thus, following the formalism of Good and Muller 关20兴, the Experimental data are often plotted in so-called Fowler-
zero energy datum exists at the cathode/vacuum interface of Fig. 2 Nordheim coordinates to determine unknown or uncertain param-
共ignoring image charge potential兲. The second term in Eq. 共1兲 eters such as emission area, work function, and field enhancement
represents the shift in potential due to image charge 关20兴. factor by rearranging Eq. 共7兲 as
2.2 Current-Voltage Behavior. The physics of field emis-
sion have been described in detail by Good and Muller 关20兴. The
development below provides some of the major results that are
central to the present work. The total emission current density J
takes the form
J⫽q 冕 ⬁
⫺W a
D 共 W 兲 N 共 W 兲 dW (3)
where ⫺W a represents the bottom of the emitter’s conduction

band, W is the x-component of electron energy 共i.e., in the direc-
tion of emission兲, the term N(W)dW is the electron supply func-
tion, and D(W) is the quantum tunneling transmission coefficient.
In the present work, the transmission coefficient is calculated us-
ing the WKB approximation 关42兴
再冕冑
D 共 W 兲 ⫽exp ⫺
x2
x1
8m
ប2
兩 V 共 x 兲 ⫺W 兩 dx 冎 (4)
where V(x) is the potential profile 关see Eq. 共1兲兴, x 1 and x 2 are the Fig. 3 Electron potential as a function of position from emitter
zeros of W⫺V(x), and m is the mass of an electron. We note that and emitter radius. All profiles produce the same current den-
the WKB approximation is, in general, limited to relatively low sity, J Ä10 AÕcm2 . ␾Ä1.7 eV. K Ä5.5. T Ä300 K.
ln 冉冊
I
F2
⫽C 0 共 A, ␤ , ␾ 兲 ⫹C 1 共 ␤ , ␾ 兲
1
F
(8)
that is useful in studying important problems that result from elec-
tron deposition in the gate or anode. For example, modeling and
simulation of anode heating in field emission electronic devices
where I is the measured current, and A is the effective emission requires an understanding of both the spatial and energetic distri-
area. A linear fit of F 1 versus ln(I/F2) enables the evaluation of the butions of electron energy emitted from the cathode.
two constants C 0 and C 1 , which depend on area A, field enhance- The average total energy of emitted electrons can be expressed
ment factor ␤, and effective work function ␾. Thus, the formula- as the sum of axial and radial components, 具 ␧ 典 ⫽ 具 W 典 ⫹ 具 ␧ ␳ 典 . The
tion is underdetermined, and an assumed value is typically taken average axial energy can be calculated as
冕
for one of the three primary variables. As discussed later, signifi- ⬁
cant opportunities for advanced techniques, such as parameter es- 1
具W典⫽ qWD 共 W 兲 N 共 W 兲 dW (9)
timation, exist for interpreting experimental data. J ⫺W a
The emitter’s radius R has a profound effect on the field re-
quired to produce emission. Figure 3 shows the potential field where the transmission function D(W) and the supply function
near the emitter for radii ranging from R⫽5 nm to infinity 共i.e., a N(W)dW are given in Eqs. 共3兲 and 共4兲, respectively. Similarly, the
planar surface兲. We choose an effective work function of ␾⫽1.7 radial component becomes
eV and a dielectric constant K⫽5.5 共i.e., that of diamond兲. The
value of the work function is a convenient choice to demonstrate
energy conversion processes. This choice is consistent with the
具 ␧ ␳典 ⫽
4␲q
Jh 3 冕冕 ⬁
⫺W a
⬁
0
D 共 W 兲 f 共 W, ␳ 兲
␳3
2m
d ␳ dW
冕
minimum Schottky barrier for nitrogen-doped diamond 关32兴 and
with recent observations and analysis of emission from graphite- 4 ␲ qm 共 kT 兲 2 ⬁
⫽ D共 W 兲
like grain boundary regions 关43,44兴. Later, results for several val- Jh 3 ⫺W a
ues of work function are considered 共see Fig. 5兲. For each curve in
Fig. 3, the average field produces a current density of 10 A/cm2, as
computed using the potential profile of Eq. 共1兲 and numerical
evaluations of the current density integral in Eq. 共3兲. We note that
⫻ 冉 1
2
␣共 W 兲2⫹
␲2
6
⫹dilog关 e ␣ 共 W 兲 ⫹1 兴 dW 冊 (10)
the potential curves for emitter radii smaller than 50 nm exhibit where
large differences, while those for radii greater than 50 nm 共includ- W⫺ ␨
ing the planar case R⫽⬁兲 are quite similar. The figure indicates ␣共 W 兲⫽ (11)
that the barrier width is slightly narrower in the high-energy re- kT
冕
gions 共i.e., above the Fermi level兲 for small radii and substantially x ln共 t 兲
larger at lower energies for small radii. Thus, the emission radius dilog共 x 兲 ⫽ dt. (12)
influences the energetic distribution of emission. The required ap- 1 1⫺t
plied 共average兲 fields range from F⫽1.08 V/ ␮ m for R⫽5 nm to
F⫽614 V/ ␮ m for a planar surface (R⫽⬁). This difference in The dilog function can be computed efficiently as a series 关45兴.
applied field strongly influences energy conversion due to the in- The sum of the axial and radial energies of Eqs. 共9兲 and 共10兲
creased losses 共e.g., from higher required voltages兲 incurred at represents the total average energy of emitted electrons. These
high fields. electrons must be replaced by others to preserve charge continuity.
The difference in total average energy between emitted and re-
2.3 Energy Transport. The starting point for energy trans- placement electrons determines whether field emission produces
port modeling is the current density integral of Eq. 共3兲, from heating or cooling of the cathode. For example, if the average
which the number of electrons emitted per unit time and area can emitted energy exceeds the average replacement energy, then
be derived. Energy moments of this integral can provide energy cooling ensues 共neglecting other possible energy transport mecha-
transport rates. Chung et al. 关27兴 reported related calculations for nisms, such as reverse emission兲. The energies of replacement
overall electron energy using numerical evaluation of double in- electrons typically fall near the chemical potential ␨. Chung et al.
tegrals. Here, we separate energy into axial 共i.e., in the emission 关27兴 and Cutler et al. 关28兴 showed that the average energy of
direction兲 and radial 共i.e., perpendicular to the emission direction兲 replacement electrons can be several tenths of an electron volt
components. The underlying free-electron model of the emitter below ␨. Again, the present formulation differs from that of Chung
material and the separability of the potential field employed here et al. by separating the axial and radial energy components. The
permit this spatial separation. The advantages of the present for- average axial ( 具 W r 典 ) and radial ( 具 ␧ ␳ r 典 ) replacement energies can
mulation are that 共1兲 the double integral can be reduced analyti- be expressed as an integral over available energy states below the
cally to a single integral and 共2兲 directional information is retained chemical potential as
具 W r典 ⫽ 冕冕␨
⫺W a
␨ ⫺W
0
W 兵 1⫺ f 共 W, ␳ 兲关 1⫺D 共 W 兲兴其 d␧ ␳ dW 冕冕
␨
⫺W a 0
␨ ⫺W
兵 1⫺ f 共 W, ␳ 兲关 1⫺D 共 W 兲兴其 d␧ ␳ dW
冕 ␨
⫺W a
W 兵共 ␨ ⫺W 兲 ⫹ 关 1⫺D 共 W 兲兴 kT 其关 ln 2⫹ ␣ 共 W 兲 ⫺ln关 e ␣ 共 W 兲 ⫹1 兴兴 dW
冕
⫽ ␨ (13)
兵共 ␨ ⫺W 兲 ⫹ 关 1⫺D 共 W 兲兴 kT 其关 ln 2⫹ ␣ 共 W 兲 ⫺ln关 e ␣ 共 W 兲 ⫹1 兴兴 dW
⫺W a
冕冕
␨
⫺W a
␨ ⫺W
0
␧ ␳ 兵 1⫺ f 共 W, ␳ 兲关 1⫺D 共 W 兲兴其 d␧ ␳ dW 冕 ␨
⫺W a
g 共 W 兲 dW
冕冕冕
具 ␧ ␳r典 ⫽ ␨ ␨ ⫺W ⫽ ␨
兵 1⫺ f 共 W, ␳ 兲关 1⫺D 共 W 兲兴其 d␧ ␳ dW 兵共 ␨ ⫺W 兲 ⫹ 关 1⫺D 共 W 兲兴 kT 其关 ln 2⫹ ␣ 共 W 兲 ⫺ln关 e ␣ 共 W 兲 ⫹1 兴兴 dW
⫺W a 0 ⫺W a
(14)
where
g共 W 兲⫽
共 ␨ ⫺W 兲 2
2
⫺ 关 1⫺D 共 W 兲兴共 kT 兲 2再␲2
12
1
⫺ln共 2 兲 ␣ 共 W 兲 ⫹ ␣ 共 W 兲 2 ⫹dilog关 e ␣ 共 W 兲 ⫹1 兴 .
2 冎 (15)
Each of the integrals in Eqs. 共3兲, 共9兲, 共10兲, 共13兲, and 共14兲 has
been evaluated numerically with adaptive quadrature to a conver-
gence to five decimal places. The combination of these integrals
provides a means of evaluating the rate of energy flux q ⬙ to or
from the cathode due to electron emission. This energy flux can be
expressed as the product of the electrical current density and the
difference in average energy between emitted and replacement
electrons:
J
q ⬙⫽ 关具 W 典 ⫹ 具 ␧ ␳ 典 ⫺ 共具 W r 典 ⫹ 具 ␧ ␳ r 典兲兴 . (16)
q
The average energies described above depend on the applied
field F, as shown in Figure 4 for an effective emitter radius of
R⫽10 nm at a temperature T⫽300 K. For low fields, the average
axial emission energies 具 W 典 are relatively high compared to the
lowest possible axial energy. As the field increases, the energy
decreases. The decrease in the emission energy is a consequence
of the narrowing barrier width at highly populated energy states.
For both axial ( 具 W r 典 ) and transverse ( 具 ␧ ␳ r 典 ) replacement ener-
gies, the slight decrease in energy is a consequence of the greater
number of empty low-energy states due to emission from those
states. In the case of emission electrons, their average transverse
energies ( 具 ␧ ␳ 典 ), which must be greater than zero, tend to increase
with increasing field due to the reduction in axial energies de-
Fig. 4 Average emitted and replacement electron energies as scribed above. This increase is a consequence of the distribution
a function applied field. Average axial emitted energyÄŠ W ‹. Av- of directional energies governed by Fermi-Dirac statistics 共i.e., as
erage radial emitted energy Š ␧ ␳ ‹. Average axial replacement axial energy decreases, higher radial energy states become avail-
energyÄŠ W r ‹. Average radial emitted energy Š ␧ ␳ r ‹. Figure inset able兲. The inset of Fig. 4 shows the net electron energy exchange
shows net electron exchange energy as a function of applied due to the foregoing mechanisms. The net exchange is positive at
field. Emitter characteristic radius R Ä10 nm. Work function low field due primarily to the emission of high-energy electrons
␾Ä1.7 eV. K Ä5.5. Temperature T Ä300 K. through the top of the potential barrier. As the field increases,
Fig. 5 Emission energy flux from the cathode as a function applied field F ,
emitter characteristic radius R , and work function ␾: „a… ␾Ä1 eV; „b… ␾Ä1.7 eV;
and „c… ␾Ä3 eV. Each part shows curves for three emitter radii, R Ä10, 25, and
100 nm. K Ä5.5. Temperature T Ä300 K.
more low-energy electrons emit, and at a field of approximately 3 tor and external emission from the semiconductor into vacuum.
V/␮m, the net exchange of energy becomes negative and causes For example, much recent work 关32–34,50,51兴 on diamond emit-
cathode heating. ters suggests that the interface between a base electrode 共either
Predictions from the theory outlined above suggest that sub- metallic or semiconductor兲 and diamond 共a wide-bandgap semi-
stantial rates of direct electrical-thermal energy conversion and conductor兲 governs the field emission process. Therefore, the
transfer are possible with field emission. The energy flux q ⬙ from analysis of semiconductor emitters can be substantially more com-
the cathode depends primarily on four parameters: 共1兲 emitter plicated than that of metals 共such as shown in Fig. 2兲. However,
work function ␾, 共2兲 characteristic emitter radius R, 共3兲 applied wide-bandgap semiconductors, such as diamond, can provide sub-
field F, and 共4兲 temperature T. As shown in Fig. 4, net exchange stantial energy filtering effects as well as favorable structural
energies greater than 0.1 eV are possible. The net energy flux is properties, and consequently, are well suited to direct energy con-
the product of this exchange energy and the electric current den- version applications.
sity 关see Eq. 共16兲兴, and consequently, even a large net energy Tunneling into the conduction band of a diamond cathode is
exchange may produce little thermodynamic effect if the current shown schematically in the one-dimensional band diagrams of
density is low. Fig. 6. Figure 6共a兲 shows the unbiased state, with slight band
Figure 5 shows the results of this balance for three emitter radii bending due to space charge effects 关52兴. The band gap, E g ⫽E C
共R⫽10 nm, 25 nm, 100 nm兲 and work functions 共␾⫽1 eV, 1.7 eV, ⫺E V , of diamond is 5.5 eV, and the difference between the base
3 eV兲. For each condition, the energy flux at low fields is negli- electrode’s Fermi energy and the cathode’s valence band energy
gible due to the low electrical current density. As the field in- for an undoped sample is E FB ⫺E V ⫽1.4 eV, which is the energy
creases, the electrical current increases, and substantial energy associated with grain boundaries. Cathode doping can alter this
flux away from the emitter is made possible by the net energy energy difference. The effective work function ␾ eff represents the
exchange between emitted and replacement electrons. As the field axial electron energy required for an electron in the base electrode
increases further, the energy exchange decreases and ultimately to emit into vacuum. The parameter ␹ ⫽E vac⫺E C is the electron
becomes negative, which causes heating. As the emitter radius affinity and represents the energy required to eject an electron
increases, the peak energy flux decreases due to the downward from the conduction band into vacuum. Numerous recent studies
shift in energy distribution that is inherent to larger radii 共see Fig. indicate that ␹ is negative 共although it is shown positive in Fig. 6兲
2兲. At the same time, the field range for positive energy flux grows when the diamond surface is terminated by hydrogen 关53–56兴.
with increasing radius due to the reduction in field enhancement. This negative electron affinity can significantly enhance emission.
The figure also shows a strong influence of the work function. For A gate electrode can be used to extract high-energy electrons
the lowest work function, ␾⫽1 eV, very high rates of local energy from the cathode. Figure 1 shows nanotip emitters surrounded by
flux of order q ⬙ ⬃106 W/cm2 are predicted for all emitter radii. gate structure. Under a voltage bias 共positive on the gate elec-
The corresponding local current densities for the maximum en- trode兲, the electric field causes the bands in the cathode and
ergy fluxes are of order J⬃107 A/cm2 for all radii. As the work vacuum to shift, as shown in Fig. 6共b兲. Near the base-electrode/
function increases to ␾⫽1.7 eV, the peak energy fluxes decrease cathode interface, significant band bending narrows the potential
significantly 共q ⬙ ⬃10 to 100 W/cm2兲, and local current densities at barrier width. This narrow barrier increases the probability of
the peak energy fluxes are of order J⬃104 A/cm2 . For the largest quantum tunneling, and field emission occurs. A second tunneling
work function, ␾⫽3 eV, maximization of energy flux is not evi- process may also occur at the cathode/vacuum barrier for ␹⬎0.
dent, and the energy flux is negligible or negative over the entire The presence of internal field emission significantly compli-
range of applied fields. These results indicate that low-work- cates the task of modeling. Solid-state scattering events can alter
function materials are necessary to enable significant energy con- the distribution of electrons that ultimately emit into vacuum. For
version. The probability of developing materials with effective polycrystalline diamond films near room temperature, Lerner
work functions as low as ␾⫽1 eV are quite remote, and therefore et al. 关31兴 showed that electrons can travel quasiballistically for
the associated heat fluxes in Fig. 1共a兲 may be practically infea- films less than several microns in thickness. Thus, for thin films,
sible. For the case of semiconductor emitters, appropriate n-type the band structure of Fig. 6 can provide the basis for a reasonable
dopants could potentially produce effective work functions near emission model. However, neither Fowler-Nordheim theory nor
␾⫽1.7 eV, and we note that, in the particular case of diamond, the the foregoing model possesses the fidelity to represent such com-
goal of finding suitable shallow donors has been elusive. How- plex profiles in detail. We note, however, that unlike basic Fowler-
ever, several recent studies indicate some optimism in this area Nordheim theory, the present model does allow for the inclusion
关46 – 49兴. of size effects in thermodynamic calculations. A more rigorous
approach generally requires numerical solutions to Schrödinger’s
3 Discussion equation 关57兴, and the inclusion of multi-dimensional effects 关58兴.
Field emission from semiconductor materials can involve both The numerical solution of Schrödinger’s equation possesses the
internal emission from a metallic conductor into the semiconduc- advantage of eliminating the WKB approximation in the evalua-
Fig. 6 Band diagrams for field emission from diamond: „a… Unbiased at ther-
mal equilibrium; and „b… Field emission via tunneling from the base electrode
to the cathode’s conduction band and from the conduction band to vacuum
tion of transmission coefficients. Further, a general numerical Estimates of these two factors enable estimation of two of the
solver would enable analysis of more complex systems, such as three unknown parameters 共A, ␾, and ␤兲 with an assumed value
resonant-tunneling structures 关6 – 8兴. Advanced, multidimensional for the third. Estimation of these values is dubious with linear
tunneling formulations have been reported 关59– 64兴 with particu- Fowler-Nordheim analysis. Also note that the estimation of the
lar application to scanning tunneling microscopy. However, to derivative of the data (dI/dV) deserves scrutiny because the dif-
date, a combined analysis, including complex band structure, ferentiation of discrete points tends to magnify experimental
multi-dimensionality, numerically computed transmission coeffi- errors 关79兴.
cients, and energy exchange calculations has not been reported A geometrically more rigorous approach to emission modeling
and would represent a significant achievement. and interpretation of experimental data has been developed by
We also anticipate that, with recent and continued progress in Jensen and co-workers 关37,72,80兴. In their work, the emission
nanoscale fabrication methods, new nanoscale materials and sur- area and field enhancement are related analytically to the geom-
faces will require the replacement of continuum models with sub- etry of the tip emitter. Further, the variability of emission from
continuum quantum models. Such models would replace the sup- multiple tips in an array is considered explicitly. It is likely that
ply function N(W)dW of Eq. 共5兲. Miskovsky and co-workers such an approach, when coupled with emission models that in-
have shown that discrete atomic-scale effects can strongly influ- clude band-bending effects, will provide greater efficacy in the
ence the local density of electron states from which emitting elec- interpretation of experiment results.
trons originate 关65兴. They have also considered strong localized In all existing and proposed models for the interpretation of
heating and thermoelectric effects near atomically sharp emitters field emission data, a statistical approach can provide estimates
关66兴. Karabutov et al. 关44兴 have postulated that two-dimensional with superior confidence as well as insights into the accuracy of
quantum wells are responsible for the low turn-on fields of poly- estimated parameters. The sensitivity of the emission current to
crystalline diamond emitters. Interestingly, the resulting current- the free parameters can indicate the identifiability of the param-
voltage relations still closely follow Fowler-Nordheim trends. eters in a given emission model. Parameter estimation techniques
Fursey and Glazanov have compared current-voltage predictions 关81,82兴 represent a powerful tool for experimental interpretation
of formulations that include nanoscale tip curvature with those and model development. Such tools are particularly suitable for
from traditional Fowler-Nordheim theory 关67兴. A group of studies field emission applications due to significant measurement noise
by He, Cutler, Miskovsky, and co-workers 关68 –71兴 and Jensen 共e.g., transient fluctuations of emission current兲 and the potential
and co-workers 关39,72兴 highlight the effects of nanoscale tip ge- complexity of new theoretical and computational models. For ex-
ometries on field emission and the commensurate shortcomings of ample, the model developed herein, with the inclusion of tempera-
using traditional Fowler-Nordheim theory for these small-scale ture and characteristic emitter radius, presents an opportunity for
devices. These studies reinforce the need for advanced treatment more rigorous determination of physics-based emission param-
of nanoscale and atomic effects in the study of field emitters. eters through advanced estimation methods.
Advanced electron spectroscopy will be required to study rig- The questions of total energy conversion capacity and effi-
orously the thermodynamics of field emission. Gadzuk and Plum- ciency are important to consider for the broad range of potential
mer reviewed the field-emission energy distribution 共FEED兲 tech- applications of field emission devices. The energy fluxes shown in
nique 关73兴. Most such spectrometers employ hemispherical
Fig. 5 result from emission from nanoscale emitters. Both the
elements 关74,75兴. Recently, researchers have developed non-
local energy flux and the number density per unit area of nanos-
hemispherical 关76兴 and simplified hemispherical 关77兴 devices that
cale emitters will ultimately determine the total energy conversion
offer enhanced resolution and simplicity. Electron spectroscopy
capacity. A variety of fabrication technologies-nanodiamond pyra-
has proven to be useful for studying field emission devices with
midal tips 关83兴, focused ion beam molds 关84兴, and carbon nano-
electronic functions. An understanding of the electron energy dis-
tubes 关85兴-exist that can produce tremendously high number den-
tribution is particularly important in the study of anode heating,
which is a significant cause of failure in high-current field emis- sities of field nanoemitters. We believe that the rapidly evolving
sion devices 关78兴. For energy-conversion field-emission devices, field of nanofabrication will ultimately present a wide range of
experimental quantification of electron energy distributions will fabrication options.
be essential. Ideally, the thermodynamic performance metrics, coefficient of
Another area for further study involves the interpretation of performance for refrigeration and thermal efficiency for power
experimental data and its comparison with increasingly complex generation, of field emission devices should approach that of the
models. Determination of emission parameters from the Fowler- limiting Carnot efficiency, in a manner similar to that of vacuum
Nordheim theory 关see Eq. 共7兲兴 has historically been a heuristic thermionic devices 关86兴. In practice, thermal radiation and con-
process where the value of the parameters is largely dependent on duction losses, as well as electrical losses 共e.g., in terminal leads
the estimation procedure. The incongruous results are a feature of and from Joule heating of tip emitters兲, would limit practical
the models because the unknown parameters of interest are prima- maximum efficiencies to approximately 50–70 percent of the Car-
rily non-physical and are, in general, correlated. The Fowler- not efficiencies. This estimate is consistent with those for thermi-
Nordheim formulation of emission involves three parameters, onic converters 关86兴. Calculations by Cutler et al. 关10兴 suggest
namely the emission area 共A兲, the work function 共␾兲, and a field that the coefficient of performance of a refrigeration field emis-
enhancement factor 共␤兲. However, by casting emission data into a sion device can exceed that of traditional thermoelectric devices.
linear form, only two parameters can be uniquely determined; two The performance can improve further through the use of gate
of the free parameters are linearly dependent on the third. As a electrodes and with retarding potentials. For power generation ap-
result, experimenters must fix one of the free parameters using plications, traditional thermionic devices using metallic emitters
independent data and then proceed to estimate the remaining pa- can be highly efficient. However, the use of these materials has
rameters using a linear regression. been limited to high temperature 共⬎1500 K兲 applications because
One approach commonly used and described by Brodie and they prevent the emission of all except the highest energy elec-
Schwoebel 关13兴 is to differentiate the Fowler-Nordheim equation trons. In effect, thermionic converters are limited by the emitter
with respect to the applied voltage. The voltage and current are material’s work function. In contrast, the development of novel
measured in experiments and the derivative of the I – V character- emitter materials with low work functions could make the thermi-
istics can also be obtained from the data. Thus, at any point, a onic process amenable to operation at lower temperatures due to
value for C 0 关see Eq. 共8兲兴 can be calculated, and a regression the allowed emission of lower-energy electrons. Consequently, we
procedure is commonly employed to estimate its value. Substitu- believe that field emission materials and devices hold great prom-
tion of this value back into the governing equation provides C 1 . ise for enabling the operation of a fundamentally efficient energy
conversion mechanism in temperature ranges where other direct ␤ ⫽ field enhancement factor, see Fig. 2
energy conversion processes lack either sufficient capacity 共e.g., ␹ ⫽ electron affinity 共eV兲, see Fig. 6
thermionics兲 or efficiency 共e.g., thermoelectrics兲. ␧ ⫽ dielectric constant, see Eq. 共2兲
␾ ⫽ work function 共eV兲, see Fig. 2
4 Summary and Conclusions ␨ ⫽ Fermi level 共eV兲, see Fig. 2
␳ ⫽ radial electron momentum 共kg m/s兲
Field emission offers an intriguing new technical approach for
direct energy conversion. This work has provided an overview of Subscripts
the physics of electron field emission with particular emphasis on r ⫽ replacement electron value
thermal and nanoscale geometric effects. The traditional model for ␳ ⫽ radial direction
field emission, based on Fowler-Nordheim theory, ignores thermal
effects and linearizes highly non-linear fields in order to achieve
an analytic solution. The new model presented herein includes the
effects of temperature and nanoscale emitter geometry. The effi- References
cacy of this approach is demonstrated by the strong effects of the 关1兴 Hicks, L. D., and Dresselhaus, M. S., 1993, ‘‘Effect of Quantum-Well Struc-
characteristic emitter radius on emission 共see Figs. 3 and 5兲. tures on the Thermoelectric Figure of Merit,’’ Phys. Rev. B, 47, pp. 12 727–
The potential for high energy capacity and high efficiency sug- 12 731.
关2兴 Mahan, G. D., 1998, ‘‘Good Thermoelectrics,’’ Solid State Phys., 51, pp. 81–
gests that field emission devices may find a broad range of appli- 157.
cations; however, many technical challenges remain, particularly 关3兴 Mahan, G. D., and Woods, L. M., 1998, ‘‘Multilayer Thermionic Refrigera-
as new nanoscale materials and fabrication technologies emerge. tion,’’ Phys. Rev. Lett., 18, pp. 4016 – 4019.
As characteristic dimensions decrease below 10 nm, the con- 关4兴 Shakouri, A., Lee, E. Y., Smith, D. L., Narayanamurti, V., and Bowers, J. E.,
1998, ‘‘Thermoelectric Effects in Submicron Heterostructure Barriers,’’ Mi-
tinuum assumption becomes suspect, and atomic-scale modeling croscale Thermophys. Eng., 2, pp. 37– 47.
will be required. Further, the complex band structure of semicon- 关5兴 Miskovsky, N. M., and Cutler, P. H., 1999, ‘‘Microelectronic Cooling Using
ductor emitters, even at continuum scales, necessitates more ad- the Nottingham Effect and Internal Field Emission in a Diamond 共Wide-Band
vanced approaches to the characterization of quantum tunneling. Gap Material兲 Thin-Film Device,’’ Appl. Phys. Lett., 75, pp. 2147–2149.
关6兴 Tsu, R., and Greene, R. F., 1999, ‘‘Inverse Nottingham Effect Cooling in
In order for models to be useful, important material properties, Semiconductors,’’ Electrochem. Solid-State Lett., 2, pp. 645– 647.
such as work functions and electron affinities, must be character- 关7兴 Korotkov, A. N., and Likharev, K. K., 1999, ‘‘Possible Cooling by Resonant
ized experimentally. Fowler-Nordheim Emission,’’ Appl. Phys. Lett., 75, pp. 2491–2493.
Another major challenge lies in the demonstration of practical 关8兴 Korotkov, A. N., and Likarev, K. K., 2000, ‘‘Cooling by Resonant Fowler-
Nordheim Emission,’’ Physica B, 284, pp. 2030–2031.
energy conversion. In order to serve useful engineering purposes, 关9兴 Fisher, T. S., 2001, ‘‘Influence of Nanoscale Geometry on the Thermodynam-
the nanoscale emission phenomenon must be scaled up to macro- ics of Electron Field Emission,’’ Appl. Phys. Lett., 79共22兲, pp. 3699–3701.
scopic dimensions. Thus geometric, chemical, and surface unifor- 关10兴 Cutler, P. H., Miskovsky, N. M., Kumar, N., and Chung, M. S., 2000, ‘‘New
mity will be important considerations in any functional device. Results on Microelectronic Cooling Using the Inverse Nottingham Effect. Low
Temperature Operation and Efficiency,’’ Electrochemical Society Proceedings,
The achievement of such uniformity is a significant challenge, 2000–28, pp. 99–111.
even in a laboratory setting. 关11兴 Fowler, R. H., and Nordheim, L. W., 1928, ‘‘Field Emission From Metallic
Finally, we note that a complete study of practical thermody- Surfaces,’’ Proc. R. Soc. London, Ser. A, 119, pp. 173–181.
namic efficiencies has not yet been reported. The similarity be- 关12兴 Spindt, C. A., 1968, ‘‘A Thin Film Field Emission Cathode,’’ J. Appl. Phys.,
39, pp. 3504 –3505.
tween field emission and thermionic emission suggests that high 关13兴 Brodie, I., and Schwoebel, P. R., 1994, ‘‘Vacuum Microelectronic Devices,’’
efficiency operation is possible. The efficiency will depend on a Proc. IEEE, 82, pp. 1006 –1018.
wide range of factors, including radiative material properties, an- 关14兴 Nation, J. A., Schachter, L., Mako, F. M., Len, L. K., Peter, W., Tang, C.-M.,
ode work functions, and electrical resistivity of electrodes. This and Rao, T., 1999, ‘‘Advances in Cold Cathode Physics and Technology,’’
Proc. IEEE, 87, pp. 865– 889.
issue is the subject of current study by the authors. 关15兴 Kang, W. P., Fisher, T. S., and Davidson, J. L., 2001, ‘‘Diamond
Microemitters—The New Frontier of Electron Field Emission and Beyond,’’
New Diamond and Frontier Carbon Technology, 11, pp. 129–146.
Acknowledgments 关16兴 Robertson, J., 1999, ‘‘Mechanisms of Electron Field Emission from Diamond,
Diamond-Like Carbon, and Nanostructured Carbon,’’ J. Vac. Sci. Technol. B,
The authors thank Professors J. L. Davidson, W. P. Kang, and 17, p. 659.
A. M. Strauss for helpful conversations. This work was supported 关17兴 Fleming, G. M., and Henderson, J. E., 1940, ‘‘The Energy Losses Attending
by the NSF CAREER program 共Grant # CTS-9983961兲 and by a Field Current and Thermionic Emission of Electrons from Metals,’’ Phys. Rev.,
Vanderbilt Discovery Grant. 58, pp. 887– 894.
关18兴 Nottingham, W. B., 1941, ‘‘Remarks on Energy Losses Attending Thermionic
Emission of Electrons from Metals,’’ Phys. Rev., 59, pp. 906 –907.
Nomenclature 关19兴 Fleming, G. M., and Henderson, J. E., 1941, ‘‘On the Energy Losses Attending
Thermionic and Field Emission,’’ Phys. Rev., 59, pp. 907–908.
D(W) ⫽ tunneling transmission coefficient, see Eq. 共4兲关20兴 Good, R. H., and Muller, E. W., 1956, ‘‘Field Emission,’’ Handbook of Phys-
F ⫽ applied electric field (V/ ␮ m)⫽V/L ics, 21, pp. 176 –231.
f ⫽ Fermi-Dirac function, see Eq. 共6兲关21兴 Dyke, W. P., and Dolan, W. W., 1956, ‘‘Field Emission,’’ Adv. Electron. Elec-
tron Phys., VIII, pp. 90–187.
G(x) ⫽ image charge potential (V), see Eq. 共2兲关22兴 Gomer, R., 1961, Field Emission and Field Ionization, Harvard University
h ⫽ Planck’s constant⫽4.1357⫻10⫺15 eV•s Press, Cambridge, MA.
I ⫽ electrical current 共A兲关23兴 Swanson, L. W., Crouser, L. C., and Charbonnier, F. M., 1966, ‘‘Energy Ex-
J ⫽ current density 共A/cm2兲 changes Attending Field Electron Emission,’’ Phys. Rev., 151, pp. 327–340.
关24兴 Engle, I., and Cutler, P. H., 1967, ‘‘The Effect of Different Surface Barrier
K ⫽ emitter’s dielectric constant Models on the Nottingham Energy Exchange Process,’’ Surf. Sci., 8, pp. 288 –
k ⫽ Boltzmann’s constant⫽8.617⫻10⫺5 eV/K 307.
L ⫽ cathode-anode separation 共␮m兲关25兴 Bergeret, H., Septier, A., and Drechsler, M., 1985, ‘‘Nottingham Effect of a
Superconducting Metal,’’ Phys. Rev. B, 31, pp. 149–153.
m ⫽ electron rest mass⫽9.1095⫻10⫺31 kg 关26兴 Miskovsky, N. M., Park, S. H., He, J., and Cutler, P. H., 1993, ‘‘Energy Ex-
N(W) ⫽ electron supply function 关s/共kg m4兲兴, see Eq. 共5兲 change Processes in Field Emission from Atomically Sharp Metallic Emit-
q ⫽ electron charge⫽1.60219⫻10⫺19 Coulomb ters,’’ J. Vac. Sci. Technol. B, 11, pp. 366 –370.
关27兴 Chung, M. S., Cutler, P. H., Miskovsky, N. M., and Sullivan, T. E., 1994,
R ⫽ emitter characteristic radius 共nm兲 ‘‘Energy Exchange Processes in Electron Emission at High Fields and Tem-
V ⫽ applied voltage 共V兲 peratures,’’ J. Vac. Sci. Technol. B, 12, pp. 727–736.
W ⫽ axial electron energy 共eV兲关28兴 Cutler, P. H., Chung, M. S., Miskovsky, N. M., Sullivan, T. E., and Weiss, B.
x ⫽ axial coordinate 共m兲 L., 1994, ‘‘A New Model for the Replacement Process in Electron-Emission at
High Fields and Temperatures,’’ Appl. Surf. Sci., 76, pp. 1– 6.
Greek Symbols 关29兴 Huang, Z.-H., Cutler, P. H., Miskovsky, N. M., and Sullivan, T. E., 1995,
‘‘Calculation of Electron Field Emission from Diamond Surfaces,’’ J. Vac. Sci. 关59兴 Lang, N. D., 1985, ‘‘Vacuum Tunneling Current from an Adsorbed Atom,’’
Technol. B, 13, pp. 526 –530. Phys. Rev. Lett., 55, pp. 230–233.
关30兴 Cutler, P. H., Huang, Z.-H., Miskovsky, N. M., D’Ambrosio, P., and Chung, 关60兴 Lang, N. D., 1986, ‘‘Theory of Single-Atom Imaging in the Scanning Tunnel-
M., 1996, ‘‘Monte Carlo Study of Hot Electron and Ballistic Transport in ing Microscope,’’ Phys. Rev. Lett., 56, pp. 1164 –1167.
Diamond: Low Electric Field Region,’’ J. Vac. Sci. Technol. B, 14, pp. 2020– 关61兴 Lang, N. D., 1986, ‘‘Spectroscopy of Single Atoms in the Scanning Tunneling
2023. Microscope,’’ Phys. Rev. B, 34, pp. 5947–5950.
关31兴 Lerner, P., Cutler, P. H., and Miskovsky, N. M., 1997, ‘‘Hot Electron and 关62兴 Lucas, A. A., Morawitz, H., Henry, G. R., Vigneron, J.-P., Lambin, Ph., Cutler,
Quasiballistic Transport of Nonequilibrium Electrons in Diamond Thin P. H., and Feuchtwang, T. E., 1988, ‘‘Scattering-Theoretic Approach to Elastic
Films,’’ J. Vac. Sci. Technol. B, 15, pp. 398 – 400. One-Electron Tunneling Through Localized Barriers: Application to Scanning
关32兴 Lerner, P., Cutler, P. H., and Miskovsky, N. M., 1997, ‘‘Theoretical Analysis of Tunneling Microscopy,’’ Phys. Rev. B, 37, pp. 10 708 –10 720.
Field Emission from a Metal Diamond Cold Cathode Emitter,’’ J. Vac. Sci. 关63兴 Lucas, A. A., Morawitz, H., Henry, G. R., Vigneron, J.-P., Lambin, Ph., Cutler,
Technol. B, 15, pp. 337–342. P. H., and Feuchtwang, T. E., 1988, ‘‘Tunneling through Localized Barriers
关33兴 Lerner, P., Miskovsky, N. M., and Cutler, P. H., 1998, ‘‘Model Calculations of with Application to Scanning Tunneling Microscopy: New Scattering Theo-
Internal Field Emission and J-V Characteristics of a Composite n-Si and
retic Approach and Results,’’ J. Vac. Sci. Technol. A, 6, pp. 296 –299.
N-Diamond Cold Cathode Source,’’ J. Vac. Sci. Technol. B, 16, pp. 900–905.
关64兴 Huang, Z. H., Feuchtwang, T. E., Cutler, P. H., and Kazes, E., 1990, ‘‘The
关34兴 Cutler, P. H., Miskovsky, N. M., Lerner, P. B., and Chung, M. S., 1999, ‘‘The
Wentzel-Kramers-Brillouin Method in Multidimensional Tunneling: Applica-
Use of Internal Field Emission to Inject Electronic Charge Carriers into the
tion to Scanning Tunneling Microscopy,’’ J. Vac. Sci. Technol. A, 8, pp. 177–
Conduction Band of Diamond Films: A Review,’’ Appl. Surf. Sci., 146, pp.
181.
126 –133.
关35兴 Everhart, T. E., 1967, ‘‘Simplified Analysis of Point-Cathode Electron 关65兴 Miskovsky, N. M., and Cutler, P. H., 1999, ‘‘Calculation of the Local Density
Sources,’’ J. Appl. Phys., 38, pp. 4944 – 4957. of States for a Discrete Pyramidal Model of a Diamond Tip Surmounted by a
关36兴 Robinett, R. W., 1997, Quantum Mechanics: Classical Results, Modern Sys- Single Atom,’’ Surf. Sci., 439, pp. 173–180.
tems, and Visualized Examples, Oxford University Press, New York, pp. 263– 关66兴 Miskovsky, N. M., Park, S. H., He, J., and Cutler, P. H., 1993, ‘‘Energy Ex-
264. change Processes in Field Emission from Atomically Sharp Metallic Emit-
关37兴 Jensen, K. L., Kodis, M. A., Murphy, R. A., and Zaidman, E. G., 1997, ‘‘Space ters,’’ J. Vac. Sci. Technol. B, 11, pp. 366 –370.
Charge Effects on the Current-Voltage Characteristics of Gated Field Emitter 关67兴 Fursey, G. N., and Glazanov, D. V., 1998, ‘‘Deviations from the Fowler-
Arrays,’’ J. Appl. Phys., 82, pp. 845– 854. Nordheim Theory and Peculiarities of Field Electron Emission from Small-
关38兴 Modinos, A., 1984, Field, Thermionic and Secondary Electron Emission Spec- Scale Objects,’’ J. Vac. Sci. Technol. B, 16, pp. 910–915.
troscopy, Plenum Press, New York. 关68兴 He, J., Cutler, P. H., Miskovsky, N. M., and Feuchtwang, T. E., 1991, ‘‘Deri-
关39兴 Jensen, K. L., and Zaidman, E. G., 1993, ‘‘Field Emission from an Elliptical vation of the Image Interaction for Non-Planar Pointed Emitter Geometries:
Boss: Exact Versus Approximate Treatments,’’ Appl. Phys. Lett., 63, pp. 702– Application to Field Emission I-V Characteristics,’’ Surf. Sci., 246, pp. 348 –
704. 364.
关40兴 Jensen, K. L., 1999, ‘‘Semianalytical Model of Electron Source Potential Bar- 关69兴 He, J., Cutler, P. H., and Miskovsky, N. M., 1991, ‘‘Generalization of Fowler-
riers,’’ J. Vac. Sci. Technol. B, 17, pp. 515–519. Nordheim Field Emission Theory for Nonplanar Metal Emitters,’’ Appl. Phys.
关41兴 Nazarov, Yu. V., 1989, ‘‘Anomalous Current-Voltage Characteristics of Tunnel Lett., 59, pp. 1644 –1646.
Junctions,’’ Sov. Phys. J., 68, pp. 561–566. 关70兴 Cutler, P. H., He, J., and Miskovsky, N. M., 1993, ‘‘Theory of Electron Emis-
关42兴 Robinett, R. W., 1997, Quantum Mechanics: Classical Results, Modern Sys- sion in High Fields from Atomically Sharp Emitters: Validity of the Fowler-
tems, and Visualized Examples, Oxford University Press, New York, pp. 229– Nordheim Equation,’’ J. Vac. Sci. Technol. B, 11, pp. 387–391.
233. 关71兴 Cutler, P. H., He, J., Miller, J., Miskovsky, N. M., Weiss, B., and Sullivan, T.
关43兴 Karabutov, A. V., Frolov, V. D., Pimenov, S. M., and Konov, V. I., 1999, E., 1993, ‘‘Theory of Electron Emission in High Fields from Atomically Sharp
‘‘Grain Boundary Field Electron Emission from CVD Diamond Films,’’ Dia- Emitters: Validity of the Fowler-Nordheim Equation,’’ Prog. Surf. Sci., 42, pp.
mond Relat. Mater., 8, pp. 763–767. 169–185.
关44兴 Karabutov, A. V., Frolov, V. D., and Konov, V. I., 2001, ‘‘Diamond/sp2 -Bonded 关72兴 Jensen, K. L., Mukhopadhyay-Phillips, P., Zaidman, E. G., Nguyen, K., Kodis,
Carbon Structures: Quantum Well Field Electron Emission,’’ Diamond Relat. M. A., Malsawma, L., and Hor, C., 1997, ‘‘Electron Emission from a Single
Mater., 10, pp. 840– 846. Spindt-Type Filed Emitter: Comparison of Theory with Experiment,’’ Appl.
关45兴 Dwight, H. B., 1961, Tables of Integrals and Other Mathematical Data, Mac- Surf. Sci., 111, pp. 204 –212.
millan Company, New York, p. 144. 关73兴 Gadzuk, J. W., and Plummer, E. W., 1973, ‘‘Field-Emission Energy-
关46兴 Collins, A. T., 1994, Properties and Growth of Diamond, G. Davies, ed., EMIS Distribution 共FEED兲,’’ Rev. Mod. Phys., 45, pp. 487–548.
Datareviews Series No. 9, INSPEC, London. 关74兴 Kuyatt, C. E., and Simpson, J. A., 1967, ‘‘Electron Monochromator Design,’’
关47兴 Koizumi, S., Kamo, M., Sato, Y., Mita, S., Sawabe, A., Reznik, A., Uzan- Rev. Sci. Instrum., 38, pp. 103–111.
Saguy, C., and Kalish, R., 1998, ‘‘Growth and Characterization of Phosphorus 关75兴 Wysocki, J. K., 1982, ‘‘Testing of a Field Electron Energy Analyser,’’ Journal
Doped n-type Diamond Thin Films,’’ Diamond Relat. Mater., 7, pp. 540–544. of Physics E-Scientific Instruments, 15, pp. 1376 –1385.
关48兴 Sakaguci, I., Gamo, M. N., Kikuchi, Y., Yasu, E., and Haneda, H., 1999, 关76兴 Oshima, C., 1997, ‘‘An Apparatus for High Resolution Field Emission Spec-
‘‘Sulfur: A Donor Dopant for n-type Diamond Semiconductors,’’ Phys. Rev. B, troscopy,’’ Adv. Colloid Interface Sci., 71, pp. 353–369.
60, pp. R2139–R2141. 关77兴 Bernhard, J. M., Rouse, A. A., Sosa, E. D., Gnade, B. E., Golden, D. E., and
关49兴 Saada, D., Adler, J., and Kalish, R., 2000, ‘‘Sulfur: A Potential Donor in Chalamala, B. R., 1999, ‘‘A Compact Electron Energy Analyzer for Measuring
Diamond,’’ Appl. Phys. Lett., 77, pp. 878 – 879. Field Emission Energy Distributions,’’ Rev. Sci. Instrum., 70, pp. 3299–3302.
关50兴 Geis, M. W., Twichell, J. W., and Lyszczarz, T. M., 1996, ‘‘Diamond Emitters 关78兴 Grant, P., Py, C., Möbner, C., Blais, A., Tran, H., and Gao, M., 2000, ‘‘Electron
Fabrication and Theory,’’ J. Vac. Sci. Technol. B, 14, pp. 2060–2067. Field Emission From Diamond-like Carbon, a Correlation with Surface Modi-
关51兴 Koenigsfeld, N., Philosoph, B., and Kalish, R., 2000, ‘‘Field Emission Con- fications,’’ J. Appl. Phys., 87, pp. 1356 –1360.
trolled by the Substrate/CVD Diamond Interface,’’ Diamond Relat. Mater., 9, 关79兴 Ehrich, F. F., 1954, ‘‘Differentiation of Experimental Data Using Least Squares
pp. 1218 –1221.
Fitting,’’ J. Aeronaut. Sci., 22, pp. 133–134.
关52兴 Silva, S. R. P., Amaratunga, G. A. J., and Okano, K., 1999, ‘‘Modeling of the
关80兴 Jensen, K. L., Yater, J. E., Zaidman, E. G., Kodis, M. A., and Shih, A., 1998,
Electron Field Emission Process in Polycrystalline Diamond and Diamond-like
‘‘Advanced Emitters for Next Generation RF Amplifiers,’’ J. Vac. Sci. Technol.
Carbon Thin Films,’’ J. Vac. Sci. Technol. B, 17, pp. 557–561.
B, 16, pp. 2038 –2048.
关53兴 Ristein, J., Stein, W., and Ley, L., 1998, ‘‘Photoelectron Yield Spectroscopy on
关81兴 Beck, J. V., and Arnold, K. J., 1977, Parameter Estimation in Engineering and
Negative Electron Affinity Diamond Surfaces: A Contactless Unipolar Trans-
port Experiment,’’ Diamond Relat. Mater., 7, pp. 626 – 631. Science, Wiley, New York.
关54兴 Baumann, P. K., and Nemanich, R. J., 1998, ‘‘Surface Cleaning, Electronic 关82兴 Haftka, R. T., Scott, E. P., and Cruz, J. R., 1998, ‘‘Optimization and Experi-
States and Electron Affinity of Diamond 共100兲, 共111兲 and 共110兲 Surfaces,’’ ments: A Survey,’’ Appl. Mech. Rev., 51, pp. 435– 448.
Surf. Sci., 409, pp. 320–355. 关83兴 Wisitsora-at, A., Kang, W. P., Davidson, J. L., Li, Q., Xu, J. F., and Kerns, D.
关55兴 Krainsky, I. L., and Asnin, V. M., 1998, ‘‘Negative Electron Affinity Mecha- V., 1999, ‘‘Efficient Electron Emitter Utilizing Boron-Doped Diamond Tips
nism for Diamond Surfaces,’’ Appl. Phys. Lett., 72, pp. 2574 –2576. with sp共2兲 Content,’’ Appl. Surf. Sci., 146, pp. 280–287.
关56兴 Nemanich, R. J., Baumann, P. K., Benjamin, M. C., Nam, O.-H., Sowers, A. T., 关84兴 Gotoh, Y., Inoue, K., Ohtake, T., Ueda, H., Hishida, Y., Tsuji, H., and Ishikawa,
Ward, B. L., Ade, H., and Davis, R. F., 1998, ‘‘Electron Emission Properties of J., 1994, ‘‘Application of Focused Ion Beam Techniques to the Fabrication of
Crystalline Diamond and III-Nitride Surfaces,’’ Appl. Surf. Sci., 130, pp. 694 – Lateral-Type Thin-Film Edge Field Emitters,’’ Japanese Journal of Applied
703. Physics Part 2–Letters, 33, pp. L63–L70.
关57兴 Lui, W. W., and Fukuma, M., 1986, ‘‘Exact Solution of the Schrödinger Equa- 关85兴 Tamtair, F. G., Chen, L. C., Wer, S. L., Hong, W. K., Chen, K. H., and Cheng,
tion Across an Arbitrary One-Dimensional Piecewise-Linear Potential Bar- H. C., 2000, ‘‘High Current Density Field Emission from Arrays of Carbon
rier,’’ J. Appl. Phys., 60, pp. 1555–1559. Nanotubes and Diamond-Clad Si Tips,’’ J. Vac. Sci. Technol. B, 18, pp. 1207–
关58兴 Fonseca, L. R. C., von Allmen, P., and Ramprasad, R., 2000, ‘‘Numerical 1213.
Simulation of the Tunneling Current and Ballistic Electron Effects in Field 关86兴 Hatsopoulos, G. N., and Gyftopoulos, E. P., 1973, Thermionic Energy Conver-
Emission Devices,’’ J. Appl. Phys., 87, pp. 2533–2541. sion, Vol. I, MIT Press, Cambridge, MA, ch. 2.
Interface and Strain Effects on the
Thermal Conductivity of
Alexis R. Abramson Heterostructures: A Molecular
Chang-Lin Tien Dynamics Study
Arun Majumdar
e-mail: majumdar@me.berkeley.edu Molecular dynamics simulations are used to examine how thermal transport is affected by
the presence of one or more interfaces. Parameters such as film thickness, the ratio of
Department of Mechanical Engineering, respective material composition, the number of interfaces per unit length, and lattice
University of California, strain are considered. Results indicate that for simple nanoscale strained heterostructures
Berkeley, CA 94720-1740 containing a single interface, the effective thermal conductivity may be less than half the
value of an average of the thermal conductivities of the respective unstrained thin films.
Increasing the number of interfaces per unit length, however, does not necessarily result
in a corresponding decrease in the effective thermal conductivity of the superlattice.
关DOI: 10.1115/1.1495516兴
Keywords: Conduction, Heat Transfer, Microscale, Molecular Dynamics, Nanoscale,

