J Heat Transfer 2001 Vol 123 N4

Transactions Journal of
of the ASME Heat Transfer ®
HEAT TRANSFER DIVISION Published Bimonthly by The American Society of Mechanical Engineers
Chair, L. C. WITTE
Vice Chair, J. H. KIM
Past Chair, R. A. NELSON VOLUME 123 • NUMBER 4 • AUGUST 2001
Secretary, Y. BAYAZITOGLU
Treasurer, R. D. SKOCYPEC
Editor, V. DHIR „2005… HEAT TRANSFER PHOTOGALLERY
Associate Editors,
C. T. AVEDISIAN „2002… 617 Heat Transfer Photogallery
H. H. BAU „2003… K. D. Kihm
C. BECKERMANN „2001…
A. BEJAN „2003… 618 Evaporating Capillary Pore Flows Ranging From 10-mm to 10-␮m
V. P. CAREY „2003… Diameter
F. B. CHEUNG „2002… H. J. Kim and K. D. Kihm
T. Y. CHU „2002…
B. T. F. CHUNG „2001… 619 Visualization of Atomic Force Microscopy From Molecular Dynamics
M. FAGHRI „2003… Simulations
J. G. GEORGIADIS „2003…
J. P. GORE „2002…
Tai-Hsi Fan and Andrei Fedorov
M. HUNT „2002… 620 Subcooled Pool Boiling Heat Transfer in Microgravity and Hi-g
D. A. KAMINSKI „2001… Jungho Kim, John Benton, John McQuillen, and Mary Vickerman
R. L. MAHAJAN „2001…
A. MAJUMDAR „2001… 621 Shear Layer Instability and Mixing in Micro Heat Spreaders
G. P. PETERSON „2003…
D. POULIKAKOS „2002…
C. Sert and A. Beskok
S. S. SADHAL „2002… 622 Oscillatory Entrained Droplet EHD Two-Phase Flow
R. D. SKOCYPEC „2003…
D. A. ZUMBRUNNEN „2001…
J. S. Cotton, M. Shoukri, and J. S. Chang
623 Local Fin-Surface Heat Transfer for Flow Around a Circular Cylinder With
BOARD ON COMMUNICATIONS and Without Vortex-Generating Winglets
Chair and Vice President J. E. O’Brien and Manohar S. Sohal
OZDEN OCHOA
TECHNICAL PAPERS
OFFICERS OF THE ASME
President, W. A. WEIBLEN Conduction Heat Transfer
Executive Director,
D. L. BELDEN 624 Thermal Resistance Models for Non-Circular Moving Heat Sources on a
Treasurer, Half Space
R. E. NICKELL Y. S. Muzychka and M. M. Yovanovich
PUBLISHING STAFF
633 The Steady Inverse Heat Conduction Problem: A Comparison of Methods
Managing Director, Engineering With Parameter Selection
CHARLES W. BEARDSLEY Robert Throne and Lorraine Olson
Director, Technical Publishing Forced Convection
PHILIP DI VIETRO
645 Film Cooling of a Cylindrical Leading Edge With Injection Through Rows
Managing Editor, Technical Publishing
of Compound-Angle Holes
CYNTHIA B. CLARK
Y.-L. Lin and T. I.-P. Shih
Managing Editor, Transactions
CORNELIA MONAHAN 655 Time-Resolved Thermal Boundary-Layer Structure in a Pulsatile
Production Coordinator Reversing Channel Flow
COLIN McATEER Sean P. Kearney, Anthony M. Jacobi, and Robert P. Lucht
Production Assistant 665 Effect of Turbulence With Different Vortical Structures on Stagnation
MARISOL ANDINO
Region Heat Transfer
Aung N. Oo and Chan Y. Ching
Transactions of the ASME, Journal of Heat Transfer (ISSN
0022-1481) is published bi-monthly (Feb., Apr., June, 675 Effects of Rib Arrangements on Heat Transfer and Flow Behavior in a
Aug., Oct., Dec.) by The American Society of Mechanical
Engineers, Three Park Avenue, New York, NY 10016. Rectangular Rib-Roughened Passage: Application to Cooling of Gas
Periodicals postage paid at New York, NY and additional Turbine Blade Trailing Edge
mailing offices. POSTMASTER: Send address changes to
Transactions of the ASME, Journal of Heat Transfer, c/o Robert Kiml, Sadanari Mochizuki, and Akira Murata
THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS, 22
Law Drive, Box 2300, Fairfield, NJ 07007-2300. 682 Energy Separation and Acoustic Interaction in Flow Across a Circular
CHANGES OF ADDRESS must be received at Society Cylinder
headquarters seven weeks before they are to be effective.
Please send old label and new address. R. J. Goldstein and Boyong He
STATEMENT from By-Laws. The Society shall not be
responsible for statements or opinions advanced in papers or
... printed in its publications (B7.1, Para. 3), COPYRIGHT
© 2001 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
„Contents continued on inside back cover…
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
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 123, Number 4 AUGUST 2001
Natural and Mixed Convection

688 Experimental and Numerical Study of Conjugate Heat Transfer in a Horizontal Channel Heated From Below
Wilson K. S. Chiu, Cristy J. Richards, and Yogesh Jaluria
698 Conjugate Mixed Convection With Surface Radiation From a Vertical Plate With a Discrete Heat Source
C. Gururaja Rao, C. Balaji, and S. P. Venkateshan
Evaporation, Boiling, and Condensation
703 Heat Transfer in Direct Contact Condensation of Steam to Subcooled Water Spray
Minoru Takahashi, Arun Kumar Nayak, Shin-ichi Kitagawa, and Hiroyuki Murakoso
711 Experimental and Theoretical Study of HeatingÕDrying of Moist Paper Sheet With a Gas-Fired Infrared Emitter
J. Seyed-Yagoobi and A. N. Husain
719 The Effect of Dissolving Gases or Solids in Water Droplets Boiling on a Hot Surface
Qiang Cui, Sanjeev Chandra, and Susan McCahan
Manufacturing Processes
729 Interface Shape and Thermally-Driven Convection in Vertical Bridgman Growth of Gallium Selenide: A
Semiconductor With Anisotropic Solid-Phase Thermal Conductivity
Hanjie Lee and Arne J. Pearlstein
Microscale Heat Transfer
741 Molecular Dynamics Simulation of Nanodroplet Evaporation
J. H. Walther and P. Koumoutsakos
749 Monte Carlo Study of Phonon Transport in Solid Thin Films Including Dispersion and Polarization
Sandip Mazumder and Arunava Majumdar
Heat Exchangers
760 Integrative Thermodynamic Optimization of the Crossflow Heat Exchanger for an Aircraft Environmental Control
System
Jose V. C. Vargas, Adrian Bejan, and David L. Siems
770 An Improved Design and Rating Analyses of Counter Flow Wet Cooling Towers
Jameel-ur-Rehman Khan and Syed M. Zubair
Heat Transfer Enhancement
779 Heat Transfer Enhancement to the Drag-Reducing Flow of Surfactant Solution in Two-Dimensional Channel With
Mesh-Screen Inserts at the Inlet
Peiwen Li, Yasuo Kawaguchi, Hisashi Daisaka, Akira Yabe, Koichi Hishida, and Masanobu Maeda
790 Using Porous Fins for Heat Transfer Enhancement
S. Kiwan and M. A. Al-Nimr
Thermal Systems
796 Precision Temperature Control of High-Throughput Fluid Flows: Theoretical and Experimental Analysis
Kevin M. Lawton, Steven R. Patterson, and Russell G. Keanini
TECHNICAL NOTES
803 Detailed Heat Transfer Coefficient Distributions on a Large-Scale Gas Turbine Blade Tip
Shuye Teng, Je-Chin Han, and G. M. S. Azad
810 Transport Phenomena of Developing Laminar Mixed Convection in Inclined Rectangular Ducts With Wall
Transpiration
Wei-Mon Yan and Pei-Yuan Tzeng
814 Clouds Over Soot Evaporation: Errors in Modeling Laser-Induced Incandescence of Soot
G. J. Smallwood, D. R. Snelling, F. Liu, and Ö. L. Gülder
ANNOUNCEMENTS
819 Call for Papers: Microgravity Transport Processes in Fluid, Thermal, Materials, and Biological Sciences
820 Call for Photographs: 2001 International Mechanical Engineering Congress and Exposition „IMECE…
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
Journal of
Heat Transfer
Heat Transfer
Photogallery
The fifth ‘‘Heat Transfer Photogallery’’ was held at the 2000 and promote their participation in 2001-IMECE Photogallery ses-
International Mechanical Engineering Congress and Exhibition in sion presentation 共refer to the Call for Photogallery for 2001-
Orlando last November. The Heat Transfer Visualization Commit- IMECE announced in this volume兲关1–5兴.
tee sponsored the session and attracted 13 photo displays that
illustrate phenomena that occur in the presence of a temperature References
gradient. Six entries were selected for publication in this special 关1兴 Bae, S., Kim, M. H., and Kim, J., 1999, ‘‘Improved Technique to Measure
section of the ASME JOURNAL OF HEAT TRANSFER. Time and Space Resolved Heat Transfer under Single Bubbles during Satu-
The purpose of publishing these photographs is to draw atten- rated Pool Boiling of FC-72,’’ Exp. Heat Transfer, 12, No. 3, pp. 265–278.
关2兴 Fedorov, A., and Viskanta, R., 1998, ‘‘Heat/Mass Transfer and Adsorption
tion to the innovative features of optical diagnostics and aesthetic Dynamics in a Honeycomb Adsorbent: Application of the Simplified Local
qualities of thermal processes. To focus on the visualization, the Density Model,’’ Therm. Sci. Eng., 6, No. 1, pp. 1–10.
text is kept to a minimum and further details should be found 关3兴 Kim, H. J., and Kihm, K. D., 2001, ‘‘Thermal and Flow Characteristics of
through the listed references or directly from the authors. The Evaporating Capillary Pore Ranging from 10-mm to 10-␮m Diameter,’’ 3rd
Pacific Symposium on Flow Visualization and Image Processing 共PSFVIP-3兲,
photographs include visualizations of the following: 共1兲 the effect Paper No. F3202, Maui.
of the pore diameter, ranging from 10-mm to 10-␮m, on the cap- 关4兴 O’Brien, J. E., and Sohal, M. S., 2000, ‘‘Heat Transfer Enhancement for
illary thermal and flow fields; 共2兲 molecular dynamics simulations Finned-Tube Heat Exchangers with Winglets,’’ Proceedings, 2000-ASME In-
for atomic force microscope visualization; 共3兲 comparison of boil- ternational Congress and Exposition, Orlando, HTD-Vol. 365, pp. 137–146.
关5兴 Sert, C., and Beskok, A., 2000, ‘‘Time Periodic Forced Convection Cooling in
ing bubble characteristics between 1-g and reduced gravity at Micro Heat Spreaders,’’ Proceedings, 2000-ASME International Congress and
0.05-g; 共4兲 oscillatory flow forced convection and mixing inside a Exposition, Orlando, MEMS-Vol. 2, pp. 571–580.
micro heat spreader; 共5兲 entrained drop formation from electrohy-
drodynamically driven oscillatory annular flow; and 共6兲 mapping K. D. Kihm
of local surface heat transfer coefficients with and without vortex-
generating winglets. It is now proposed that the journal readers Department of Mechanical Engineering,
enjoy viewing these collections, acquire knowledge of the state- Texas A&M University,
of-the-art features potentially applicable for their own research, College Station, TX 77843-3123
Journal of Heat Transfer Copyright © 2001 by ASME AUGUST 2001, Vol. 123 Õ 617
Downloaded 07 Dec 2010 to 194.27.225.72. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 1 Calculated flow and temperature distributions inside sidewall-heated capillary pores of different diameters
EVAPORATING CAPILLARY PORE FLOWS RANGING FROM 10-mm TO 10-␮m DIAMETER
H. J. Kim and K. D. Kihm

Texas A&M University, College Station, Texas
Temperature and velocity profiles are numerically predicted for mined as part of the iterative solutions. The pore diameters are
evaporating capillary pore flows that are driven by both thermo- ranged from 10-mm to 10-␮m for water at 1 atm. With decreasing
capillary interfacial phoresis and buoyant convection of the pore pore diameter, the convection-driven circulating flows diminish
bulk driven by the superheated 共2°C兲 pore side wall. A finite vo- and vertically migrating flows, driven by the interfacial evapora-
lume method 共FVM兲 is used with a generalized boundary-fitted tion, dominate. Note that the magnitudes of the velocity vectors
coordinate 共BFC兲 system to handle the concave meniscus surface are decreasing and then increasing with decreasing pore diameter.
geometry. The interfacial boundary conditions, the meniscus Nearly stratified temperature distribution prevails the pores smal-
shape and the rate of evaporation, had to be simultaneously deter- ler than 1-mm diameter indicating negligible convection 共Fig. 1兲.
618 Õ Vol. 123, AUGUST 2001 Copyright © 2001 by ASME Transactions of the ASME
Fig. 1 Schematic of the AFM action
Fig. 4 Image of the atomic normal stress „axial… forces
Fig. 2 Instantaneous snapshot of polymer molecular motion
Fig. 5 Image of the atomic shear stress „friction… forces

Fig. 3 Topology of the sample surface atomic structure
VISUALIZATION OF ATOMIC FORCE MICROSCOPY FROM MOLECULAR DYNAMICS SIMULATIONS

Tai-Hsi Fan and Andrei Fedorov
School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, Georgia
A computer visualization method is used to present the details the substrate via mapping the local height field by the color image
of the imaging action by the surface scanning equipment—Atomic to generate the rubbersheet-like surface. This is virtual represen-
Force Microscope 共AFM兲 from the Molecular Dynamics 共MD兲 tation of the instantaneous topographic image of the surface gen-
simulations. The purpose of the scanning process is to resolve erated by AFM. Figure 4 presents computer imaging of the short
共image兲 the surface details, to observe the tip-substrate interac- range axial intermolecular forces 共i.e., atomic normal stresses兲
tions, and to record the dynamics of transport processes in the film induced by atoms located nearby to the cutting plane, while the
down to the atomic or molecular level. The device uses a nonde- atomic level friction forces induced by the tip and polymer mol-
structive tip to probe the interatomic and intermolecular forces ecules on the sample surface are demonstrated in Fig. 5. For the
and is schematically shown in Fig. 1. Figures 2 through 5 show purpose of visualization of the unstructured data irregularly lo-
results of MD simulations of an AFM cantilever tip that probes a cated in space, the mapping surfaces are created to interpolate the
sample covered with the lubricant film. The cantilever tip is made scattering data points into correct field values on the mapping
of nickel 共Ni, shown in red color兲, the sample is gold 共Au, shown surface 共Fig. 4兲.
in gold color兲, and the lubricant film consists of several long chain
molecules of the polymer 共shown in green color兲. Figure 2 pre- Acknowledgments
sents an instantaneous snapshot of the polymer molecular motion The authors would like to thank Dr. David Luedtke, Professor
around the AFM tip induced by intermolecular interactions be- Uzi Landman from GIT School of Physics, and Dr. William Rib-
tween the molecules of the substrate, film, and the cantilever tip. arsky from GIT College of Computing for providing the MD
In Fig. 3, we show the topology of the surface atomic structure of dataset.
Fig. 1 False color images of boiling heat transfer in microgravity and Hi-g
SUBCOOLED POOL BOILING HEAT TRANSFER IN MICROGRAVITY AND Hi-g

Jungho Kim and John Benton
University of Maryland
John McQuillen and Mary Vickerman

NASA Glen Research Center
An array of microscale heaters each 0.27 mm ⫻ 0.27 mm acceleration vector within the plane of the image 共the circle
in size were held at constant temperature using electronic feed- corresponds to 0.05 g兲. The x-direction is perpendicular to the
back loops, enabling the heat transfer from each heater to be image.
determined. The above images were obtained by combining In microgravity, a large ‘‘primary’’ bubble surrounded by nu-
video images of boiling taken from below and adding false color merous smaller bubbles was observed. The primary bubble moved
according to the heat transfer from each heater in the array. The in a circular pattern on the heater array as it coalesced with the
boiling behavior in microgravity and high-g 共about 1.8 g兲 is smaller bubbles. A dry spot formed underneath this primary
shown on Fig. 1 at two wall superheats. The microgravity bubble, as indicated by the low heat transfer. The size of the
environment was provided by a KC-135. The bulk fluid primary bubble increased with wall superheat. In high-g, numer-
共FC-72兲 was at 1 atm and subcooled by 35 K. The circle at ous small bubbles that nucleate, grow, and detach very rapidly
the bottom right of each image shows the direction of the were observed.
Fig. 1 A schematic view of the MHS
Fig. 2 Boundary conditions used in numerical modeling of MHS devices „insulated q Ä0, constant heat flux q̇ Ä1 and constant
temperature T Ä0…
Fig. 3 Snapshots of temperature contours during ejection to the right reservoir. Red and blue show high and low temperatures,
respectively.
SHEAR LAYER INSTABILITY AND MIXING IN MICRO HEAT SPREADERS

C. Sert and A. Beskok
Texas A&M University, Micro-Fluidics Laboratory, College Station, Texas
The micro heat spreader 共MHS兲 is a micro-fluidic device designed for ␣⫽5.6. and Pr⫽1, respectively. Since Pr⫽1, the temperature field
thermal management of microelectronic components. It connects two res- closely follows the flow. Figs. 3共a–d兲 show snapshots of temperature con-
ervoirs by a set of micro-channels 共Fig. 1兲. The bottom surfaces of the tours during ejection to the right reservior, where shear layer instability
reservoirs are membranes that are driven with a phase difference of ␲, at the channel expansion is observed. The vortex rolls due to the shear
either by electrostatic or piezoelectric actuation. The idea is to minimize layer instability are visible in the temperature contours of Figs. 3 共b, c兲.
the chip surface temperature by oscillatory flow forced convection and The flow rapidly turns towards the oscillating membrane due to the
mixing. Numerical simulations are performed for an MHS device with influence of the end wall and the membrane motion. The temper-
channel to reservoir expansion ratio H/h⫽25. The boundary conditions ature contours show two primary counter-rotating zones with various
and the MHS geometry are shown in Fig. 2. Both the flow and temperature smaller rotating structures, which promote further mixing in the MHS
fields are time-periodic, and Womersley and Prandtl numbers are system.
Fig. 1 High speed video images of oscillatory entrained droplet EHD two-phase flow
OSCILLATORY ENTRAINED DROPLET EHD TWO-PHASE FLOW

J. S. Cotton, Ph.D.†,* and M. Shoukri, Ph.D.*
†
Long Manufacturing, Dana Corp., Oakville, Ontario, Canada
J. S. Chang, Ph.D.*
*McMaster University, Engineering, Hamilton, Ontario, Canada
In an experimental investigation of electrohydrodynamic two- the oscillations suggests that the fluctuations in the flow pattern
phase flow, the application of a 60 Hz AC voltage potential to are actually the continuous transition between flow regimes due to
convective boiling and condensation systems has led to a regime the approximate ‘‘on/off’’ electric field applied to the electrode. It
unique to any flow pattern previously observed 关1–3兴. It is an is postulated that the resultant flow regime is the continuous con-
oscillatory flow, where droplets as large as 2 mm in diameter are struction and destruction of two separate flow regimes. For ex-
entrained in a vapor core that is surrounded by an annular liquid ample, at Re⫽3500 the flow pattern in the absence of the electric
film around the circumference of the tube and electrode, resem- field is dominantly stratified flow and intermittent annular or en-
bling a multi-layered annular flow. The droplets oscillate radially trained droplet flow at high applied DC voltage levels 关1,4兴.
in the lower portion of the annulus at a frequency of approxi- Hence, the oscillatory-entrained droplet flow pattern is believed to
mately 120 Hz, twice the frequency of the applied field, occasion- be a result of the continuous flow pattern transition between these
ally being entrained by the inner or outer annular film. In addition two regimes 共Fig. 1兲.
to the droplet formation, small spouts or jets of liquid were ob-
served on the upper half of the annular film surrounding the elec- References
trode. These spouts seemed to form randomly on the crest of a 关1兴 Cotton, J. S., 2000, ‘‘Mechanisms of Electrohydrodynamic 共EHD兲 Flow and
Heat Transfer in Horizontal Convective Boiling Channels,’’ Ph.D. thesis, Mc-
wave created by interfacial instabilities present in the film and Master University, Hamilton, Ont., Canada.
would spray a fine mist into the upper portion of the vapor core, as 关2兴 Cotton, J. S., Chang, J. S., and Shoukri, M., 2001, ‘‘Numerical Simulation of
shown in the upper figure. The size of the droplets, intensity of Electric Field Distributions in Electrohydrodynamic Two-Phase Flow Re-
motion, rate of deposition and the occurrence of spouts were gimes,’’ IEEE/DEIS Transactions on Dielectric and Electrical Insulation, sub-
mitted for review.
highly dependent on the amplitude of the 60 Hz applied voltage. 关3兴 Cotton, J. S., Chang, J. S., and Shoukri, M., 2001, ‘‘Mechanisms of AC Elec-
The interaction between the phases and the extremely high in- trohydrodynamic Flow and Convective Boiling Heat Transfer in Horizontal
terfacial area, when coupled with the increased turbulent mixing Annular Channels,’’ ASME J. of Heat Transfer, to be submitted.
created by the oscillatory motion of the flow, led to significant 关4兴 Cotton, J. S., Shoukri, M. M., Chang, J. S., and Smith-Pollard, T., 2000,
‘‘Electrohydrodynamic 共EHD兲 Flow and Convective Boiling Augmentation in
enhancement of the overall Nusselt number 共⬃300 percent兲 and Single-Component Horizontal Annular Channels,’’ Proceedings of the ASME
overall pressure drop in both condensation and evaporation when Heat Transfer Division, Heat Transfer Enhancement of Multi-Phase Flow,
applied under the appropriate conditions 关1–3兴. The frequency of HTD-366, pp. 177–184.
Fig. 1 Direct comparison of local fin-surface heat transfer distributions for flow around circular cylinder without „a…
and with „b… vortex-generating winglets at ReH È1200
LOCAL FIN-SURFACE HEAT TRANSFER FOR FLOW AROUND A CIRCULAR CYLINDER

WITH AND WITHOUT VORTEX-GENERATING WINGLETS
J. E. O’Brien and Manohar S. Sohal
Idaho National Engineering and Environmental Laboratory „INEEL…, Idaho Falls, ID 83415
A direct comparison of local fin-surface heat transfer coeffi- transient in a lexan substrate. The transient localized heating is
cients for flow around a circular tube without 共a兲 and with 共b兲 quantitatively recorded using an imaging infrared camera. Values
winglets at ReH⬃1200 is presented in Fig. 1. The comparison of local heat transfer coefficient are then determined from a one-
reveals that, for this winglet location, the horseshoe vortex pro- dimensional inverse heat conduction analysis. The winglets had a
duced by the interaction of the flow with the circular cylinder is 1:2 height/length aspect ratio and were oriented at a 45 degree
disrupted by the winglets. There is a reduction in the width of the angle to the flow. The height of the winglets was 90 percent of the
low-heat-transfer wake region, but heat transfer coefficients di- channel height. The quantitative thermal visualization images are
rectly downstream of the cylinder are actually slightly reduced for
obtained using a precision imaging infrared camera 共FLIR PRISM
the winglet case compared to the no-winglet case. Stagnation-
region heat transfer coefficients are slightly higher for the winglet DS兲. The camera detector has a 12-bit digital dynamic range and
case compared to the no-winglet case. The experiments were per- a minimum discernible temperature difference 共MDT兲 of 0.1° C at
formed in a narrow rectangular duct designed to simulate a single 30° C. It is equipped with a 25-mm standard lens, which provides
passage of a fin-tube heat exchanger. A transient heat transfer a 17 deg⫻13 deg field of view. Infra-red thermography has sev-
measurement technique was employed for obtaining detailed local eral advantages over thermochromic liquid crystals for surface
heat transfer measurements on the model fin surface. Using this temperature mapping, including wide available temperature range,
technique, the room-temperature fin/tube model is suddenly ex- high spatial resolution, excellent thermal resolution, and full-field
posed to a uniformly heated airflow, initiating a heat conduction direct digital data acquisition and processing.
Thermal Resistance Models
Y. S. Muzychka
Assistant Professor
for Non-Circular Moving Heat
Faculty of Engineering and Applied Science,
Memorial University of Newfoundland,
St.John’s, NF, Canada, A1B 3X5
Sources on a Half Space
e-mail: yuri@engr.mun.ca Solutions to stationary and moving heat sources on a half space are reviewed for rect-
angular and elliptic contacts. The effects of shape, heat flux distribution, and orientation
with respect to the direction of motion are examined. The dimensionless thermal resis-
M. M. Yovanovich tance is shown to be a weak function of heat source shape if the square root of contact
Distinguished Professor Emeritus area is used as a characteristic length scale. Simple expressions are developed for cal-
Fellow ASME culating total thermal resistances of non-circular moving heat sources by combining
Department of Mechanical Engineering asymptotic solutions for large and small values of the Peclet number. Both uniform and
University of Waterloo parabolic heat flux distributions are examined. A model is developed for predicting av-
Waterloo, ON, Canada, N2L 3G1 erage or maximum flash temperatures of real sliding contacts. Comparisons of the pro-
e-mail: mmyov@mhtl.uwaterloo.ca posed model are made with numerical solutions for two cases involving non-circular
contacts. 关DOI: 10.1115/1.1370516兴
Keywords: Conduction, Contact Resistance, Heat Transfer, Modeling, Tribology
Introduction et al. 关9兴 was that the conventional approach adopted in most
tribology references was not applicable to non-circular heat
The analysis of heat transfer from sliding and rolling contacts is sources.
important in many tribological applications such as ball bearing The present work discusses various aspects of heat transfer in
and gear design. In these applications heavily loaded contacts are tribological applications involving stationary and sliding contacts.
typical and knowledge of the contact temperatures which result In all cases heat is either supplied to the contact or is generated
from frictional heat generation is required for minimizing thermal through contact friction. This paper has four objectives. These are
related problems such as scoring, lubricant breakdown, and adhe- 共i兲 provide a comprehensive review of the literature related to
sive wear due to flash welding. stationary and moving heat sources on half space, 共ii兲 examine the
A review of typical tribology books such as the texts by Halling effect that heat source shape and heat flux distribution have on the
关1兴 and Williams 关2兴, and Handbook sections by Winer and Cheng thermal resistance, 共iii兲 develop a model which is applicable to a
关3兴 and Cowan and Winer 关4兴 shows that the analysis of heat heat source of arbitrary shape and flux distribution, and 共iv兲 use
transfer from sliding or rolling contacts has not been extensively the proposed model to predict the flash temperature in a non-
modelled. These reviews generally present equations and results circular contact for real surfaces. In addressing these issues, a
for only one configuration, the circular contact. Although this con- number of gaps in the literature have been filled. In addition, a
clear and consistent approach to modeling arbitrary contacts has
tact geometry arises quite frequently in tribology applications,
been developed. Presently, the field of tribology has only adopted
others such as the elliptic contact are also quite common in ball a simplified approach in the prediction of contact temperatures
bearing and gear applications where non-conforming contacts due to sliding. The present approach does not allow for the effect
prevail 关5–7兴. of shape, aspect ratio, and flux distribution to be modelled easily.
The analysis for moving heat sources which is presented in a This was the primary motivation of the development of a hybrid
number of tribology references 关1–4兴, is based upon the assump- numerical scheme by Neder et al. 关9兴. The expressions and
tion that one of the contacts can be modelled as a stationary heat method developed in the present work have been validated against
source and the other as a fast moving heat source. In many prob- a small set of numerical data for real and ideal contacts. The
lems the assumption of a fast moving heat source may not be valid results of Neder et al. 关9兴 are readily computed using the present
and the analysis will incorrectly predict the average or maximum approach with significantly less effort.
contact temperature. With this in mind, Tian and Kennedy 关8兴
developed accurate correlations for the circular and square heat
source which predict the temperature for any speed. These corre- Governing Equations
lations were then used to formulate models for predicting flash
temperatures in sliding asperities. A review of the literature 关3,4,8,10–15兴, reveals that extensive
In a recent paper 关9兴, a hybrid computational method for non- analysis of the problem has been undertaken for various contact
circular heat sources was developed. For this method, a numerical spot shapes and thermal boundary conditions for both stationary
and moving heat sources.
approach based upon the superposition of point heat sources was
The governing equation for a moving heat source may be ob-
employed for the stationary portion and a transient finite element
tained from the transient heat conduction equation with a trans-
method was employed for the moving portion. This new approach formation of variables 关13兴. The resulting equation for steady state
was then used to predict temperatures in a steel/bronze sliding conditions, is
contact problem, with sliding motion normal and parallel to the
grinding direction. The primary motivation for the work of Neder ⳵ 2T ⳵ 2T ⳵ 2T V ⳵ T
⫹ ⫹ ⫽ , (1)
⳵x2 ⳵y2 ⳵z2 ␣ ⳵x
Contributed by the Heat Transfer Division for publication in the JOURNAL OF
HEAT TRANSFER. Manuscript received by the Heat Transfer Division April 24, where the coordinate system is fixed to the heat source and the
2000; revision received January 8, 2001. Associate Editor: A. Bejan. half space moves beneath it with velocity V, see Figs. 1 and 2.
over the region of contact, where r⫽ 冑x 2 ⫹y 2 ⫹z 2 . Solution of the
moving heat source by means of Eq. 共2兲 is rather involved, requir-
ing numerical integration. Solutions for the square and circular
contact are tabulated in Tian and Kennedy 关8兴. A simpler ap-
proach based upon the combination of asymptotic solutions is
presented in a later section for the arbitrarily shaped heat source.
Asymptotic Solutions
Stationary Heat Sources. If the velocity of the heat source is
small (V/ ␣ →0), the governing equation reduces to Laplace’s
equation
⳵ 2T ⳵ 2T ⳵ 2T
⫹ ⫹ ⫽0 (3)
⳵x2 ⳵y2 ⳵z2
with the same boundary conditions prescribed earlier.
Many solutions for stationary heat source problems have been
obtained by superposition of the point heat source 关11兴 on a half
space
Q
T⫺T b ⫽ , (4)
2 ␲ kr
where r⫽ 冑x 2 ⫹y 2 ⫹z 2 . Solutions for various heat flux distribu-
tions and source shapes have been found 关14–16兴. Of particular
Fig. 1 Rectangular heat source interest are the solutions for the rectangular and elliptical
heat sources which contain the limiting cases for the square and
circular contacts.
The thermal boundary conditions are constant or zero tempera- Moving Heat Sources. If the velocity of the heat source is
ture in regions remote from the source, i.e., r⫽ 冑x 2 ⫹y 2 ⫹z 2 large (V/ ␣ →⬁), Eq. 共1兲 simplifies to give
→⬁, T→T b , or T(x→⫾⬁,y→⫾⬁,z→⬁)⫽T b ⫽0 and pre-
scribed heat flux q over the source area ⳵ T/ ⳵ z 兩 z⫽0 ⫽⫺q(x,y)/k ⳵ 2T V ⳵ T
⫽ . (5)
while the region outside of the source area is assumed to be ⳵z2 ␣ ⳵x
adiabatic ⳵ T/ ⳵ z 兩 z⫽0 ⫽0.
Equation 共5兲 is essentially the one dimensional diffusion equation
Solution to Eq. 共1兲 is usually obtained by superposition of the
for a half-space with t⫽x/V. This equation assumes that heat
point heat source 关11兴
conduction into the half space is one-dimensional and the solution
T⫺T b ⫽ 冉冊
Q
2 ␲ kr
e ⫺V/2␣ 共 r⫺x 兲 (2)
may be approximated by the equation for heat flow at the surface
of a half space with flux specified boundary conditions 关11,12兴
2q
T⫺T b ⫽ 冑␣ t, (6)
k 冑␲
where t must be replaced by the effective traverse time t
⫽2x ⬘ /V, and x ⬘ is the distance from an arbitrary point within the
source to the leading edge of the source.
This approach was applied by Jaeger 关12兴 for the strip and
square heat sources, by Archard 关17兴 for the circular source for the
uniform heat flux distribution, and by Francis 关18兴 for the circular
heat source having a parabolic heat flux distribution. Later, it will
be applied to obtain a solution for an elliptical heat source and
comparisons will be made with the solution of Jaeger 关12兴 for a
rectangular source.
No solution was found for the equivalent isothermal moving
heat source for a circular contact. A solution for this boundary
condition may be obtained by extending the work of Francis 关18兴
or Tian and Kennedy 关8兴 for the parabolic heat flux distribution.
The solution for a moving elliptic heat source with uniform and
parabolic heat flux distribution will be obtained in a later section.
The analysis based on Eq. 共6兲 is only valid for large values of
the dimensionless group Pe⫽Va/ ␣ , or Peclet number. This group
may be interpreted as a measure of the relative thermal penetra-
tion depth, ␦ /a, of heat into the half space. Beginning with the
definition, ␦ ⫽ 冑␲ ␣ t, which is the thermal penetration depth for
heat flow into a half space, the relative penetration depth for a
circular contact is
␦ 冑␲ ␣ t
⬃ . (7)
Fig. 2 Elliptic heat source a a
Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 625
If the traverse time for a moving circular heat source is taken to Table 2 Effect of shape on isoflux stationary heat sources †11‡
be t⫽2a/V, then Eq. 共7兲 may be written as
␦
a
⬃ 冑 2 ␲ ␣ 冑2 ␲
Va
⬃
冑Pe
. (8)
Thus if Pe→⬁, the penetration ␦ is small, and may be taken to

be one-dimensional since the spreading of heat into the half-space
is negligible. On the other hand, if Pe→0, the spreading of heat
into the substrate will be significant. A solution for all values of
Peclet number can only be obtained numerically.
Review and Solution of Stationary and Moving Heat