Thin Films
Introduction decrease with decreasing period thickness from 300 Å to 45 Å

关10兴. Some investigations suggest that interface quality may also
As the demand for miniaturization of microelectronics devices
significantly affect the thermal conductivity of superlattices
increases, superior thermal control becomes even more important
关11,12兴. However, researchers have reported seemingly contradic-
to the future success of the microelectronics and other industries.
tory experimental results regarding the effect of interface rough-
Reduced feature sizes lead to an increase in power dissipation per ness. In InSb/AlSb superlattices, thermal conductivity decreases
unit area, and therefore thermal management is a major concern. as interfacial roughness becomes more pronounced 关13兴, but the
However, research in this area has shown that thermal behavior in trend may be opposite for Si/Ge superlattices 关14兴. Although the
micro/nano-sized devices often cannot be predicted by conven- effect of lattice strain on electronic properties in superlattices has
tional macroscopic theory. Nonetheless, the discovery of unique been studied extensively over the past decade 关15,16兴, there has
micro/nanoscale thermal phenomena has led to an interest in new been little or no research investigating how lattice strain may af-
devices and applications that take advantage of the unusual phys- fect thermal conductivity in heterostructures. Theoretical studies
ics, and in particular has stimulated new research in the area of 关17–19兴 have demonstrated that phonon spectra are altered by
thermoelectrics 关1,2兴. An efficient thermoelectric device must strain in superlattices, but there is conflicting evidence describing
maximize the ratio of electrical to thermal conductivity in addition the extent of the effect. Presently, questions regarding the cause of
to providing a high Seebeck coefficient. Hence, engineering a ma- the unusual and somewhat perplexing thermal behavior in super-
terial with a low thermal conductivity is one manner in which to lattices have remained unanswered. A variety of mechanisms such
improve the functionality of the device. The ability to predict, as acoustic impedance 关20兴 and phonon spectra mismatch 关21兴,
control and manipulate the thermal transport in micro/ mini-band formation and a corresponding phonon group velocity
nanostructures will most likely result in future progress in the area reduction 关22–25兴, phonon tunneling 关26兴, and interface scattering
of thermoelectrics. due to roughness, defects or dislocations 关11,27兴 have been pro-
One of the first observations of unusual thermal characteristics posed as contributors. This work further suggests that lattice
of micro/nanostructures involved the appearance of a reduced strain, commonly exhibited by symmetrically and asymmetrically
thermal conductivity in superlattices 关3,4兴, which are two- strained superlattices, may also play a significant role in affecting
dimensional structures of alternating thin films. Capinski et al. thermal transport. There is limited understanding of these phonon
关5,6兴 demonstrated that the thermal conductivity of GaAs/AlAs heat conduction mechanisms and their relation to material proper-
superlattices can be less than that of an alloy of the two materials. ties, individual film thickness, the ratio of respective material
Lee et al. 关7兴 also reported a reduced thermal conductivity asso- composition, and superlattice configuration and quality. A goal of
ciated with Si/Ge superlattices. They showed there is an increase this research is to further reflect the complexity of the situation by
in thermal conductivity with period thickness, resulting in a peak using computer simulations to reveal interesting thermal charac-
at ⬃200 Å, followed by a significant downward trend. In contrast, teristics of heterostructures.
Yamasaki et al. 关8兴 and Venkatasubramanian et al. 关9兴 discovered A variety of distinctly different theoretical and computational
the opposite trend in Bi2 Te3 /Sb2 Te3 superlattices; for periods approaches have been used to determine the effective thermal con-
shorter than 60 Å, the thermal conductivity of these heterostruc- ductivity of a superlattice structure. Many studies have employed
tures actually decreases with increasing period thickness, fol- modified bulk formulas for the lattice thermal conductivity, using
lowed by a trend of increasingly higher values. Moreover, an appropriately adapted acoustic phonon dispersion relation for
Huxtable et al. studied Si/Si0.7Ge0.3 superlattices and uncovered the structure 关22,28兴. Other investigators have applied the Boltz-
no peak in thermal conductivity; rather, they revealed a non-linear mann transport equation 共BTE兲 for phonons 关29,30兴 or Monte
Carlo simulations 关31,32兴. Often, experimentally fitted values for
the phonon relaxation time and acoustic phonon dispersion of
HEAT TRANSFER. Manuscript received by the Heat Transfer Division September 24, bulk media are still employed for these methods. By including a
2001; revision received May 13, 2002. Associate Editor: C. T. Avedisian. specularity parameter in the solution of the BTE, researchers have
been able to capture the effect of a partially diffuse interface instantaneous temperature at a point x is calculated classically
关11,30兴. This study, however, takes a different approach by using from the planar average of the kinetic energy of the atoms in each
classical nonequilibrium molecular dynamics 共MD兲 simulations to y-z plane such that
investigate the effective thermal conductivity of bi-material struc-
N
tures including simple representations of superlattices. Due to its m 兺 i x v 2i
classical nature, this technique cannot capture effects related to T x,inst⫽ (1)
3N x k B
vibrational energy quantization that control heat capacity at tem-
peratures much below the Debye temperature. However, phonon
where m is the mass and v the speed of the atom, N x is the total
characteristics such as dispersion, interface reflection, confine-
number of atoms in the specific y-z plane and k B is the Boltzmann
ment and tunneling are accurately captured without assuming em-
constant. After the simulation has reached steady state, this tem-
pirically derived physical values, provided the interatomic poten-
perature is averaged over an adequate time period, and this time-
tial is well known. Employing MD for this type of analysis is a
averaged value is used in the calculation of effective thermal con-
less common approach, but previous research on two 关33兴 and
ductivity. An appropriate imposed heat flux is chosen for each
three-dimensional 关34兴 superlattices using non-equilibrium and
structure to ensure that the same temperature difference 共cold bath
equilibrium MD techniques, respectively, have uncovered some
at ⬃60 K and hot bath at ⬃90 K兲 is maintained across all struc-
interesting observations regarding thermal transport in
tures. Argon 共Ar兲 and krypton 共Kr兲 are chosen as the materials for
superlattices.
the analysis whose Debye temperatures are approximately 92 K
There are various factors that may potentially influence the ef-
and 72 K and melting temperatures are approximately 84 K and
fective thermal conductivity of a heterostructure, including thin
116 K, respectively. To avoid melting conditions, the Kr film, with
film thickness and the ratio of the respective material composition,
the higher of the two melting temperatures, is always nearest the
numbers of interfaces, period thickness, and strain throughout the
hot bath, and the simulation checks to ensure that melting does not
structure. This study offers a systematic examination of how these
occur. It is important to note that across this temperature range,
factors might influence the effective thermal conductivity and
molecular dynamics simulations have revealed that thermal con-
may, therefore, help to predict how to better engineer these struc-
ductivity is not a strong function of temperature 关36兴, and there-
tures for use in thermoelectric devices or other related applica-
fore, a linear temperature profile is expected, provided the thick-
tions. Lennard-Jones materials are simulated, and although their
ness of the film/structure is greater than the phonon mean free
applicability in real applications is limited, the results should be
path. The actual heat flux across each plane is calculated from
used as a benchmark from which further study of other material
particle positions and velocities using the equations of Irving and
systems will benefit.
Kirkwood 关37兴. At steady state, a time average of this calculated
flux is then spatially averaged across the structure to provide an
Computational Model effective heat flux. This value can be compared with the imposed
The computational model is based on the classical MD subrou- heat flux to assess the degree to which heat spreading may occur
tines given by Allen and Tildesley 关35兴 and on the non- in the structure in the y- and z-directions. The total simulation
equilibrium simulations of thin films performed by Lukes et al. time is determined by estimating the thermal diffusion time for
关36兴. However, essential modifications to account for more than a each structure and using a run time approximately 5–20 times this
single solid material have been made. A schematic of the three- characteristic time scale. Femtosecond time steps are used, and
dimensional molecular dynamics computational structure is the total simulation time ranges from 1–30 nanoseconds. The
shown in Fig. 1. Boundary conditions consist of an applied heat probable errors of the planar temperature, the effective heat flux
flux in the x-direction and are periodic over eight planes in the and the effective thermal conductivity are determined in the man-
other two dimensions 共the y and z-directions兲. The periodic ner outlined in the appendix and in more detail by Lukes et al.
boundary conditions simulate an infinite film in these directions, 关36兴. For large structures 共⭓24 unit cells in the x-direction兲, the
although their use does remove the existence of long-wavelength calculated effective heat flux differed by 8 –12 percent of the im-
fluctuations in the system. Nonetheless, certain repeated simula- posed heat flux, while for smaller structures, the value ranged
tions with up to four times the number of planes in each direction between 3– 8 percent. The errors in thermal conductivity reported
demonstrated no significant deviation from the results presented throughout this work are associated with the convergence criteria
here. Fixed walls are used at the ends of the structure in the of the simulation. For the sake of computational time, the in-
x-direction to discourage evaporation of the constituent atoms to creased accuracy of the calculation is sacrificed for large struc-
vacuum. The heat flux is simulated by adding energy to the four tures that may require unreasonably long simulation times. Hence,
planes adjacent to the leftmost fixed walls 共the hot bath兲 and re- runs that are too short may be to blame for larger errors, approxi-
moving the same amount of energy from the four planes adjacent mately 15 percent in some cases, associated with simulating thick
to the rightmost fixed walls 共the cold bath兲. The atoms between structures.
these two baths are referred to as ‘‘regular’’ atoms. The effective Utilizing a true representation of the intermolecular potential in
cross-plane thermal conductivity of a structure is simply deter- these MD simulations is critical for ensuring that the physics un-
mined by the quotient of the heat flux and the temperature differ- derlying the micro/nanoscale thermal phenomena is accurate.
ence across the regular atoms in the structure, multiplied by the Therefore, only Lennard-Jones 共LJ兲 solids, for which the intermo-
total thickness over which the thermal gradient is imposed. The lecular potential is well established, are chosen. The LJ 12-6 po-
tential is
⌽ 共 r i j 兲 ⫽4␧ 冋冉冊冉冊册
␴ LJ
rij
12
⫺
␴ LJ
rij
6
(2)
where r i j is the distance between atoms i and j, ␧ is the well depth