Sources
A discussion of a number of important solutions for sliding heat
sources is now presented. In many cases, gaps existed in the lit- heat generation may be represented by Case C in Table 1 关15兴. In
erature, and the present authors have developed new solutions most analyses the assumption of a uniform heat flux distribution is
for a number of problems. These are discussed throughout the often made. The effect of heat flux distribution on the thermal
sections that follow. resistance based upon the average contact temperature is small.
The variation from the uniform flux distribution is ⫺7.4 percent
Stationary Heat Sources „Pe\0…. Extensive analysis of
for the isothermal heat source and ⫹4.1 percent for the parabolic
heat conduction from isolated heat sources on a half space has
heat source. Thus the uniform heat flux distribution may be taken
been performed by a number of researchers 关11,14–16,19兴. The
as representative of the mean value if the exact flux distribution is
simplest contact geometry is the circular contact. The analysis has
not known. If the resistance is based upon the maximum source
been performed for three heat flux distributions: the uniform heat
temperature, the variation from the uniform flux distribution is
flux, parabolic heat flux, and the inverse parabolic heat flux. The
⫺21.4 percent for the isothermal heat source and ⫹17.9 percent
inverse parabolic heat flux represents a uniform temperature dis-
for the parabolic heat source. In both cases the maximum and
tribution over the contact area. The solutions for the dimension-
minimum values for the average or maximum source temperature
less thermal resistance for these three cases are summarized in
are bounded by the solutions for Case A and Case C.
Table 1.
Table 2 presents a comparison of the dimensionless thermal
The thermal resistance may be defined with respect to the
resistance for a circular and square heat source with uniformly
average surface temperature such that
distributed heat flux 关11兴. In both cases, the dimensionless thermal
T̄ c ⫺T b resistance is greater for the circular heat source than for the square
R̄⫽ (9) source. The relative differences are 12.2 percent for the resistance
Q based upon the average source temperature and 11.3 percent for
or with respect to the maximum surface temperature such that the resistance based upon the maximum source temperature.
Later, it will be shown that if L⫽ 冑A, the effects of source shape
T̂ c ⫺T b and aspect ratio are minimized. This will eventually lead to a
R̂⫽ . (10) simplified model for an arbitrarily shaped moving heat source.
Q
If the shape of the heat source shown in Figs. 1 and 2 is allowed
A dimensionless thermal resistance may be defined as to vary with aspect ratio ⑀ ⫽b/a, then the solutions are somewhat
R * ⫽RkL, (11) more complex than those given in Table 2. The solution for the
dimensionless thermal resistance of a stationary rectangular
where L is an appropriate characteristic length related to the heat uniform heat source 关11兴 is
source area 关5,6,14–16兴. This thermal resistance is a spreading
resistance due to the transfer of heat through a finite discrete point
of contact. Spreading resistance concepts appear in any analysis R̄ka⫽
2␲ 再
1 sinh⫺1 共 ⑀ 兲
⑀
⫹sinh⫺1 共 1/⑀ 兲
冋册冎
of stationary or sliding contact problems in heat transfer and
tribology, and form the basis for the field of thermal contact 1 1 共 1⫹ ⑀ 2 兲 3/2
⫹ 2⫹ ⑀ ⫺ (12)
conductance. 3 ⑀ ⑀2
In most tribological applications involving frictional heat gen-
eration, the average heat flux q̄⫽Q/A is known. What may not be for the average contact temperature, and
known precisely, is the distribution of heat flux q(x,y) over the
contact. If the contact is Hertzian, the distribution of frictional R̂ka⫽
2␲ 再
1 sinh⫺1 共 ⑀ 兲
⑀
⫹sinh⫺1 共 1/⑀ 兲冎 (13)
for the maximum contact temperature.

Table 1 Effect of boundary conditions on stationary circular The solution for the elliptic heat source was obtained by
heat source †15‡ Yovanovich 关5,14–16兴 and is given by
1
R̂ka⫽ K共 ⑀ ⬘ 兲 (14)
2␲
for the isothermal contact, and
16
R̄ka⫽ K共 ⑀ ⬘ 兲 (15)
3␲3
and
2
R̂ka⫽ K共 ⑀ ⬘ 兲 (16)
␲2
626 Õ Vol. 123, AUGUST 2001 Transactions of the ASME
for the isoflux contact, where ⑀ ⬘ ⫽ 冑1⫺ ⑀ 2 , and K( ⑀ ⬘ ) is the com- Table 4 Effect of shape on isoflux moving heat sources
plete elliptic integral of the first kind of complementary modulus †12,17‡
⑀⬘. Equations 共15兲 and 共16兲 were obtained by comparing the lim-
iting case of the circular contact from Table 1 with the result for
the isothermal contact. Equations 共15兲 and 共16兲 accurately predict
the numerical results presented in 关16兴 for the elliptic contact
which were obtained using the method of superposition of point
heat sources. No solution was available for the parabolic flux
distribution. A solution for this configuration is easily obtained by
analogy with the elastic contact problem 关20兴, or by comparison
with the solutions presented above. The effect of aspect ratio on a
stationary elliptic heat source is
2
f 共 ⑀ 兲⫽ K共 ⑀ ⬘ 兲 (17)
␲
Thus the solution for the parabolic heat flux distribution is the Finally, Table 4 presents a comparison of the asymptotic solutions
function f ( ⑀ ) multiplied by the values for the resistance for Case for the fast moving heat source for the circular and square heat
C in Table 1: sources. The results are 16.4 percent and 21.5 percent higher for
9 the circular heat source for the thermal resistance based upon the
R̄ka⫽ K共 ⑀ ⬘ 兲 (18) average and maximum source temperatures, respectively.
16␲
If the contact is rectangular the thermal resistance will vary
and with aspect ratio ⑀ ⫽b/a, where 0⬍ ⑀ ⬍⬁. The solution obtained
by Jaeger 关12兴 for the strip source is applicable to a rectangular
3
R̂ka⫽ K共 ⑀ ⬘ 兲 . (19) heat source since the solution assumes one dimensional heat flow
4␲ into the half space, i.e., the penetration depth is small compared
Equation 共19兲 may also be derived from the analogous elastic with the characteristic dimension of the contact zone. The solution
contact problem discussed in 关20兴, for the Hertzian pressure for the finite rectangular source 关12兴 is
distribution.
Moving Heat Sources „Pe\ⴥ…. Solutions for moving heat
sources have been obtained for a number of configurations and
R̄ka⫽
3 冑␲
&
冉冊冑
a
b
1
Pe
(21)
boundary conditions. All of the moving source solutions are writ- for the average contact temperature, and
冉冊冑
ten in terms of the Peclet number. The Peclet number is defined as
& a 1
R̂ka⫽ (22)
2 冑␲
VL b
Pe⫽ , (20) Pe
␣
for the maximum contact temperature, where Pe⫽Va/ ␣ , is based
where L is a characteristic length scale representative of the con- upon the half width of the rectangle in the direction of motion, see
tact geometry. If the contact geometry is circular or square then Fig. 1.
L⫽a, the radius of the contact or the half side length of the No solution was found for the fast moving elliptical contact. In
square. Later, it will be shown that if L⫽ 冑A, the area of the heat order to obtain a solution for the elliptical contact, the approach
source, the effect of shape and aspect ratio on the dimensionless developed by Jaeger 关12兴 for the square source and by Archard
resistance is small. 关17兴 for the circular contact was applied. In this case the effective
The effect of heat flux distribution 共uniform or parabolic兲 on the contact time is
冑冉冊
thermal resistance for a moving circular heat source is given in
Table 3. The solution for the uniform heat flux distribution was y2
obtained by Archard 关17兴 and the solution for the parabolic heat 2 a 2 1⫺
2x ⬘ b2
flux distribution was obtained by Francis 关18兴. These solutions are t⫽ ⫽ . (23)
V V
only valid for large values of the Peclet number. The effect of flux
distribution on the thermal resistance based upon the average con- Applying the approach of Jaeger 关12兴 and Archard 关17兴 gives
冉冊冑
tact temperature is small. The relative difference being only 1.6
percent. The relative difference increases to 15.9 percent for the 1 a 1
R̄ka⫽ (24)
thermal resistance based upon the maximum source temperature. ␲ b Pe
for the dimensionless thermal resistance based upon the average
contact temperature, and
Table 3 Effect of boundary condition on a moving circular
heat source †17,18‡
R̂ka⫽
2& a
␲ 3/2 b 冉冊冑 1
Pe
(25)
for the dimensionless thermal resistance based upon the maximum

contact temperature. In both cases, Pe⫽Va/ ␣ is based upon the
half width of the heat source in the direction of motion, see Fig. 2.
Comparison of Eqs. 共21兲 and 共22兲 and Eqs. 共24兲 and 共25兲 with
the solutions for the square and circular heat source provided in
Table 4, shows that the solutions are identical except for the term
(a/b). This factor accounts for the effect of heat source aspect
ratio with respect to the direction of motion. These results may be
applied to infer the following solutions for a fast moving elliptic
heat source with parabolic heat flux distribution:
冉冊冑
Table 5 Dimensionless resistance for stationary isoflux hy-
a 1 perelliptic contacts †16‡
R̄ka⫽0.323 (26)
b Pe
and
R̂ka⫽0.589 冉冊冑
a
b
1
Pe
. (27)
In the next section, the results will applied to develop new

models applicable to a real contacts of non-circular shape.
Analysis of Real Contacts