of the potential and ␴ LJ is the equilibrium separation parameter.
The force experienced by an atom due to the presence of another
atom is simply the first derivative of Eq. 共2兲. Using Newton’s
equations of motion, a relationship between force and acceleration
can be established. Therefore, an indirect relation exists between
the interatomic potential and the vibrational motion of the atoms
Fig. 1 Schematic of molecular dynamics simulation cell in the system. Using a finite difference approach, or the com-
monly employed velocity Verlet algorithm 关38兴, the acceleration
may be expressed in terms of velocity, position and time, and the
solution then marches accordingly in time.
The phonon mean free path is generally less than ⬃2 nm in LJ
materials 关39兴 as compared to ⬃100 nm in ‘‘real’’ engineering
materials such as semiconductors. Consequently, a much shorter
simulation time is required to accurately capture the physics. To
further reduce computation time, the common convention of using
a cutoff radius outside of which atomic interactions are negligible
is employed. Argon 共Ar兲 and krypton 共Kr兲 are chosen as the ma-
terials for the analysis because their interatomic potentials are
well known. Both of these materials have a face-centered cubic
unit cell 共UC兲 structure, and therefore, one UC along the
x-direction corresponds to two planes of atoms. Lorentz-Berthelot
mixing rules 关40兴 are used to account for interaction between Ar
and Kr atoms. For simulations of bi-material films, the non-
dimensional units are based on the parameters for Kr. For some Fig. 2 One-dimensional temperature distribution from an MD
simulations, a structure with a coherent interface is desired. To simulation of a bi-material film composed of 16 unit cells Kr
achieve this, the equilibrium interatomic distance is chosen for „solid diamond… adjacent to 16 unit cells Ar „solid circle…. The
both materials to be the same and that of Kr (a⫽5.64 Å). How- temperature jump indicates the presence of interfacial thermal
ever, the equilibrium interatomic distance for Ar is actually resistance. The inset illustrates the instantaneous and best-fit
slopes of the lines corresponding to the two materials normal-
slightly less (a⫽5.31 Å), thereby forcing the Ar film to be in
ized using the best-fit slope of Kr.
tension. By varying parameters such as structure and individual
film thicknesses, the ratio of the respective material composition,
the number of interfaces, and lattice strain, the simulations per-
formed here investigate the reduction in the effective thermal con- of regular atoms adjacent to the hot bath. The best-fit lines for
ductivity of various bi-material films and simplified superlattices. both the Kr and Ar halves are shown, and the respective slopes are
Computational power limits the capability of simulating complex different because the thermal properties of the materials vary.
superlattices, but simple structures with up to twelve periods are From graphs such as this one, the effective cross-plane thermal
analyzed. conductivity is simply determined by the ratio of the effective heat
For all cases, the effective thermal conductivity of the entire flux and the temperature difference between the first and last plane
structure is compared to an ‘‘average’’ thermal conductivity using of regular atoms 共using values from the best fit lines as shown兲
a simple series resistance calculation involving the respective re- multiplied by the total thickness of the regular atoms in the struc-
sistances of the individual Ar and Kr thin films in their natural ture. Since the MD simulation of a single film calculates a non-
lattice states such that dimensional thermal conductivity for any Lennard-Jones solid, the
R ave⫽R Kr⫹R Ar dimensionalized values for individual films of Kr or Ar may be
(3) determined using the results from a single simulation and apply-
共 t Kr⫹t Ar兲 k Krk Ar ing the respective transformation as given by Lukes et al. 关36兴.
k ave⫽ Comparing the two transformations, the dimensionalized thermal
t Krk Ar⫹t Ark Kr
conductivity of a Kr film is theoretically about 70 percent that of
where R is the resistance, k is the thermal conductivity, and t is the Ar for any MD simulation. This is also evident from the simula-
individual film thickness. The subscript ave refers to the average tions, and the normalized slopes of the two best-fit lines are illus-
property, while Kr and Ar describe the respective individual prop- trated on the inset to Fig. 2.
erties of each material. This equation represents the expected ef- To initially assess the effect of a single interface on various thin
fective thermal conductivity if there were no interfacial thermal film configurations, the thermal conductivities of lattice-strained
effects present to alter the property. Researchers often compare bi-material films of differing overall thicknesses but with equal
thermal conductivity of superlattices to an average of unstrained individual thicknesses of Kr and Ar are analyzed and shown in
bulk values, assessing that the reduction in the thermal conductiv- Fig. 3. The lattice strain is imposed by initially forcing the Ar
ity of the superlattice is due to a mechanism such as acoustic
impedance mismatch. However, such a comparison may not be
sufficient to further one’s physical understanding of interfacial
thermal effects since thin film and bulk properties may be mark-
edly different. Therefore, this study utilizes an average of the thin
film thermal conductivities at the appropriate corresponding tem-
peratures for a more appropriate comparison to reveal that the
interface can play a substantial role in altering thermal transport.
Note that the average thermal conductivity is estimated using un-
strained thermal conductivities of the individual films, which al-
lows one to additionally compare the effect of strain in certain
simulations.

Effect of Single Interface. An example of a temperature
profile for a bi-material film 共16 UC in the x-direction of Kr ad-
jacent to 16 UC of Ar兲 is illustrated in Fig. 2. The temperature
profiles for the simulations were reasonably linear for each indi- Fig. 3 Effective thermal conductivity of strained bi-material
vidual film, and a temperature jump occurred at the interface be- KrÕAr films as a function of increasing thickness „solid square…
tween materials. Note that the x-axis corresponds to the atomic and compared with thin film average „solid circle…. Note m
plane number, and plane number 1 corresponds to the first plane Än.
Fig. 5 Effective thermal conductivity of simple asymmetrically
strained superlattices as a function of numbers of interfaces
per unit thickness „solid square… and compared with thin film
Fig. 4 Effective thermal conductivity of strained bi-material average „solid circle…
KrÕAr films as a function of individual film thickness ratio, but
the same overall thickness „solid square… and compared with
thin film average „solid circle…. Note n Ä24À m .
differences in thermal conductivities due to the disparity between
film thicknesses. For example, in the 4 UC/20 UC averaged case,
the structure is composed of 83 percent Ar, and the Ar film is
atoms to lattice spacings equal to that of the Kr atoms, imposing a ‘‘thick’’. Together, these effects result in a high thermal conduc-
coherent interface. Note that the structure consists of only two tivity. However, for the 20 UC/4 UC averaged case, the structure
films and is not representative of a superlattice. The standard error is composed of only 17 percent Ar, but the thickness of the Kr film
analysis 关36兴 for these simulations resulted in errors in the effec- is large. Even though the thermal conductivity of Kr is theoreti-
tive thermal conductivity between 8 and 15 percent of the calcu- cally 70 percent that of Ar, a 20 UC film of Kr has a thermal
lated values. Error bars are shown for the first and last points only conductivity similar to that of a 4 UC film of Ar, and there is a
for clarity. The two different trends correspond to the MD simu- flattening out of values rather than a continued decrease. There is,
lation for the bi-material films 共bottom trend兲 and the calculation albeit questionably due to the relative magnitude of the results in
of the ‘‘average’’ thermal conductivity 共top trend兲. The latter as- Fig. 4, a corresponding trend for the case of the bi-material film
sumes a simple series resistance analysis given by Eq. 共3兲, using 共bottom兲. There is no conclusive evidence from this study that
the MD calculated values for the individual film thermal conduc- variation in the ratio of the constituent materials influences inter-
tivities at the appropriate corresponding temperatures. The first facial thermal effects. Nonetheless, it is important to consider this
observation made from the data is that as individual thin film ratio in the design of engineering structures, by accounting not
thickness increases from 4 UC to 32 UC, the effective thermal only for expected differences due to a disparity between thermal
conductivity also appears to increase. The phonon mean free path properties, but also for thin film boundary scattering effects if one
of Ar is estimated to be approximately 2–3 UC 关39兴共and is as- or both of the films are of the order of or less than the phonon
sumed to be of the same order for Kr兲, and therefore boundary mean free path.
scattering is dominant for very thin films. Consequently, as film
thickness increases, boundary effects become less pronounced, Effect of Multiple Interfaces. In engineering applications,
and thermal conductivity also increases. However, even for the structures consisting of multiple interfaces are often of great con-
thickest films, the thermal resistance due to interfacial effects and cern. However, there is limited understanding of how to best de-
strain still cause a disparity between the thin film average and sign a multilayer structure for superior control of thermal trans-
bi-material film values. Using the same series resistance formula- port characteristics. Presently, a general rule of thumb relies on
tion of previously reported experimental results for bulk Kr 关41兴 the hypothesis that increasing the number of interfaces per unit
and Ar 关39,41兴, the effective thermal conductivity calculates to length will result in more phonons being reflected and therefore, a
approximately 0.30 W/mK, which is comparable to the averaged reduction in the effective thermal conductivity of the structure.
MD results for the thicker films. Moreover, a comparison of the However, some researchers have hypothesized that mini-band for-
two trends illustrates that the presence of a single interface results mation results in an overall decrease in the phonon group velocity,
in a reduction of the thermal conductivity by a factor of ⬃2. This thereby causing a reduction in the thermal conductivity 关22–25兴.
discrepancy may be attributed to one or more of the thermal If this were true, increasing the numbers of interfaces per unit
mechanisms responsible for interfacial thermal resistance except length would effectively decrease the period thickness, resulting
for the following: 共i兲 interface scattering due to roughness, since in fewer locations along the dispersion relation where the group
the interface is atomically smooth, and 共ii兲 mini-band formation, velocity approaches zero and a corresponding increase in thermal
since there is no additional periodicity. conductivity. Thus, if both mechanisms were significantly present,
To assess the impact of the thickness ratio of the films, the they would compete against each other to affect thermal transport.
effective thermal conductivities of lattice-strained bi-material For this study, simple superlattices, all approximately 75 nm thick
films of the same overall thickness but varying thickness ratios are 共48 UC兲, are used to assess the effect of systematically increasing
compared in Fig. 4. The standard error analysis for these simula- the number of interfaces per unit length. The results for superlat-
tions resulted in errors between 10 and 13 percent of the calcu- tice configurations consisting of a single interface and up to 23
lated values. For the series resistance calculation 共top trend兲, as interfaces 共not including the boundaries兲 are shown in Fig. 5. This
the amount of Ar is reduced in the structure, there is a correspond- corresponds to structures comprised of one, two, four, six, eight
ing reduction in the overall thermal conductivity since the thermal and twelve periods but with the same overall thickness. The stan-
conductivity of Ar is approximately 1.4 times that of Kr. However, dard error analysis for these simulations resulted in errors in the
the decline flattens out for the 16 UC Kr/8 UC Ar and 20 UC Kr/4 effective thermal conductivity between 8 and 10 percent of the
UC Ar structures. It is important to recognize that there are two calculated values. The top trend in the figure, indicating the thin
competing effects here: the fraction of Ar in the structure and the film average, demonstrates again that as the individual film thick-
ness decreases 共as it does with increasing number of periods/
interfaces兲, the thermal conductivity decreases. However, the su-
perlattice thermal conductivity 共bottom trend兲 does not
significantly change over the range investigated here. Nonethe-
less, there appears to be a larger disparity between the thin film
average and the superlattice thermal conductivity for smaller num-
bers of periods. This may imply that for film thicknesses much
greater than the dominant phonon wavelengths, a single interface
and the effect of strain may be responsible for a considerable
decrease in thermal conductivity. However, adding more inter-
faces while maintaining the same overall thickness will not nec-
essarily result in a large decrease in thermal conductivity if the
individual film thicknesses become very small. When the thin
films are of the order of the dominant phonon wavelengths, pho-
non tunneling may counteract any additional interface reflection
effects that would otherwise result with the addition of more in-
terfaces. Furthermore, mini-band formation may emerge when
multiple periods are present, acting to further counteract any in-
crease in interface reflection. Interestingly, some analytical predic-
tions indicate that the thermal conductivity in superlattices should
decrease with increasing period thickness due to a corresponding
decrease in phonon tunneling, and then approach a constant value
at a critical thickness 关21,42兴. However, these analytical ap-
proaches cannot entirely capture competing interfacial thermal re-
sistance effects. There is conflicting experimental evidence that
affirms but also contradicts the finding that thermal conductivity
decreases with increasing period thickness 关7–9兴, but the super- Fig. 6 „a… initial and „b… final positions of Kr „light circles… and
lattices in the referenced experiments are of markedly different Ar „dark circles… atoms for a molecular dynamics simulation of
materials, qualities and configurations. Certain factors such as a bi-material film with a semi-coherent or relaxed interface. The
phonon relaxation time variation, phonon coherence and interface initial conditions are set to the exact lattice parameters of Kr
and Ar. After the simulation, the atoms rearranged themselves
scattering due to roughness may differ significantly among such that the lattice parameter at the interface is approximately
samples and are often too complicated to evaluate, making an the average of a Kr and a Ar . The interatomic distance grows
accurate comparison of results difficult. Although MD inherently smaller for the Ar atoms and larger for the Kr atoms away from
accounts for many of these factors, limits in computational power the interface.
did not permit a simple analysis of superlattices with a greater
number of interfaces. Future analysis of larger superlattices may
reveal interesting characteristics. For many of the superlattices
fabricated and proposed as potential thermoelectric components, constraints of the relatively small dimensions of the simulation.
the thin film thicknesses may approach the order of the dominant Consequently, it is favorable for the atomic spacing on either side
phonon wavelengths. Even though there are many complex fac- of the interface to either slightly contract 共Kr atoms兲 or expand
tors to consider, improvements in the engineering design of these 共Ar atoms兲 to align with adjacent atoms, while the atoms farthest
structures require that at the very least, the competing effects of away from the interface maintain their respective lattice param-
phonon tunneling, mini-band formation and interfacial thermal re- eters. Since the structure exhibits a smooth transition from the Kr
sistance be considered. to Ar film, the interfacial strain is minimized and there is rela-
tively no lattice strain in either film just a few atomic planes away
Effect of Strain. It is well known that a lattice mismatch at from the interface. This behavior is similar to what one would
the interface between conducting materials can influence charge expect at the interface between heterostructures of nanowires in
transport due to the presence of defects or misfit dislocations, and which there are only a relatively small number of atoms along the
due to strain-induced changes in the electronic band structure. The interface as opposed to in an actual superlattice where an interface
effects of interfacial strain on phonon transport and thermal con- may consist of many millions of atoms.
ductivity are not well understood. For the aforementioned MD The thermal conductivities of bi-material structures with re-
simulations, the lattice parameter of 5.64 Å 共that of Kr兲 was im- laxed interfaces and of differing overall thicknesses are compared
posed on both the Kr and Ar films to ensure a coherent interface. with their strained counterparts and calculated average thermal
Similar to an asymmetrically strained superlattice, the Ar film was conductivities in Fig. 7. The strained and average values for the
kept in tension while the Kr film was not strained. Allowing the thermal conductivity come directly from Fig. 3. The probable er-
natural lattice state of both materials to exist during the simula- ror of the effective thermal conductivity for the unstrained case is
tions resulted in an interesting reconfiguration of the lattice struc- approximately 10 percent for all structures. The results demon-
ture. Figure 6 illustrates the 共a兲 initial and 共b兲 final atomic lattice strate that for the MD simulations, allowing the two materials to
positions in a single x – y plane of a 12 UC Kr/12 UC Ar structure exist in their natural lattice state rather than forcing a coherent
共only the first 12 atomic planes on either side of the interface are interface results in an increase in thermal conductivity. In fact, the
shown for clarity兲. These snapshots were taken prior to imposing effective thermal conductivity of the strained structure is ⬃35
a heat flux. The natural lattice constants of the respective materials percent lower than when the entire structure is in a relaxed state.
were initially set in this simulation and were purposely chosen to Using the thermal conductivity results from the strained and re-
necessitate a large and unnatural degree of lattice mismatch. After laxed simulations along with the average calculation for the films
an adequate time period 共⬃2 ns兲, the atoms rearranged themselves given by Eq. 共3兲, the boundary resistance 共BR兲 for both the
to, presumably, their lowest energy state, and formed a semi- strained and relaxed cases can be determined and compared using
coherent or a relaxed interface to accommodate the disparity be- a simple series resistance relationship, R eff⫽RKr⫹R Ar⫹R BR .
tween lattice parameters. Even though periodic boundary condi- Consequently, the boundary resistance in the case of the strained
tions are imposed along the planes perpendicular to the interface structure is calculated to be more than three times that of the
to simulate an infinite film, the structure experiences the spatial relaxed structure. Not only does strain throughout the Ar film
Conclusions
There is limited understanding of the many complex mecha-
nisms affecting thermal transport in heterostructures with one or
multiple interfaces. This study served to assess the feasibility of
using molecular dynamics as a tool to further elucidate the effects
of varying parameters in these types of structures such as overall
thickness and the ratio of the respective material composition, the
number of interfaces per unit length, and lattice strain. The results
indicate that for simple asymmetrically strained bi-material films
where a single interface is present, the thermal conductivity may
be less than half the value of an average of two similar thin films.
Moreover, one would expect that varying the ratio of the two
materials in a simple bi-material film would also affect thermal
transport, particularly if the thermal conductivities of the two ma-
Fig. 7 Effective thermal conductivity of bi-material KrÕAr films terials are markedly different. However, when one or both of the
as a function of overall thickness „thickness ratioÄ1… for films approach the phonon mean free path, the possible competing
strained „solid square…, relaxed „solid triangle… and thin film effects of thin film boundary scattering and the ratio of film com-
average „solid circle… cases. Note m Ä n . position must be considered. Nonetheless, there is no conclusive
evidence from this study that variation in the ratio of the constitu-
ent materials significantly influences interfacial thermal effects.
There are competing theories that suggest that increasing the num-
result in a decrease in effective thermal conductivity, but the prop- ber of interfaces per unit length will influence the thermal conduc-
erty mismatch that results from the strain also appears to contrib- tivity of the structure. However, this study demonstrated that a
ute to a further reduction. Comparing the three trends in Fig. 7, it single interface results in a considerable decrease in thermal con-
seems that the strain due to forcing a coherent interface may ac- ductivity, but additional interfaces do not necessarily cause a fur-
tually be a dominant contributor to the reduction in thermal con- ther significant reduction in the property. Care must be taken to
ductivity. Reflection due to other mechanisms may only result in a account for the competing increase in effective thermal conduc-
reduction of ⬃20 percent of the average thermal conductivity tivity due to phonon tunneling in superlattices with thin films of
value as compared to a reduction of ⬃50 percent for the strained the order of or less than the dominant phonon wavelength, and it
heterostructures. is important to also examine whether mini-band formation might
Performing a simple calculation to estimate the influence of be present to counteract phonon reflection at the interface. In one
lattice strain on the thermal conductivity provides elementary in- of the first investigations into the effect of lattice strain on thermal
sight to the degree of its effect. Taking the second derivative of conductivity in heterostructures, this study revealed that the im-
the interatomic potential 关Eq. 共2兲兴 gives the spring stiffness, g posed tensile strain caused by forcing a coherent interface resulted
⫽d 2 ⌽/dr i2j of the interatomic bond. Forcing the Ar film in tension in a significant decrease in the thermal conductivity. A more ac-
to match the lattice constant of Kr, results in a strain, ␥ ⫽(a s curate theoretical analysis must quantify the effects of the im-
⫺a 0 )/a 0 ⫽6.2 percent experienced throughout the Ar film. For posed strain on the dispersion relation and phonon relaxation time
this calculation, a s is the strained lattice constant, or in this case before a thorough understanding of its effects on thermal conduc-
the lattice constant corresponding to Kr, and a 0 is the equilibrium tivity can be fully appreciated. Nonetheless, results provide some
lattice constant, or the natural lattice constant of Ar. The expres- insight into the mechanisms that may influence thermal transport,
sion for the spring stiffness as a function of lattice strain is derived and should be carefully considered in the design of superlattices
for the L-J potential to be used for the purpose of manipulation of heat flow.
g⫽g 0 冋 13 1
⫺
7 1
6 共 1⫹ ␥ 兲 14 6 共 1⫹ ␥ 兲 8 册 (4)
Acknowledgments
The authors gratefully acknowledge the financial support of the
National Science Foundation and the Department of Energy.
where the spring stiffness at equilibrium, g 0 ⫽72␧/21/3␴ 2 . Be-
cause of anharmonicity in the interatomic potential, the spring Nomenclature
stiffness of the strained lattice will be different from that of the
relaxed state, and for ␥⫽6.2 percent, g/g 0 ⫽0.21. Ar ⫽ argon
The change in spring stiffness fundamentally alters the phonon C ⫽ specific heat 共J/kg-K兲
dispersion relation, which in turn modifies the phonon group ve- Kr ⫽ krypton
locity as well as the lifetime of phonon-phonon interaction, ␶ ph . M ⫽ total number of planes
The effect of both these contributions to thermal conductivity can Nx ⫽ atomic plane number at location x
be significant. As a demonstration, consider only the effect of R ⫽ thermal resistance 共K/W兲
modifying the phonon group velocity and disregard a change in Tx ⫽ atomic plane temperature at location x
␶ ph . Then, applying kinetic theory and substituting the velocity, a ⫽ lattice constant 共䊊兲
v ⬇r i j 冑g/m, into the expression for thermal conductivity, b ⫽ slope
g ⫽ spring stiffness 共N/m兲
k ⫽ thermal conductivity 共W/m-K兲
1 2 g
k⫽ Cr ␶ (5) kB ⫽ Boltzmann’s constant (1.38⫻10⫺23 J/K)
3 i j m ph m ⫽ mass 共kg兲
q ⫽ heat flux 共W/m2兲
where C is heat capacity. Using the above expressions, the thermal rij ⫽ distance between atoms i and j
conductivity of the relaxed Ar film is reduced by ⬃80 percent t ⫽ thickness 共m兲
when a strain of 6.2 percent is imposed. However, molecular dy- v ⫽ velocity 共m/s兲
namics simulations of single Ar films in the strained and relaxed
states revealed a reduction of only ⬃30 percent in thermal con- Greek Symbols
ductivity. ␧ ⫽ well-depth parameter 共J兲
␥ ⫽ strain References
⌽ ⫽ interatomic potential 共J兲关1兴 Mahan, G., Sales, B., and Sharp, J., 1997, ‘‘Thermoelectric materials: New
␴ ⫽ probable error Approaches to an Old Problem,’’ Phys. Today, 50, pp. 42– 47.
␴ sd ⫽ probable error 关2兴 Dresselhaus, M. S., Dresselhaus, G., Sun, X., Zhang, Z., Cronin, S. B., Koga,
␴ 具典 ⫽ standard error T., Ying, J. Y., and Chen, G., 1999, ‘‘The Promise of Low-Dimensional Ther-
moelectric Materials,’’ Microscale Thermophys. Eng., 3, pp. 89–100.
␴ LJ ⫽ Lennard-Jones equilibrium separation parameter 共䊊兲关3兴 Yao, T., 1987, ‘‘Thermal Properties of AlAs/GaAs Superlattices,’’ Appl. Phys.
␶ ph ⫽ phonon relaxation time 共s兲 Lett., 51, pp. 1798 –1800.
␶ run ⫽ run time 共s兲关4兴 Weisbuch, C., and Vinter, B., 1991, Quantum Semiconductor Structures, Aca-
␶ corr ⫽ autocorrelation time 共s兲 demic Press, Boston, MA.
关5兴 Capinski, W. S., and Maris, H. J., 1996, ‘‘Thermal Conductivity of GaAs/AlAs
Subscript Superlattices,’’ Physica B, 220, pp. 699–701.
关6兴 Capinski, W. S., Maris, H. J., Ruf, T., Cardona, M., Ploog, K., and Katzer, D.
0 ⫽ equilibrium S., 1999, ‘‘Thermal Conductivity Measurements of GaAs/AlAs Superlattices
BR ⫽ boundary resistance Using a Picosecond Optical Pump-and-Probe Technique,’’ Phys. Rev. B, 59,
eff ⫽ effective pp. 8105– 8113.
关7兴 Lee, S. M., Cahill, D. G., and Venkatasubramanian, R., 1997, ‘‘Thermal Con-
f ⫽ film ductivity of Si-Ge Superlattices,’’ Appl. Phys. Lett., 70, pp. 2957–2959.
inst ⫽ instantaneous 关8兴 Yamasaki, I., Yamanaka, R., Mikami, M., Sonobe, H., Mori, Y., and Sasaki, T.,
s ⫽ strained 1998, ‘‘Thermoelectric Properties of Bi2 Te3 /Sb2 Te3 Superlattice Structures,’’
x ⫽ plane number Proceedings 17th International Thermoelectrics Conference ICT ’98, IEEE,
CA, pp. 210–213.
关9兴 Venkatasubramanian, R., 2000, ‘‘Lattice Thermal Conductivity Reduction and
Phonon Localization Like Behavior in Superlattice Structures,’’ Phys. Rev. B,
61, pp. 3091–3097.
Appendix 关10兴 Huxtable, S. T., Abramson, A. R., Majumdar, A., Tien, C. L., LaBounty, C.,
Fan, X., Zeng, G., Abraham, P., Bowers, J. E., Shakouri, A., and Croke, E. T.,
The error analysis of the effective thermal conductivity in- 2001, ‘‘Thermal Conductivity of Si/SiGe Superlattices,’’ Proceedings IMECE
volves a series of calculations that examine the statistical fluctua- ’01, ASME, New York.
tions of the simulation. Lukes et al. 关36兴 provide additional details 关11兴 Chen, G., and Neagu, M., 1997, ‘‘Thermal Conductivity and Heat Transfer in
of the error analysis. The standard deviation of the planar tem- Superlattices,’’ Appl. Phys. Lett., 71, pp. 2761–2763.
关12兴 Rosenblum, I., Adler, J., Brandon, S., and Hoffman, A., 2000, ‘‘Molecular-
perature, ␴ sd,Tx , is first used to determine the standard error of the Dynamics Simulation of Thermal Stress at the 共100兲 Diamond/Substrate Inter-
planar temperature 关35,43兴, face: Effect of Film Continuity,’’ Phys. Rev. B, 62, pp. 2920–2936.
关13兴 Borca-Tasciuc, T., Achimov, D., Liu, W. L., Chen, G., Lin, C. H., Delaney, A.,
␴ 具 Tx 典 ⫽ ␴ sd,Tx 冑2 ␶ corr / ␶ run (6) and Pei, S. S., 2001, ‘‘Thermal Conductivity of InAs/AlSb Superlattices,’’
Proceedings International Conference on Heat Transfer and Transport Phe-
where ␶ run is the run time of the simulation after reaching steady nomena in Microscale, Banff, Canada, Begell House, New York, pp. 369–371.
state, and ␶ corr is the autocorrelation time of the instantaneous 关14兴 Borca-Tasciuc, T., Liu, W. L., Liu, J. L., Zeng, Song, D. W., Moore, C. D.,
planar temperatures given approximately by the ratio of phonon Chen, G., Wang, K. L., Goorsky, M. S., Radetic, T., Gronsky, R., Sun, X., and
Dresselhaus, M. S., 1999, ‘‘Thermal Conductivity of Si/Ge Superlattices,’’
mean free path 关39兴 to the speed of sound 关44兴. For simulations Proceedings 18th International Conference on Thermoelectrics ICT ’99 IEEE,
involving bi-material structures, the average of the speeds of CA.
sound is used. The probable error associated with the slope of the 关15兴 Rieger, M. M., and Vogl, P., 1993, ‘‘Electronic-Band Parameters in Strained
best fit line of the temperature profile of the regular atoms is found Si共1-x兲Ge共x兲 Alloys on Si共1-y兲Ge共y兲 Substrates,’’ Phys. Rev. B, 48, pp.
14276 –14287.
using a weighted least squares method 关45兴, 关16兴 Tserbak, C., Polataoglou, H. M., and Theodorou, G., 1993, ‘‘Unified Approach
冉冊
to the Electronic-Structure of Strained Si/Ge Superlattices,’’ Phys. Rev. B, 47,
冪冉
M 1 pp. 7104 –7124.
兺 x⫽1
f
关17兴 Ghanbari, R. A., White, J. D., Fasol, G., Gibbings, C. J., and Tuppen, C. G.,
␴ 具2Tx 典
冊冉冊冉冊
1990, ‘‘Phonon Frequencies for Si-Ge Strained Layer Superlattices Calculated
␴bf⫽ (7)
Mf 1 Mf x2 Mf 1 2 in a Three-Dimensional Model,’’ Phys. Rev. B, 42, pp. 7033–7041.
兺 x⫽1 兺 x⫽1 ⫺ 兺 x⫽1 关18兴 Qteish, A., and Molinari, E., 1990, ‘‘Interplanar Forces and Phonon Spectra of
␴ 具2Tx 典 ␴ 具2Tx 典 ␴ 具2Tx 典 Strained Si and Ge: Ab initio Calculations and Applications to Si/Ge Superlat-
tices,’’ Phys. Rev. B, 42, pp. 7090–7096.
where x is the plane number along the x-direction and M f is the 关19兴 Sui, Z., and Herman, I. P., 1993, ‘‘Effect of Strain on Phonons in Si, Ge, and
total number of planes in the film. For bi-material films, the slope, Si/Ge Heterostructures,’’ Phys. Rev. B, 48, pp. 17938 –17953.
关20兴 Little, W. A., 1959, ‘‘The Transport of Heat Between Dissimilar Solids at Low
b f , and its respective error, ␴ b f , are calculated for each indi- Temperatures,’’ Can. J. Phys., 37, pp. 334 –349.
vidual film. For example, a structure composed of four alternating 关21兴 Stoner, R. J., and Maris, H. J., 1993, ‘‘Kapitza Conductance and Heat Flow
layers of Kr and Ar will require the calculation of the four differ- Between Solids at Temperatures From 50 to 300 K,’’ Phys. Rev. B, 48, pp.
ent slopes and their respective errors. Each individual film’s ther- 16373–16387.
关22兴 Tamura, S., Tanaka, Y., and Maris, H. J., 1999, ‘‘Phonon Group Velocity and
mal conductivity, k f , is also calculated as the quotient of the Thermal Conduction in Superlattices,’’ Phys. Rev. B, 60, pp. 2627–2630.
effective heat flux, q eff , and the respective slope, b f , to be used in 关23兴 Tamura, S., Hurley, D. C., and Wolfe, J. P., 1988, ‘‘Acoustic Phonon Propaga-
subsequent error calculations. There will also be a probable error tion in Superlattices,’’ Phys. Rev. B, 38, pp. 1427–1449.
associated with the effective heat flux, q eff 关45兴, 关24兴 Simkin, M. V., and Mahan, G. D., 2000, ‘‘Minimum Thermal Conductivity of
Superlattices,’’ Phys. Rev. Lett., 84, pp. 927–930.
冑冉兺冊
M 关25兴 Narayanamurti, V., Stormer, H. L., Chin, M. A., Gossard, A. C., and Wieg-
⳵ q eff mann, W., 1979, ‘‘Selective Transmission of High-Frequency Phonons by a
␴ q eff⫽ ␴ 具2q x 典 (8) Superlattice: The ‘‘Dielectric’’ Phonon Filter,’’ Phys. Rev. Lett., 43, pp. 2012–
x⫽1 ⳵ 具 q x典 2016.
关26兴 Chen, G., 1999, ‘‘Phonon Wave Heat Conduction in Thin Films and Superlat-
where ␴ 具 qx 典 is the standard error of the planar heat flux, q x , cal- tices,’’ ASME J. Heat Transfer, 121, pp. 945–953.
culated for each plane along the x-direction in the same manner as 关27兴 Swartz, E. T., and Pohl, R. O., 1987, ‘‘Thermal Resistance at Interfaces,’’ Appl.
Eq. 共7兲. Finally, the formulation for the total probable error of Phys. Lett., 51, pp. 2200–2202.
关28兴 Balandin, A., and Wang, K. L., 1998, ‘‘Significant Decrease of the Lattice
each film’s thermal conductivity is Thermal Conductivity Due to Phonon Confinement in a Free-Standing Semi-
冑冉冊冉冊
conductor Quantum Well,’’ Phys. Rev. B, 58, pp. 1544 –1549.
⳵k f 2 ⳵k f 2
关29兴 Chen, G., 1997, ‘‘Size and Interface Effects on Thermal Conductivity of Su-
␴kf⫽ ␴ q2 eff ⫹ ␴ 2b f (9)
⳵ q eff ⳵b f perlattices and Periodic Thin-Film Structures,’’ ASME J. Heat Transfer, 119,
pp. 220–229.
关30兴 Chen, G., 1998, ‘‘Thermal Conductivity and Ballistic-Phonon Transport in the
The largest value of ␴ k f from those calculated for each film is Cross-Plane Direction of Superlattices,’’ Phys. Rev. B, 57, pp. 14958 –14973.
used to represent the probable error of the effective thermal con- 关31兴 Peterson, R. B., 1994, ‘‘Direct Simulation of Phonon-Mediated Heat Transfer
ductivity of the bi-material structure. in a Debye Crystal,’’ ASME J. Heat Transfer, 116, pp. 815–1994.
关32兴 Mazumdar, S., and Majumdar, A., 2001, ‘‘Monte Carlo Study of Phonon the Formation of Physical Clusters of Molecules: Application to Small Water
Transport in Solid Thin Films Including Dispersion and Polarization,’’ ASME Clusters,’’ J. Chem. Phys., 76, pp. 637– 649.
J. Heat Transfer, 123, pp. 749–759. 关39兴 Dobbs, E. R., and Jones, G. O., 1957, ‘‘Theory and Properties of Solid Argon,’’
关33兴 Liang, X. G., and Shi, B., 2000, ‘‘Two-Dimensional Molecular Dynamics Rep. Prog. Phys., 20, pp. 516 –564.
Simulation of the Thermal Conductance of Superlattices,’’ Mater. Sci. Eng., A, 关40兴 Reid, R. C., Prausnitz, J. M., and Poling, B. E., 1987, The Properties of Gases
292, pp. 198 –202. and Liquids, Mc-Graw Hill, New York.
关34兴 Volz, S., Saulnier, J. B., Chen, G., and Beauchamp, P., 2000, ‘‘Molecular 关41兴 White, G. K., and Woods, S. B., 1958, ‘‘Thermal Conductivity of the Solidified
Dynamics of Heat Transfer in Si/Ge Superlattices,’’ High Temp.-High Press., Inert Gases: Argon, Neon and Krypton,’’ Philos. Mag., 3, pp. 785–797.
32, pp. 709–714. 关42兴 Bao, Y., and Chen, G., 2000, ‘‘Lattice Dynamics Study of Anisotropy of
关35兴 Allen, M. P., and Tildesley, D. J., 1987, Computer Simulation of Liquids, Heat Conduction in Superlattices,’’ Proceedings of MRS Spring Meeting,
Clarendon Press, Oxford. Symposium Z, Materials Research Society, PA.
关36兴 Lukes, J. R., Li, D. Y., Liang, X. G., and Tien, C. L., ‘‘Molecular Dynamics 关43兴 Jacucci, G., and Rahman, A., 1984, ‘‘Comparing the Efficiency of Metropolis
Study of Solid Thin-Film Thermal Conductivity,’’ ASME J. Heat Transfer, 122, Monte Carlo and Molecular Dynamics Methods for Configuration Space Sam-
pp. 536 –543. pling,’’ Nuovo Cimento, D4, pp. 341–356.
关37兴 Irving, J. H., and Kirkwood, J. G., 1950, ‘‘The Statistical Mechanical Theory 关44兴 Weast, R. C., Astle, M. J., and Beyer, W. H., eds., 1996, CRC Handbook of
of Transport Processes A. The Equations of Hydrodynamics,’’ J. Chem. Phys., Chemistry and Physics, CRC Press, Boca Raton.
18, pp. 817– 829. 关45兴 Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, B. P., 1992,
关38兴 Swope, W. C., Anderson, H. C., Berens, P. H., and Wilson, K. R., 1982, ‘‘A Numerical Recipes in FORTRAN: The Art of Scientific Computing, 2nd edi-
Computer Simulation Method for the Calculation of Equilibrium Constants for tion, Cambridge University Press, Cambridge.
Journal of
Heat Transfer Technical Notes
This section contains shorter technical papers. These shorter papers will be subjected to the same review process as that for full papers.
Wavelets and the Discrete Ordinate ⳵ I ␭⬘ ⳵ I ␭⬘

␮ ⫹␰ ⫹a ␭ •I ␭⬘ ⫽a ␭ •I ␭b
⬘ , (1)
Method for Radiative Heat ⳵x ⳵y
Transfer in a Two-Dimensional where I ␭⬘ ⫽I ␭⬘ (␭,x,y, ␮ , ␰ ), I ␭b

⬘ ⫽I ␭b
⬘ (␭,x,y), a ␭ ⫽a ␭ (␭,x,y), and
共␮,␰兲 are the direction cosines.
Rectangular Enclosure With a Nongray The nondimensional form of the above equation is obtained
Medium ⳵ I ␭⬘ * ⳵ I ␭⬘ *
␮ ⫹r• ␰ ⫹ ␬ L •I ␭⬘ * ⫽ ␬ L •I ␭b
⬘* (2)
⳵x* ⳵y*
Ye Wang and Yildiz Bayazitoglu by introducing the following nondimensional parameters.
Department of Mechanical Engineering and Materials x * ⫽x/L, y * ⫽y/H, r⫽L/H
Science, Rice University, Houston, TX 77005
␬ L ⫽a ␭ •L, ␬ H ⫽ ␣ ␭ •H
␭ * ⫽ 共 ␭⫺␭ 0 兲 /⌬␭ (3)
Introduction
I ␭⬘ * ⫽I ␭⬘ •⌬␭/ 共 ␴ T 41 兲
The mathematical model for describing a radiation field in a
participating medium is the radiative transfer equation 共RTE兲. The ⬘ * ⫽I ␭b
I ␭b ⬘ •⌬␭/ 共 ␴ T 41 兲 ,
dependent variable in the RTE, the spectral radiative intensity, is a
function of location, wavelength, and solid angle. This makes it where L and H are the width and height of the two-dimensional
very difficult to be solved, especially for multi-dimensional geom- rectangular enclosure, ␬ L and ␬ H are the optical width and optical
etry. Therefore previously the gray medium assumption was usu- height, ␭ 0 and ⌬␭ are the starting wavelength and the wavelength
ally used to simplify the problem 关1–3兴, and combined width of the absorption band, respectively, and T 1 is the tempera-
conductive-radiative heat transfer in multi-dimensional geometry ture of surface 1. Thus x * , y * 苸关 0,1兴 , ␭*苸关0,1兴.
was also studied 关4 – 6兴. In order to incorporate the nongray ef- The nondimensional radiative intensity I ␭⬘ * in the RTE Eq. 共2兲
fects, many approximate methods have been developed. One of is a function of the wavelength, location, and solid angle. Many
them is to apply the picket fence model 关7兴, where the band- successful methods to solve the location dependent problems
distributed radiative properties are approximated by a box model. 共such as finite difference and finite element method兲 and the solid
Another approach is the weighted-sum-of-gray-gases model 关8兴, angle dependent problems 共such as DOM and P-N approxima-
where the nongray gas is replaced by a number of gray gases. tion兲 already exist. Studies show that the absorption coefficients of
Other researchers 关9–10兴 introduced the wavelet expansion to most participating media are strongly dependent on wavelength. A
evaluate the spectral radiative intensity in their one-dimensional similar behavior is expected for the radiative intensity. Since
nongray analyses. This paper will extend the use of wavelets wavelet is relatively new and very useful tool for function ap-
in the solution of nongray problems 关10兴 to the two-dimensional proximation, the wavelet expansion in the wavelength domain is
geometries. The results of pure radiation without internal heat
generation will be presented for nongray medium in a rectangular
enclosure.
Numerical Analysis
In general, the governing equations for radiative heat transfer in
the field of a participating medium are the radiative transfer equa-
tion and the energy conservation equation. By simultaneously
solving these two-coupled equations, the temperature distribution
and heat flux in the field can be obtained. For simplicity, the
absorbing, emitting, and nonscattering medium in a two-
dimensional rectangular enclosure is considered here, as shown in
Fig. 1. The surfaces of the enclosure are assumed to be black and
isothermal.
1 Radiative Transfer Equation. The RTE in a two-
dimensional rectangular enclosure is
8
HEAT TRANSFER. Manuscript received by the Heat Transfer Division February 5, Fig. 1 Geometry and coordinate system of two-dimensional
1999; revision received April 16, 2001. Associate Editor: P. Menguc. rectangular enclosure
introduced to evaluate the wavelength dependence of radiative merical methods such as finite volume 关13兴 when boundary con-
intensity. ditions are given.
Applying the discrete wavelet transform method 关11兴, the radia- Boundary conditions for the DOM are generated by expressing
tive intensity is first expanded in terms of the wrapped Daubechies the intensity leaving the surface along the ordinate direction m as
wavelet W i as, the sum of emitted and reflected intensities. In a two-dimensional
N rectangular enclosure the boundary conditions are the following:
I ␭⬘ * 共 ␭ * ,x * ,y * , ␮ , ␰ 兲 ⫽ 兺 c 共 x * ,y * , ␮ , ␰ 兲 •W ,
i i (4)
at surface 1 ⬘ * ⫽␧I ␭b
I ␭m ⬘*⫹
1⫺␧
兺 w 兩␰ 兩I ⬘* ␰ m ⬎0
␲ ␰ m ⬘ ⬍0 m ⬘ m ⬘ m ⬘
i⫽1
where (11a)
W 1 ⫽1 (5a) 1⫺␧
W i ⫽2 •W 共 2 ␭ * ⫺k 兲
j/2 j
i⫽2,3, . . . ,N. (5b)
I ␭m ⬘*⫹ 兺 w 兩␰ 兩I ⬘*
␲ ␰ m ⬘ ⬎0 m ⬘ m ⬘ m ⬘
␰ m ⬍0
(11b)
Here j and k are integers and describe the scaling and dilation of
1⫺␧
兺
the wavelets, respectively. Daubechies wavelets are a set of com-
I ␭m ⬘*⫹ w 兩␰ 兩I ⬘*
pact supported base functions. They constitute an orthonormal ␲ ␮ m ⬘ ⬍0 m ⬘ m ⬘ m ⬘
system, that is,
冕 1 ␮ m ⬎0 (11c)
W i •W j •d␭ * ⫽ ␦ i j , (6)
1⫺␧
0
I ␭m ⬘*⫹ 兺 w 兩␰ 兩I ⬘*
␲ ␮ m ⬘ ⬎0 m ⬘ m ⬘ m ⬘
where ␦ is Kronecker ␦-function.
Now we substitute the expansion Eq. 共4兲 back into the RTE Eq. ␮ m ⬍0 (11d)
共2兲 and apply the Galerkin method, considering the orthonormal
property Eq. 共6兲 of the wavelet functions. A set of equations about Since all the surfaces are assumed to be black, the boundary
the expansion coefficients, c i (x * ,y * , ␮ , ␰ ), are created as conditions come down to following:
冕
N
1 ⬘ * ⫽I ␭b,w
I ␭m ⬘* . (12)
dc j dc j
␮
dx *
⫹r• ␰ ⫹ c•
dy * i⫽1 i 兺 0
␬ L •W i •W j •d␭ * Now we apply Galerkin method to the boundary conditions by
substituting the expansion Eq. 共4兲 into Eq. 共12兲, multiplying by
⫽ 冕 0
1
⬘ * •W j •d␭ * ,
␬ L •I ␭b (7)
the wavelet bases on both sides, and integrating over the wave-
length domain. With the consideration of the orthonormal prop-
erty of wavelet functions, a set of boundary conditions for Eq.
where j⫽1,2, . . . ,N. 共10兲 can be obtained as
In matrix form, it becomes
␮
dC
⫹r• ␰
dC
⫹A•C⫽B (8)
c j,m ⫽ 冕 1
0
⬘ * •W j •d␭ * ,
I ␭b,w (13)
dx * dy *
where j⫽1,2, . . . ,N.
where The expansion coefficients in a particular direction, c i,m , can
be solved from Eq. 共10兲 when the temperature distribution of the
C⫽ 关 c 1 共 x * ,y * , ␮ , ␰ 兲 , . . . ,c N 共 x * ,y * , ␮ , ␰ 兲兴 T (9a) medium is given 共since A and B in Eq. 共9b,c兲 are functions of the
冕 1 temperature distribution兲. Once the expansion coefficients are de-

A i, j ⫽ ␬ L •W i •W j •d␭ * , i, j⫽1,2, . . . ,N (9b) termined, the radiative intensity and heat flux in the direction m
0 can be calculated as follows:
B⫽ 冋冕 0
1
⬘ * •W 1 •d␭ * , . . . ,
␬ L •I ␭b 冕 1
0
册
⬘ * •W N •d␭ * .
␬ L •I ␭b
T
*⫽
Im 冕
0
1
⬘ * •d␭ * ⫽c 1,m
I ␭m (14)
(9c) M M
The unknown vector C here is not dependent on the wavelength. It

can be solved with any numerical method developed for gray
*⫽
q r,x 兺
m⫽1
w m • ␮ m •I m
*⫽ 兺
m⫽1
w m • ␮ m •c 1,m (15a)
media. M M
Among the numerical methods developed for solving gray me-
dium radiative heat transfer, the DOM is considered to be a prom-
*⫽
q r,y 兺
m⫽1
w m • ␰ m •I m
*⫽ 兺
m⫽1
w m • ␰ m •c 1,m (15b)
ising one with high accuracy and little computer time consump-
tion. We use this method in this work to solve the set of equations and the divergence of the radiative heat flux is
冉冊
about the expansion coefficients C, which are the functions of M
⳵ q r,x
* ⳵ q r,y
* ⳵ c 1,m ⳵ c 1,m
location and solid angle.
Applying the DOM, Eq. 共8兲 is replaced by a set of equations of
ⵜ•q r* ⫽
⳵x*
⫹r ⫽ w ␮ 兺
⳵ y * m⫽1 m m ⳵ x *
⫹r• ␰ m
⳵y*
冉冊
discrete ordinate directions, M N
␮m
dC m
⫹r• ␰ m
dC m
⫹A•C m ⫽B, (10)
⫽
m⫽1
兺 w m B 1⫺ 兺A
i⫽1
1i •c i,m . (16)
dx * dy *
Note that these are values inside the absorption band. Outside the
where C m ⫽ 关 c 1,m (x * ,y * , ␮ m , ␰ m ), . . . ,c N,m (x * ,y * , ␮ m , ␰ m ) 兴 T , band, the radiative intensity and heat flux will not change through-
with m⫽1,2, . . . ,M , A and B are as above in Eq. 共9b,c兲. The out the medium since no absorption occurs there.
selection of the ordinate directions ( ␮ m , ␰ m ) and their quadrature
weights w m was given in 关12兴. Equation 共10兲 is a set of differential 2 Energy Equation. For the case of pure radiation, the en-
equations about the coefficients c i,m and can be solved with nu- ergy equation is
ⵜ•q r ⫽q ⵮ . (17) Table 2 Nondimensional surface heat fluxes q r Õ ␴ T 42 in two-
dimensional rectangular enclosures with nongray medium CO2
Nondimensionalize the above equation as for various aspect ratios „Case 1…
ⵜ•q r* ⫽0 if q ⵮ ⫽0 (18a)
ⵜ•q r* ⫽1 if q ⵮ ⫽0, (18b)
and r⫽L/H is the aspect ratio where L and H are width and
height of two-dimensional enclosure, respectively. The dimen-
sionless parameters are
q r* ⫽q r / 共 ␴ T 41 兲 if q ⵮ ⫽0
(19)
q r* ⫽q r / 共 q ⵮ •L 兲 if q ⵮ ⫽0.
Table 1 Convergence of nondimensional temperature distri-

bution „ E À E 2 …Õ„ E 1 À E 2 … at centerline in two-dimensional
square enclosure filled with nongray medium CO2 „Case 1…
Table 3 Nondimensional surface heat fluxes q r Õ ␴ T 42 in two-

dimensional rectangular enclosures with nongray medium CO2
for various aspect ratios „Case 2…
Fig. 2 Nondimensional temperature distribution at centerline

in two-dimensional rectangular enclosure filled with nongray
medium CO2 for various aspect ratios „Case 1…
Fig. 3 Nondimensional temperature distribution at centerline

in two-dimensional rectangular enclosure filled with nongray
medium CO2 for various aspect ratios „Case 2…
To solve the RTE and energy equation simultaneously, the Nomenclature
modified quasi-linearization algorithm 共MQA兲关14兴 is applied by
treating the energy equation as the constraint condition for the a ␭ ⫽ spectral absorption coefficient
RTE. c i ⫽ wavelet expansion coefficients of spectral radiative
In summary, the over-all solution procedure is as follows: 共1兲 intensity I ␭⬘
Assume the initial temperature distribution; 共2兲 calculate the ab- d ⫽ spectral line spacing in Eq. 共20兲
sorption coefficient a ␭ ; 共3兲 calculate A and B according to expres- E ⫽ emissive power (⫽ ␴ T 4 )
sion 共10b,c兲; 共4兲 solve Eq. 共10兲 for the coefficients c i,m ; 共5兲 Cal- H ⫽ height of the two-dimensional rectangular enclosure
culate the radiative intensity and heat flux by expression 共14兲 and I ⫽ radiative intensity
共15兲; 共6兲 if the energy equation 共18兲 is satisfied, stop; otherwise, I ␭⬘ ⫽ spectral radiative intensity in a particular direction
adjust the temperature distribution; and 共7兲 repeat steps 共2兲 to 共6兲 ⬘ ⫽ spectral blackbody intensity in a particular direction
I ␭b
until the energy equation is satisfied. L ⫽ width of the two-dimensional rectangular enclosure
M ⫽ number of discrete ordinates in DOM
N ⫽ number of wavelet expansion terms of radiative in-
tensity
Results and Discussion q r ⫽ radiative heat flux
A square enclosure filled with a nongray medium is considered. q ⵮ ⫽ internal heat generation rate
It is assumed that surface 1 is hot at temperature T 1 , the other r ⫽ aspect ratio of two-dimensional rectangular enclosure
three surfaces are cold at zero temperature. No heat generation (⫽L/H)
takes place inside the medium, so Eq. 共18a兲 is the energy equation S ⫽ mean line intensity in Eq. 共20兲
where the nondimensional parameter is the emissive power of the T ⫽ temperature
hot wall. In this work, as an example, CO2 is chosen and its 4.3 T i ⫽ temperature of surface i(i⫽1⫺4)
␮m absorption band is studied. The absorption coefficient is W i ⫽ wrapped Daubechies wavelet functions
evaluated by w i ⫽ quadrature weight of discrete ordinate ␻ i in DOM
x,y ⫽ coordinates
sinh共 2 ␲ • ␥ /d 兲
a ␭ ⫽ 共 ␳ •S/d 兲 , (20) Greek Symbols
cosh共 2 ␲ • ␥ /d 兲 ⫺1
␥ ⫽ line half width in Eq. 共20兲
where ␳ is density, S/d is mean-line-intensity-to-spacing ratio, and ␦ i j ⫽ Kronecker ␦-function
␥ /d is line-width-to-spacing ratio for the considered narrow band. ␧ ⫽ emissivity
The parameters S/d and ␥ /d incorporate the effects of wave- ␬ H ⫽ optical height for two-dimensional rectangular enclo-
length, temperature, and pressure and can be found in 关15兴. sure
The examples considered are ␬ L ⫽ optical width for two-dimensional rectangular enclo-
sure
case 1: T 1 ⫽1500 K, T 2 ⫽400 K, ␭ ⫽ wavelength
␭ 0 ⫽ start wavelength of the absorption band
P CO2⫽0.2 atm, P N2⫽10 atm, H⫽1 cm ⌬␭ ⫽ wavelength width of the absorption band
␮,␰,␩ ⫽ direction cosines
␳ ⫽ density
case 2: T 1 ⫽1500 K, T 2 ⫽400 K, ␴ ⫽ Stefan-Boltzmann constant
␻ ⫽ solid angle
P CO2⫽1.0 atm, P N2⫽0 atm, H⫽10 cm
Subscripts
These examples are chosen because there are one-dimensional H ⫽ height
results to compare with. L ⫽ width
The convergence trend of the centerline temperature distribu- m,m ⬘ ⫽ directions of the discrete ordinates
tion for case 1 for an increasing number of wavelet expansion r ⫽ radiative
terms is presented in Table 1. Considering the current limit of w ⫽ wall
computer capacity, all subsequent results for nongray medium are ␭ ⫽ spectrally dependent
obtained with N⫽16. The temperature distributions at the center- Superscripts
line are shown in Figs. 2 and 3. The results for the limit r→⬁ are
taken from one-dimensional case 关10兴. The results follow the ex- * ⫽ nondimensional quantity
pected trend as the aspect ratio r increases. The nonsmoothness
for optically thin medium in Fig. 2 is due to the ray effect that is References
inevitable while using discrete ordinate method. The surface heat
关1兴 Modest, M. F., 1975, ‘‘Radiative Equilibrium of a Gray Medium in a Rectan-
fluxes are given in Tables 2 and 3. Compared to the case of trans- gular Enclosure Bounded by Gray Walls,’’ J. Quant. Spectrosc. Radiat. Transf.,
parent gas, the existence of the absorbing gas results in the de- 15, p. 445.
creases of surface heat flux. 关2兴 Yuen, W. W., and Wong, L. W., 1984, ‘‘Analysis of Radiative Equilibrium in a
The present work successfully solved the two-dimensional non- Rectangular Enclosure With Gray Medium,’’ ASME J. Heat Transfer, 106, p.
433.
gray radiation problem. The same method can be easily extended 关3兴 Yuen, W. W., and Ho, C. F., 1985, ‘‘Analysis of Two-Dimensional Radiative
to three-dimensional problems. Conclusively, the introduction of Heat Transfer in a Gray Medium With Internal Heat Generation,’’ Int. J. Heat
wavelet functions makes it possible to study radiative transfer Mass Transf., 28, p. 17.
through a nongray medium with an arbitrary distribution of the 关4兴 Ratzel, A. G., 1981, ‘‘P-N Differential Approximation for Solution of One- and
Two-Dimensional Radiation and Conduction Energy Transfer in Gray Partici-
absorption coefficient. pating Media,’’ Ph.D. dissertation, Department of Mechanical Engineering,
University of Texas, Austin.
关5兴 Yuen, W. W., and Takara, E. E., 1988, ‘‘Analysis of Combined Conductive-
Radiative Heat Transfer in a Two-Dimensional Rectangular Enclosure With a
Acknowledgment Gray Medium,’’ ASME J. Heat Transfer, 110, p. 468.
关6兴 Kim, T. Y., and Baek, S. W., 1991, ‘‘Analysis of Combined Conductive and
This work was in part supported by Texas Advanced Technol- Radiative Heat Transfer in a Two-Dimensional Rectangular Enclosure Using
ogy Program under Grant No. 003604-041. the Discrete Ordinates Method,’’ Int. J. Heat Mass Transf., 34, pp. 2265.
关7兴 Modest, M. F., and Sikka, K. K., 1992, ‘‘The Stepwise Gray P-1 Approxima- study its effect on the system performance, but the detailed analy-
tion for Multi-Dimensional Radiative Transfer in Molecular- Gas-Particulate
Mixtures,’’ J. Quant. Spectrosc. Radiat. Transf., 48, pp. 159.
sis on heat transfer aspects was not carried out. Hence, an attempt
关8兴 Modest, M. F., 1991, ‘‘The Weighted-Sum-of-Gray-Gases Model for Arbitrary has been made in the present study to improve the heat transfer
Solution Methods in Radiative Transfer,’’ ASME J. Heat Transfer, 113, pp. coefficient in coaxial tube-in-tube heat-exchanger by incorporat-
650. ing certain modifications, which result in a forced convective va-
关9兴 Bayazitoglu, Y., and Wang, B. Y., 1996, ‘‘Wavelets in the Solution of Nongray
Radiative Heat Transfer Equation,’’ HTD-Vol. 332, Proceeding of the ASME
por absorption process. Since the development of analytically
Heat Transfer Division, 1, pp. 129. based universal correlation for prediction of heat transfer is lim-
关10兴 Wang, Y., and Bayazitoglu, Y., 1998, ‘‘Wavelets and Discrete Ordinates ited to the inherent complications of two-phase flow phenomena,
Method in Solving One-Dimensional Nongray Radiation Problem,’’ Int. J. use of semi-empirical correlation is widely accepted in practice by
Heat Mass Transf., in press.
关11兴 Newland, D. E., 1993, An Introduction to Random Vibrations, Spectral and
design engineers. This paper also presents a correlation for the
Wavelet Analysis, John Wiley & Sons Inc., New York. forced convection heat transfer in horizontal tubes in an
关12兴 Truelove, J. S., 1987, ‘‘Discrete-Ordinate Solutions of the Radiation Transport absorption/compression heat pump, with the obtained experimen-
Equation,’’ ASME J. Heat Transfer, 109, p. 1048. tal results.
关13兴 Fiveland, W. A., 1984, ‘‘Discrete-Ordinates Solutions of the Radiative Trans-
port Equation for Rectangular Enclosures,’’ ASME J. Heat Transfer, 106,
p. 699.
关14兴 Miele, A., and Iyer, R. R., 1970, ‘‘General Technique for Solving Test Facility and Procedure
Nonlinear Two-Point Boundary-Value Problems via the Method The present experimental investigation is carried out on a test
of Particular Solutions,’’ Optimization Theory and Applications, 5,
p. 382. rig which has been designed and built for measurement of in-tube,
关15兴 Edwards, D. K., 1976, ‘‘Molecular Gas Band Radiation,’’ Advances in Heat two-phase-flow heat transfer coefficient of the absorption process.
Transfer, 12, Academic Press, New York p. 115. The schematic diagram of the test rig used in the present study is
shown in Fig. 1. R-22/DMA is chosen as the working pair, in
which R-22 serves as the refrigerant. The organic fluid DMA 共N,
Experimental Study on Forced N, dimethylacetamide兲 is used as an absorbent. The test facility
Convection Heat Transfer consists of six main parts: a test section 共absorber兲, an absorption/
compression loop, a hot and cold water loop, a R-22 concentration
Inside Horizontal Tubes in an regulating circuit, a cooling circuit of low capacity and a data
AbsorptionÕCompression Heat Pump acquisition system.
The Test-Section. A coaxial tube-in-tube heat-exchanger has
been used as the absorber. The absorber tube has an OD of 19 mm
Li Yong and K. Sumathy and an ID of 16 mm and the annulus surrounding the absorber
Department of Mechanical Engineering, University tube is also of the same length with an ID of 27 mm. The test
of Hong Kong, Hong Kong section is of 24 m in length and is subdivided into four equal parts
to detect the local heat transfer, which is transferred from the
two-phase refrigerant mixture 共R-22 vapor⫹weak solution兲 to wa-
ter in the annulus. The exterior of the annulus is insulated with a
Quasi-local absorption heat transfer coefficients and pressure thick layer of insulation to ensure adiabatic condition. The flow
drop inside a horizontal tube absorber have been investigated rates of the weak solution and R-22 vapor are controlled using a
experimentally, with R-22/DMA as the working pair. The absorber bypass valve V3 and compressor. Since the weak solution pump is
is a counterflow coaxial tube-in-tube heat-exchanger with the very sensitive to cavitation, a cooling circuit of low capacity is
working fluid flowing in the inner tube while the water moves used to cool the weak solution.
through the annulus. A large temperature gliding has been expe-
rienced during the absorption process. Experimental results show The R-22 Concentration Regulating Circuit. The circuit
that the heat transfer coefficient of the forced convective vapor contains two valves and a R-22 accumulator with cooling water
absorption process is higher compared to the vertical falling film flowing through it. In order to test the effect of R-22 concentration
absorption. A qualitative study is made to analyze the effect of on the heat transfer process, the R-22 concentration regulating
mass flux, vapor quality and solution concentration on pressure circuit is used. Similarly, to increase the R-22 concentration, the
drop and heat transfer coefficients. On the basis of the experimen- system is halted temporarily such that the pressure in the recu-
tal results, a new correlation is proposed whereby the two-phase perator as well as regenerator 1 reduces, and R-22 is added into
heat transfer is taken as a product of the forced convection of the the accumulator through valve V13. To reduce concentration, the
absorption and the combined effect of heat and mass transfer at pump circulating the weak solution is stopped, while the compres-
the interface. The correlation is found to predict the experimental sor continues to operate. Meanwhile, the R-22 in the accumulator
data almost within 30 percent. 关DOI: 10.1115/1.1473142兴 gets cooled and condensed by the cooling water.
Keywords: Experimental, Forced Convection, Heat Transfer, Experimental Procedure. A typical experimental run begins
Heat Pump by first turning on the water pumps and fixing the mass flux and
temperature at the desired level. Next, the compressor is turned on
and valve V6 is opened for compressing the vapor into the ab-
Introduction sorber. The expansion valve V5 is also opened slightly for throt-
Several large-scale absorption/compression heat pumps have tling the condensed liquid. On realizing the cooling effect in the
been operated to identify the appropriate cycle components, but cooling circuit, the weak solution pump is turned on and valve V1
there is still considerable work to be done before the absorption/ is opened to circulate the weak solution in the system. Meanwhile,
compression technology can be considered a viable alternative the expansion valve V5 and by-pass valve V3 are regulated to
solution to the vapor-compression technology. Most of the experi- achieve desired pressure and flow rate. The system is then allowed
mental works reported have used the vertical or horizontal shell to reach the steady-state condition before the data-acquisition pro-
and tube heat-exchanger as absorber and desorber 关1兴. Torstensson cess begins. Achieving ‘‘steady-state’’ conditions is defined by
and Nowacki 关2兴 used coaxial tube-in-tube heat-exchangers to requiring less than 2 percent fluctuation in system pressure and
temperatures within 10 minutes. Once the steady-state condition is
reached, the data acquisition system records measurements of
HEAT TRANSFER. Manuscript received by the Heat Transfer Division July 13, 2001; temperature, pressure, and mass flow rates at various points
revision received February 25, 2002. Associate Editor: B. T. F. Chung. around the experimental loop. The experimental work were con-
tures in each sub-section. The energy transferred 共Q兲 in each sub-
section is calculated from an energy balance on the water flowing
in the annulus.
The quasi-local heat transfer coefficient (h t p ) of the two phase
absorption process for each subsection of the absorber is deter-
mined as:
1 d ii d ii d ii d io
⫽ ⫺ ⫺ ln (3)
h t p d io •U d io •h w 2␭ c d ii
where d ii denotes the inside diameter of the inner tube. Physical
properties of the working pairs are calculated using the database
developed by Guo et al. 关4兴.
The value of ‘‘Q’’ is checked by comparing the energy balance
on the absorption side, Q l , during single-phase flow experiments.
The two energy balances—Q and Q l agree within 2.2 percent for
all runs. Temperatures of the fluid are measured by pairs of ther-
mocouples at each location and have experimental uncertainties
by about ⫾0.2 K. The pressures in test-section and pressure drop
Fig. 1 Schematic diagram of the experimental setup of each sub-section are measured respectively with strain-gauge-
type pressure transducer and differential pressure transducer accu-
rate to ⫾0.1 percent. The mass flow rates have uncertainties of
ducted under certain test conditions: 共i兲 solution ⫾0.5 percent and the water-side thermal resistance is about
concentration—64 percent to 90 percent weight of R-22; 共ii兲 weak 22 percent⬃49 percent of the total thermal resistance and
solution mass flux—91 to 280 kg/共m2 s兲; and 共iii兲 absorption propagation-of-error 关5兴 analyses provide the uncertainty as ⫾9.3
temperature—293 to 323 K. percent for h t p .
Data Reduction Results