In this section, application of the theory of moving heat sources
to real contacts is discussed. A simple approach to modelling the
effects of shape, aspect ratio, orientation, and heat flux distribu-
tion is presented. It will be assumed that the shape of a real con-
tact is elliptic, and that classic Hertzian analysis for elastic contact
of non-conforming surfaces may be used to predict the contact
zone dimensions 关21兴.
Effect of Contact Shape. Hertzian theory may be used to
predict the contact size for elastic contact. However, this assumes
the shape of the contact is elliptic. Depending on the surface to-
pography, this assumption may not be valid. Thus, it is desirable
to examine the effect that shape and aspect ratio have on the
overall resistance of moving heat sources.
Yovanovich et al. 关16兴 examined the effect of the shape and
aspect ratio of an isolated stationary contact having a uniform flux
distribution. The geometry examined by Yovanovich et al. 关16兴
was the hyperellipse, defined as
R̄ s k 冑A⫽
冑⑀ s
␲ 再 sinh⫺1 共 ⑀ s 兲
⑀s
⫹sinh⫺1 共 1/⑀ s 兲
冉冊冉冊
x
a
␥
⫹
y
b
␥
⫽1, (28) ⫹
1 1
冋
2 ⫹ ⑀ s⫺
3 ⑀s
共 1⫹ ⑀ 2s 兲 3/2
⑀ 2s 册冎 (33)
where a and b are the semi-major and semi-minor axis lengths, for the average contact temperature, and
再冎
respectively. The parameter ␥ determines the shape of the contact.
The values of the parameter ␥ which were examined by Yovanov-
冑⑀ s sinh⫺1 共 ⑀ s 兲
R̂ s k 冑A⫽ ⫹sinh⫺1 共 1/⑀ s 兲 (34)
ich et al. 关16兴 were ␥ ⫽1/2, ␥ ⫽1, and ␥ ⫽2. If ␥ →⬁ the hyper- ␲ ⑀s
ellipse becomes a rectangular contact. Yovanovich et al. 关16兴 for the maximum contact temperature. Since the effect of aspect
showed that if the thermal resistance is non-dimensionalized using ratio is small, average values of the dimensionless resistance
the square root of the contact area, the solutions are weak func- given in Table 5 may be used to approximate the resistance for a
tions of shape and aspect ratio. heat source with variable aspect ratio.
Table 5 presents the dimensionless resistance based upon the In the case of the moving heat source the Peclet number should
average and centroidal values of temperature for different values also be based upon the square root of the contact area, i.e., L
of the parameter ␥, Yovanovich et al. 关16兴. It is clearly seen that ⫽ 冑A in Eq. 共20兲. Table 6 summarizes the results for the rectan-
the dimensionless resistance varies very little with aspect ratio gular and elliptic heat sources for different heat flux distributions,
⑀ ⫽b/a and shape parameter ␥. The solutions for the isoflux when the resistance is non-dimensionalized using the square root
stationary elliptic heat source become
16
R̄ s k 冑A⫽ 冑␲ ⑀ s K共 ⑀ s⬘ 兲 (29)
3␲3 Table 6 Dimensionless resistance of moving heat sources on
a half-space
for the average contact temperature, and
2
R̂ s k 冑A⫽ 冑␲ ⑀ s K共 ⑀ s⬘ 兲 (30)
␲2
for the maximum contact temperature. If the flux distribution is
parabolic, then the solutions presented earlier become
9
R̄ s k 冑A⫽ 冑␲ ⑀ s K共 ⑀ s⬘ 兲 (31)
16␲
and
3
R̂ s k 冑A⫽ 冑␲ ⑀ s K共 ⑀ s⬘ 兲 . (32)
4␲
If the heat source is rectangular, then the dimensionless thermal
resistance, Eqs. 共12, 13兲 become
and small values of the Peclet number. If the contact is moving at
moderate speeds 0.1⬍Pe⬍10, a composite solution is required.
Tian and Kennedy 关8兴 combined the asymptotic results for the
circular heat source using an equation which is a special case of
the more general form
1 1 1
⫽ ⫹ . (36)
R nt R ns R nm
Equation 共36兲 is one form of the asymptotic correlation method
proposed by Churchill and Usagi 关22兴. This method allows the
combination of asymptotic solutions, to generate a model which is
valid for all values of the dependent parameter.
A single value of n⫽2 was found to give excellent agreement
between the approximate model and numerical results of Tian and
Kennedy 关8兴 over the entire range of Peclet numbers for the cir-
Fig. 3 R̂ a* versus Pea for a circular and square moving heat cular heat source having a uniform or parabolic heat flux distribu-
source tion and a square heat source having a uniform heat flux distribu-
tion. Thus, the parameter n does not appear to depend upon the
shape of the source or the flux distribution. The models developed
by Tian and Kennedy 关8兴 are specifically for the circular heat
of the contact area. Comparisons of the dimensionless resistance source for uniform and parabolic flux distributions. They are not
R̂ * are provided in Figs. 3 and 4 using the data for the exact applicable to elongated contacts such as elliptic or rectangular
analytical solutions of a isoflux heat sources from Tian and contacts. In addition, Tian and Kennedy 关8兴 presented their corre-
Kennedy 关8兴. Figure 3 shows a comparison of the results for a lations in terms of contact temperatures; rather than thermal resis-
circular and square heat source when the characteristic length is tance. The use of thermal resistance facilitates the calculation of
L⫽a. In Figure 4, the results for each geometry have virtually the partition of heat into the contacting bodies.
collapsed onto one another when L⫽ 冑A. Thus the effect of the A general model for a moving heat source will now be obtained
shape of the heat source is not a significant factor, when the re- by combining the dimensionless resistances for a stationary and
sults are appropriately non-dimensionalized. fast moving heat sources in the form of Eq. 共36兲. As noted earlier,
If the heat source is rectangular or elliptical the Peclet number the definition of aspect ratio is different for the moving and sta-
must be replaced with a modified Peclet number Pe* defined as tionary heat sources. The aspect ratio of the stationary heat source
is now denoted by ⑀ s ⫽b/a such that 0⬍ ⑀ s ⬍1, and the aspect
冑A ⫽ 共 ⑀ m 兲 Pe冑A .
1/2
Pe* (35)
ratio of the moving heat source is now denoted by ⑀ m such that
The aspect ratio ⑀ m ⫽b/a now accounts for the effect of the 0⬍ ⑀ m ⬍⬁. Also, since the effect of shape has been shown to be
shape and orientation of the heat source. Since a rectangular or negligible, only the solution for the elliptic heat source will be
elliptic heat source may be oriented in the direction of motion considered in the model development.
parallel to the short or long axis of the heat source, the resistance Combining the stationary and moving heat source solutions for
must change based on orientation. Given the same source area and both the average and maximum contact surface temperatures gives
velocity, the Peclet number remains unchanged, but the resistance
will decrease if the direction of motion is parallel to the short axis 0.750
of the heat source. Thus, for a moving heat source, 0⬍ ⑀ m ⬍⬁. R̄ t k 冑A⫽ (37)
冑共 ⑀ m 兲 1/2
Pe冑A ⫹6.05/共 ⑀ s K2共 ⑀ s⬘ 兲兲
This is quite important for the moving heat source since the resis-
tance will increase with decreasing ⑀ m , i.e., a⬎b, and decrease and
with increasing ⑀ m , i.e., b⬎a. If the heat source is stationary, the
orientation of the contact is not important and 0⬍ ⑀ s ⬍1. 1.200
R̂ t k 冑A⫽ (38)
Models for 0ËPeËⴥ. In the previous sections the thermal 冑共 ⑀ m 兲 1/2
Pe冑A ⫹11.16/共 ⑀ s K2 共 ⑀ s⬘ 兲兲
resistances of isolated stationary and fast moving contacts were
presented. These solutions represent asymptotic solutions for large for the uniform flux distribution, and
0.762
R̄ t k 冑A⫽ (39)
冑共 ⑀ m 兲 1/2Pe冑A ⫹5.77/共 ⑀ s K2共 ⑀ s⬘ 兲兲
and
1.390
R̂ t k 冑A⫽ (40)
冑共 ⑀ m 兲 1/2Pe冑A ⫹10.79/共 ⑀ s K2 共 ⑀ s⬘ 兲兲
for the parabolic flux distribution. These expressions can now be
applied to arbitrarily shaped heat sources for all values of the
Peclet number.
Bounds on Thermal Resistance. The expressions developed
previously for the elliptic heat source assume that it is oriented
with one of the axes parallel to the direction of motion, see Fig. 2.
Equations 共37–40兲 are not valid for a heat source oriented at an
angle to the direction of motion. The solution for this case re-
quires integration over the surfaces of the oblique orientation.
Fig. 4 R̂ *
冑A versus Pe冑A for a circular and square moving heat
However, Eqs. 共37–40兲 may be used to bound the values by con-
source sidering the results for the two extreme cases of ⑀ m . The limiting
cases may then be averaged in a number of ways. It is proposed Table 7 Comparison of average flash temperature results †12‡
that the bounding values be weighted using following expression
x cos 共 ␪ 兲 ⫹R *
R * ⫽R * y sin 共 ␪ 兲 ,
2 2
(41)
where R *x denotes the dimensionless resistance based upon the
aspect ratio in the x-direction, and R *
y denotes the dimensionless
resistance based upon the aspect ratio in the y-direction. This for-
mulation is proposed since it reduces to the limiting cases for ␪
⫽0 deg and ␪ ⫽90 deg, and returns the arithmetic average at ␪
⫽45 deg. In the absence of an exact solution, Eq. 共41兲 should
provide good results for the average or maximum temperature
prediction.
Prediction of Flash Temperatures. The concept of maxi-
mum or average flash temperature is discussed in detail in Ar-
chard 关17兴, Blok 关23兴, Winer and Cheng 关3兴, and Cowan and
Winer 关4兴. The computation of the flash temperature assumes that
one of the contacting surfaces is a stationary heat source and the
other a moving heat source. By accounting for the partition of heat
into each of the surfaces, an estimate for the average or maximum
temperature may be obtained. ing heat source Eqs. 共37–40兲 which were presented in the previ-
In the previous section a general model for an isolated moving ous section. Thus, T c may be computed for either the maximum or
source for 0⬍Pe⬍⬁ was developed. This model may now be average value which occurs within the contact for either the isof-
used to predict the average or maximum flash temperatures. The lux or parabolic flux distribution. The validity of Eq. 共46兲 may be
analysis begins by defining the total heat flow and the partition of questioned on the grounds that the temperature distribution of the
each into the two contacting surfaces. The total heat generated by real contact will be skewed, however, models for the stationary
sliding friction is denoted Q g ⫽ ␮ FV, where ␮ is the coefficient of heat source resistance assume a symmetric temperature profile.
friction, while the heat which flows into the stationary and moving For a slow moving contact Pe⬍0.1 the profile is nearly symmetric
surfaces are denoted Q s and Q m , respectively. Through conser- and the partition of heat into each of the surfaces is equal assum-
vation of energy, the total heat flow is then ing that each surface has the same thermal properties. For a fast
Q g ⫽Q s ⫹Q m , (42) moving heat source Pe⬎10, the maximum temperature is located
at or near the trailing edge. Most of the heat will be conducted
which may be written in terms of the temperature excess and into the moving surface since it has a lower thermal resistance. In
resistance in each surface the transition region 0.1⬍Pe⬍10, the effect of temperature distri-
bution shape should be small since the maximum temperature is
共 T s ⫺T b,s 兲共 T m ⫺T b,m 兲 located between the centroid and the trailing edge. Thus, Eq. 共46兲
␮ FV⫽ ⫹ . (43)
Rs Rm may be applied for either the average or maximum contact tem-
perature basis. Also, due to the relatively short contact times and
Now if perfect thermal contact is assumed at the interface, then size of asperities, the penetration depth will be small and the
T s ⫽T m ⫽T c at all points within the contact, and the expression assumption of a half-space is then reasonable.
given above may be solved for T c Table 7 presents a comparison of results computed by Jaeger
T b,s T b,m 关12兴 for mild steel k⫽60.3 W/mK, ␣ ⫽17.7⫻10⫺6 m2/s, F
␮ FV⫹ ⫹ ⫽400 g, and ␮ ⫽0.23 for a square source with half side length
Rs Rm
T c⫽ . (44) a⫽1⫻10⫺5 m. The maximum difference between the model and
1 1 the data of Jaeger 关12兴 is 2.1 percent at V⫽15 m/s. This error is
⫹
Rs Rm small considering that the values presented by Jaeger 关12兴 were
based upon graphical results which are also subject to round off
The general expression given above may be applied to any errors.
combination of slow, moderate or fast moving sources using the In the final example, the proposed model is compared with
expressions developed earlier. If the bulk temperatures are equal, numerical results reported by Neder et al. 关9兴. The system exam-
Eq. 共44兲 may be further simplified. In the case of a typical sliding ined by Neder et al. 关9兴 consisted of a bronze substrate ( ␣
asperity contact, the system is modelled as a stationary heat
⫽13.8⫻10⫺6 m2/s, k⫽50 W/mK) and a steel slider ( ␣ ⫽20.0
source and a moving heat source in parallel. Equation 共44兲 may be
written in terms of the dimensionless thermal resistance ⫻10⫺6 m2/s, k⫽62 W/mK). The maximum pressure considered
was P⫽450 MPa and the coefficient of friction ␮ ⫽0.25. Neder
R * ⫽Rk 冑A to give
et al. 关9兴 considered sliding in directions perpendicular and paral-
T b,s 冑Ak s T b,m 冑Ak m lel to the grinding direction of a real surface. Some difficulty was
␮ FV⫹ ⫹ encountered interpreting the data reported by Neder et al. 关9兴. The
R s* Rm * authors reported a range for the equivalent diameter of the largest
T c⫽ , (45)
冑Ak s 冑Ak m real contact spot in the direction parallel and perpendicular to the
⫹ direction of sliding along with the contact width, 2a, in the slid-
R s* *
Rm
ing direction only. The equivalent diameters which were tabulated
which may be further simplified to give by Neder et al. 关9兴 have different values for each direction. It is
assumed that if the same surface was considered, the equivalent
* 兲 / 冑A⫹T b,s k s R m* ⫹T b,m k m R s*
共 ␮ FVR s* R m diameter should be the same in both sliding directions, and that
T c⫽ , (46)
* ⫹k m R s*
k sR m the contact spot aspect ratio may be computed assuming an ellip-
tic contact. The area is given by A⫽ ␲ D 2e /4⫽ ␲ ab. Given D e and
where the value for R s* is taken to be the appropriate value of the 2a, 2b is computed using (8 ␮ m⬍D e ⬍10 ␮ m) and (6 ␮ m
stationary dimensionless resistance Eqs. 共29–32兲 and R m * is the ⬍2a⬍8 ␮ m). These dimensions were examined for both sliding
appropriate expression for the dimensionless resistance of a mov- directions and the maximum flash temperatures were computed
Table 8 Comparison of maximum flash temperature results T b ⫽ bulk temperature, K
†9‡ T c ⫽ contact temperature, K
V ⫽ velocity, m/s
Greek Symbols
␣ ⫽ thermal diffusivity, m2/s
␦ ⫽ thermal penetration depth, m
⑀ ⫽ ellipticity or aspect ratio, ⬅b/a
⑀ ⬘ ⫽ complementary modulus, ⬅ 冑1⫺ ⑀ 2
⑀ m ⫽ aspect ratio, moving source, 0⬍b/a⬍⬁
⑀ s ⫽ aspect ratio, stationary source, 0⬍b/a⬍1
␾ ⫽ angle between principal planes of contact, rad
␥ ⫽ hyperellipse parameter
using Eq. 共54兲 along with the expressions for the dimensionless ␪ ⫽ temperature excess, K
resistance based upon the maximum contact temperature. Results ␮ ⫽ coefficient of friction
are summarized in Table 8. The predicted temperature range is in ␯ ⫽ Poisson’s ratio
excellent agreement with the reported values given by Neder et al. ␪ ⫽ angle, rad
关9兴. Neder et al. 关9兴 also reported values of T̂⫽32.8°C and T̂ Superscripts
⫽37.5°C on two plots for the V⫽10 m/s case, in the direction
perpendicular and parallel to the grinding direction, respectively. (•) ⫽ based on the average temperature
ˆ ⫽ based on maximum temperature
(•)
These results are also within the range of temperatures computed
using the proposed model. Subscripts
Summary and Conclusions 1,2 ⫽ surface 1, surface 2
冑A ⫽ based on L⫽ 冑A
A review of the important literature for stationary and moving b ⫽ bulk
heat sources was presented. The effects of shape and heat flux c ⫽ contact
distribution for elliptic and rectangular heat sources were exam- g ⫽ generated
ined. It was shown that the dimensionless thermal resistance is a L ⫽ arbitrary length scale L
weak function of shape for stationary and moving heat sources m ⫽ moving
when the results are non-dimensionalized using the square root of s ⫽ stationary
the heat source area. A simple model for all values of the Peclet t ⫽ total
number was developed by combining the asymptotic solutions for
stationary and moving heat sources. A method was proposed for
predicting the thermal resistance of an elliptic heat source oriented
at any angle with respect to the direction of motion. These results
References
were then applied to develop a general expression for determining 关1兴 Halling, J., 1975, Principles of Tribology, MacMillan Education Ltd.
关2兴 Williams, J. A., 1994, Engineering Tribology, Oxford University Press.
the flash temperature for real surfaces in sliding contact. The pro- 关3兴 Winer, W. O., and Cheng, H. S., 1980, ‘‘Film Thickness, Contact Stress and
posed model was compared with recent numerical data for real Surface Temperatures,’’ in Wear Control Handbook, ASME Press, New York,
contacts of non-circular shape. Excellent agreement between the pp. 81–141.
model and data was obtained for both elliptic and square contacts. 关4兴 Cowan, R. S., and Winer, W. O., 1992, ‘‘Frictional Heating Calculations,’’ in
ASM Handbook, Volume 18 Friction, Lubrication, and Wear Technology,
ASM International, pp. 39–44.
Acknowledgments 关5兴 Yovanovich, M. M., 1971, ‘‘Thermal Constriction Resistance Between Con-
tacting Metallic Paraboloids: Application to Instrument Bearings,’’ AIAA
The authors acknowledge the financial support of the Natural Progress in Astronautics and Aeronautics: Heat Transfer and Spacecraft Con-
Sciences and Engineering Research Council of Canada under op- trol, 24, J. W. Lucas, ed., MIT Press, pp. 337–358.
关6兴 Yovanovich, M. M., 1978, ‘‘Simplified Explicit Elastoconstriction Resistance
erating grant A7455. The first author would also like to thank Dr. Expression for Ball/Race Contacts,’’ AIAA Paper 78–84.
J.B. Medley of the Department of Mechanical Engineering, 关7兴 Bejan, A., 1989, ‘‘Theory of Rolling Contact Heat Transfer,’’ ASME J. Heat
University of Waterloo. Transfer, 111, pp. 257–263.
关8兴 Tian, X., and Kennedy, F. E., 1994, ‘‘Maximum and Average Flash Tempera-
Nomenclature tures in Sliding Contacts,’’ ASME J. Tribol. 116, pp. 167–174.
关9兴 Neder, Z., Varadi, K., Man, L., and Friedrich, K., 1998, ‘‘Numerical and Finite
a ⫽ semi-major axis of ellipse or rectangle, radius of Element Contact Temperature Analysis of Steel-Bronze Real Surfaces in Dry
Sliding Contact,’’ ASME/STLE Tribology Conference, Toronto, Canada.
circle, m 关10兴 Cameron, A., Gordon, A. N., and Symm, G. T., 1968, ‘‘Contact Temperatures
A ⫽ area, m2 in Rolling/Sliding Surfaces,’’ Proc. R. Soc., London, Ser. A A268, pp. 45–61.
b ⫽ semi-minor axis of ellipse or rectangle, m 关11兴 Carslaw, H. S., and Jaeger, J. C., 1959, Conduction of Heat in Solids, Oxford
E(•) ⫽ complete elliptic integral of the second kind University Press.
关12兴 Jaeger, J. C., 1942, ‘‘Moving Sources of Heat and Temperature at Sliding
F ⫽ applied load, N Contacts,’’ Proceedings of the Royal Society, New South Wales, 76, pp. 203–
k ⫽ thermal conductivity, W/mK 224.
K(•) ⫽ complete elliptic integral of the first kind 关13兴 Rosenthal, D., 1946, ‘‘The Theory of Moving Sources of Heat and Its Appli-
L ⫽ arbitrary length scale, m cation to Metal Treatments,’’ Trans. ASME, 68, pp. 849–866.
关14兴 Yovanovich, M. M., 1988, ‘‘Chapter 8: Theory and Applications of Constric-
P ⫽ pressure, MPa tion and Spreading Resistance Concepts for Microelectronic Thermal Manage-
PeL ⫽ Peclet number, ⬅VL/ ␣ ment,’’ in Advances in Cooling Techniques for Computers, Hemisphere Pub-
PeL* ⫽ modified Peclet number, ⬅ 冑⑀ Pe lishing.
q ⫽ heat flux, W/m2 关15兴 Yovanovich, M. M., 1998, ‘‘Chapter 3: Conduction and Thermal Contact Re-
sistance 共Conductance兲,’’ in Handbook of Heat Transfer, W. M. Rohsenow, J.
q̄ ⫽ average heat flux, ⬅Q/A P. Hartnett, and Y. L. Cho, eds., McGraw-Hill, New York.
Q ⫽ heat flow rate, W 关16兴 Yovanovich, M. M., Burde, S. S., and Thompson, J. C., 1977, ‘‘Thermal
R ⫽ thermal resistance, K/W Constriction Resistance of Arbitrary Planar Contacts with Constant Heat
Flux,’’ AIAA Progress in Astronautics and Aeronautics: Thermophysics of
R * ⫽ dimensionless thermal resistance, ⬅RkL Spacecraft and Outer Planet Entry Probes, 56, A. M. Smith, ed., pp. 127–139.
t ⫽ time, s 关17兴 Archard, J. F., 1958, ‘‘The Temperature of Rubbing Surfaces,’’ Wear, 2, pp.
T ⫽ temperature, K 438–455.
关18兴 Francis, H. A., 1970, ‘‘Interfacial Temperature Distribution Within a Sliding Proceedings of the Eighth International Heat Transfer Conference, 1, pp. 35–
Hertzian Contact,’’ ASLE Trans., 14, pp. 41–54. 45.
关19兴 Holm, R., 1968, Electric Contacts, Springer-Verlag. 关22兴 Churchill, S. W., and Usgai, R., 1972, ‘‘A General Expression for the Corre-
关20兴 Johnson, K. L., 1985, Contact Mechanics, Cambridge University Press. lation of Rates of Transfer and Other Phenomena,’’ American Institute of
关21兴 Yovanovich, M. M., 1986, ‘‘Recent Developments in Thermal Contact, Chemical Engineers, 18, pp. 1121–1128.
Gap, and Joint Conductance Theories and Experiment,’’ Heat Transfer 1986, 关23兴 Blok, H., 1963, ‘‘The Flash Temperature Concept,’’ Wear, 6, pp. 483–494.
The Steady Inverse Heat
Conduction Problem: A
Robert Throne
Comparison of Methods With
Department of Electrical Engineering,
University of Nebraska, Parameter Selection
Lincoln, NE 68588
e-mail: rthrone1@unl.edu In the past we have developed the Generalized Eigensystem 共 GES L 兲 techniques for solv-
ing inverse boundary value problems in steady heat conduction, and found that these
Lorraine Olson vector expansion methods often give superior results to those obtained with standard
Fellow ASME Tikhonov regularization methods. However, these earlier comparisons were based on the
Department of Mechanical Engineering and optimal results for each method, which required that we know the true solution to set the
Department of Engineering Mechanics, value of the regularization parameter (t) for Tikhonov regularization and the number of
University of Nebraska, mode clusters 共 N clusters 兲 for GES L . In this paper we introduce a sensor sensitivity
Lincoln, NE 68588 method for estimating appropriate values of N clusters for GES L . We compare those
results with Tikhonov regularization using the Combined Residual and Smoothing Opera-
tor (CRESO) to estimate the appropriate values of t. We find that both methods are quite
effective at estimating the appropriate parameters, and that GES L often gives superior
results to Tikhonov regularization even when N clusters is estimated from measured data.
关DOI: 10.1115/1.1372193兴
Keywords: Finite Element, Heat Transfer, Inverse, Numerical Methods
Introduction order Tikhonov regularization on two two-dimensional test geom-