As mentioned earlier, since the absorber test section is very Experimental in-tube forced convective absorption heat transfer
long (length⫽24 m), it is divided into four subsections, in order coefficient data and pressure drop results are reported for the
to detect the local parameters. Hence, the average heat transfer working pair R-22/DMA in an absorption/compression heat pump.
coefficient in each subsection is treated as quasi-local heat transfer The results are compared to the conventional falling film situation.
coefficient, which in turn is determined from the overall heat It is observed that the heat transfer coefficient is in the range of
transfer coefficient and the water-side heat transfer coefficient by 800⬃2200 W/(m2 K), higher than that of a falling film 共200
using the procedures described below. ⬃900 W/(m2 K), 关6兴兲. Also, a temperature gliding of about 25 K
The water-side convective heat transfer coefficient (h w ) is cal- can be realized in the present experiments, which can be utilized
culated using the correlation of heat transfer in the annulus 关3兴. to enhance the system COP.
h w ⫽0.01032共 Re0.87
w ⫺280 兲 Prw 共 d oi /d io 兲
0.4 0.16
关 ␭ w / 共 d oi ⫺d io 兲兴 Since the in-tube forced convective absorption was demon-
(1) strated to be more effective, experiments were also conducted to
study the effect of various parameters that influence the heat and
where d io and d oi denote the inside diameter of the outer tube and mass transfer coefficient as well as pressure drop of each sub-
outside diameter of the inner tube, respectively. The overall heat section, such as the mass flux and the solution concentration in the
transfer coefficient 共U兲 is found from the energy balance on the absorption section. Since the concentration could not be con-
test sub-section. trolled directly, R-22 concentration is calculated based on conser-
Q vation of species and energy. Hence, only certain range of R-22
U⫽ . (2) concentration has been chosen to represent the quasi-constant con-
A•LM TD
centration for the different mass flux 共Fig. 2共a兲兲. In the figures,
The log mean temperature difference 共LMTD兲 is determined the vapor quality 共Y兲 represents the average quality of each sub-
from the water-side and absorption-side inlet and exit tempera- section.
Fig. 2 Absorption pressure drop for DMAÕR-22 mixture showing the effect of „a… mass flux, and „b… R-22 concentration
Fig. 3 Quasi-local heat transfer coefficient versus „a… mass flux, and „b… R-22 concentration
The absorption pressure drop of each sub-section for DMA/

R-22 mixture is shown in Fig. 2. The data in the Fig. 2共a兲 corre-
sponds to two average quality conditions corresponding to certain
htp
hl
⫽C 1 冉冊
1
X tt
C2
C
1
1⫹ P E 3 R x Le2/3
(4)
R-22 mass concentration within a range of 73 percent In the above equation, P E is a new dimensionless parameter that
⬃77 percent. It is seen that, the pressure drop increases with the represents the combined effect of heat mass transfer at the moving
increase in mass flux and also as expected, the pressure drop gas-liquid interface. The heat transfer coefficient with total mass
increases with the increase in the vapor quality, in the test section. flow rate as liquid (h l ), for each subsection of the absorber is
Besides the influence from mass flux and vapor quality, another determined using Dittus-Boelter equation 关7兴.
factor affecting the pressure drop might be the R-22 concentra- The coefficients C 1 , C 2 , C 3 that appear in Eq. 共4兲 are obtained
tion. Hence, the influence of R-22 concentration on the pressure from the experimental data by using a least-squares technique.
drop is plotted in Fig. 2共b兲. It is seen that, a higher pressure drop Based on present experimental study the optimal values turn out
is experienced at lower R-22 concentrations. This is because of to be 5.897, 0.149, and 1.016 respectively. The predicted heat
the fact that at any given temperature the viscosity of DMA transfer coefficients are then compared with the experimental val-
共1.02⫻10⫺3 kg/(m s) at 298 K兲 is higher than R-22. Therefore, ues as shown in Fig. 4. It can be seen that the correlation 共5兲 could
when R-22 concentration is lower, it implies that the fraction of predict the experimental data almost within 30 percent.
冉冊
DMA in the mixture is more which results in higher frictional 0.149
losses, leading to a higher pressure drop in the test-section. htp 1 1
⫽5.897 . (5)
Figure 3 shows typical results of the effect of mass flux and hl X tt 1⫹ P 1.016
E R x Le2/3
R-22 concentration on the quasi-local absorption heat transfer for
DMA/R-22 mixtures. It is found from the Fig. 3共a兲 that, other Conclusion
conditions remaining the same, high vapor-liquid mixture mass
In-tube forced convective absorption for working pair R-22/
flux results in a higher absorption heat transfer coefficient. This is
DMA has been studied experimentally in an absorption/
attributed to greater convection effects at higher mass fluxes. It is
also apparent from the Fig. 3共a兲 that high average vapor quality
produces higher coefficients, because the average film thickness is
thinner. Thus, as absorption progresses, the rate of heat transfer
decreases owing to the increase in the liquid film thickness and
decrease in flow velocity. From Fig. 3共b兲, it can be seen that the
heat transfer coefficient increases slightly with increase in average
solution concentration. This is because the viscosity of the work-
ing pair decreases with the increase in R-22 concentration.
The Proposed Correlation

In the absorption process, the heat and mass transfer resistance
mainly occurs in the liquid film. The absorption process of a mix-
ture can exhibit significant differences from condensation of a
single, pure component. The differences are related to the phase
equilibrium effects which are referred to as mixture effects. The
heat transfer is not only influenced by the physical properties of
the mixture 共such as thermal conductivity, viscosity, specific heat
etc.兲, but also influenced by the phase equilibrium and mass dif-
fusion resistance. With the heat and mass transfer coupled to-
gether, the mixture effects are determined by the combined effects
of phase transition properties and mass transfer resistance.
Since there is a similarity between the in-tube forced convec-
tion absorption process and generation process, the model sug-
gested by Guo et al. 关7兴 for in-tube forced convection generation Fig. 4 Comparison of proposed correlation results with the
process has been used to include the mixture effects. experimental data
compression heat pump. The experimental data indicate that the 关5兴 Moffat, R. J., 1988, ‘‘Describing the Uncertainties in Experimental Results,’’
Exp. Therm. Fluid Sci., 1, pp. 3–17.
coaxial tube-in-tube heat-exchanger can be used successfully in 关6兴 Guo, K. H., Shu, B. F., and Chen, L. P., 1994, ‘‘Experimental and Theoretical
the absorption/compression heat pump and the heat transfer coef- Study on Falling Fil Absorption Process,’’ Engineering Thermal Physics, 15,
ficient can be enhanced in the forced convection absorption pro- pp. 160–170 共in Chinese兲, ISSN 0253-231X.
cess compared to the falling film absorption. Results demonstrate 关7兴 Guo, K. H., Chen, Z., Shu, B. F., and Meng, Z. X., 1996, ‘‘Theoretical and
Experimental Study on Forced Convection In-tube Generation Process,’’ En-
that the pressure drop in the test section increases with the mass gineering Thermal Physics, 17, Suppl., pp. 187–190, 共in Chinese兲, ISSN 0253-
flux, the quality and the R-22 concentration. It is also found that 231X.
the heat transfer coefficient increases with the mass flux, the qual-
ity, the R-22 concentration. In the present study, an empirical cor-
relation based on the model of in-tube generation process is also On the Accuracy of Beam-Averaged
developed for the forced convective absorption process which
agrees well with the experimental data in the given experimental
Interferometric Heat Transfer
range. Measurements
Nomenclature D. Naylor
A ⫽ heat transfer surface area, m2 e-mail: dnaylor@ryerson.ca
cp ⫽ specific heat, kJ/共kg K兲 Department of Mechanical, Aerospace and Industrial
COP ⫽ Coefficient of Performance Engineering, Ryerson University, 350 Victoria Street,
d ⫽ diameter, m
D ⫽ diffusion coefficient, m2/s Toronto, Ontario, Canada, M5B 2K3
h ⫽ heat transfer coefficient, W/共m2 K兲
hfg ⫽ latent heat of condensation, J/kg
LMTD ⫽ log mean temperature difference Laser interferometry can be used in a three-dimensional tempera-
Q ⫽ heat transfer rate, W ture field to measure the average fluid temperature and heat trans-
Rx ⫽ the ratio of absorbent mass to the absorbate mass, fer rate, integrated over the length of an experimental model.
关 (1⫺x)/x 兴 However, such measurements are inherently approximate when
T ⫽ temperature, K the surface temperature varies in the direction of the test/object
U ⫽ overall heat transfer coefficient, W/共m2 K兲 beam. In this study an analysis is performed to determine the
x ⫽ average R-22 concentration accuracy of beam-averaged heat transfer rate measurements
Y ⫽ average vapor quality made in ideal gases. Two analysis methods are considered. The
Subscripts first method is based on extrapolation of the near-wall tempera-
ture field to obtain the surface gradient. In the second method, the
a ⫽ absorption temperature gradient at the surface is obtained directly from the
g ⫽ gas gradient of the fringe field. The results show that the intrinsic
i ⫽ inner error in the measurements depends strongly upon the form and
l ⫽ liquid magnitude of the temperature variation in the light beam direc-
o ⫽ outer tion. Although the error in the measured heat transfer rate is
s ⫽ solution shown to be small for many commonly encountered conditions, it
tp ⫽ two phase can be greater than ten percent in extreme cases.
w ⫽ water 关DOI: 10.1115/11482400兴
wa ⫽ tube wall
Greek Keywords: Convection, Heat Transfer, Interferometry, Measure-
ment Techniques, Temperature
␣ ⫽ thermal diffusivity, m2/s
␤ ⫽ the slope of the curve in the phase equilibrium dia-
gram for DMA/R-22, ⫺(dT/dx)
␮ ⫽ dynamic viscosity, kg/共m s兲 Introduction
␳ ⫽ density, kg/m3 A laser interferometer can be used in a three dimensional tem-
␭ c ⫽ thermal conductivity of the tube, W/共m K兲 perature field to measure the average fluid temperature, integrated
␭ l ⫽ thermal conductivity of the liquid solution, W/共m K兲 over the length of the phase object. Such beam-averaged measure-
Dimensionless Numbers ments have been made for a variety of three dimensional problems
关1–5兴. However, Frank 关6兴 has shown that the average tempera-
P E ⫽ P E number, ␤ c p x/h f g ture measurements are inherently approximate when the tempera-
Pr ⫽ Prandtl number, c p ␮ /␭ ture varies in the direction of the test/object beam. The error is
Re ⫽ Reynolds number, ␳ ud/ ␮ caused by the nonlinear variation of refractive index with fluid
Le ⫽ Lewis number, ␣ /D temperature. In the present work, an analysis is performed to de-
X tt ⫽ Martinelli parameter, ( ␮ l / ␮ g ) 0.1( ␳ g / ␳ l ) 0.5关 (1⫺Y )/ termine the accuracy of beam-averaged heat transfer measure-
Y 兴 0.9 ments made in ideal gases. An improved understanding of intrin-
sic error associated with this technique, may promote its wider
application to three-dimensional heat transfer problems.
References
关1兴 Groll, E. A., 1997, ‘‘Current Status of Absorption/Compression Cycle Tech- Problem Definition
nology,’’ ASHRAE Trans., 103, Part 1, pp. 361–374.
关2兴 Torstensson, H., and Nowacki J. E., 1991, ‘‘A Sorption/Compression Heat As shown in Fig. 1, consider a thermally developing flow of an
Pump Using Exhaust Air a Heat Source,’’ Proceedings of Absorption Heat ideal gas over a non-isothermal surface of length L. The surface
Pump Conference, Tokyo, pp. 103–108. temperature T s varies in the direction of the light beam. The light
关3兴 VDI-Warmeatlas1977, ‘‘Berechnungsblatter fur den warmeubergang,’’ Siebte,
erweiterte Auflage. VDI-Verlag 共in German兲, ISBN 15063.3434.
关4兴 Guo, K. H., Chen, H. R. Mei, J. B., Shu, B. F. and Meng, Z. X., 1994, Contributed by the Heat Transfer Division for publication in the JOURNAL OF
‘‘Physical Properties Database of Absorption Working Pairs,’’ project report, HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 14,
submitted to Chinese Academy of Sciences 共in Chinese兲. 2001; revision received March 20, 2002. Associate Editor: M. L. Hunt.
beam enters the test section at temperature T ⬁ and exits at tem- the error in the beam-averaged heat transfer rate calculated from
perature T h . This beam corresponds to the test beam of a Mach- the approximate temperature measurements will also be small. It
Zehnder interferometer or the object beam of an holographic in- will be shown in the next section that this assumption is not al-
terferometer. ways correct.
As mentioned previously, when there is a temperature variation
in the direction of the light beam, the temperature that should be Analysis of Beam-Averaged Heat Transfer Rates
assigned to an interference fringe is not exactly the beam-
averaged temperature. The effective fringe temperature can be de- The heat transfer rate, averaged in the direction of the beam (q̄)
rived as follows 共Naylor and Machin 关7兴兲: Consider a ray of light can be expressed in terms of gradient of the average fluid tem-
in the test/object beam passing at a distance y above the surface perature (T̄) as
冏冕
shown in Fig. 1. Neglecting refraction effects, for an ideal gas the
fringe shift order ⑀ (y) is related to the refractive index field ⳵ T̄ 1
q̄⫽⫺k s where T̄⫽ T 共 x * ,y 兲 dx * (5)
N(x,y) as follows: ⳵y y⫽0 0
⑀共 y 兲⫽
1
␭0 冕
0
L
共 N 共 x,y 兲 ⫺N ref兲 dx⫽
PGL
R␭ 0 冕冉 1
0
1
⫺
1
冊
T 共 x * ,y 兲 T ref
dx *
It is clear from Eq. 共3兲 that, in general, the gradient of the effec-
tive fringe temperature ( ⳵ T̄ f / ⳵ y 兩 y⫽0 ) will not be exactly equal to
(1) the average fluid temperature gradient ( ⳵ T̄/ ⳵ y 兩 y⫽0 ). To assess this
where T ref is the reference temperature, L is the length of the test source of error, two analysis methods for determining the heat flux
model in the light beam direction, P is the gas pressure, R is the from an interferogram have been considered. The first is the tra-
gas constant, G is the Gladstone-Dale constant, and ␭ 0 is the ditional approach of calculating the near-wall fluid temperature
wavelength of the laser light source. profile normal to the surface and using extrapolation to get the
Now consider the hypothetical uniform temperature along the surface temperature gradient. The second approach is based on
measuring the surface gradient of the fringe shift, ⳵ ⑀ / ⳵ y 兩 y⫽0 .
light ray, T̄ f , that would produce same the fringe shift ⑀ (y) as the
non-uniform distribution, T(x * ,y). With this uniform tempera- Temperature Profile Extrapolation Method. To gain an un-
ture, the fringe shift is derstanding of the accuracy of the beam-averaged heat flux mea-
冉冊
surements, two dimensional laminar forced convection from a
PGL 1 1 non-isothermal plate has been simulated. Fringe fields have been
⑀共 y 兲⫽ ⫺ (2)
R␭ 0 T̄ f T ref calculated using the temperature field from a similarity solution
and the simulated measurements have been compared to the exact
Equating Eqs. 共1兲 and 共2兲 and simplifying gives average heat transfer rates. Details of the similarity solution are
T̄ f ⫽ 冋冕
0
1 1
T 共 x * ,y 兲
dx * 册 ⫺1
(3)
given by Levy 关8兴. It will be outlined only briefly here.
The surface temperature of the plate was assumed to have a
power function variation:
Equation 共3兲 is the effective temperature that should be as- T s ⫽ 共 T L ⫺T ⬁ 兲共 x * 兲 n ⫹T ⬁ (6)
signed to a fringe when the temperature varies in the test/object
beam direction. But, in general, T(x * ,y) is not known and cannot where T L is the surface temperature of the plate at x * ⫽x/L⫽1.
be determined from a single interferogram. The approach taken to The governing equations for constant property, incompressible,
overcome this difficulty in several previous studies has been to boundary layer flow over a zero incidence flat plate can be trans-
assume that T̄ f is equal to the arithmetic average fluid tempera- formed into the following ordinary differential equations:
ture, T̄. With this approximation, it follows from Eq. 共2兲 that 2 f ⵮ ⫹ f f ⬙ ⫽0 (7)
T ref Pr
T̄⯝T̄ f ⫽ (4) ␪ ⬙ ⫹nPrf ⬘ 共 1⫺ ␪ 兲 ⫹ ␪ ⬘ f ⫽0 (8)
⑀ R␭ 0 T ref 2
1⫹
PGL where Pr is the Prandtl number and the following dimensionless
But, it is evident from Eq. 共3兲 that the fringe temperature T̄ f is not parameters have been introduced:
the arithmetic average of T(x * ,y). Recently, Naylor and Machin y ⳵f u T s ⫺T
关7兴 have analyzed this source of error for a range of conditions ␩ ⫽ 冑Re f ⬘⫽ ⫽ ␪⫽ (9)
x ⳵␩ u⬁ T s ⫺T ⬁
that are used in interferometric heat transfer experiments. The
study considered power function and periodic temperature varia-
tions in the beam direction, with water and ideal gases as the test
fluids. The results show that the difference between T̄ f and T̄ is
less than two percent of the overall beam temperature difference
(T h ⫺T ⬁ ), even for the most adverse form of temperature distri-
bution in the light beam direction. For many commonly encoun-
tered experimental conditions, the intrinsic error was less than one
percent. From these results, one might be tempted to assume that
Fig. 2 „a… Temperature contours of a forced convection

boundary layer on a plate with a linear surface temperature
variation „ n Ä1… for ReLÄ104 , PrÄ0.7, „b… Simulated beam-
averaged infinite fringe interferogram „T L À T ⴥ Ä20 K, P
Fig. 1 Thermally developing flow over a surface with a tem- Ä100 kPa, G Ä2.256Ã10À4 m3 Õkg, ␭ 0 Ä632.8 nm, L Ä0.4 m, R
perature variation in the testÕobject beam direction Ä287 JÕkgK, T ⴥ Ä300 K….
lower limits of each curve correspond to a total fringe shift 5
⭐ ⑀ max⭐20. This is a typical range of total fringe shift for inter-
ferometric heat transfer measurements. It should be mentioned
that smaller phase differences can be measured using modern op-
tical methods 关9,10兴. But, when interference patterns are recorded
on film, several fringes are usually needed to characterize the
near-wall temperature profile.
It can be seen in Fig. 3 that the error in the gradient is less than
two percent for n⫽0, 0.1, and 0.5, over the full range of surface
temperature differences. In fact for n⫽0.5, which corresponds to
a constant heat flux surface, the maximum error is only about 0.2
percent. But, for larger values of the exponent (n⫽1,2,4), the
error increases rapidly with the overall surface temperature differ-
ence (T L ⫺T ⬁ ) and can exceed ten percent in the most extreme
conditions. So, for these exponents, beam-averaged interferomet-
ric measurements could significantly under predict the average
Fig. 3 Effect of surface temperature difference on the percent- heat flux. These results were found to be independent of Reynolds
age error in the surface temperature gradient, calculated using
the actual mean temperatures and the effective fringe tempera-
number and weakly dependent upon the parameter PGL/␭ o R,
tures „PGLÕ R ␭ 0 Ä4.97Ã104 K, T ⴥ Ä300 K… which affects the optical sensitivity.
In Fig. 3, results have been calculated up to a temperature dif-
ference of T L ⫺T ⬁ ⫽80 K. At this temperature difference, the non-
linear variation of refractive index with temperature can cause
In Eq. 共9兲, u ⬁ is the free stream velocity and Re is the Reynolds substantial measurement error. Whenever possible, the experimen-
number based on the distance from the leading edge 共x兲. The tal model length 共L兲 should be extended to decrease the tempera-
boundary conditions for Eqs. 共7兲 and 共8兲 are: ture difference needed to produce the desired fringe shift. How-
␩ ⫽0 f ⫽ f ⬘ ⫽0 ␪ ⫽0 ever, in many situations the model design requires a compromise
(10) between optical sensitivity and achieving the desired experimental
␩ →⬁ f ⬘ →1 ␪ →1 conditions 共Reynolds number, Rayleigh number, aspect ratio,
etc.兲. As a result, it may be necessary to operate at higher tem-
Equations 共7兲 and 共8兲 have been solved subject to the above perature differences than are optimal based on strictly optical
boundary conditions using the shooting method for temperature considerations.
exponents in the range 0⭐n⭐4. It is interesting to note in Fig. 3 that for an isothermal surface
To illustrate the procedure used to calculate the fringe tempera- (n⫽0) the inherent error in the beam-averaged gradient is not
ture gradient ( ⳵ T̄ f / ⳵ y 兩 y⫽0 ), Fig. 2共a兲 shows the isotherms in the zero. The error is not zero because gradient was calculated using
thermal boundary layer for a linear surface temperature distribu- effective fringe temperatures (T̄ f ) that were calculated above the
tion (n⫽1.0), obtained from the similarity solution. The beam- surface 共at the locations of ⑀⫽1 and ⑀⫽2兲. At these y-locations
averaged fringe field was calculated from the temperature field
using Eq. 共1兲. This fringe field is shown as a simulated interfero- there is a slight temperature variation along the light beam. So, T̄ f
gram in Fig. 2共b兲. This interference pattern was calculated for the will not be exactly the average fluid temperature (T̄). For an
following set of conditions, which are typical for the application isothermal surface, T̄ f ⫽T̄ in the limit as y→0. For this reason, the
of interferometry: G⫽2.256⫻10⫺4 m3 /kg 共air兲, ␭ o ⫽632.8 nm Fringe Gradient Method should be used on isothermal surfaces.
共He-Ne laser兲, R⫽287 J/kgK 共air兲, P⫽100 kPa, L⫽0.4 m, T L This is discussed in the next section.
⫺T ⬁ ⫽20 K. These parameters correspond to PGL/␭ o R⫽4.97
Fringe Gradient Method. The average heat flux can be de-
⫻104 K. termined using an alternate method, which makes direct use of the
Once the simulated fringe field was obtained, the surface tem- fringe gradient at the surface. This is done by differentiating Eq.
perature gradient was calculated using the two fringes closest to 共1兲 and evaluating the result at the surface, as follows
the plate surface. Effective fringe temperatures were calculated
numerically at the y-locations corresponding to fringe shift orders
of ⑀⫽1 and ⑀⫽2, using Eq. 共3兲. In this calculation, the fringe shift
order was taken to be zero at the surface. The surface gradient was
⳵⑀
⳵y
冏 y⫺0
⫽
⫺PGL
R␭ 0 冕 0
1 1 ⳵T
T 2s ⳵ y
冏 y⫽0
dx * (12)
then calculated using the effective fringe temperature at the first

For an isothermal surface, the above equation can be rearranged to
fringe T̄ f 兩 ⑀ ⫽1 and effective fringe temperature at the second fringe give the beam-averaged surface temperature gradient:
T̄ f 兩 ⑀ ⫽2 , as follows
⳵ T̄ f
⳵ y y⫽0
⫽ 共 T̄ f 兩 ⑀ ⫽2 ⫺T̄ f 兩 ⑀ ⫽1 兲 / 共 y 2 ⫺y 1 兲 (11)
⳵ T̄
⳵y
冏 y⫽0
⫽
⫺R␭ 0 T 2s ⳵ ⑀
PGL ⳵y
冏 y⫽0
(13)
where y 1 and y 2 are the locations of the first and second fringes. Using the above equation, the beam-averaged heat flux on an
This approximate gradient was compared to the actual gradient isothermal surface can be calculated with no intrinsic beam-
calculated in the same manner, using the actual average tempera- averaging error. This observation appears to have been first made
tures at the same y locations. In practice, higher order extrapola- by Papple and Tarasuk 关4兴.
tion schemes are sometimes used to determine the surface gradi- Equation 共12兲 can also be used to get the exact beam-averaged
ent from interferograms. But, since the same method of heat transfer rate on an isoflux surface. For an isoflux surface Eq.
extrapolation has been applied in both cases, the difference be- 共12兲 can be rearranged to give:
冏冏冋冕册
tween these gradients will give a meaningful indicator of the in- ⫺1
trinsic error associated with the interferometric measurements. ⳵ T̄ ⫺R␭ 0 ⳵ ⑀ 1 1
⫽ 2 dx * (14)
Figure 3 shows the percentage difference in the surface gradient ⳵y y⫽0
PGL ⳵ y y⫽0 0 Ts
calculated using the actual mean temperatures and the effective
fringe temperatures. Data are presented for various values of the In applying Eq. 共14兲, the integral of the surface temperature varia-
surface temperature profile exponent, n. Note that the upper and tion could be obtained from thermocouple measurements.
If the surface is neither isothermal nor isoflux, the average sur-
face gradient ( 兰 ⳵ T/ ⳵ y 兩 y⫽0 dx * ) cannot be isolated in Eq. 共12兲 and
the beam-averaged heat flux cannot be evaluated exactly. The re-
mainder of this work will consider the use of the following ap-
proximation of Eq. 共12兲:
冋冏册
⳵ T̄
⳵y y⫽0 approx
⫽
⫺R␭ 0 T̄ 2s ⳵ ⑀
PGL ⳵y
冏 y⫽0
where T̄ s ⫽ 冕 0
1
T s 共 x * 兲 dx *
(15)
In this case, T̄ s could be obtained from thermocouple measure-
ments or from the optically measured beam-averaged surface tem-
perature. The accuracy of this approximate method to obtain av-
erage heat transfer rates will now be considered.
Substituting Eq. 共15兲 into Eq. 共12兲, it can be seen that the ap-
proximate mean surface temperature gradient is related to the ex-
act surface temperature gradient as follows:
冋冏册冉冕
⳵ T̄
⳵y y⫽0 approx
⫽
1
0
T s dx * 冊冕
2 1 1 ⳵T
0 Ts ⳵y
2 冏 y⫽0
dx * (16)
For forced convection over a flat plate with a power function

temperature variation 共Eq. 共6兲兲, the local surface temperature gra-
dient is 共Levy 关8兴兲:
⳵T
⳵y
冏 y⫽0
⫽ 共 T ⬁ ⫺T s 兲
⳵␪
⳵␩
冏 ␩ ⫽0
⳵␩
⳵ y
⫽C 1 共 x * 兲 n⫺1/2 (17)
where C 1 is a constant for a fixed set of experimental conditions.

The error in the beam-averaged surface temperature gradient can
be calculated as:
冋冏册冋冏册
⳵ T̄
⳵y y⫽0 approx
⫺
⳵ T̄
⳵y y⫽0 exact
冋冏册 ⳵ T̄
⳵y y⫽0 exact
Fig. 4 Effect of the surface temperature difference on the per-
centage error in the surface temperature gradient, calculated
冉冕冊冕冏
using the fringe gradient for „a… forced convection and „b… free
1
0
T s dx *
2 1
0
T ⫺2
s
⳵T
⳵y y⫽0
dx * ⫺ 冕冏0
1 ⳵T
⳵y y⫽0
dx *
convection
冕冏
⫽ 1 ⳵T
dx * this case, if the surface temperature variation is a power function
0 ⳵y y⫽0 given by Eq. 共6兲, the surface gradient varies with x as follows:
where the local gradient at the surface ( ⳵ T/ ⳵ y 兩 y⫽0 ) and the sur-
(18) ⳵T
⳵y
冏 y⫽0
⫽C 2 共 x * 兲共 5n⫺1 兲 /4 (19)
face temperature (T s ) vary as given in Eq. 共17兲 and Eq. 共6兲. where C 2 is a constant for a fixed set of experimental conditions.
Equation 共18兲 has been evaluated analytically using a commer- The derivation of Eq. 共19兲 can be found in the similarity solution
cial symbolic computation program for n⫽0 to 4. Calculations of the boundary layer equations by Sparrow and Gregg 关11兴.
have been done for a surface temperature difference range of 0 Equation 共18兲 has been evaluated analytically for the local sur-
K⭐T L ⫺T ⬁ ⭐80 K. The results are given in Fig. 4共a兲. It can be face temperature gradient and surface temperature variations
seen that the results are very similar to those in Fig. 3, which were given in Eq. 共19兲 and Eq. 共6兲. The results, which apply to free
obtained using linear temperature profile extrapolation. As previ- convection, are shown in Fig. 4共b兲. It can be seen that the results
ously discussed, for this method the error is always zero for an for free convection have a similar form the those for forced con-
isothermal surface (n⫽0), when the surface temperature gradient vection. Again, for an isoflux boundary condition (n⫽0.2), the
is calculated from the fringe gradient. As in the previous analysis, error in the gradient is small over the full range of surface tem-
for the constant heat flux case (n⫽0.5) the error is very small perature difference. For larger values n (n⫽0.5,1,2,4), the error is
over the complete range of surface temperature difference. But, greater than the forced convection case.
comparing Fig. 4共a兲 to Fig. 3, it can be seen that the Fringe Gra-
dient Method gives slightly higher level of error than the Tem-
perature Profile Extrapolation Method for n⫽4.0. But, it would be Concluding Remarks
unusual to encounter such an extreme surface temperature varia- The intrinsic error associated with beam-averaged measure-
tion in a practical convection problem. For most applications, both ments is often inadequately addressed or simply ignored in many
methods will have a similar level of error. Note that the results in previous interferometric studies. In the current work, an analysis
Fig. 4共a兲 are for forced convection, but are independent of Prandtl has been performed to assess this measurement error, using solu-
number, Reynolds number and the parameter that affects the options for two classical convective heat transfer problems. The re-
tical sensitivity (PGL/␭ o R). sults show that the measurement error in the beam-averaged sur-
This approximate ‘‘fringe gradient method’’ has also been ap- face temperature gradient will not always be small. For both
plied for free convection from a vertical nonisothermal plate. In methods considered, the error in beam-averaged temperature gra-
dient was less than one percent, when measurements were made popular techniques to obtain high resolution, full field heat trans-
on an isothermal or isoflux surface. However, the error in the fer coefficient distributions for many applications. Especially for
beam-averaged heat flux was found to be greater than ten percent, the investigation of internal cooling problems, for example in gas
under some extreme conditions. turbine blades, this technique has become a powerful experimen-
tal tool 关1兴. Thereby, the question of the correct choice and deter-
References mination of the fluid reference temperature to be used for the
关1兴 Yousef, W. W., and Tarasuk, J. D., 1981, ‘‘An Interferometric Study of Com- calculation of the heat transfer coefficient has received increasing
bined Free and Forced Convection in a Horizontal Isothermal Tube,’’ ASME J. attention particularly for passages with large length-to-diameter
Heat Transfer, 103, pp. 249–256.
关2兴 Kato, S., and Maruyama, N., 1989, ‘‘Holographic Interferometric Measure- ratios and significant heat exchange upstream of the position of
ments of the Three-Dimensional Temperature Field with Thermally Develop- interest. Some of the models and approaches developed are given
ing Flow in the Measuring-beam Direction,’’ Exp. Therm. Fluid Sci., 2, pp. in 关2– 4兴 and are reviewed by Ireland and Jones 关1兴. The above
333–340. cited approaches are related to the transient techniques using hot
关3兴 Li, J., and Tarasuk, J. D., 1992, ‘‘Local Free Convection Around Inclined
Cylinders in Air: an Interferometric Study, Experimental Thermal and Fluid 共heated兲 fluid flow and cold model starting temperatures to estab-
Science,’’ Exp. Therm. Fluid Sci., 5, pp. 235–242. lish a heat flux between the flow and the model surface. Another
关4兴 Papple, M. L. C., and Tarasuk, J. D., 1987, ‘‘An Interferometric Study of kind of transient heat transfer measurements uses electrically
Developing Natural Convective Flow in Inclined Isothermal Ducts,’’ AIAA heated foils or coatings, where it can be assumed that an uniform
Paper 87-1589.
关5兴 Fehle, R., Klas, J., and Mayinger, F., 1995, ‘‘Investigation of Local Heat heat flux is generated in the thin isothermal foil 关5– 8兴. Consider a
Transfer in Compact Heat Exchangers by Holographic Interferometry,’’ Exp. model of low conductivity which can be considered as semi-
Therm. Fluid Sci., 10, pp. 181–191. infinite during the test run and is covered by a heater foil. This
关6兴 Frank, M. E., 1970, ‘‘Interferometer Measurements in Free Convection on a model is initially at an uniform temperature equal to the flow
Vertical Plate with Temperature Variation in the Light-beam Direction,’’
Fourth International Heat Transfer Conference, Vol. 4, pp. 1–12. temperature T 0 when the power to the heater foil is switched on
关7兴 Naylor, D., and Machin, A. D., 2001, ‘‘The Accuracy of Beam-Averaged In- and the model wall temperature rises. The response in wall tem-
terferometric Temperature Measurements in a Three-Dimensional Field,’’ Exp. perature to this heating input q can be calculated by considering
Heat Transfer, 14, pp. 217–228. the wall of the model as a semi-infinite body:
关8兴 Levy, S., 1952, ‘‘Heat Transfer to Constant Property Laminar Boundary-Layer
冉冊冉冊
h 冑t
Flows with Power-Function Free-Stream and Wall-Temperature Variation,’’ J.
Aeronaut. Sci., 19, pp. 341–348. T w ⫺T 0 h 2t
h ⫽1⫺exp erfc (1)
关9兴 Slepicka, J. S., and Cha, S. S., 1995, ‘‘Stabilized Nonlinear Regression for
Interferogram Analysis,’’ Appl. Opt., 34, pp. 5039–5044.
q ks 冑k s
关10兴 Breuckmann, B., and Thieme, W., 1985, ‘‘Computer-Aided Analysis of Holo- In Eq. 共1兲 h is the local heat transfer coefficient and k s
graphic Interferograms Using the Phase-Shift Method,’’ Appl. Opt., 24, pp.
2145–2149. ⫽␭ s ␳ s c p s . The local heat flux q, which is dissipated in the heater
关11兴 Sparrow, E. M., and Gregg, J. L., 1958, ‘‘Similar Solutions for Free foil is divided into the convective heat flux at the wall q w and the
Convection From a Nonisothermal Vertical Plate,’’ Trans. ASME, 80, pp. 379– conduction flux q s into the model 共Fig. 1兲.
386.
If this technique is applied to a long internal coolant channel,
the heat input into the fluid will increase the fluid temperature and
Bulk Temperature Development in therefore the reference bulk temperature is a function of stream-
wise position and time. The use of Eq. 共1兲 where the fluid tem-
Transient Heat Transfer Measurements perature is taken as constant 共equal T 0 兲 with space and time will
Using Heater Foils lead in this case to a wrong result for the heat transfer coefficient.
It is the aim of the present paper to describe a simplified model for
the fluid bulk temperature development during this kind of tran-
Jens von Wolfersdorf sient experiment similar to the one described in 关4兴 for the case of
heated flow. The first step is the derivation of an analytical solu-
University of Stuttgart, Institute for Aerospace
tion in case of uniform heat transfer coefficient. Then the simpli-
Thermodynamics, Pfaffenwaldring 31, 70569 Stuttgart, fied model will be described leading to a simple expression for the
Germany bulk temperature space and time development. Using this expres-
e-mail: Jens.Vonwolfersdorf@itlr.uni-stuttgart.de sion in the boundary condition for the semi-infinite wall-
conduction-convection problem, solutions for the local wall tem-
perature history are obtained, which should be used in transient
test analysis. The comparison between the analytical solution, nu-
The time and space development of the fluid bulk temperature in a merical solutions and the simplified model is used to show the
transient heat transfer experiment for internal channel cooling applicability of the approximations made in the simplified model’s
investigations using heater foils is addressed. An analytical solu- development.
tion for uniform heat transfer coefficients is derived which shows
the effect of wall heating on the bulk temperature during a tran-
sient test run. A simplified model is proposed for characterizing 2 Analytical Solution
the bulk temperature development by introducing an upstream The analytical solution for the conjugate convection problem
heat transfer parameter. With this, analytical solutions for the will be determined for the case of an uniform heat transfer coef-
local wall temperature history can be derived. The presented so- ficient h. The fluid enters the channel 共perimeter U, which is as-
lution can be used for data reduction of transient tests of this kind. sumed to be constant兲 with a mass flow ṁ. The energy equation
关DOI: 10.1115/1.1482403兴 for the fluid can be written as:
Keywords: Heat Transfer, Measurement Techniques ⳵T f
ṁc p ⫽hU 共 T w 共 x,t 兲 ⫺T f 共 x,t 兲兲 (2)
⳵x
1 Introduction where axial heat conduction effects have been neglected with the
Transient heat transfer measurement techniques using Thermo- initial condition:
chromic Liquid Crystals 共TLC兲 are nowadays one of the most T f 共 x,0兲 ⫽T 0 (3)
Contributed by the Heat Transfer Division for publication in the JOURNAL OF and the boundary condition at the channel entrance:
HEAT TRANSFER. Manuscript received by the Heat Transfer Division April 25, 2001;
revision received March 22, 2002. Associate Editor: H. O. S. Lee. T f 共 0,t 兲 ⫽T 0 (4)
Fig. 2 Description of simplified model
1.e-10 leading to about 30 terms in the sums. Nevertheless for the

range of parameters usually found in the transient experiments
considered (Bi2 Fo⬍6), the solution is suitable for comparison
with the following simplified model.
3 Simplified Model
Fig. 1 Energy balance at the model surface For the determination of the time variation of the fluid tempera-
ture at a given position x a simplified one-dimensional model
similar to the one given in 关4兴 is proposed. The fluid enters the
The solid wall will be considered as a semi-infinite body. Neglect- channel at x⫽0 with a constant temperature T 0 and is heated up
ing again axial heat conduction effects leads to: to the position x by the convective heat flux at the wall q w . The
total amount of heat transferred up to the position x is for a con-
⳵Ts ⳵ 2T s stant perimeter U:
⫽a for 0⭐y⬍⬁ (5)
with the initial condition:

⳵t ⳵y2
Q 共 x,t 兲 ⫽U 冕0
x
q w 共 ␰ ,t 兲 d ␰ ⫽U 冕 0
x
h 共 ␰ 兲共 T w 共 ␰ ,t 兲 ⫺T f 共 ␰ ,t 兲兲 d ␰
T s 共 x,y,0兲 ⫽T 0 (6) (9)

and the boundary condition at the fluid-solid-interface: This heat flux is described by a time dependent, space-averaged
heat flux q f and a single averaged heat transfer parameter ␣ (x),
⫺␭ s
⳵Ts
⳵y
冏 ⫽q⫺h 共 T s 共 y⫽0 兲 ⫺T f 兲 ⫽h 冉 q
h
⫹T f ⫺T w 冊 (7)
both of which are characteristic for the measurement location x
共see Fig. 2兲.
冕冕
y⫽0
A small times solution to this system of equations can be derived 1 x ␣共 x 兲 x

q f 共 x,t 兲 ⫽ q w 共 ␰ ,t 兲 d ␰ ⫽ 共 T w 共 ␰ ,t 兲 ⫺T f 共 ␰ ,t 兲兲 d ␰
using Laplace Transforms and power series expansions as de- x 0 x 0
scribed in 关4兴 for the fluid and the solid wall temperatures. Since (10)
the main interest is the fluid temperature development, only this
solution will be given here: where
⌰f⫽
T f ⫺T 0
⫽
2
T ref⫺T 0 冑␲
Bi 冑Fo⫺exp ⫺4St
x
dh
Bi 冑Fo 冉冊冕 0
x
h 共 ␰ 兲共 T w 共 ␰ ,t 兲 ⫺T f 共 ␰ ,t 兲兲 d ␰
冕
␣共 x 兲⫽
再
x . (11)
兺 n! 冉 4St d 冊
2
⬁
1 x n
1 共 T w 共 ␰ ,t 兲 ⫺T f 共 ␰ ,t 兲兲 d ␰
共 Bi 冑Fo兲 n
冉冊
⫻ ⫹ 0
冑␲ n⫽1 h 3 n
⌫ ⫹ This definition implies that the fluid entering the channel experi-
2 2 ences at any time t a constant heat flux boundary condition and a
再冉冊冎
⬁ j⫺1 n constant lumped heat transfer parameter up to the position of in-
1 x
⫹ 兺兺共 ⫺1 兲 j⫺n
共 n⫺1 兲 !n! 共 j⫺n 兲 !
4St
dh
terest. Therefore one obtains from a simplified energy balance
equation for the time history of the fluid temperature at the posi-
冎
j⫽2 n⫽1
tion x the simple expression.
1
共 Bi 冑Fo兲 j
冉冊
⫻ 共 j⫺1 兲 ! (8) U
3 j T f 共 x,t 兲 ⫽T 0 ⫹q f 共 x,t 兲 x (12)
⌫ ⫹ ṁc p
2 2
From Eq. 共1兲 one knows the wall temperature response to the
where heating based on the inlet temperature T 0 . Using a heat transfer
q h hU coefficient h 0 based on the inlet temperature and not on the local
⫽T ref⫺T 0 , ␤⫽ , b⫽ , fluid temperature one obtains:
␭s
冉冊冉冊
h ṁc p
T w ⫺T 0 h 20 t h 0 冑t
x h0 ⫽1⫺exp erfc (13)
Bi Fo⫽ ␤ at
2 2
and 4St ⫽bx.
dh
q ks 冑k s
As mentioned, this equation is valid for sufficiently small ␤ 2 at as The left hand side of Eq. 共13兲 is the ratio of the local convective
usually found in the transient experiments considered. heat flux into the fluid to the total heat dissipated in the heating
Numerical investigations of this equation have shown, that for layer per area. Therefore one can approximate the time history of
the heat flux up to the position x in Eq. 共12兲 by using a space-
Bi2 Fo⬎6 the solution starts to deviate, probably due to numerical
averaged heat transfer coefficient based on the inlet temperature.
accuracy in calculating the infinite sums 共very large and small
terms cancel out each other兲 as well as due to the validity of the
small times approximation. In these investigations each summa-
tion was performed until the individual contributors were less than
qf
q
⫽1⫺exp
h̄ 20 t
ks
erfc 冉冊冉冊
h̄ 0 冑t
冑k s
(14)
This average heat transfer coefficient is defined as:
h̄ 0 ⫽
1
x 冕0
x
h 0共 ␰ 兲 d ␰ (15)
For the case of a constant heat flux and a constant ␣ (x) up to the
position x as implied by Eqs. 共10兲 and 共12兲, the relationship which
exists between h 0 ( ␰ ) and ␣ (x) is given by:
␣
h 0⫽ (16)
␣U
1⫹ ␰
ṁc p
Integrating Eq. 共16兲 according to Eq. 共15兲 one obtains:
h̄ 0
⫽
冉
ln 1⫹
␣U
ṁc p
x
⫽A
冊 (17)
Fig. 3 Comparison of the simplified model with analytical and
␣ ␣U numerical solutions „uniform heat transfer coefficient distribu-
x tion…
ṁc p
using the parameter A for brevity.
With this the simplified model of the fluid bulk temperature boundary condition Eq. 共7兲 applying Duhamel’s Theorem for the
development is given from Eqs. 共12兲, 共14兲, and 共17兲 by: time dependency. From this, the heat input into the fluid was
再冉冊冉冊冎
determined for a given mass flow rate assuming a constant area
qU ␣ 2t 2 ␣ 冑t channel with constant perimeter and therefore hydraulic diameter.
T f ⫺T 0 ⫽ x 1⫺exp A erfc A (18)
ṁc p ks 冑k s For this comparison the following data are used. A 500 mm
long square cooling channel with a hydraulic diameter of 20 mm
Now one can solve for the wall temperature history using Eq. 共18兲 is considered. A mass flow rate of 0.01 kg/s with an inlet tempera-
as the fluid temperature boundary condition. Solving Eqs. 共5兲, 共6兲, ture of 293 K enters the channel. The wall material is assumed to
and 共7兲 one obtains for the general case where ␣ A⫽h: be Perspex ( 冑k s ⫽569 W冑s/m2K) with an initial temperature
冉冊冋冉冊冉冑冑冊册
equal to the fluid inlet temperature. For the heat transfer coeffi-
qU q h 2t h t
T w ⫺T 0 ⫽ x⫹ 1⫺exp erfc cient a value of 250 W/m2K is used for both, the local heat trans-
ṁc p h ks ks fer coefficient and the upstream heat transfer parameter. The
冉再冊冉冊冉冑冑冊
heater foil is taken to dissipate an area-specific heat flux of
qU 1 ␣ 2t 2 ␣ t 10 000 W/m2 leading to q/h⫽40 K, which is typical for the tran-
⫹ x exp A erfc A
ṁc p ␣ ks ks sient experiments considered. Figure 3 shows the time develop-
A ⫺1 ment of the fluid bulk temperature at the stream-wise positions
h
x⫽0.1 m, x⫽0.3 m, and x⫽0.5 m. The agreement between the
⫺exp 冉冊冉冑冑冊冎
h 2t
ks
erfc
h t
ks
(19)
analytical and the approximative solution is generally good with
some small differences at the last station. The maximum differ-
ence between the solutions is within 5 percent.
For the special case ␣ A⫽h the solution is given by: As a second comparison a heat transfer coefficient distribution
冊冋冉冊冉冑冑冊册
is randomly generated with values between 100 W/m2 K and
T w ⫺T 0 ⫽ 冉 qU
ṁc p
x⫹
q
h
1⫺exp
h 2t
ks
erfc
h t
ks
⫹
qU
ṁc p
x
400 W/m2K. The simplified model is compared to the numerical
solution in Fig. 4, where for the heat transfer parameter ␣ at the
respective positions the average heat transfer coefficients up to x
⫻ 再 2 h 冑t
冑␲ 冑k s
⫺
2h 2 t
ks
exp
h 2t
ks 冉冊冉冑冑冊冎
erfc
h t
ks
(20)
⫽0.1 m, x⫽0.3 m, and x⫽0.5 m from the random distribution
have been used. Also in this case the agreement is very good,
which shows the applicability of the assumptions made in the
These equations should be used for data reduction purposes in the simplified model’s development. The model 共Eq. 共18兲兲 is therefore
transient tests considered for the determination of the heat transfer well suited for pre-test estimation of the importance of the fluid’s
coefficients. Therefore more than one local wall temperature indi-
cation using, e.g., liquid crystals is needed to determine the two
unknowns h, the local heat transfer coefficients and ␣, the up-
stream heat transfer parameter.

The simplified model of the fluid bulk temperature development
共Eq. 共18兲兲 will be compared to the analytical solution 共Eq. 共8兲兲 for
the case of uniform heat transfer coefficient and with numerical
solutions using a one-dimensional method. In this method the one-
dimensional thermal energy equation for an incompressible, con-
stant property fluid as a function of space and time with convec-
tive heat transfer to a semi-infinite solid was numerically
integrated taking into account the heat generating layer on the
solid-fluid interface. The heat flux split according to Fig. 1 was
determined using the prescribed local heat transfer coefficients
and the calculated wall temperature from the analytical solution of Fig. 4 Comparison of the simplified model with numerical so-
transient heat conduction in a semi-infinite wall with the heat flux lutions „random heat transfer coefficient distribution…
upstream heating as well as for the final data reduction using Eqs.
共19兲 and 共20兲.
Solutions of Radiative Heat Transfer
in Three-Dimensional Inhomogeneous,
5 Conclusions Scattering Media
A simplified model describing the time and space development
of the fluid bulk temperature in a transient test with heater foils at
the wall is derived for internal cooling channel applications. This
L. M. Ruan
model, which compares well to an analytical solution and numeri- e-mail: ruanlm@hope.hit.edu.cn
cal solutions is suited to estimate the heating effect on the mea-
surement results. Using the model in a transient test, two un- H. P. Tan
knowns 共the local heat transfer coefficient and an upstream heat School of Energy Science and Engineering,
transfer parameter兲 have to be determined from wall temperature Harbin Institute of Technology,
indications.
Harbin, PR China, 150001
Nomenclature
Symbols In the present study, we use the Monte-Carlo (MC) method to
a ⫽ thermal diffusivity simulate radiative heat transfer in three-dimensional inhomoge-
A ⫽ parameter, Eq. 共17兲 neous scattering unit cube with black or gray walls. The results
Bi ⫽ Biot number show that the averaging method of non-uniform radiative proper-
cp ⫽ specific heat at constant pressure ties in each medium element has influence on the results. One
dh ⫽ hydraulic diameter reasonable averaging method has been employed in our model. In
Fo ⫽ Fourier number mean while, several characters of exchange factor have been em-
h ⫽ heat transfer coefficient ployed to estimate the performance of pseudo-random numbers
k ⫽ material parameter generator and the numerical uncertainty of MC simulation.
ṁ ⫽ mass flow rate 关DOI: 10.1115/1.1495519兴
q ⫽ heat flux per area
Q ⫽ heat flux
St ⫽ Stanton number
Keywords: Heat Transfer, Monte Carlo, Participating Media, Ra-
t ⫽ time
T ⫽ temperature diation, Scattering, Three-Dimensional
U ⫽ perimeter
x ⫽ stream-wise position
y ⫽ vertical coordinate
␣ ⫽ heat transfer parameter
␭ ⫽ thermal conductivity 1 Introduction
␳ ⫽ density Radiative heat transfer in three-dimensional inhomogeneous,
␰ ⫽ axial coordinate
scattering media has played an important role in engineering ap-
⌰ ⫽ dimensionless temperature
plications. Because of the non-uniform distribution of temperature
Subscripts and absorbing gaseous species, and scattering particles concentra-
0 ⫽ entrance or inlet tions, the flame and combustion system should be treated as inho-
f ⫽ fluid mogeneous media. Some researchers have studied this problem
s ⫽ solid successfully by means of several methods, such as YIX method
w ⫽ wall 关1兴 and MC method 关2兴. Hsu et al. 关3,4兴 developed a benchmark
solution set using MC method and YIX method, while Guo and
References Maruyama 关5,6兴 studied the same problem by REM method and
关1兴 Ireland, P. T., and Jones, T. V., 2000, ‘‘Liquid Crystal Measurements of Heat scaling technique. The benchmarking results for radiative heat
Transfer and Surface Shear Stress,’’ Meas. Sci. Technol., 11, pp. 969–986. transfer in a unit cube with black walls were given in their papers.
关2兴 Chyu, M. K., Ding, H., Downs, J. P., Van Sutendael, A., and Soechting, F. O.,
1998, ‘‘Determination of Local Heat Transfer Coefficient Based on Bulk Mean
In present study, firstly, we use MC method to simulate the
Temperature Using a Transient Liquid Crystals Technique,’’ Exp. Therm. Fluid same problem as that in the paper by Hsu and Farmer 关4兴. In
Sci., 18, pp. 142–149. numerical simulation, a piecewise constant interpolation of radia-
关3兴 Saabas, J., Arora, S. C., and Abdel Messeh, W., 1987, ‘‘Application of the
Transient Test Technique to Measure Local Heat Transfer Coefficients Associ- tive properties should be used. However, analyzing the results of
ated With Augmented Airfoil Cooling Passages,’’ ASME Paper 87-GT-212. simulating cases in this paper, we find that the averaging method
关4兴 von Wolfersdorf, J., Hoecker, R., and Hirsch, C., 1998, ‘‘A Data Reduction of non-uniform radiative properties in each element has influence
Procedure for Transient Heat Transfer Measurements in Long Internal Cooling
Channels,’’ ASME J. Heat Transfer, 120, pp. 314 –321. on the results, especially for non-radiative equilibrium problem.
关5兴 von Wolfersdorf, J., Hoecker, R., and Sattelmayer, T., 1993, ‘‘A Hybrid Tran- The detailed conclusion needs study in the future. We think that
sient Step-Heating Heat Transfer Measurement Technique Using Heater Foils
and Liquid-Crystal Thermography,’’ ASME J. Heat Transfer, 115, pp. 319–
some mean properties adopted by Hsu and Farmer 关4兴 is not pre-
324. cise, though the influence is not obvious in some cases. Then the
关6兴 Turnbull, W. N., and Oosthuizen, P. H., 1999, ‘‘A New Experimental Tech- radiative heat transfer in a unit cube with gray walls is calculated,
nique for Measuring Surface Heat Transfer Coefficients Using Uncalibrated
Liquid Crystals,’’ Proceedings of the ASME Heat Transfer Divison. HTD—Vol. as well. In mean while, several characters of exchange factor have
364-4, pp. 121–126. been employed to estimate the performance of pseudo-random
关7兴 Turnbull, W. N., and Oosthuizen, P. H., 1999, ‘‘Theoretical Evaluation of New numbers generator and the uncertainty of Monte-Carlo simulation.
Phase Delay Methods for Measuring Local Heat Transfer Coefficients,’’ Trans.
Can. Soc. Mech. Eng., 23共3 and 4兲, pp. 361–376.
关8兴 Farmer, J. P., Seager, D. J., and Liburdy, J. A., 1997, ‘‘The Effect of Shaping Contributed by the Heat Transfer Division for publication in the JOURNAL OF
Inclined Slots on Film Cooling Effectiveness and Heat Transfer Coefficient,’’ HEAT TRANSFER. Manuscript received by the Heat Transfer Division September 14,
ASME Paper 97-GT-339. 2001; revision received May 13, 2002. Associate Editor: K. S. Ball.
2 Analysis Table 1 Conditions of different cases
2.1 Monte-Carlo Simulation for the Exchange Factor.