etries and four temperature/flux patterns. We found that GESL
Inverse problems have been studied for many years and steady
generally gave more stable inverse solutions than the other tech-
inverse boundary value problems, in which we attempt to identify
niques for all of the test cases examined. For the more compli-
boundary values on a portion of the boundary, have arisen as a
cated test patterns examined in that paper, zero order Tikhonov
particular class of inverse problems. These steady inverse prob-
regularization generally outperformed higher order Tikhonov
lems are generally associated with elliptic forward problems, and
regularization. Hence, in this paper we only examine the results
typical applications include electrostatics, magnetostatics, elastic-
for zero order Tikhonov regularization.
ity, flow in porous media, or steady heat transfer.
However, Tikhonov regularization has an adjustable parameter
In this paper we are considering two-dimensional steady heat
t which must be chosen appropriately in order to obtain stable
conduction, as shown schematically in Fig. 1. The governing
inverse solutions. GESL also has a scalar adjustable parameter
equation in the domain is Laplace’s equation
which must be chosen appropriately. All of the results in our
ⵜ• 共 kⵜT 兲 ⫽0, (1) previous work 关1兴 used the true solution to choose the adjustable
parameter; thus, these prior results were the optimal possible re-
where k is the conductivity which may depend on direction and sults for each case.
position. For the forward problem, either a temperature or a heat
flux must be specified at each location on all surfaces 共inside and
outside兲, and then the temperature and heat flux at all other points
in the domain may be calculated. For the inverse problems under
consideration in this paper, both the heat flux and the temperature
are assumed known on some boundaries, and we wish to infer the
temperature and heat flux on the rest of the boundaries.
For the inhomogeneous steady heat conduction equation with a
known source, we would solve the inverse problem in two steps
using superposition, since the equation is linear. We would first
solve for the heat fluxes and temperatures due to the known
source, then solve the homogeneous equation separately. How-
ever, if the source is unknown, then there are an infinite number of
possible solutions and the problem can not be solved uniquely.
In a previous paper 关1兴 we extended the Generalized Eigensys-
tem (GESL ) methods we developed for inverse electrocardio-
graphy 关2–7兴 to problems in steady inverse heat conduction. 共A
review of related work in steady inverse heat conduction is avail-
able in that paper.兲 We compared the GESL techniques to trun-
cated singular value decomposition and zero, first, and second

HEAT TRANSFER. Manuscript received by the Heat Transfer Division February 18, Fig. 1 Typical geometry for inverse boundary value problem
2000; revision received February 1, 2001. Associate Editor: D. A. Kaminski. in steady heat conduction
In this paper we introduce a sensor sensitivity method for esti-
mating appropriate values of the adjustable parameter N clusters for
GESL . We compare those results with Tikhonov regularization
冋 Koo
KToi
Koi To
Kii Ti
⫽册冋册冋
Moo
0
0 Qo
册冋册
Mii Qi
, (5)
using the Combined Residual and Smoothing Operator 共CRESO兲 where

to estimate the appropriate values of t. First, we discuss the in-
⫺1 T
verse algorithms and the methods for choosing the adjustable pa- Koo ⫽K̃oo ⫺K̃or K̃rr K̃or (6)
rameters. Next, we describe the test cases employed, and the re- ⫺1
sults obtained with each technique. Finally, we summarize our Koi ⫽K̃oi ⫺K̃or K̃rr K̃ri (7)
work and give concluding remarks. and
T ⫺1
Kii ⫽K̃ii ⫺K̃ri K̃rr K̃ri . (8)
Algorithms for Steady Inverse Heat Conduction
Problems Naturally, the computation is performed in the computationally
efficient manner indicated by Wilson 关9兴 rather than by brute force
This section first discusses finite element methods for typical inversion, but the concept is clearly illustrated here. In what fol-
inverse boundary value problems in steady heat conduction. Next, lows therefore, we will deal primarily with the reduced equation
we describe the inverse methods compared in this paper: set described by Eq. 共5兲.
Tikhonov zero-order regularization 共TIK-0兲 and the Generalized For the inverse problems under consideration in this paper, both
Eigensystem method (GESL ). Both of these methods have a the heat flux and the temperature are assumed known on some
single scalar adjustable parameter which must be chosen based on boundaries, and we wish to infer the temperature and heat flux on
measured data. the other boundaries. The methods developed in the subsequent
sections can be modified to estimate temperatures and heat fluxes
Numerical Discretization Using Finite Element Methods from measured temperatures alone. However, using measured
Using standard finite element methods 关8兴, the relationship be- heat fluxes alone will not uniquely determine the temperatures
tween the temperatures and heat fluxes may be schematically rep- since fluxes are based on gradients of temperatures.
冋册冋册冋
resented by
册冋册
Tikhonov Regularization. Because the structure of Eq. 共5兲 is
K̃oo K̃or K̃oi Moo 0 0 now identical to that which would be obtained using a boundary
To Qo
element method 共BEM兲, we employ the Tikhonov regularization
K̃Tor K̃rr K̃ri Tr ⫽ 0 0 0 0 , (2) methods developed by Martin and Dulikravich 关10–12兴 for steady
Ti Qi inverse heat transfer problems derived from BEM formulations as
K̃Toi T
K̃ri K̃ii 0 0 Mii
a comparison.
where To are the temperatures at nodes on the outer boundary, Ti Let us consider the specific case when the outer fluxes and
are the temperatures at nodes on the inner boundaries, and Tr are temperatures are known and we wish to estimate the inner fluxes
the temperatures at the remaining nodes. The matrices K̃ are the and temperatures. The basic forward Eq. 共5兲 is rearranged to place
conductivity 共or ‘‘stiffness’’兲 matrices resulting from the finite the known outer fluxes and temperatures on the right-hand-side of
element approximation to ⵜ•(kⵜT). Qo is the normal heat flux at the inverse equation, and the unknown inner fluxes and tempera-
the nodes on the outer boundary, with the assumption that the tures are moved to the left:
冋册冋册冋册冋册
normal heat flux is interpolated between the nodes with the shape
functions which were used to create the finite elements. 共That is, if Kii ⫺Mii Ti ⫺KToi 0 To
the elements are linear the normal heat flux is assumed to vary ⫽ . (9)
Koi 0 Qi ⫺Koo Moo Qo
linearly between the boundary nodes, etc.兲 Qi is the normal heat
flux at the nodes on the inner boundaries under the same interpo- Since the right-hand-side is now completely known, we can write
lation assumptions. Moo therefore is a consistent ‘‘area’’ matrix the resulting equation as
for the outer boundary, of the form Ax⫽f, (10)
Moo ⫽ 冕 NT Nd⌫ o , (3) where x T

⫽ 关 TTi QTi 兴 is the vector of unknowns, and
where N is the vector of shape functions for the outer boundary

and ⌫ o is the outer surface. Mii is a consistent ‘‘area’’ matrix for
A⫽ 冋 Kii
Koi
⫺Mii
0
册
冋册冋册
the inner boundaries. If there are n o nodes on the outer boundary,
⫺KToi 0 To
n i nodes on the inner boundary, and n r nodes in the rest of the f⫽ .
domain, then the total number of nodes is n tot⫽no⫹ni⫹nr . There ⫺Koo Moo Qo
are then n o entries in the vectors To and Qo , n r entries in Tr , and
Notice that there may be fewer unknowns than knowns 共or more,
n i entries in Ti and Qi . The matrices K̃ and M are square and for that matter兲 so that A is generally not square. 共A has n i ⫹n o
symmetric, so that Koo is n o ⫻n o , K̃or is n o ⫻n r , Koi is n o rows and 2n i columns, and f is a vector with n i ⫹n o entries.兲
⫻n i , etc. For any given forward problem the temperature or the Tikhonov regularization stabilizes the inverse by balancing the
heat flux must be specified on the inner and outer boundaries, so error in the solution to the equation with a measure of some func-
that Eq. 共2兲 reduces to a well-posed and well-conditioned system tion of the unknowns, such as the magnitude or slope. In general,
containing n r equations and n r unknowns. we minimize ⌸ with respect to x̂:
For computational purposes, it is useful to recognize that the
boundary temperatures completely determine the remaining tem- ⌸⫽ 储 Ax̂⫺f储 ⫹t 储 Rx̂储 , (11)
peratures, so that static condensation 关9兴 may be used to eliminate where x̂ is the estimate of the solution, R is called the regulariza-
the volume temperatures before any inverse processing occurs. 共If tion 共matrix兲 operator, and t is the regularization 共or smoothing兲
internal sensors are used, nodes corresponding to those locations parameter. In zero-order Tikhonov regularization R⫽I and so we
should not be condensed out.兲 Solving the middle row of Eq. 共2兲 penalize the magnitude of the unknowns. In first-order regulariza-
gives tion R is chosen so that Rx̂ is an estimate of the gradient of the
⫺1 unknowns, while in second-order regularization Rx̂ is an estimate
Tr ⫽⫺K̃rr 共 K̃ri Ti ⫹K̃Tor To 兲 . (4)
of the second derivative of the unknowns. First-order regulariza-
Inserting this expression back into Eq. 共2兲 gives tion penalizes solutions which have large slopes, while second-
order regularization penalizes solutions which have large second
derivatives—i.e. solutions which are not smooth. In a previous
paper 关1兴 we investigated three choices for R, but in this paper
we will employ only the most common choice: zero-order
regularization.
The value of x̂ which minimizes ⌸ is given by the solution to
the matrix equation
共 AT A⫹tRT R兲 x̂⫽AT f. (12)

When t is zero we have a pure least-squares solution, but this is
generally unstable. When t is large we have a smooth solution, but
that solution may have little relationship to the original problem.
By solving this equation with an appropriately chosen value of t a
stable inverse solution may be obtained.
Generalized Eigensystem Methods. The Generalized Eigen-
system (GESL ) methods take a different approach than Tikhonov
regularization. The basic idea in GESL is to choose basis vectors
in a systematic way, yet choose vectors which will accurately
represent the expected temperature and heat flux profiles with a
small number of modes. This is in contrast to zero order Tikhonov
regularization, where the solution is based solely on the A matrix
without regard to the expected nature of the temperature and heat
flux profiles.
For the GESL method, we first assume that we have a set of
N modes linearly independent basis vectors X, which are solutions to
the forward problem. 共We will address the selection of these basis
vectors below.兲 We partition the vectors as:
冋册
Xo
X⫽ Xr ,
Xi
(13)
where Xo contains the part of the basis vectors used to represent

the temperature solution at the outer boundary nodes 共Xo is n o
⫻N modes兲, Xi contains the part of the basis vectors used to repre-
sent the temperature solution at the inner boundary nodes 共Xi is
n i ⫻N modes兲, and Xr contains the basis vectors at the rest of the
nodes.
Next, we calculate the normal heat fluxes on the boundaries
which correspond to the temperature basis vectors. Inserting Xo
and Xi into Eq. 共5兲 allows us to calculate the normal boundary
fluxes, Yi and Yo :
冋册冋
Yo
Yi
⫽
M⫺1
oo
0
0
Mii⫺1
册冋 Koo
KToi
册冋册
Koi Xo
Kii Xi
. (14)
We assume that our estimate for the temperature solution T̂ and

the flux solution Q̂ can be represented as a sum of the basis
vectors:
T̂⫽ 冋册冋册
T̂o
T̂i
⫽
Xo ␣
Xi ␣
(15)
Q̂⫽ 冋册冋册
Q̂o
Q̂i
⫽
Yo ␣
Yi ␣
, (16)
where ␣ is the vector of expansion coefficients 共containing N modes

entries兲 for both the outer and inner temperatures and heat fluxes.
If we again consider the case where the outer boundary data are Fig. 2 First cluster for the GESL method for the square with
measured, the GES method simply minimizes the error between holes geometry
the measured data on the outer boundary and the estimate evalu-
ated at the outer boundary:
⌸⫽ 储 To ⫺Xo ␣储 2 ⫹ 储 Qo ⫺Yo ␣储 2 . (17) This is a system of linear algebraic equations for ␣, with N modes
equations in N modes unknowns. Solving this equation system gives
By minimizing ⌸ with respect to ␣ we obtain a matrix equation the expansion coefficients ␣. The estimate at the inner surface is
which allows us to solve for ␣: then given by
共 XTo Xo ⫹YTo Yo 兲 ␣⫽XTo To ⫹YTo Qo . (18) T̂i ⫽Xi ␣ (19)
Fig. 3 Second cluster for the GESL method for the square with holes geometry
Q̂i ⫽Yi ␣. (20) problems. Notice also, however, that we might have greater con-
fidence in the temperature measurements than the heat flux mea-
Notice temperature errors and flux errors are directly summed surements. Then Eq. 共17兲 could be adjusted with confidence
in Eq. 共17兲. For this to be reasonable, we should nondimensional- weighting factors.
ize the problem so that both T and Q are non-dimensional. In If we use many modes in the expansion, the inverse solution
what follows we only apply the technique to dimensionless test will still be unstable. Again, the goal is to choose a good set of
Fig. 4 Third cluster for the GESL method for the square with holes geometry
trial forward solutions, and to order them carefully so that the

truncation to N modes vectors yields stable inverse solutions.
The method used in this study for selecting the vectors seeks
冋 KGo
0
0
KGi
册冋册冋
Xo
Xi
⫽
Moo
0
0
Mii
册冋册冋
Xo ⌳oo
Xi 0
0
⌳ii
册 . (21)
solutions which are smooth on the boundaries, and which are

ordered by the amount of oscillation in the boundary solution. We Here KG is a surface gradient matrix 关1兴 and M is the surface area
introduce a generalized eigenproblem: matrix. Because the eigenproblems for the boundaries are now
Fig. 6 L-curve used for identifying appropriate N clusters values
Fig. 5 L-curve used for identifying appropriate t values for for GESL
zero order Tikhonov regularization
particular way to estimate the location of the bend: we choose t to