The total radiative exchange factor 共exchange factor兲 RD i j is de-
fined as: fraction of emissive power of element i which is ab-
sorbed by element j. The MC technique is employed to calculate
this factor. Solving a thermal radiation problem by MC implies
simulating a series of stochastic process. There are a great number
of bundles emitting from each element, then the algorithm follows
every bundle till it is absorbed by one element or goes out of the
system. If total n i bundles are emitted from element i in which n i j
bundles are absorbed by element j, so the exchange factor can be
calculated as: RD i j ⫽n i j /n i . 2. Reciprocity relationship: ␧ i F i RD i j ⫽␧ j F j RD ji , for two sur-
Several textbooks 共such as Modest 关7兴 and Yang 关8兴兲 had de- face element i and j; ␧ i F i RD i j ⫽4 ␬ a j V j RD ji , for surface
scribed the MC simulation detailed. In general, the radiative prop- element i and volume element j; and 4 ␬ ai V i RD i j
erties have little relation with temperature, so RD i j will be a con- ⫽4 ␬ a j V j RD ji , for two volume element i and j.
stant for a certain sizing grid and radiative property profile in 3. Conservation relationship: For an enclosure consisting of
medium. Because exchange factor is directly related with the M1 surface elements and M2 volume elements:
兺Mj⫽1 ␧ j F j RD ji ⫹ 兺 k⫽1 4 ␬ ak V k RD ki ⫽␧ i F i , where i denotes
1 M2
emitting power of element, we can calculate many radiative trans-
the surface element or 兺 M j⫽1 ␧ j F j RD ji ⫹ 兺 k⫽1 4 ␬ ak V k RD ki
1 M2
fer properties of the medium, by means of the factor.
⫽4 ␬ ai V i , where i denotes the volume element.
2.2 Energy Equation. In order to carry out numerical cal-
culations, the enclosure should be divided into M1 finite surface- In general, the summation relationship can be guaranteed ex-
elements and M2 finite volume-elements, so the energy equation actly, if the MC model has no theoretical errors. However, the
of element i at radiative equilibrium is expressed as follows: other two relationships cannot be agreed exactly at any time, be-
M2 M1 cause of the inevitable statistical error of MC method. If a set of
Q i⫽ 兺 4␬
j⫽1
a j V j RD ji E b 共 T j 兲 ⫹ 兺 ␧ F RD
k⫽1
k k ki E b 共 T k 兲 (1)
truly random numbers is used for the simulation, then the errors
will decrease as the number of sample increase.
One homogeneous, absorbing and scattering square medium
where Q i is the emissive energy of element i. If element i is a with optical thickness ␬ L x ⫽ ␬ L y ⫽1 and bounded by gray bound-
sub-volume, then Q i ⫽4 ␬ ai V i E b (T i ), and if element i is a sub- aries is treated as an example. The scattering albedo ␻⫽0.5 and
surface, Q i ⫽␧ i F i E b (T i ). V i and ␬ ai are the volume and the ab- linearly scattering phase function ⌽共␪兲⫽1⫹cos共␪兲 is adopted. The
sorption coefficient of volume element i; F k and ␧ k are the area emissivities of four boundaries are 0.5, 0.6, 0.7, and 0.8, respec-
and the emissivity of surface element k. tively. The 5⫻5 square grid is used with side length of 0.2. Some
results about reciprocity relationship and conservation relationship
of RD are shown in Fig. 1. We can see that both mean errors of
3 Results and Discussion reciprocity and conservation relationship are decreased with the
increase of samples emitted from every sub-element. Our MC
3.1 The Uncertainty Analysis. The exchange factor RD
code runs on Legend 2000 PIII/933. The total CPU time is
has some properties that are similar to the view factor. These can
be used as benchmarks to estimate the performance of pseudo- 1364.51 共sec兲 for 5⫻106 bundles per sub-element. The CPU time
random numbers generator, which is very important to MC pro- for MC simulation is strongly relative with the optical thickness
gram, and the uncertainty of MC simulation. For an enclosure and scattering albedo of medium, the emissivity of boundaries and
consisting of N elements, the features are shown as follows: the simulated bundles. Except the characters mentioned above, the
symmetry of RD in symmetric system is one other criterion of
1. Summation relationship: 兺 Nj⫽1 RD i j ⫽1 MC model.
Table 2 Relative difference between ␬¯ 1 and ␬¯ 2 „percent… „„ a …

Ä0.9, „ b …Ä0.1, z Ä0…
Table 3 Divergence of radiative heat fluxes at y Ä z Ä0 and sur-

face heat fluxes at x ÄÀ0.5 and y Ä0 „Case E2…
Fig. 1 The satisfaction with reciprocity and conservation rela-

tionship of RD
Table 4 Emissive power at y Ä z Ä0 „Case E3, E4 and E7… Table 8 Surface heat fluxes at x ÄÀ0.5 and y Ä0
冉
␬ 共 x,y,z 兲 ⫽a 1⫺
兩x兩
0.5 冊冉 1⫺
兩y兩
0.5 冊冉 1⫺ 冊
兩z兩
0.5
⫹b (3)
Table 5 Surface heat fluxes at x ÄÀ0.5 and y Ä0 „E3, E4 and The constants a and b for several cases are listed in Table 1. A
E7…
linear anisotropic scattering phase function is employed: ⌽( ␪ )
⫽1⫹B cos(␪), the constant B is listed in Table 1 as well. For case
E2, unity blackbody emissive power in the medium is given. For
case E3, E4, and E7, the radiative equilibrium is assumed and
only one of the black walls (x⫽⫺0.5) has unity emissive power.
The emissive power and surface flux, or radiative flux diver-
gence and surface flux for several cases are calculated. In actual
computation, 9⫻9⫻9 cubic grid with side length of 1/9 is used,
the bundles that emitted from one sub-element are 106 for case2
Table 6 Emissive power at z Ä0 „Case E3…
and 107 for all other cases, so the total bundles are 1.215⫻109 for
case2 and 1.215⫻1010 for other cases. Each element is assumed to
be homogenous for numerical calculation, although the medium
as whole is inhomogeneous. The mean extinction coefficient in
each element must be calculated firstly. In Hsu and Farmer’s paper
关4兴, we cannot find the formula for the mean radiative properties.
Comparing with the results of our calculation 共see Table 3兲, we
believe that the extinction coefficient ¯␬ 2i at center of each element
i was treated as the mean value by Hsu and Farmer. However, we
think that the average extinction coefficient should be calculated
as follows:
¯␬ li ⫽ 冕冕冕 vi
␬ 共 x,y,z 兲 d v 冒冕冕冕 vi
dv (4)
The extinction coefficients that calculated by two kinds of av-

Table 7 Emissive power at y Ä z Ä0 erage methods are different. The relative difference between them
are shown in Table 2, in which the coordinates at center of volume
element are represented as this element. The maximum relative
difference is appeared at the center of the cube. Certainly, this
kind of deviation will be reduced as the grids increasing. For case
E2, we calculated the divergence of radiative heat flux based on
¯␬ 2 and ¯␬ 1 respectively 共see Table 3兲. The relative difference pro-
file is also shown in Table 3, which is just close to the difference
distribution between ¯␬ 2 and ¯␬ 1 . In mean time, the surface heat
flux was calculated based on ¯␬ 2 and ¯␬ 1 respectively, and relative
differences between them are almost same and close to the differ-
ence between ¯␬ 2 and ¯␬ 1 in the medium-elements which are near
the surfaces. For cases E3, E4, and E7, suppose ¯␬ ⫽ ¯␬ 1 , the cal-
culating values are shown in Table 4 and 5. In these tables, the
One of the most straightforward ways of estimating the error ‘‘MC’’ indicates Hsu and Farmer’s values 关4兴, and ␴ 10 is defined
associated with the sampling result S(N) is to break up the result in Eq. 共2兲 for I⫽10.
into a number of I sub-samples S(N i ), then the variance or ad- For case E2, the energy source is unity blackbody emissive
justed mean square deviation can be obtained as follows 关7兴: power in medium and all six walls are blackbody and cold. The
I
walls only absorb energy, which can be treated as heat sink, and
1
兺关 S 共 N 兲 ⫺S 共 N 兲兴
have no reflection and emission. So this is a non-radiative-
␴ 2I ⫽ 2
(2) equilibrium problem. The extinction coefficient ␬ i 共or absorbing
I 共 I⫺1 兲 i⫽1
i
coefficient ␬ ai ⫽ ␻ ␬ i 兲 has influence on not only transferring and
where S(N)⫽1/I 兺 i⫽1
I
S(N i ) is the mean value of I S(N i ). absorbing power in medium, but also emitting power (4 ␬ ai V i E b )
of each source element. There is different emissive power in dif-
3.2 Inhomogeneous Media With Black Walls. The radia- ferent volume element due to the different absorbing coefficient.
tive heat transfer in a unit cube with black wall was calculated by So the value of the average extinction coefficient has great influ-
YIX and MC methods in Hsu and Farmer’s paper 关4兴. We simulate ence on radiative heat transfer for case E2.
the same problem in this paper. The coordinate origin lies at the However, for case E3 and E4, the source is one of the black
center of the cube and ⫺0.5⭐(x,y,z)⭐0.5(m) is the domain of walls with unity emissive power and radiative equilibrium is as-
computation. The inhomogeneous extinction coefficient (m ⫺1 ) sumed. The extinction coefficient has no influence on the source,
distribution is given by no matter which mean extinction coefficient is adopted, there is
little difference in results. Even when we simplify the inhomoge-
neous unit cube as homogeneous one, in which the mean extinc-
Electric Network Representation of the
tion coefficient ¯␬⫽0.2125 is averaged by Eq. 共3兲 in the whole Unsteady Cooling of a Lumped
cube, there is no obvious difference in emissive power of medium.
The emissive power at z⫽0 of case E3 with different extinction
Body by Nonlinear Heat Transfer
coefficients by three average methods are shown in Table 6. For Modes
case E7, the optical thickness is large, how to calculate mean
extinction coefficient has little influence on radiative transfer.
Francisco Alhama
Dpto. de Fı́sica Aplicada, E.T.S. Ingenieros Industriales,
3.3 Inhomogeneous Media With Gray Walls. In present
study, the radiative transfer in a unit cube with gray and diffuse Universidad Politécnica de Cartagena,
wall is simulated. The conditions of different cases are listed in 30203 Cartagena, Murcia, Spain
Table 1. The inhomogeneous extinction coefficient distribution is
same as above. For case E13, E14, E23, and E24, the radiative Antonio Campo1
equilibrium is assumed and one of the gray walls (x⫽⫺0.5) has College of Engineering, Idaho State University,
unity blackbody emissive power. The results are shown in Tables
7 and 8. It can be seen that the decrease of wall emissivity will
Pocatello, ID 83209
decrease the emissive power in the medium near hot surface and Keywords: Analog Techniques, Cooling, Forced Convection,
increase it near cold surface. For all our cases, a constant value of Natural Convection, Radiation
emissive power is kept as about 1/6 at x⫽y⫽z⫽0, which is just
the exact solution. It proves the precision of our model once more.
Introduction
The solution of the unsteady heat conduction equation in a solid
body is difficult because of the dependence of the temperature on
4 Conclusions the space variable共s兲 and time. In most situations, the solution
In this paper, the Monte-Carlo method is used to simulate ra- sought is approached by establishing a distributed model consist-
diative heat transfer in three-dimensional inhomogeneous, aniso- ing of a partial differential equation, the boundary conditions and
tropically scattering media. In numerical simulation, a piecewise an initial condition.
constant interpolation of radiative properties should be used. Bas- In general, the rate of heat conduction in a solid body absent of
ing upon our calculation, how to average the inhomogeneous ra- internal heat generation is dependent upon two resistances: the
diative properties in one medium element will have influence on internal resistance inside the body and the surface resistance be-
the results. Although both property evaluation methods are ap- tween the body surface and the surrounding fluid. Within this
proximate, it can obtain more reasonable results to simulate by ample framework, there are two limiting cases of importance, case
means of one suitable averaging property. The performance of 共a兲 negligible internal resistance and case 共b兲 negligible surface
pseudo-random numbers generator is very important to the preci- resistance. The former is associated with a small temperature
sion of MC model. Several characters of exchange factor have variation inside the body and a large temperature difference be-
been employed to estimate the performance of pseudo-random tween the body surface and the fluid. Putting this statement in
numbers generator and the numerical uncertainty of MC simula- perspective, the body can be considered as a ‘‘lump’’ with nearly
tion. Finally, the radiative heat transfer in a unit cube with gray uniform temperature at any instant of time. This simplified ap-
walls is simulated, and the results can be treated as a benchmark. proach establishes the basic assumption in the lumped model.
From a physical standpoint, there is an electric analogy to the
lumped model, owing to the mathematical equivalence of the
lumped heat equation to the equation that governs the voltage in a
Acknowledgments resistance-capacitance electric circuit 共the so-called RC circuit兲.
However, it is important to stress that the electric analogy in ques-
This research is supported by the National Natural Science
tion applies exclusively when the heat exchange between a solid
Foundation of China 共59806003兲.
body and a fluid occurs by a linear heat transfer mode, such as
forced convection, where the forced convection coefficient re-
mains constant during the entire cooling period.
References The objective of this technical note is to establish for the first
time the basis for the construction of advanced electric analogies
关1兴 Hsu, P. F., Tan, Z., and Howell, J. R., 1993, ‘‘Radiative Transfer by YIX
Method in Nonhomogeneous, Scattering and Non-Gray Media,’’ AIAA Journal
to the lumped model embodying heat exchange at the solid-fluid
of Thermophysics and Heat Transfer, 7共3兲, pp. 487– 495. interface by a nonlinear heat transfer mode, for instance natural
关2兴 Farmer, J. T., and Howell, J. R., 1994, ‘‘Monte Carlo Prediction of Radiative convection or thermal radiation. Due to the intrinsic nonlinearity
Heat Transfer in Inhomogeneous, Anisotropic, Nongray Media,’’ AIAA Journal of these two modes, the natural convection coefficient and the
of Thermophysics and Heat Transfer, 8共1兲, pp. 133–139.
关3兴 Hsu, P. F., and Tan, Z., 1995, ‘‘Recent Benchmarks of Radiative Heat Transfer
equivalent radiation coefficient in the lumped heat equation are no
Within Nonhomogeneous Participating Media and the Improved Method,’’ longer constants, but vary with the instantaneous temperature of
Proceedings of the Int. Symp. On Radiation Transfer, pp. 107–126. the body during the cooling period.
关4兴 Hsu, P. F., and Farmer, J. T., 1997, ‘‘Benchmark Solution of Radiative Heat
Transfer Within Nonhomogeneous Participating Media Using the Monte Carlo Mathematical Modeling
and YIX Method,’’ ASME J. Heat Transfer, 119共1兲, pp. 185–188.
关5兴 Guo, Z. X., and Maruyama, S., 1999, ‘‘Scaling Anisotropic Scattering in Ra- Fundamentally, when a solid body is immersed in a fluid at a
diative Transfer in Three Dimensional Nonhomogeneous Media,’’ Int. Com- different temperature, the heat exchange between the body surface
mun. Heat Mass Transfer, 26共7兲, pp. 997–1007. and the fluid may occur by three distinct heat transfer modes: 共1兲
关6兴 Guo, Z. X., and Maruyama, S., 2000, ‘‘Radiative Heat Transfer in Inhomoge-
neous, Nongray, and Anisotropically Scattering Media,’’ Int. J. Heat Mass
1
Transf., 43共13兲, pp. 2325–2336. Current address: Mechanical Engineering Dept., The University of Vermont,
关7兴 Modest, M. F., 1993, Radiative Heat Transfer, McGraw-Hill, Singapore. Burlington, VT 05405
关8兴 Yang, W. J., Taniguchi, H., and Kudo, K., 1995, ‘‘Radiative Heat Transfer by Contributed by the Heat Transfer Division for publication in the JOURNAL OF
the Monte Carlo Method,’’ Advances in Heat Transfer, 27, Academic Press, HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 21,
New York. 2001; revision received May 8, 2002. Associate Editor: M. L. Hunt.
forced convection, 共2兲 natural convection, and 共3兲 thermal radia-
tion. Also, there is a possibility that two of these modes are com-
bined.
For the sake of generality, the solid body in this work has been
conceived as a small metallic sphere whose volume/surface area
ratio V/A amounts to D/6, but the theory is equally applicable to
plates and cylinders. For simplicity, the chosen coolant is air at
standard atmospheric conditions. The thermal properties of both
metal and air are taken as uniform and are evaluated at the proper
film temperature T f ⫽(T i ⫹T ⬁ )/2.
Forced Convection Cooling. The cooling of a lumped sphere
by a forced flow of a single-phase viscous fluid is described by the
lumped heat equation along with the initial condition
dT
␳ sc sD ⫽⫺6h̄ 共 T⫺T ⬁ 兲 , T 共 0 兲 ⫽T i (1)
dt
In this differential equation of first order, the forced convection
coefficient, h̄, is constant and its magnitude may be taken from
the experimental-based correlation equation recommended by
Whitaker 关1兴:
Nu⫽2⫹ 共 0.4 Re1/2⫹0.06 Re2/3兲 Pr0.4 (2) Fig. 1 Electric circuit for forced convection cooling of a me-
tallic sphere
which is valid for gases and common liquids 共0.7⬍Pr⬍380兲 op-
erating within the range of 3.5⬍Re⬍7.6⫻104 . Other correlation
equations for higher Re have been proposed by Ahmed and Yo-
vanovich 关2兴. The validity of Eq. 共5a兲 covers the interval 0⬍GrPr⬍1011 and the
As the fluid velocity approaches zero, u ⬁ →0, Re→0 in Eq. 共2兲, spectrum of gases and common liquids with Pr⭓0.7. An equally
so that the fluid motion tends to disappear giving way to a no-flow handy correlation equation for GrPr⬎1011 is found in 关4兴.
condition. This limiting condition yields a constant Nusselt num- As the instantaneous temperature T of the sphere approaches
ber Nu→2 共a diffusive limit兲. Since h̄→2k a /D, the heat transfer the air temperature T ⬁ , the fluid motion tends to disappear giving
mode switches from forced convection to conduction in a station- way to a no-flow condition equivalent to Gr→0. Similar to the
ary infinite air medium around the sphere. case of forced convection, this limiting condition is characterized
The forced convection coefficient, h̄, may be isolated for air by a constant Nusselt number Nu→2 共a diffusive limit兲 in Eq. 共5a兲
共Pr⫽0.7兲 in Eq. 共2兲, leading to the function and the heat transfer mode switches from natural convection to
冋冉冊冉冊册
conduction in a stationary infinite air medium around the sphere.
1/2 2/3
ka ka u ⬁D u ⬁D The natural convection coefficient, h̄, for air 共Pr⫽0.7兲 may be
h̄⫽2 ⫹ 0.35 ⫹0.05 (3)
D D va va extracted from Eq. 共5a兲 resulting in the function
where the subscript ‘‘a’’ in the thermal properties k a and v a iden-
tifies the air. Once the air velocity u ⬁ is assigned, h̄ turns out to be
a constant quantity in the temperature domain 关 T ⬁ ,T i 兴 .
h̄⫽2
ka
D D
ka g␤a
⫹0.413 1/4
v 2a
冉冊 1/4
共 T⫺T ⬁ 兲 1/4 (6)
Introducing Eq. 共3兲 into Eq. 共1兲 and separating variables sup- where again the subscript ‘‘a’’ in the thermophysical properties
plies the exponential temperature distribution k a , v a , and ␤ a are indicative of air. Contrary to the case of linear
冉冊
forced convection cooling emblematic of a constant h̄ over the
T 共 t 兲 ⫺T ⬁ 6h̄ temperature domain 关 T ⬁ ,T i 兴 , now for natural convection Eq. 共6兲
⫽exp ⫺ •t (4)
T i ⫺T ⬁ ␳ sc sD indicates that h̄ is by no means a constant quantity over 关 T ⬁ ,T i 兴 .
In actuality, since the diameter D and the air temperature T ⬁ are
which has been the trademark of the lumped model under the
fixed quantities, h̄ is a derived quantity that varies with the instan-
assumption that the forced convection coefficient, h̄, is constant
taneous temperature, T. Specifically, for laminar natural convec-
共Kreith and Bohn 关3兴兲. In Eq. 共4兲, the quantity ␳ s c s D/6h̄ may be
tion of air, h̄ is regulated by the nonlinear temperature function
viewed as the product of the capacitance and the resistance, i.e.,
the thermal time constant, t c . h̄(T) in Eq. 共6兲1 whose range is 关 h̄ min ,h̄max兴. Substituting Eq. 共6兲
into Eq. 共1兲 leads to
冉冊
Natural Convection Cooling. The lumped heat equation for
the sphere, along with the initial condition, is given again by Eq. dT ka ka g␤a 1/4
␳ sc sD ⫽⫺12 共 T⫺T ⬁ 兲 ⫺2.478 1/4 共 T⫺T ⬁ 兲 5/4
共1兲. The natural convection mode implies that the cooling of the dt D D v 2a
sphere is driven by a buoyant upflow of a single-phase viscous (7)
fluid. In this regard, the natural convection coefficient, h̄, may be
In all likelihood, the nonlinear first-order differential equation 共7兲
quantified by the experimental-based correlation equation credited
does not admit an analytic solution. Notwithstanding, adoption of
to Churchill 关4兴:
the temperature transformation
h̄D 共 GrPr兲 1/4 ␪ ⫽T⫺T ⬁ (8)
Nu⫽ ⫽2⫹0.589 (5a)
ka f 共 Pr兲
converts Eq. 共7兲 into Bernoulli equation 共Polyanin and Zaitsev
where f (Pr) stands for a ‘‘universal’’ Prandtl number function 关5兴兲
冋冉冊册
f 共 Pr兲 ⫽ 1⫹
0.469
Pr
9/16 4/9
(5b)
1
When the buoyant air flow is turbulent, the temperature function is of the form
(T⫺T ⬁ ) 1/3, but the solution procedure is unaffected.
d␪ 6 ⑀ s␴ 1
⫹a ␪ ⫽⫺b ␪ 5/4, ␪ 共 0 兲 ⫽T i ⫺T ⬁ (9) e⫽ • (14)
dt ␳ sc s D
where the coefficients ‘‘a’’ and ‘‘b’’ are computed from the ratios The denominator of the integrand on the LHS of Eq. 共13兲 may be
broken into partial fractions. Subsequently, after evaluating the
a⫽
12k a 1
•
␳ sc s D 2
, b⫽
␳ sc s 冉冊
2.478k a g ␤ a
v 2a
1/4
•
1
D 5/4
(10)
integrals and rearranging terms, the solution of Eq. 共13兲, though in
implicit form, delivers the analytic temperature distribution
Despite that the presence of ␪ 5/4 prevents Eq. 共9兲 from being lin-
ear, the equation does admit an analytic solution via a variable
transformation. Letting u⫽ ␪ ⫺1/4, Eq. 共9兲 is modified into a linear
ln 冏冏冏
1 T⫺T ⬁ 1 T i ⫺T ⬁
⫺ ln
2 T⫹T ⬁ 2 T i ⫹T ⬁
⫺tan⫺1
T
T⬁
冏
⫹tan⫺1
Ti
T⬁ 冉冊
⫽⫺2et 冉冊
(15)
differential equation in the new variable u which is amenable to
the method of separation of variables. In the end, the analytic In this equation, it is apparent that the numerical evaluation of t in
temperature distribution may be written as terms of T is direct. Thermal radiation cooling does not lead to an
冋册
exponential solution, so a time constant, as such does not exists.
T 共 t 兲 ⫺T ⬁ a 4
冉冊
⫽ (11)
T i ⫺T ⬁ at
共 a⫹b 兲 exp ⫺b Electric Circuits
4
It should be noted that the structure of Eq. 共11兲 deviates markedly Forced Convection Cooling. When the cooling occurs by
from the structure of Eq. 共4兲 for forced convection cooling. Natu- forced convection 共a linear heat transfer mode兲, the forced con-
ral convection cooling does not lead to an exponential solution, so vection coefficient h̄ is a constant quantity and the electric circuit
a time constant as such does not exists. analogous to the thermal system is shown in Fig. 1. In the thermal
system, the thermal resistance is R⫽1/h̄, the thermal capacitance
Thermal Radiation Cooling. If a hot sphere is cooled by is C⫽ ␳ s c s D/6, the thermal potential is T⫺T ⬁ and the heat flow is
thermal radiation in air, the lumped heat equation, together with Q. In the electric circuit, the electrical resistance is R e , the elec
the initial condition is described by trical capacitance is C e , the electrical potential is E⫺E ⬁ , and the
current is j. To construct an electric circuit that would behave
dT exactly similar to the thermal system, E(t)⫺E ⬁ ⫽T(t)⫺T ⬁ , the
␳ sc sD ⫽⫺6 ⑀ s ␴ 共 T 4 ⫺T 4⬁ 兲 , T 共 0 兲 ⫽T i (12)
dt ratio ␳ s c s D/6h̄ needs to be equal to the time constant R e C e , and
the initial voltage of the capacitor is identical to the initial tem-
where T and T ⬁ are absolute temperatures. For the thermal radia- perature T i . In this way, the set of equivalences that surfaces up
tive analysis, the two key assumptions are: 共1兲 the sphere surface
is: R e ⫽1/h̄, C e ⫽ ␳ s c s D/6, E(0)⫽T i and E ⬁ ⫽T ⬁ .
is gray, and 共2兲 the configuration factor between the sphere and
When the switch S w in the electric circuit of Fig. 1 is closed at
the air space is one 关3兴.
t⫽0, the energy stored in the capacitor C e is discharged through
Although Eq. 共12兲 is nonlinear, fortunately it does admit an
the resistance R e . As a result, internal energy is stored in the
analytic solution. Thus, separating variables and integrating from
thermal system, whereas electric charge is stored in the electric
the initial condition (T i ,0) to a given arbitrary condition (T,t),
circuit.
yields
Current textbooks on heat transfer 共for instance 关3兴兲 report the
冕 T
Ti T
4
dT
⫺T 4⬁
⫽⫺e 冕
0
t
dt (13)
RC electric circuit for forced convection cooling with a constant
forced convection coefficient h̄ that involves a capacitor and a
resistor in parallel to the voltage source. Conversely, the old text-
where the coefficient ‘‘e’’ is computed from the ratio: book by Giedt 关6兴 presents a RC electric circuit containing a ca-
Fig. 2 Electric circuit for natural convection cooling of a metallic sphere
Fig. 3 Electric circuit for thermal radiation cooling of a metallic sphere
pacitor and a resistor in series to the voltage source. We opted for ⫻(g ␤ a / ␯ 2a ) 1/4(T⫺T ⬁ ) 5/4, turns out to be a function of the voltage
the latter and the electric circuit sketched in Fig. 1 is identical to 共synonymous with the temperature兲 in the capacitor.
the one used in 关6兴.
Thermal Radiation Cooling. Let us state Eq. 共12兲 in the al-
Natural Convection Cooling. For natural convection cooling ternate form
共a nonlinear heat transfer mode兲, the natural convection coeffi-
cient h̄ is no longer constant; that is, h̄ depends on the instanta- dT
neous temperature T of the sphere. For the design of the electric ␳ sc sD ⫽⫺6 ⑀ s ␴ h r 共 T 兲共 T⫺T ⬁ 兲 , T 共 0 兲 ⫽T i (18)
dt
circuit analogous to the thermal system cooled by natural convec-
tion a different avenue needs to be taken. The nonlinear, non-
homogeneous first-order ordinary differential equation 共7兲 is re- where the radiation coefficient h r (T) is given by the function
written in terms of the current j in the following form
h r 共 T 兲 ⫽ ⑀ s ␴ 共 T 2 ⫹T 2⬁ 兲共 T⫹T ⬁ 兲 (19)
j nc ⫺ j nc,1⫺ j nc,2⫽0 (16)
since T and T ⬁ are absolute temperatures.
where these elements are defined by The analogous electric circuit for the case of thermal radiation
cooling is designed by way of a voltage dependent resistor 共VDR兲
dT in series with a capacitor and a constant voltage source as pictured
j nc ⫽ ␳ s c s D (17a) in Fig. 3共a兲. The respective values of the capacitance C e
dt ⫽␳scsD and the resistance R e,r ⫽6 ⑀ s ␴ h r (T) may be readily de-
rived. The function for the VDR device is designated by G r 共see
j nc,1⫽⫺12共 k a /D 兲共 T⫺T ⬁ 兲 (17b) Fig. 3共b兲兲. The current of the source j r ⫽(T⫺T ⬁ )/R r
⫽6 ⑀ s ␴ h r (T)(T⫺T ⬁ ) is just the same current in the capacitor j c
⫽ ␳ s c s D(dT/dt). The resistance ends are the voltage difference
j nc,2⫽⫺2.478共 k a /D 1/4兲共 g ␤ a / ␯ 2a 兲 1/4共 T⫺T ⬁ 兲 5/4 (17c) T⫺T ⬁ . The initial temperature T i provides the input for the initial
voltage of the capacitor. The discharge of the capacitor is con-
In the spirit of an electric analogy, Eq. 共17a兲 is the current of a trolled by the source which in real time gives a current that de-
capacitor whose capacitance is C e ⫽ ␳ s c s D and Eq. 共17b兲 is the pends on the voltage in the capacitor. The constant voltage source
current of a resistor whose resistance is R e,1⫽D/(12k). The re- may be eliminated in the electric circuit of Fig. 3共b兲 since its
sistance ends are the voltage difference T⫺T ⬁ . Meanwhile, Eq. influence in the solution is taken into account in the equation that
共17c兲 is the current of a voltage-dependent resistor 共VDR兲 whose specifies the current in the source G r .
When the switch S w in the electric circuit of Fig. 3 is closed at
resistance is R e,2⫽2.478(k a /D 1/4)(g ␤ a / ␯ 2a ) 1/4(T⫺T ⬁ ) 1/4. The
t⫽0, the energy stored in the capacitor C e is discharged through
initial voltage of the capacitor corresponds to the initial tempera-
the resistance R e,r . As a result, internal energy is stored in the
ture T i . Now, taking into account the signs of each of the currents
thermal system, whereas electric charge is stored in the electric
according to Eq. 共16兲, the analogous electric circuit is illustrated
circuit.
in Fig. 2共a兲.
When the switch S w is closed at t⫽0 in the electric circuit of Computer Tools for Electric Circuit Analysis. Electric cir-
Fig. 2共a兲 the energy stored in the capacitor C e is discharged cuits containing linear elements, such as voltage sources, currents
through two resistors R e,1 and R e,2 . As a result, internal energy is and resistances can be analyzed in straightforward mathematical
stored in the thermal system, whereas electric charge is stored in way. Clearly, these three elements are associated with the forced
the electric circuit. The variable resistor in Fig. 2共a兲 is substituted convection mode. Important circuit elements, such as diodes, tran-
by a VDR device which for short is denoted by the symbol G nc in sistors, and voltage-dependent resistors 共VDR兲 are nonlinear and
Fig. 2共b兲. In this device, the current j nc,2⫽2.478(k a /D 1/4) computer tools for the analysis of circuits containing these ele-
ments are readily available in the market. One of these nonlinear
elements, the voltage-dependent resistor may be connected to
Experimental Validation of a
natural convection and thermal radiation modes. The best-known Combined Electromagnetic and
family of software tools for electric circuit analysis that include
linear and nonlinear elements is named Sគ imulation Pគ rogram with
Thermal Model for a Microwave
Iគntegrated Cគ ircuit Eគ mphasis 共SPICE兲共Nagel 关7兴兲. In order to use Heating of Multi-Layered Materials
SPICE, the analyst must first provide a complete description of
the electric circuit. This step is done through a series of element Using a Rectangular Wave Guide
statements, with one such statement for every element in the cir-
cuit. Second, these statements are followed by control statements,
which instruct the program what needs to be calculated. P. Rattanadecho
e-mail: phadu@blue.nagaokaut.ac.jp
Conclusions Faculty of Engineering, Thammasat University
共Rangsit Campus兲, Prathumthani, 12121, Thailand
Under the premises of the powerful lumped model, the RC
electric circuit that simulates forced convection cooling 共linear
heat transfer mode兲 with a constant convection coefficient has K. Aoki and M. Akahori
been a staple in the theory of heat conduction for many decades Department of Mechanical Engineering, Nagaoka
关3兴. However, for the equally important situations of cooling by University of Technology, 1603-1, Kamitomioka,
natural convection 共nonlinear heat transfer mode兲 or thermal ra- Nagaoka, Niigata, 940-2188, Japan
diation 共nonlinear heat transfer mode兲 the appropriate electric cir-
cuits do not exist in the heat transfer literature. These shortcom-
ings were precisely the source of motivation for undertaking this
study on advanced electro-thermal analogies. The heating of multi-layered materials by microwave heating with
When the cooling occurs by natural convection, the controlling rectangular wave guide has been investigated numerically and
natural convection coefficient is susceptible to the shape of the experimentally. The multi-layered materials, which consist of the
solid body and its orientation. Because of this, the Nu correlation layer of higher dielectric material (antireflection layer) and lower
equations may involve one, two or even three terms and the ap- dielectric material (sample), have the convergent effect of the in-
propriate electric circuits may vary slightly from the one sketched cident microwave in sample, and it can change the heating pattern
in Fig. 2. In contrast, the electric circuit for thermal radiation in the sample with ease. In this study, the effect of an antireflection
cooling regardless of the body shape is given by Fig. 3. layer thickness on the heating process is clarified in detail, con-
sidering the interference between incidents and reflected waves in
the dielectric materials. Based on a model combining the Maxwell
and heat transport equations, the results showed that when a layer
Nomenclature of lower dielectric material is attached in front of sample, the
a,b ⫽ constants in Eq. 共10兲 microwave energy absorbed and distribution of temperature
C ⫽ thermal capacitance, J/K within the sample are enhanced. The predicted results are in
Ce ⫽ electric capacitance, farads agreement with experimental results for microwave heating of
e ⫽ constant in Eq. 共14兲 multi-layered materials using a rectangular wave guide.
E⫺E ⬁ ⫽ electric potential, volts 关DOI: 10.1115/1.1495521兴
h̄ ⫽ forced or natural convection coefficient, W/m2•K
h r (T) ⫽ radiation coefficient, W/m2•K Keywords: Heat Transfer, Materials, Microwave, Modeling, Nu-
j ⫽ current flow, amperes merical Methods, Thermal
Q ⫽ heat flow, W
R ⫽ thermal resistance, K/W 1 Introduction
Re ⫽ electric resistance, ohms The microwave heating of layered materials has been studied
T⫺T ⬁ ⫽ thermal potential, K by many investigators, including Etternberg 关1兴, Gori et al. 关2兴,
Greek letters Nikawa et al. 关3兴, Ayappa et al. 关4兴 and Lin and Ghandhi 关5兴.
Although most previous investigations are deplete with numerical
␪ ⫽ temperature transformation for natural convection, T simulations of simple model in one-dimensional form, there are
⫺T ⬁ , K only few papers have been reported on microwave heating of
layered materials inside a rectangular wave guide, especially a full
comparison of the space-time evolution of temperature between
References simulated results with experimental heating data. Part of the rea-
关1兴 Whitaker, S., 1972, ‘‘Forced Convection Heat Transfer Correlations for Flow son may be that analysis of microwave heating of layered mate-
Past Flat Plates, Single Cylinders and Single Spheres,’’ AIChE J., 18, pp. rials is considerably more challenging due to the influence of di-
361–372. electric properties in each layer, resulting to the complex
关2兴 Ahmed, G. R., and Yovanovich, M. M., 1994, ‘‘Approximate Analytical Solu-
tion of Forced Convection Heat Transfer From Isothermal Spheres for all interactions of microwave field with layered materials.
Prandtl Numbers,’’ ASME J. Heat Transfer, 116, pp. 838 – 843. For microwave heating process, deep and localized heating
关3兴 Kreith, F., and Bohn, M. S., 1993, Principles of Heat Transfer, Fifth Edition, technique for the dielectric material layer are sometimes de-
West, New York, pp. 120–126. manded. When microwave fields are used, the depth of penetra-
关4兴 Churchill, S. W., 1990, ‘‘Free Convection Around Immersed Bodies,’’ in Hemi-
sphere Heat Exchanger Design Handbook, G. F. Hewitt, ed., Section 2.5.7, tion is generally shallow and it is difficult to heat deep-lying layer
Hemisphere, New York. and relatively large layered volumes without excessive surface
关5兴 Polyanin, A. D., and Zaitsev, V. F., 1995, Handbook of Exact Solutions for heating. In order to overcome this difficulty, using the multi-
Differential Equations, CRC Press, Boca Raton, FL. layered material approach is one of the best means. The authors
关6兴 Giedt, W. H., 1957, Principles of Engineering Heat Transfer, Van Nostrand,
Princeton, NJ, pp. 283–286.
关7兴 Nagel, L. W., 1975, ‘‘SPICE2: A Computer Program to Simulate Semi- Contributed by the Heat Transfer Division for publication in the JOURNAL OF
Conductor Circuits,’’ Memo No. UCB/ERL M520, Electronic Research Labo- HEAT TRANSFER. Manuscript received by the Heat Transfer Division October 10,
ratory, University of California, Berkeley, CA. 2001; revision received April 22, 2002. Associate Editor: P. S. Ayyaswamy.
suggest that the microwave energy absorbed and distribution of x-z plane. Corresponding to electromagnetic field, temperature
temperature within the sample are enhanced when a layer of lower fields also can be considered in two-dimensional model. The
dielectric material 共called antireflection layer兲 is attached in front model proposed is based on the following assumptions:
of sample 共called higher dielectric material兲.
The characteristics of microwave heating of multi-layered ma-
terials studied here are potentially applicable to design of electro- 1兲 The absorption of microwave by air in rectangular wave
magnetic hyperthermia system for the treatment of cancer. The guide is negligible.
using of lower dielectric material as a mean of antireflection layer 2兲 The walls of rectangular wave guide are perfect conductors.
allowing more of the microwave energy can deposit deep inside 3兲 All materials are non-magnetic.
the human tissues to keep them at the desired temperature eleva- 4兲 The packed bed is an isotropic medium and thermal proper-
tion. Furthermore, the concept of microwave heating of multi- ties of packed bed are constant.
layered materials can be also useful for explaining the drying 5兲 The liquid phase is incompressible fluid.
phenomenon in a fundamental level, particularly the complex in- 6兲 The effect of the sample container on the electromagnetic,
teractions of microwave field with multi-layered materials. Con- velocity and temperature fields can be neglected.
sidering microwave drying of porous materials 共Ratanadecho Maxwell’s Equation. Assuming the microwave of TE 10
et al. 关6兴兲, the drying layer takes place on a front retreating from mode, the governing equations for the electromagnetic field can
the surface into the interior of the sample dividing it into two be written in term of the component notations of electric and
layers, dry and wet layers. Inside the drying front, the sample is magnetic field intensities 共Ratanadecho et al. 关6,7兴兲
wet, i.e., the voids contain liquid water and this layer acts as
higher dielectric material. Outside the drying front, no liquid wa- ⳵Ey ⳵Hx
ter exists, all water is in vapor state and the dry layer acts as the ⫽␮ (1)
⳵z ⳵t
lower dielectric material. The changing of dry layer thickness 共or
lower dielectric material兲 would change the intensity of electric ⳵Ey ⳵Hz
field, wavelength and location of maximum microwave energy ⫽⫺ ␮ (2)
⳵x ⳵t
absorbed with respect to drying times. This phenomenon explains
why the understanding of interactions of microwave field with
layered materials much be carefully performed.
Since the microwave heating of layered material is very com-
⫺ 冉 ⳵Hz ⳵Hx
⳵x
⫺
⳵z 冊
⫽ ␴ E y ⫹␧
⳵Ey
⳵t
(3)
plicated, consequently, the study in microwave heating of layered where, permittivity or dielectric constant ␧, magnetic permeability
materials should be systematically studied. This work, the micro- ␮ and electric conductivity ␴ are given by
wave heating of multi-layered material based on a two-
dimensional model with experimental data, in which the micro- ␧⫽␧ 0 ␧ r , ␮⫽␮0 , ␴ ⫽2 ␲ f ␧ tan ␦ (4)
wave of TE 10 mode operating at a frequency of 2.45 GHz, is
completely presented in order to validate the possibility for using The dielectric properties are assumed to vary with temperature
the multi-layered materials as a heating sample. The result pre- during the heating process. To determine the functional depen-
sented here provides a basis for fundamental understanding of dence of the temperature 关8兴, the theory of mixing formulas is
microwave heating of multi-layered materials. used 共Wang and Schmugge 关9兴兲. Further, all boundary conditions
for solving Maxwell’s equations have been already presented in
previous work 共Ratanadecho et al. 关6兴兲.
2 Experimental Configuration
Heat Transport Equation. The temperature of the material
Figure 1 shows the experimental apparatus. The microwave
exposed to incident wave is obtained by solving the conventional
system was a monochromatic wave of TE 10 mode operating at a
heat transport equation with the microwave power included as a
frequency of 2.45 GHz. Microwave energy was generated by
local electromagnetic heat generation term:
magnetron 共Micro Denshi Co., model UM-1500兲, it was transmit-
ted along the z-direction of the rectangular wave guide with inside
dimensions of 110 mm⫻54.61 mm toward a water load that was
situated at the end of the wave guide. The water load 共lower
absorbing boundary兲 ensured that only a minimal amount of mi-
crowave was reflected back to the sample. The sample heated was
a packed bed of 50 mm in thickness, which was composed of
glass beads and water, called a higher dielectric material. The
antireflection layer was also a packed bed (d⫽1.0 mm) which
was composed of glass beads (d⫽1.0 mm) and air, called a lower
dielectric material. The sample and the antireflection layer are
arranged in series perpendicular to direction of irradiation via a
rectangular wave guide. During the experiment, output of magne-
tron was adjusted at 1000 W. The powers of incident, reflected and
transmitted waves were measured by a wattmeter using a direc-
tional coupler 共Micro Denshi Co., model DR-5000兲. The tempera-
ture was measured with a Luxtron fluroptic thermometer model
790 共accurate to ⫾0.5°C兲.
3 Analysis of Microwave Heating Using a Rectangular