decoupled, solving Eq. 共21兲 will find sequences of vectors which maximize the difference between the derivative of the smoothing
are orthogonal over each boundary and which are ordered by their term t 储 x̂储 2 and the derivative of the fit to the data 储 Ax̂⫺f储 2 .
smoothness on that boundary. There are as many zero eigenvalues In order to estimate the appropriate number of clusters to use in
for this problem as there are surfaces, because each surface can the GESL method, we now introduce the sensor sensitivity
have one mode which has a constant temperature and hence no method, which is also based on an L-curve. Let us define E as the
surface gradient. error in the fit to the measured data:
Once the modes are identified, they are sorted into ‘‘clusters’’.
The first cluster includes all of the modes with a zero eigenvalue. E⫽ 储 T̂o ⫺To 储 2 ⫹ 储 Q̂o ⫺Qo 储 2 . (22)
The second cluster includes the lowest non-zero modes from each Also, define S s as the sensitivity of the solution vector to the
surface. The third cluster includes the second-lowest non-zero sensor data:
modes from each surface, and so on. Figures 2–4 show an ex-
Ns
ample: the first three clusters for the square with holes test case
used later in this paper. The first cluster for this geometry includes S s⫽ 兺关储 T̂ ⫺T̂ 储 ⫹ 储 Q̂ ⫺Q̂ 储兴 ,
j⫽1
i
j 2
i i
j 2
i (23)
three modes with zero eigenvalues. The second cluster includes
two modes 共sine-like and cosine-like兲 for each surface, for a total where N s is the total number of sensors, T̂i is the estimate of the
of six modes. The third cluster also has 6 vectors, but now they
inside temperature using all of the sensors, T̂ij is the estimate of
oscillate more rapidly on the surfaces. As many clusters as desired
are added to the set of vectors, and we identify the number of the inside temperature using all sensors except sensor j, Q̂i is the
clusters used in the solution as N clusters . estimate of the inside heat flux using all of the sensors, and Q̂ij is
Notice that this method finds only the trial functions for the the estimate of the inside heat flux using all sensors except sensor
temperature. The trial functions for the heat flux are found by j. Figure 6 schematically shows the L-curve for this method: the
inserting the temperature functions into Eq. 共14兲, and then the trial error in the fit to the measured data is plotted against the sensitiv-
vector selection is complete. In addition, although the temperature ity to the sensor data. Although the curve is shown as continuous
expansion vectors we chose are uncoupled, the heat flux vectors for clarity, in practice it is a discrete set of points. For each pos-
will be coupled. sible N clusters there is a value for E and a value for S s which lies on
the curve. To choose the most appropriate N clusters we choose the
Estimation of Appropriate Adjustable Parameters. In value which falls closest to the origin. This eliminates solutions
zero-order Tikhonov regularization, the parameter t must be se- which match the measured data extremely well, and also elimi-
lected in order to adjust the balance between the magnitude of the nates solutions which are very sensitive to the data measurements.
error in solving the matrix equation and the magnitude of the This method is similar to Generalized Cross Validation 共GCV兲
estimated solution. In the GESL method the number of clusters 关20兴. However, with GCV one attempts to minimize the overall
N clusters must be chosen to balance errors in fitting the measured error between predicted sensor values 共estimated using all the
data against the level of oscillation in the estimated solution. Al- other sensors兲 and the correct 共measured兲 sensor values. Hence,
though we can find the optimal parameters t and N clusters for test GCV focuses on the predicted errors on the boundaries with
cases where the solution is known, in realistic problems we must known 共measured兲 values. Our method attempts to examine the
estimate the parameters from known data. sensitivity of the estimated values on the surfaces where there are
In order to estimate an appropriate value of t for Tikhonov no measured values. GCV was tested as a method for choosing the
regularization, we will use the Combined Residual and Smoothing number of clusters in GESL , but performed poorly.
Operator 共CRESO兲 technique developed by Colli-Franzone 关13兴
in their studies of inverse electrocardiography, and widely used
关14–16兴 in that field. This technique is based on L-curve concepts
共see, e.g., 关16–19兴兲 Figure 5 schematically shows the L-curve, Test Cases
which plots the error in the fit to the measured data versus the We have examined two planar test geometries: the first geom-
magnitude of the estimated solution. Appropriate values of t are etry is an annulus, and the second geometry is a square with two
taken to be those at the bend, or elbow, of the curve. CRESO is a rounded rectangular holes. All the test cases have been nondimen-
sionalized. For the annulus geometry we considered two tempera- Table 1 Inverse relative error in temperature for the first tem-
ture cases, and for the square we investigated three temperature perature distribution for the annulus test case „meanÁstd.…
cases.
In computing the errors for the inverse solutions, we will use
the percent relative error. When we are estimating the inner tem-
perature values this is defined as
储 T̂i ⫺Ti 储
RE⫽100 percent⫻ (24)
储 Ti 储
Annulus ate the errors in heat flux and the errors in temperature with equal
weighting, which was assumed in deriving the inverse algorithms.
Figure 7 shows the geometry and finite element mesh for the For the forward problem the norm of the relative error in the
annulus tests considered here, which was also studied by Maillet, computed heat flux for the mesh employed was 0.13 percent on
Degiovanni, and Pasquetti 关21兴, Truffart, Jarny, and Delaunay the outer circle and 0.10 percent on the inner circle, indicating a
关22兴, Dennis and Dulikravich 关23,24兴 and Martin and Dulikravich well-converged solution.
关10兴. We added white random Gaussian noise to the analytical solu-
The geometry consists of a circle of dimensionless radius 1.0 tion on the outer boundary to examine the effects of noise on the
containing a concentric circular hole with dimensionless radius inverse solution. For ␴ 2 ⫽0.01, 0.02, 0.05, 0.10, 0.20, 0.50 共cor-
0.5. The conductivity in the region between the two circles is 1.0. responding to noise levels of 1, 2, 5, 10, 20, and 50 percent兲, we
The finite element mesh was generated with the I-DEAS finite applied temperatures and heat flux perturbations with normal dis-
element package 共SDRC, Ohio兲 and contains 12,521 linear trian- tribution and standard deviation ␴. Twenty-five trials were per-
gular elements and 6496 total nodes, with an element side length formed for each noise level, and the relative error was computed
of approximately 0.02. There were 314 nodes on the outer circle for each inverse technique.
and 157 nodes on the inner circle. Tables 1 and 2 show the mean and standard deviation of the RE
Two temperature distributions were considered for this in temperature 共Table 1兲 and heat flux 共Table 2兲 on the inner
geometry. boundary for each noise level and for both of the inverse algo-
First Temperature Distribution. The first temperature dis- rithms under consideration. The tables indicate the optimal param-
tribution is that studied by Dennis and Dulikravich 关23,24兴. For eter and estimated parameter solutions for each method: the opti-
the forward problem the temperature on the inner circle is taken to mal solution uses the best possible choice of the free parameter
be 1.0, while the temperature on the outer circle is 0.0. For the for each algorithm, while the estimated uses only measured data
inverse problem the temperatures and heat fluxes are specified on to estimate a good value for the free parameter. GESL performs
the outer boundary, and we wish to calculate the temperatures and well for this simple test case, while Tikhonov regularization is not
heat fluxes on the inner boundary. The analytical solution to the as effective. Since the true temperature is extremely smooth on
forward problem is easily found to be the boundaries 共in fact, a constant兲 it is not surprising that GESL ,
which orders the clusters by their smoothness, works best here. In
ln r fact, the first cluster for GESL is all that is required to represent
T⫽ , (25) the exact temperature pattern completely, so that all of the bound-
ln 0.5
ary data can be used to estimate the amplitudes of those modes.
for the dimensionless temperature, where r is the radial coordi-
nate, and Second Temperature Distribution. The second temperature
distribution was studied 共in dimensional form兲 by Maillet, Degio-
1 vanni, and Pasquetti 关21兴, and Truffart, Jarny, and Delaunay 关22兴.
q⫽ , (26)
r ln 0.5 The nondimensionalized forward problem is to solve ⵜ 2 T⫽0 in
for the dimensionless heat fluxes. It should be reiterated that a the domain subject to the boundary conditions T⫽0 on the inner
proper nondimensionalization of the problem allows one to evalu- circle (r⫽0.5) and ⳵ T/ ⳵ r⫽h(T⫺1) on the outer circle (r
⫽1.0). The heat transfer coefficient h is given on the top surface
of the cylinder by
h⫽ 再 15.3846
linear from 15.3846 to 6.1538
linear from 6.1538 to 12.3077
for ␪ ⫽0 to 40 deg
(27)
and is symmetric for the bottom surface of the cylinder. 共␪ is
measured counterclockwise from the right-hand-side of the cylin-
der.兲 No analytical solution for the forward problem is available,
but Fig. 8 shows the temperature contours for the numerical for-
ward solution.
Notice that this problem arises when internal measurements are
used to estimate the heat transfer coefficient on the surface of a
Table 2 Inverse relative error in heat flux for the first tempera-
ture distribution for the annulus test case „meanÁstd.…
Fig. 7 Geometry and finite element mesh for annulus test

case. „Inner circle has radius 0.5, outer circle has radius 1.0.…
Fig. 8 Forward computed temperature contours for second
annulus test case. „Temperature is zero on inner boundary; Fig. 9 Geometry and finite element mesh for square with
contour level intervals are 0.1.… holes. The outer square is 1.0Ã1.0. The first inner rectangle is
centered at „0.275,0.375… and is 0.15Ã0.25. The second inner
Table 3 Inverse relative error in temperature for the second rectangle is centered at „0.65,0.65… and is 0.2Ã0.2. The radii of
temperature distribution for the annulus test case „meanÁstd.… the corners on the inner rectangles is 0.01.
inner boundary, a constant temperature on the outer boundary, and

a specified internal heat generation in the solid region. Since the
internal heat generation was considered known, it can be omitted
Table 4 Inverse relative error in heat flux for the second tem- and included by superposition after the remaining fields are com-
perature distribution for the annulus test case „meanÁstd.… puted. However, the analytical solution to a constant temperature
on the inner boundaries in combination with 90 deg corners on the
internal edges will result in infinite heat fluxes at the corners.
Hence any finite solution computed as an inverse would have an
infinite error in the heat flux, which is misleading. Consequently,
we have rounded the internal corners to force the heat flux at those
locations to be large but finite.
The finite element mesh was generated with the I-DEAS finite
cylinder, but that 共following Truffart, Jarny, and Delaunay 关22兴兲 element package 共SDRC, Ohio兲 and contains 21,278 linear trian-
we are not attempting to estimate the coefficient but only the gular elements and 10,638 total nodes, with an average element
surface temperatures1. side length of 0.01. There were 400 nodes on the outer boundary
For the inverse problems with noise, we added random Gauss- and 80 nodes on each of the inner rectangles.
ian noise to the forward-computed solution on the inner boundary We examined three temperature distributions for this geometry.
to examine the effects of noise on the inverse solution. As before
First Temperature Distribution. In the first forward prob-
we applied temperature and heat flux perturbations, and twenty-
lem, the temperature on the outer boundary was nondimensional-
five trials were performed for each noise level.
ized to 0.0 while the temperature on the inside surface of the holes
Tables 3 and 4 show the mean and standard deviation of the RE
was taken to be 1.0. No analytical solution for the forward prob-
in temperature 共Table 3兲 and heat flux 共Table 4兲 on the outer
lem is available, but Fig. 10 shows the temperature contours for
boundary for each noise level and for each of the inverse algo-
the numerical forward solution.
rithms under consideration. The tables indicate the optimal param-
We added random Gaussian noise to the forward-computed so-
eter and estimated parameter solutions for each method. For this
lution on the outer boundary to examine the effects of noise on the
more complicated temperature pattern both methods give less ac-
inverse solution. As before we applied temperature and heat flux
curate inverses. The difference between the optimal and estimated
perturbations, and twenty-five trials were performed for each
parameter results is similar to the difference in the first test case.
noise level.
Square with Holes. Figure 9 shows the geometry and finite Tables 5 and 6 show the mean and standard deviation of the RE
element mesh for the second test geometry. This geometry is simi- in temperature 共Table 5兲 and heat flux 共Table 6兲 on the outer
lar to that studied by Dennis and Dulikravich 关23兴 but has been boundary for each noise level and for each of the inverse algo-
modified slightly. The original problem proposed by Dennis and rithms under consideration, and optimal parameter and estimated
Dulikravich was identical to that shown, except that the inside parameter solutions for each method are indicated. Although this
corners of the rectangular holes were not rounded. They chose a was a simple temperature pattern on the boundaries, the geometry
temperature pattern which included a constant temperature on the makes the temperature and heat flux patterns within the domain
more complicated. Tikhonov regularization performs poorly here,
1
While h can then be estimated from the surface data for temperatures and fluxes, and GESL performs ten to twenty times better. This superiority is
this may not be efficient, and we will postpone that question for a subsequent study. due to the fact that the solution for the forward problem can be
Fig. 10 Temperature contours for first square with holes test
case. „Temperature is 1.0 on inner boundaries, 0.0 on outer
boundary. Contours are at temperature intervals of 0.1.…
represented exactly in terms of the first GESL cluster—those cor-

responding to constant temperatures on each of the surfaces. At all
noise levels, the sensor sensitivity method is able to correctly
identify the first cluster as the stopping point, and so the optimal
and estimated results for GESL are exactly equal. Results with
Tikhonov regularization on this problem are similar to those seen
with the two annulus test cases.
Second Temperature Distribution. The second temperature
distribution tested was specifically chosen to be more difficult for
the GESL method to detect. For the forward problem, on the outer
surface of the square the temperature was given by T⫽1 on the
left edge, T⫽0 on the right edge, and a linear variation from 1 to
0 on the top and bottom edges. On the rectangular hole 共lower
left兲 the temperature was T⫽1 on the top edge, T⫽0 on the
bottom edge, and a linear variation from 1 to 0 on the left and
right edges. On the square hole 共upper right兲 the temperature was
T⫽1 on the bottom edge, T⫽0 on the top edge, and a linear
variation from 0 to 1 on the left and right edges. Figure 11共a兲
shows temperature contours for the numerical forward 共‘‘true’’兲
solution. This test case was treated in the same manner as the first
temperature distribution for the square with holes. We added ran-
dom Gaussian noise to the forward-computed solution on the
outer boundary to examine the effects of noise on the inverse
Table 5 Inverse relative error in temperature for the first tem-