Wave Guide
Physical Model. Figure 2 shows the physical model for the
microwave heating of multi-layered materials using rectangular
wave guide. Since microwave of TE 10 mode which propagates in
rectangular wave guide is uniformed in y-direction, the electro-
magnetic field can be considered in two-dimensional model on Fig. 1 Schematic of experimental facility
Fig. 2 Physical model
⳵T j
⳵t
⫽a j 冉
⳵ 2T j ⳵ 2T j
⳵x2
⫹ 2 ⫹
⳵z 冊
Qj
␳ j •C p j
(5)
where, T is temperature, a is thermal diffusivity, ␳ is density, c p is

heat capacity at constant pressure, and subscript j denotes lower or Fig. 3 Variation of the reflection rate versus thickness of anti-
reflection layer
higher dielectric materials. The local electromagnetic heat genera-
tion term Q j is directly depended upon the electric field distribu-
tion defined as:
along the center axis (x⫽54.61) of rectangular wave guide, for
Q j ⫽2 ␲ • f •␧ 0 •␧ r 共 tan ␦ 兲 E 2y (6) the cases of without and with antireflection layer, is presented.
The boundary conditions for solving heat transport equation are The vertical axis represents the intensity of the electric field E y ,
shown in Fig. 2. which is normalized to the amplitude of the input electromagnetic
wave, E yin .
4 Results and Discussion In Fig. 4, corresponding to that case without antireflection layer
共␦⫽0 mm兲, since the microwave passing through cavity having
The experimental results for the microwave heating of multi- low permittivity is directly irradiated to the sample having high
layered materials were compared with predictions from math- permittivity, large part of microwaves are reflected from the sur-
ematical model. Some of the input data for electromagnetic and face of the sample 共as referred to Fig. 3兲 and a stronger standing
thermo-physical properties are given in Table 1. wave with larger amplitude is formed within the cavity by the
In order to predict the electromagnetic and temperature fields, a superimposition of forward wave propagated from the wave input
finite difference time domain 共FDTD兲 method is employed. The region and of the reflected wave from the surface of the sample.
total 110⫻250 cells in computational domain are used in the nu-
merical calculation. Since the propagating velocity of microwave
is very fast compared with the rate of heat transfer, different time
steps of dt⫽1 关 ps兴 and 1 关s兴 are used for the computation of the
electromagnetic field and temperature profile and the spatial step
size of dx⫽dz⫽1 关 mm兴 is selected.
4.1 The Variation of the Reflection Rate. Figure 3 shows
the variation of the reflection rate versus the thickness of antire-
flection layer. It is found that the reflection rate strongly depends
on the thickness of antireflection layer and displays a wavy be-
havior with respect to the thickness. The maximum and minimum
reflection rates that occur in each thickness of antireflection can be
clearly seen in that figure. This variation of the reflection rate is
caused due to the interference between transmitted and reflected
waves in an antireflection layer. The predicted results are in good
agreement with the experimental results.
4.2 The Distribution of Electric Field. To understand the Fig. 4 Distribution of electric field in case without antireflec-
distribution of electrical field inside the rectangular wave guide tion layer on the sample
and the multi-layered materials, the simulation analysis is re-
quired. For different heating sample configuration, the electric
field pattern even with the same microwave power level. In Figs.
4 – 6, the simulation of the typical electric field of TE 10 mode
Table 1 The electromagnetic and thermo-physical properties

used in computations
Fig. 5 Distribution of electric field in the case that antireflec-

tion layer is attached on the sample „␦Ä32 mm…
Fig. 6 Distribution of electric field in case that antireflection
layer is attached on the sample „␦Ä16 mm… Fig. 8 Temperature distributions in case that antireflection
layer „␦Ä16 mm… is attached on the sample
However, the electric field within the sample is almost extin-

guished where the electric field attenuates owing to microwave the peaks of the temperature distribution decay slowly along the
energy absorbed, and thereafter the microwave energy absorbed is propagation direction. The distribution of temperature for case of
converted to the thermal energy. antireflection layer with thickness of 32 mm is nearly the same as
Figure 5 shows the distribution of electric field for the case of that without antireflection layer 共Fig. 7兲, except at the surface part
antireflection layer with thickness of 32 mm is attached in front of of the sample where the temperature drops due to the conduction
sample. It can be seen that the distribution of electric field for this of heat to the antireflection layer.
case is nearly the same as that without antireflection layer 共Fig. 4兲. When the antireflection layer with thickness of 16 mm is at-
Therefore, the presence of antireflection layer at thickness of 32 tached in front of sample, the distribution of temperature is higher
mm slightly affects the distribution of electric field, due to the than those cases without antireflection layer and attaching the an-
large part of microwaves are reflected from the surface 共as re- tireflection layer with thickness of 32 mm. This implies that with
ferred to Fig. 3兲. the presence of suitable antireflection thickness, a stronger stand-
Figure 6 shows the distribution of electric field for the case of ing wave with larger amplitude will be formed within the sample,
antireflection layer with thickness of 16 mm is attached in front of consequently, a higher maximum temperature will be produced.
sample. With the presence of antireflection layer, the small part of Compared to the results for other cases 共␦⫽0 mm and ␦⫽32兲 at
microwave is reflected from the surface of the sample 共as referred
to Fig. 3兲. Since the large part of microwaves can further penetrate
into the sample, the wave reflected from the sample-air interface
at the lower surface has the same order as propagating wave. The
propagating and reflected waves will contribute to the standing
wave pattern with the larger amplitude and wavelength inside the
sample, such pattern can lead to higher microwave energy ab-
sorbed in the interior in comparison with other cases.
4.3 The Distribution of Temperature Field. The predic-
tions from mathematical model are compared with experimental
microwave heating data in Figs. 7 and 8, which corresponds to
that of initial temperature of 0°C at heating times of 30 s, along
with the center axis (x⫽54.61 mm) of rectangular wave guide. It
is clearly seen that the distribution of temperature in the sample is
wavy behavior corresponding to that of electromagnetic field and
Fig. 9 Comparison between simulated results „a… and experi-

mental results „b… for microwave heating in the case that anti-
reflection layer is attached on the sample „␦Ä16 mm., t Ä30 s…:
Fig. 7 Temperature distributions as a parameter of thickness „a… simulated temperature distributions „Units: °C…; and „b…
of antireflection layer measured temperature distributions „Units: °C….
heating times of 30 s, the maximum temperature is about 78°C. Nomenclature
For other cases 共␦⫽0 mm and ␦⫽32兲 with the same microwave
power level at the same heating time, it is about 50°C, which is a ⫽ thermal diffusivity 关m2/s兴
about 56 percent higher. The distribution of temperature with re- E ⫽ electric field intensity 关V/m兴
spect to times for the case of antireflection layer with thickness of f ⫽ frequency of incident wave 关Hz兴
16 mm is also shown in Fig. 8. The wavy behavior of temperature H ⫽ magnetic field intensity 关A/m兴
distributions appear and rapidly raise with elapsed time because of P ⫽ power 关W兴
a stronger standing wave with larger amplitude is formed in the Q ⫽ electromagnetic heat generation 关W/m3兴
sample and the antireflection layer protects the reflected wave tan ␦ ⫽ dielectric loss tangent coefficient
from the surface. The predicted results are in agreement with ex- t ⫽ time 关s兴
perimental results for microwave heating of multi-layered materi- Greek letters
als using a rectangular wave guide. ␦ ⫽ layer thickness 关m兴
Furthermore, the distribution of temperature within the multi- ␧ ⫽ permittivity or dielectric constant 关F/m兴
layered materials in the vertical plane (x-z) is shown in Fig. 9. ␧r ⫽ relative permittivity or relative dielectric constant
The result shows the greatest temperature at the center of heating ␭ ⫽ thermal conductivity of materials 关W/m2K兴
sample where the microwave energy absorbed is maximum. It can ␮ ⫽ magnetic permeability 关H/m兴
be seen that the agreement between the two heating patterns is ␴ ⫽ electric conductivity 关S/m兴
good, particularly concerning the location of the hot region. ␻ ⫽ angular frequency 关rad/s兴
From this study, the result predicts the possibility of effective
depth and localized heating in a dissipative medium such as di- Subscripts
electric material. It seems that a proper selection of the suitable 0 ⫽ free space
antireflection layer thickness could lead to the increase in the pen- a ⫽ air
etration depth and the heat generation in the dielectric materials. in ⫽ input
p ⫽ particle
r ⫽ relative
w ⫽ water
x,y,z ⫽ coordinates
5 Conclusions Superscripts
The experimental and numerical of two-dimensional model for ⬘ ⫽ interfacial position
microwave heating of multi-layered materials is presented in order
to validate the possibility for using the multi-layered materials as References
a heating sample. In this study, the influence of the thickness of 关1兴 Ettenberg, M., 1985, ‘‘Microwave Hypothermia and Radiometry: One-
Dimensional Computer Models,’’ RCA Rev., 46, pp. 510–527.
antireflection layer on the microwave energy absorbed and heating 关2兴 Gori, F., Gentili, G., and Matini, L., 1987, ‘‘Microwave Heating of Porous
pattern that develop within multi-layered materials is clarified in Media,’’ ASME J. Heat Transfer, 109, pp. 522–525.
detail. The results showed that the microwave energy absorbed as 关3兴 Nikawa, Y., Katsumata, T., Kikuchi, M., and Mori, S., 1986, ‘‘An Electric
well as wavelength and distribution of temperature within the Field Converging Applicator with Heating Pattern Controller for Microwave
Hyperthermia,’’ IEEE Trans. Microwave Theory Tech., 34共5兲, pp. 631– 635.
sample are enhanced when a suitable thickness of antireflection 关4兴 Ayappa, K. G., Davis, H. T., Davis, E. A., and Gordon, J., 1991, ‘‘Analysis of
layer 共lower dielectric material兲 is attached in front of sample Microwave Heating of Materials with Temperature-Dependent Properties,’’
共higher dielectric material兲. This is due to attaching the suitable AIChE J., 37, pp. 313–322.
thickness of antireflection layer decreases the reflected wave from 关5兴 Lin, J. C., and Gandhi, Om P., 1996, ‘‘Computer Methods for Predicting Field
Intensity,’’ Handbook of Biological Effects of Electromagnetic Fields, 2nd ed.,
the surface of sample 共higher dielectric material兲, the latter arises Polk and Postow, eds., CRC Press, Boca Raton, FL.
from the fact that the large parts of microwaves can penetrate 关6兴 Ratanadecho, P., Aoki, K., and Akahori, M., 2002, ‘‘Influence of Irradiation
further into the sample. The propagating and reflected waves at Time, Particle Sizes and Initial Moisture Content During Microwave Drying of
each interface will contribute to the stronger standing wave with Multi-Layered Capillary Porous Materials,’’ ASME J. Heat Transfer, 124, pp.
151–161.
the larger amplitude and wavelength, resulting in a higher micro- 关7兴 Ratanadecho, P., Aoki, K., and Akahori, M., 2002, ‘‘The Characteristics of
wave energy absorbed in the interior in comparison with other Microwave Melting of Frozen Packed Bed Using a Rectangular Wave Guide,’’
cases. On the other hand, in other cases that the reflected wave is IEEE Trans. Microwave Theory Tech., pp. 1487–1494.
still higher resulting in a lower microwave energy absorbed in the 关8兴 Von Hippel, A. R., 1954, Dielectric Materials and Applications, MIT Press,
Boston.
interior. The predicted results are in agreement with experimental 关9兴 Wang, J., and Schmugge, T., 1980, ‘‘An Empirical Model for the Complex
results for microwave heating of multi-layered materials using a Dielectric Permittivity of Soil as a Function of Water Content,’’ IEEE Trans.
rectangular wave guide. Geosci. Remote Sens., GE-18共4兲, pp. 288 –295.

J Heat Transfer 2002 Vol 124 N5

Uploaded by

Copyright:

Available Formats

J Heat Transfer 2002 Vol 124 N5

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

J Heat Transfer 2002 Vol 124 N5

Uploaded by

Copyright:

Available Formats

Transactions Journal of

of the ASME Heat Transfer ®

OFFICERS OF THE ASME

Journal of Heat Transfer Volume 124, Number 5 OCTOBER 2002

Heat Transfer Enhancement

Keywords: Finned Surfaces, Heat Transfer, Radiation

Introduction optimum dimensions are obtained directly from explicit relations

dimensional tubular space radiator and gave correlations for the

On the basis of the previously stated assumption regarding the one

806 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

and remembering that the parameters ⑀ and ␥ are specified by the

dD 共 ⑀ u 2opt兲 ⫽ 共 u 2opt兲 s f ⫹a u exp共 ⫺b u ␥ 兲 (40)

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 807

⌬x n⫹1 ⫽0.9⌬x n 冉 TOL

is in common use and this formula is used whether or not the

冉 冊 2/3 tions which resulted in a variation of the discretization parameter

808 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

Results and Discussion

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 809

810 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 811

Keywords: Computational, Finite Difference, Flow, Heat Transfer, Numerical Methods

Introduction factor. Such numerical difficulties were encountered also in a

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 813

814 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 815

816 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

r k⫹n Û k⫹n ⫺r k⫺n Û k⫺n r k⫹1 V̂ k⫹1 ⫺r k⫺1 V̂ k⫺1

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 817

the experimental difficulties in delineating the reattachment point

818 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 819

Keywords: Convection, Cooling, Film Cooling, Heat Transfer, Turbines

Transient Liquid Crystal Technique for Film Cooling

While the above approach is generally valid for resolving heat

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 821

Since ⌬T r is related directly to the variations of the mainstream

822 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

tively. The density ratio in the present study is nominally equal to

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 823

Results and Discussion

824 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 825

826 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 827

Keywords: Combustion, Energy, Exergy, Heat Transfer, Second Law, Sprays

where, turbulent kinetic energy production rate P is given by

Fig. 1 Physical model Turbulent Kinetic Energy Dissipation Rate (␧).

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 829

S C j ⫽⫺ ␻ M f 共 1⫹ ␥ 兲 where, G ⬘ (d i ) is the mass fraction of the spray having diameter

830 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

hexane (C6 H14), kerosene(C10H20) and n dodecane(C12H26). The

Exergy Model of Spray Combustion

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 831

⫽0.0303 as recommended by Moran and Shapiro 共25兲.

832 Õ Vol. 124, OCTOBER 2002 Transactions of the ASME

Results and Discussion

Table 2 Influences of inlet pressure and swirl on combustion efficiencies and

Journal of Heat Transfer OCTOBER 2002, Vol. 124 Õ 833

冉冊 2/3 tions which resulted in a variation of the discretization parameter