perature distribution for the square with holes test case „mean
Ástd.…
Table 6 Inverse relative error in heat flux for the first tempera- Fig. 11 Temperature contours for second square with holes
ture distribution for the square with holes test case „mean test case: „a… ‘‘true’’ forward computed solution; „b… a GESL
Ástd.… inverse solution when 5 percent noise is added to the outer
temperatures and fluxes; and „c… a zero-order Tikhonov inverse
solution when 5 percent noise is added to the outer tempera-
tures and fluxes. „Temperature contours at 0.0, 0.1, . . . 1.0.…
Table 7 Inverse relative error in temperature for the second Table 9 Inverse relative error in temperature for the third tem-
temperature distribution for the square with holes test case perature distribution for the square with holes test case „mean
„meanÁstd.… Ástd.…
Table 8 Inverse relative error in heat flux for the second tem- Table 10 Inverse relative error in heat flux for the third tem-
perature distribution for the square with holes test case „mean perature distribution for the square with holes test case „mean
Ástd.… Ástd.…
solution. As before we applied temperature and heat flux pertur- T⫽⫺ln r/a, (28)
bations, and twenty-five trials were performed for each noise
level. where
Tables 7 and 8 show the mean and standard deviation of the RE r⫽ 冑共 x⫺x o 兲 2 ⫹ 共 y⫺y o 兲 2 .
in temperature 共Table 7兲 and heat flux 共Table 8兲 on the outer
boundary for each noise level and for each of the inverse algo- Here a⫽2, x o ⫽0.215, and y o ⫽0.375. This is the temperature
rithms under consideration, and optimal parameter and estimated pattern for a pure source located at x o , y o , and we chose the
parameter solutions for each method are indicated. On this prob- source point to lie just inside the rectangular hole. Hence, the
lem the errors in estimated temperature with GESL do not vary temperature everywhere in the domain will be given by Eq. 共28兲.
widely over all noise levels, ranging only from 13 percent at 1 The temperature on the boundaries near the source is high but not
percent noise to 17 percent at 50 percent noise. Although it does infinite, and the temperature drops off logarithmically as we move
not always choose the optimal number of clusters, the sensor sen- away from the source point. Figure 12 shows the temperature
sitivity method does an excellent job at estimating the number of contours for the numerical forward solution.
clusters for this case. Tikhonov zero-order regularization does not We added random Gaussian noise to the forward-computed so-
perform particularly well here, especially at high noise levels. At lution on the outer boundary to examine the effects of noise on the
modest noise levels 共⭐10 percent兲 the CRESO technique is quite inverse solution. As before we applied temperature and heat flux
good at estimating the appropriate t for Tikhonov regularization. perturbations, and twenty-five trials were performed for each
Figure 11 shows contour plots for the forward computed solu- noise level.
tion 共‘‘true’’兲 and a typical 5 percent noise inverse solution using Tables 9 and 10 show the mean and standard deviation of the
GESL and zero-order Tikhonov. Notice that the GESL inverse RE in temperature 共Table 9兲 and heat flux 共Table 10兲 on the outer
solution is quite reasonable, and the zero-order Tikhonov inverse boundary for each noise level and for each of the inverse algo-
solution 共despite its relative error of 26 percent兲 gives a good rithms under consideration, and optimal parameter and estimated
qualitative representation of the true solution. parameter solutions for each method are indicated. In this case
GESL , which is seeking smooth solutions, performs poorly 共al-
Third Temperature Distribution. In the third forward prob- though the solution is quite similar regardless of the noise level兲.
lem, we chose a source-like temperature distribution. The tem- However, the sensor sensitivity method predicts the optimal
perature on the boundaries of the domain was taken to be N clusters perfectly at each noise level. Tikhonov zero-order regular-
ization performs quite well on this case, and CRESO performs
very well at estimating t for the inverse. Tikhonov zero-order
regularization seeks only to limit the magnitude of the inverse, so
that the steep temperature gradients present in this test case pose
no problems. However, both methods produce very poor estimates
of the heat flux.
Concluding Remarks
We have introduced a new method for estimating the appropri-
ate number of clusters N clusters to use with Generalized Eigensys-
tem (GESL ) method inverses. In addition, we have compared op-
timal and estimated parameter results for GESL with Tikhonov
zero-order regularization. In four of the test cases, GESL outper-
forms Tikhonov regularization. In the fifth test case, a square with
holes and a logarithmic temperature pattern, Tikhonov zero-order
regularization is substantially better than GESL . However, in all
five test cases the parameter estimation techniques worked well.
The sensor sensitivity method for estimating the number of clus-
ters N clusters to be used with GESL produced excellent results on
these test cases. Also, the CRESO technique for estimating t for
Fig. 12 Temperature contours for third square with holes test Tikhonov zero-order regularization gave good results.
case. „Minimum temperature 0.69, maximum temperature 4.84. However, we should note that these results were derived from
Contour 1:1.11 2:1.52 3:1.94 4:2.35 5:2.77 6:3.18 7:3.60 8:4.01 simulated ‘‘data’’ computed by adding random zero-mean uncor-
9:4.43…. related noise to the true solutions. Hence we have not examined
the effects of biased measurements, correlated errors, geometric n r ⫽ number of nodes in the finite element mesh in the
errors, or other conditions which may arise in actual experiments. rest of the domain 共not on the outside boundary and
It is clear from this study that vector expansion techniques such not on the inside boundary兲
as the Generalized Eigensystem method show great promise for n tot ⫽ total number of nodes in the finite element mesh
steady inverse heat transfer applications. In general, the effective- N clusters ⫽ the number of clusters used in the GESL expansion
ness of the GESL method is directly related to the number of N modes ⫽ the number of modes used in the GESL expansion
expansion modes needed to accurately represent the unknown N s ⫽ the number of sensors
temperatures and heat fluxes. If a large number of the GESL ex- N ⫽ vector of finite element shape functions
pansion vectors are needed relative to the number of expansion q ⫽ heat flux vector
vectors required for zero order Tikhonov regularization 共obtained Qi ⫽ normal nodal heat fluxes on inside boundary of
from the singular value decomposition of the transfer matrix兲, domain
then GESL will generally not work as effectively as zero order Qo ⫽ normal nodal heat fluxes on outside boundary of
Tikhonov regularization. Hence, for the GESL method, it is very domain
important to chose expansion vectors in which the expected tem- Q̂ ⫽ inverse estimate for normal nodal heat flux solution
perature and heat flux patterns can be represented with only a few Q̂i ⫽ inverse estimate for normal nodal heat flux solution
modes. In a similar manner, the GESL method can be extended to on inside boundary of domain
solve coupled thermal/stress problems by choosing modes which Q̂ij ⫽ inverse estimate for normal nodal heat flux solution
include appropriate temperature and deformation patterns. on inside boundary of domain using all sensors ex-
In this paper we have studied the effectiveness of the various cept sensor j
techniques, with no attempt to optimize the algorithms for com-
putational speed. Hence, we have not attempted to compare the Q̂o ⫽ inverse estimate for normal nodal heat flux solution
computational efficiency of the algorithms. on outside boundary of domain
In the future, we hope to further develop these methods for r ⫽ radial coordinate
applications in on-line machine tool monitoring. In particular, the R ⫽ regularization matrix
inverse techniques discussed in this paper could be extremely t ⫽ penalty parameter for Tikhonov method
valuable in estimating tool tip/workpiece interface temperatures T ⫽ temperature
during turning. Here the temperature profile in the tool is quasi- Ti ⫽ nodal temperatures on inside boundary of domain
steady, and the interface temperature can be treated as an un- To ⫽ nodal temperatures on outside boundary of domain
known boundary condition on the tool. Flux and/or temperature Tr ⫽ nodal temperatures in the rest of the domain
sensors placed relatively far from the tip can then be used to T̂ ⫽ inverse estimate for nodal temperature solution
estimate the interfacial temperatures, since the problem is reduced T̂i ⫽ inverse estimate for nodal temperature solution on
to an inverse boundary value problem in steady heat transfer. inside boundary of domain
T̂ij ⫽ inverse estimate for nodal temperature solution on
inside boundary of domain using all sensors except
Acknowledgment sensor j
This work was supported by the National Science Foundation T̂o ⫽ inverse estimate for nodal temperature solution on
Grant #0001315. outside boundary of domain
x ⫽ vector of unknowns at nodal points
x̂ ⫽ estimate of vector of unknowns at nodal points
X ⫽ basis vectors for temperature for GESL method
Nomenclature Xi ⫽ basis vectors for temperature on inside boundary for
A ⫽ matrix relating knowns and unknowns for Tikhonov GESL method
method Xr ⫽ basis vectors for temperature for the rest of the do-
E ⫽ magnitude of the error in the fit to the measured data main for GESL method
f ⫽ vector of knowns for Tikhonov method Xo ⫽ basis vectors for temperature on outside boundary for
h ⫽ convection heat transfer coefficient GESL method
I ⫽ the identity matrix Yi ⫽ basis vectors for normal heat flux on inside boundary
k ⫽ conductivity tensor for GESL method
K ⫽ finite element conductivity matrix Yo ⫽ basis vectors for normal heat flux on outside bound-
KG ⫽ finite element boundary gradient matrix ary for GESL method
KGi ⫽ finite element boundary gradient matrix for inside ␣ ⫽ vector of expansion coefficients for GESL method
boundary of domain ⌫o ⫽ surface of the outer boundary
KGo ⫽ finite element boundary gradient matrix for outside ⌳ii ⫽ diagonal matrix of eigenvalues associated with inside
boundary of domain boundary
Kii ⫽ finite element conductivity matrix for inside boundary ⌳oo ⫽ diagonal matrix of eigenvalues associated with out-
of domain side boundary
Koi ⫽ finite element conductivity matrix coupling inside and ␪ ⫽ circumferential coordinate
outside boundary of domain ⵜ ⫽ gradient operator
Koo ⫽ finite element conductivity matrix for outside bound-
ary of domain References
M ⫽ finite element mass matrix 关1兴 Olson, L. G., and Throne, R. D., 2000, ‘‘A Comparison of the Generalized
Mii ⫽ finite element mass matrix for inside boundary of Eigensystem, Truncated Singular Value Decomposition, and Tikhonov Regu-
larization for the Steady Inverse Heat Conduction Problem,’’ Inverse Problems
domain in Engineering, 8, pp. 193–227.
Moo ⫽ finite element mass matrix for outside boundary of 关2兴 Throne, R., and Olson, L., 1994, ‘‘A Generalized Eigensystem Approach to
domain the Inverse Problem of Electrocardiography,’’ IEEE Trans. Biomed. Eng., 41,
n i ⫽ number of nodes in the finite element mesh on the pp. 592–600.
关3兴 Throne, R. D., and Olson, L. G., 1995, ‘‘The Effects of Errors in Assumed
inside boundary of domain Conductivities and Geometry on Numerical Solutions to the Inverse Problem
n o ⫽ number of nodes in the finite element mesh on the of Electrocardiography,’’ IEEE Trans. Biomed. Eng., 42, pp. 1192–1200.
outside boundary of domain 关4兴 Olson, L., and Throne, R., 1995, ‘‘Computational Issues Arising in Multidi-
mensional Elliptic Inverse Problems: The Inverse Problem of Electrocardio- 关15兴 Messinger-Rapport, B., and Rudy, Y., 1989, ‘‘Computational Issues of Impor-
graphy,’’ Eng. Comput., 12, No. 4, pp. 343–356. tance to the Inverse Recovery of Epicardial Potentials in a Realistic Heart-
关5兴 Throne, R. D., Olson, L. G., Hrabik, T. J., and Windle, J. R., 1997, ‘‘Gener- Torso Geometry,’’ Math. Biosci., 97, pp. 85–120.
alized Eigensystem Techniques for the Inverse Problem of Electrocardio- 关16兴 Johnston, P. R., and Gulrajani, R., 1997, ‘‘A New Method for Regularization
graphy Applied to a Realistic Heart-Torso Geometry,’’ IEEE Trans. Biomed. Parameter Determination in the Inverse Problem of Electrocardiography,’’
Eng., 44, pp. 447–454. IEEE Trans. Biomed. Eng., 44, pp. 19–39.
关6兴 Throne, R. D., Olson, L. G., and Hrabik, T. J., 1999, ‘‘A Comparison of 关17兴 Hansen, P., 1992, ‘‘Numerical Tools for the Analysis and Solution of Fred-
Higher-Order Generalized Eigensystem Techniques and Tikhonov Regulariza- holm Integral Equations of the First Kind,’’ Inverse Probl., 8, pp. 849–872.
tion for the Inverse Problem of Electrocardiography,’’ Inverse Problems in 关18兴 Hansen, P., 1992, ‘‘Analysis of Discrete Ill-Posed Problems by Means of the
Engineering, 7, pp. 143–193. L-Curve,’’ SIAM Rev., 34, pp. 561–518.
关7兴 Throne, R. D., and Olson, L. G., 2000, ‘‘Fusion of Body Surface Potential and 关19兴 Hansen, P., and O’Leary, D. P., 1993, ‘‘The Use of the L-Curve in the Regu-
Body Surface Laplacian Signals for Electrocardiographic Imaging,’’ IEEE larization of Discrete Ill-Posed Problems,’’ SIAM 共Soc. Ind. Appl. Math.兲 J.
Trans. Biomed. Eng., 47, pp. 452–462.
Sci. Stat. Comput., 14, pp. 1487–1503.
关8兴 Bathe, K., 1982, Finite Element Procedures in Engineering Analysis, Prentice-
关20兴 Trujillo, D., and Busby, D., 1997, Practical Inverse Analysis in Engineering,
Hall, Englewood Cliffs, NJ.
CRC Press.
关9兴 Wilson, E., 1974, ‘‘The Static Condensation Algorithm,’’ Int. J. Numer. Meth-
关21兴 Maillet, D., Degiovanni, A., and Pasquetti, R., 1991, ‘‘Inverse Heat Conduc-
ods Eng., 8, pp. 198–203.
关10兴 Martin, T. J., and Dulikravich, G. S., 1995, ‘‘Finding Unknown Surface Tem- tion Applied to the Measurement of Heat Transfer Coefficient on a Cylinder:
peratures and Heat Fluxes in Steady Heat Conduction,’’ IEEE Trans. Compon. Comparison Between an Analytical and a Boundary Element Technique,’’
Packag. Manuf. Technol., Part A, 18, pp. 540–545. ASME J. Heat Transfer, 113, pp. 549–557.
关11兴 Martin, T. J., and Dulikravich, G. S., 1996, ‘‘Inverse Determination of Bound- 关22兴 Truffart, B., Jarny, Y., and Delaunay, D., 1993, ‘‘A General Optimization
ary Conditions and Sources in Steady Heat Conduction With Heat Genera- Algorithm to Solve 2-D Boundary Inverse Heat Conduction Problems Using
tion,’’ ASME J. Heat Transfer, 118, pp. 546–554. Finite Elements,’’ In N. Zabaras, K. A. Woodbury, and M. Raynaud, eds.
关12兴 Martin, T. J., Halderman, J. D., and Dulikravich, G. S., 1995, ‘‘An Inverse Inverse Problems in Engineering: Theory and Practice, Palm Coast, FL, pp.
Method for Finding Unknown Surface Tractions and Deformations in Elasto- 53–60.
statics,’’ Comput. Struct., 56, pp. 825–835. 关23兴 Dennis, B. H., and Dulikravich, G. S., 1998, ‘‘A Finite Element Formulation
关13兴 Colli-Franzone, P., Guerri, L., Taccardi, B., and Viganotti, C., 1985, ‘‘Finite for the Detection of Boundary Conditions in Elasticity and Heat Transfer,’’ in
Element Approximation of Regularized Solutions of the Inverse Potential M. Tanaka and G. S. Dulikravich, eds., Proceedings of the International Sym-
Problem of Electrocardiography and Applications to Experimental Data,’’ Cal- posium on Inverse Problems in Engineering Mechanics-ISIP ’98, Nagano,
colo, 22, pp. 91–186. Japan, pp. 61–70.
关14兴 Rudy, Y., and Messinger-Rapport, B. J., 1988, ‘‘The Inverse Problem in Elec- 关24兴 Dennis, B. H., and Dulikravich, G. S., 1999, ‘‘Simultaneous Determination of
trocardiography: Solutions in Terms of Epicardial Potentials,’’ CRC Critical Temperatures, Heat Fluxes, Deformations, and Tractions on Inaccessible
Reviews in Biomedical Engineering, 16, pp. 215–268. Boundaries,’’ ASME J. Heat Transfer, 121, pp. 537–545.
Film Cooling of a Cylindrical
Leading Edge With Injection
Through Rows of Compound-
Angle Holes
Y.-L. Lin
Computations, based on the k-␻ shear-stress transport (SST) turbulence model in which
T. I.-P. Shih all conservation equations were integrated to the wall, were performed to investigate the
Fellow ASME, Professor three-dimensional flow and heat transfer about a semi-cylindrical leading edge with a flat
afterbody that is cooled by film-cooling jets, injected from a plenum through three stag-
Department of Mechanical Engineering, gered rows of compound-angle holes with one row along the stagnation line and two rows
Michigan State University, along ⫾25 deg. Results are presented for the surface adiabatic effectiveness, normalized
East Lansing, temperature distribution, velocity vector field, and surface pressure. These results show
MI 48824-1226 the interactions between the mainstream hot gas and the cooling jets, and how those
interactions affect surface adiabatic effectiveness. Results also show how ‘‘hot spots’’ can
form about the stagnation zone because of the flow induced by the cooling jets. The
computed results were compared with experimental data generated under a blind test.
This comparison shows the results generated to be reasonable and physically meaningful.
With the SST model, the normal spreading was under predicted from 20 to 50 percent.
The lateral spreading was over predicted above the surface, but under predicted on the
surface. The laterally averaged surface effectiveness was well predicted.
关DOI: 10.1115/1.1370513兴
Keywords: Computational, Film Cooling, Heat Transfer, Turbines, Turbulence
Introduction turbulence, film-cooling-hole geometry, and coolant-to-

mainstream mass-flux and momentum ratios. Computational fluid
Advanced gas-turbine stages are being designed to operate at
dynamics 共CFD兲 studies on leading-edge-film cooling have fo-
increasingly higher gas temperatures at their inlets to improve cused on the ability of CFD to predict adiabatic effectiveness and
thermal efficiency. To accommodate these high temperatures, surface heat transfer by using mostly low-order turbulence mod-
which can be higher than allowable material temperatures, film els, such as Baldwin-Lomax, k-␧ and k-␻. Of the CFD studies,
cooling in addition to internal cooling is needed to maintain ac- Garg and Gaugler 关11兴 and Heidmann et al. 关12兴 simulated real-
ceptable temperature distributions in components such as vanes, istic airfoil configurations with emphasis on film-cooling-hole exit
blades, and endwalls that come in contact with the high- profiles. Bohn et al. 关13兴 also accounted for conjugate heat trans-
temperature mainstream flow. Film cooling lowers material tem- fer. See Kercher 关14兴 for a bibliography of all papers on leading-
perature by forced convection inside film-cooling holes and by edge film cooling from 1972 to 1998.
forming a protective layer of cooler air next to the vane or blade On film cooling of leading edges, one unresolved question is
surface to insulate it from the hot gas. On film coverage, it is why hot spots form despite seemingly more than adequate cooling
important to design film-cooling jets that do not lift off or move flow. To address this question, Pratt and Whitney GESP organized
away from surfaces in need of thermal protection. Since the air a study involving a carefully designed and calibrated experimental
used for cooling 共here referred to as coolant兲 is extracted from the study 关8兴 and a series of CFD simulations performed by a number
compressor, which can be harnessed for useful work, efficiency of invited participants 共Chernobrovkin and Lakshminarayana 关15兴,
considerations demand that acceptable temperature distributions Lin, Stephens, and Shih 关16兴, Martin and Thole 关17兴, and Thakur,
in the turbine material be achieved by using minimal amount of Wright, and Shyy 关18兴兲. All CFD simulations were carried out in
coolant flow while enabling a higher turbine inlet temperature. a blind-test fashion in which experimental data are not released
To cool effectively with minimal cooling flow requires a good until the CFD simulations are completed and submitted. This pa-
understanding of the details of the fluid mechanics and heat trans- per continues the effort of Lin et al. 关16兴 to understand the inter-
fer processes involved. The importance of this problem has led actions between the mainstream and film-cooling jets in the ther-
many investigators to study film cooling 共see review by Simoneau mal protection of the leading-edge region.
and Simons 关1兴兲. One region that is especially difficult to film cool
efficiently is the leading edge of the first-stage stator airfoils, Problem Description
where there are stagnation zones and the approaching freestream
temperature is the highest. Experimental studies on leading-edge Schematic diagrams of the problem studied are given in Figs. 1
film cooling have been reported by Hanus and L’Ecuyer 关2兴, and 2. Figure 1 shows the experimental setup of Cruse et al. 关8兴,
Karni and Goldstein 关3兴, Camci and Arts 关4兴, Mayle and Anderson and Fig. 2 shows the computational model of the experi-
关5兴, Ou and Han 关6兴, Salcudean et al. 关7兴, Cruse et al. 关8兴, Ames mental setup. The experimental setup consists of a square duct
关9,10兴, and others. Parameters investigated include mainstream (149⫻149 mm2) in which hot gas approaches an obstruction
made up of two perpendicular plane walls connected by a quarter-
cylindrical wall 共radius of 50.8 mm兲. Within the cylindrical wall
HEAT TRANSFER. Manuscript received by the Heat Transfer Division April 28, are a plenum and two rows of film-cooling holes arranged in a
1999; revision received January 9, 2001. Associate Editor: J.-C. Han. staggered fashion with 9 holes in each row 共see Fig. 2兲. The di-
domain also assumes periodicity in the y-z plane so that fewer
holes need to be simulated. Measurements by Cruse et al. 关8兴
show that these assumptions are reasonable.
As noted, computations were carried out in a blind-test fashion.
Measurements made by Cruse et al. 关8兴 on adiabatic effectiveness
共␩兲 and normalized temperature distributions 共⌰兲 along selected
planes for validation were provided only after submission of the
computed results.
Problem Formulation
The flow and heat transfer problem about the film-cooled, semi-
Fig. 1 Schematic of experimental setup †7‡
cylindrical leading edge is modeled by the ensemble-averaged
conservation equations of mass 共continuity兲, momentum 共com-
pressible Navier-Stokes兲, and total energy for a thermally and
calorically perfect gas. Though the Mach number of the flow is
ameter of each hole is 6.32 mm. The inclination of the holes is low, the compressible form of the conservation equations was
along the radius of the cylindrical wall and 20 deg relative to the used because of the non-negligible density variations that result
spanwise direction. The spacing between holes in each row is 48.3 from temperature differences between the cooling air and the
mm. The separation between the two rows of holes in the x-y mainstream hot gas. The ensemble-averaged equations were
plane is 25 deg. All hole edges are essentially sharp 共radius less closed by Menter’s k-␻ shear-stress transport 共SST兲 turbulence
than 0.1 mm兲. The material of the cylindrical wall is polystyrene model 共Menter 关19兴兲, which can be applied in both the high and
with a thermal conductivity of 0.025 W/m-K so that the walls are low Reynolds number region of the turbulent flow field. The SST
nearly adiabatic 共in the measurements, a correction was made for model is a blend of the k-␻ model of Wilcox 关20兴 in the near-wall
the heat loss by conduction兲. Also, all surfaces are smooth 共rough- region and the standard k-␧ model further away from the wall.
ness is less than 100 ␮m in the holes and less than 25 ␮m else- Thus, like the k-␻ model, the SST model does not require explicit
where兲. damping near the wall because ␻ becomes extremely large as the
The hot gas approaching the quarter-cylindrical leading edge wall is approached. The blending of the k-␧ model is to eliminate
and the cooler gas entering through the plenum are both 50 per- the dependence on freestream values of ␻. Also, there is a limiter,
cent dry air plus 50 percent pure N2 with properties that are within which controls the eddy viscosity so that the turbulent shear stress
1.5 percent of dry air alone. At a distance of 140 mm upstream of is bounded by a constant times the turbulent kinetic energy 共a
the leading edge, the approaching flow has a nearly uniform pro- realizability constraint兲, which improves prediction of flows with
file with velocity (U ⬁ ) of 10 m/s, turbulent intensity less than 0.5 strong adverse pressure gradients and flow separation 共Kandula
percent, static pressure of 1 atm, and static temperature of 300 K. and Wilcox 关21兴 and Bardina et al. 关22兴兲. The details of the
The cooling flow that enters through the plenum will produce an ensemble-averaged equations along with the SST model used are
average coolant-to-mainstream density ratio 共DR兲 of 1.8 and an not given here since they are well documented in the literature
average coolant-to-mainstream mass-flux ratio 共M兲 of 2.0. The 共see, e.g., Shih and Sultanian 关23兴兲.
width of the diverter and the flow through it are adjusted to ensure The boundary conditions employed in this study for the differ-
that the stream surface approaching the stagnation line on the ent boundaries shown in Fig. 2 are as follows. At the inflow
cylinder is flat and parallel to the duct axis in the absence of film boundary, velocity and temperature profiles were kept uniform at
cooling 共measurements showed variations to be less than 1 mm兲. 10 m/s and 300 K, respectively. Profiles of the turbulence quanti-
The reader is referred to Cruse et al. 关8兴 for additional details. ties 共k and ␻兲 were also kept uniform 共turbulent intensity set at 0.5
For the problem just described, the computational domain is percent with effective viscosity equal to laminar viscosity兲. Only
taken to be the region bounded by the dash lines shown in Fig. 1 pressure was extrapolated. At the outflow boundary, an average
and reproduced in Fig. 2. As can be seen, this domain assumes back pressure of 0.999047 atm was imposed, but the pressure
symmetry about the stagnation stream surface so that the leading gradient was extrapolated. The velocity and density at that bound-
edge is semi-cylindrical with three rows of film-cooling holes in- ary were also extrapolated. At the plenum inflow, velocity and
stead of quarter-cylindrical with two rows of cooling holes. This temperature 共167.26 K兲 profiles were assumed to be uniform at
Fig. 2 Schematic of computational model
0.8591 m/s and 167.26 K, respectively. The temperature was se- formula. The system of nonlinear equations that resulted from the
lected to give a DR of 1.8. Since the flow rates through holes in aforementioned approximations to the space- and time-derivatives
different rows differ because of differences in pressure around the were analyzed by using a diagonalized alternating-direction im-
cylinder, the value of the plenum inflow velocity was iterated until plicit scheme 共Pulliam and Chaussee 关28兴兲 with local time-
an overall coolant-to-mainstream mass-flux ratio of 2.0 was ob- stepping and three-level V-cycle multigrid 共Ni 关29兴, Anderson
tained. The turbulence levels at the plenum inflow were taken to et al. 关30兴兲.
be negligibly small. Standard symmetry and periodic boundary The domain of the problem 共region within the dash lines in
conditions were imposed on the symmetry and periodic bound- Figs. 1 and 2兲 was represented by a multi-block grid system with
aries. seven blocks of structured grids, some patched and some over-
In this study, wall functions were not used. All equations in- lapped 共Fig. 3兲. The patched/overlapped approach makes it pos-
cluding those for the turbulence quantities were integrated to the sible to concentrate grid points where most needed and still main-
wall. At all walls, the no-slip and adiabatic wall conditions were tain a high quality grid system in smoothness and nearly
imposed; the turbulent kinetic energy 共k兲 was set to zero; and ␻ orthogonality. The seven blocks are as follows: a boundary-fitted
was set to 60v /( ␤ Y 2 ), where v is kinematic viscosity and Y is the H-H grid for the plenum, a C-H grid for the cylindrical leading
normal distance from the wall. edge and flat afterbody, an H-H grid for the rest of the flow
domain about the leading edge, two O-H grids for the two film
Numerical Method of Solution cooling holes, and two H-H grids to eliminate the centerline sin-
Solutions to the governing equations were obtained by using a gularity associated with the two O-H grids. The O-H and H-H
research code called CFL3D, which is based on a cell-center grids for each film-cooling hole are patched to the grid in the
finite-volume method 共see Thomas et al. 关24兴 and Rumsey and plenum and the grid about the leading edge. The C-H grid about
Vatsa 关25兴兲. In this study, the spatial differencing are all second- the leading edge overlaps the H-H grid for the rest of the domain.
order accurate. The inviscid terms were approximated by flux- The computations were carried out in one block at a time until all
difference splitting of Roe 关26,27兴. All diffusion terms were ap- blocks are analyzed, and this is repeated until a converged solu-
proximated conservatively by differencing derivatives at cell tion is obtained. Data transfer between patched and overlapped
faces. Since only steady-state solutions are sought, time deriva- grids is obtained by trilinear interpolation.
tives were approximated by the first-order accurate Euler implicit In this study, three different grid systems were used. For all
Fig. 3 Grid system used: overall grid and grid about film-cooling holes
In order to satisfy the aforementioned rules, information about
the solution 共such as boundary-layer thickness at leading edge as
well as location and size of the mainstream-jet interaction regions兲
is needed but unavailable for the first simulation. Thus, the first of
the three grids used was generated by using estimates. The
leading-edge boundary-layer thickness was estimated by using the
similarity solution for stagnation flow. To estimate the
mainstream-jet interaction regions, film-cooling studies on flat
plates were used with a factor of safety added 共i.e., used a very
fine grid with grid-aspect ratio near one over a region larger than
suggested by the flat plate studies兲. After obtaining a converged
solution with the first grid, a second grid was generated that made
use of the solution from the first grid. This grid system has 1.78
million grid points 共Fig. 3兲. The converged solution obtained with
the second grid is deemed excellent because all key features of the
flow were well within regions where grid points were clustered
and all rules were satisfied. Since the rules were quite stringent, a
third grid was constructed to examine if the rules can be relaxed.
The third grid system was generated by removing every other grid
Fig. 4 Coarsest grid used for grid-independence study in the C-H grid that surrounded the leading edge and the after-
body, which contained the majority of the grid points, but has a
layer of fine grid embedded over each film-cooling hole to resolve
the hole geometry 共Fig. 4兲. This grid system—with 0.24 million
three, the number of grid points and their distributions were ob- grid points—still has 3 grid points within a y ⫹ of 5. The con-
tained by using a very stringent set of rules established by devel- verged solution obtained with the ‘‘coarser’’ third grid was within
opers of numerical methods and turbulence models to ensure ac- 2 to 5 percent of the solution obtained by using ‘‘finer’’ second
curacy. These rules include having at least five grid points within grid. Since this was a blind test, only the ‘‘best solution’’ obtained
a y ⫹ of five and having grid spacing next to walls kept constant is reported, namely that generated by using the second grid.
for at least three grid points to resolve turbulent boundary layers. On a Cray C-90 computer, where all computations were per-
Other rules include aligning grid with flow direction as much as formed, the memory and CPU time requirements are 83.4 mega-
possible and keeping grid-aspect ratio near unity in regions of words and 33.3 hours. The CPU time given is for a converged
recirculating flow 共e.g., in mainstream-jet interaction roll-up re- steady-state solution, which typically involved 5000 iterations.
gions兲. For further details on the rules, see Stephens et al. 关31兴. The number of iterations needed to converge was excessively high
Fig. 5 Computed and measured normalized temperature ⌰ as a function of Z Õ D and Y Õ D at three X Õ D locations
because the Mach number of the flow was very low, less than Results and Discussion
0.01. If the mainstream Mach number is 0.05 or higher, which is
more typical of realistic engine conditions, then only about 1500 In this section, we first assess the capability of the CFL3D code
to 2000 iterations would be needed. Here, a solution is said to be with the SST turbulence model in predicting leading-edge film
converged when all of the following are satisfied: 共1兲 residual cooling in a blind test. Afterwards, we interrogate the flow field
drops at least three orders of magnitude; 共2兲 the second norm of by analyzing the computed results with focus on how fluid me-
the residual is less than 10⫺13; and 共3兲 normalized surface tem- chanics affects adiabatic effectiveness and the formation of hot
perature ⌰ on the leading edge varies by less than 2 percent for spots.
over 500 iterations. At this point, it is also noted that if the
‘‘coarser’’ third grid is used, then only 4.5 C-90 hours would be Comparison With Experiments. Figure 5 shows computed
needed to obtain a converged solution. and measured normalized temperature ⌰ at three X/D locations.
Fig. 6 Adiabatic effectiveness ␩: „a… computed; „b… computed, averaged over

0.43D Ã0.43D ; and „c… measured with resolution of 0.43D Ã0.43D †7‡.
Fig. 7 Adiabatic effectiveness, local and laterally averaged
Since the temperature is a scalar transported by the cooling jet, over an area of 0.43 D⫻0.43 D, and another in which no averag-
this figure shows the spread of the cooling jets in the normal 共Y兲 ing was used. This averaging was made because the measurement
and lateral 共z or Z兲 directions. At X/D⫽1.24, the computed and resolution on the wall is 0.43 D⫻0.43 D. From this figure, it can
measured spreads are comparable. At X/D⫽4.86, the computa- be seen that the computed adiabatic effectiveness is qualitatively
tions under predict the spread in the normal direction by about 20 similar to the measured ones. Basically, the topology of the highs
percent. The predicted lateral spread of the coolant from the and the lows is in the same locations with similar shapes, and the
second-row holes range from about 1⬍Z/D⬍4.5 above the sur- order of magnitude of the numbers is similar. From this figure, it
face to 1.5⬍Z/D⬍4.5 on the surface. In the measurements, it is can be seen that the computations under predict the spread of the
about 1⬍Z/D⬍4.5. Note that cooler gas at Z/D greater than coolant from the symmetry-plane holes in the X/D direction. It
approximately 4.5 is due to coolant from the symmetry-plane also under predicts the spread of the coolant from the second-row
holes in the Z/D direction. By comparing the ␩ ⫽0.2 contour line
holes. Thus, the lateral spreading is predicted more accurately
in Figs. 6共b兲 and 6共c兲 in the region on the right of the second-row
above the surface than on the surface. This is because the com- holes, one can see that computations under predict the mixing of
puted results predict more entrainment of hot gas to the surface the coolant from the symmetry and the second-row holes.
共see hot gas beneath the roll-up next to surface兲 than does the A quantitative comparison of adiabatic effectiveness is shown
measurements. This error can be attributed to the turbulence in Fig. 7. This figure shows the adiabatic effectiveness induced by
model used, which cannot account for the effects of eddy flatten- the symmetry-plane holes to be computed reasonably accurately at
ing in the near wall region as the rolled up flow impinges on the X/D⫽1.24. At X/D⫽4.86 and 9.98, the adiabatic effectiveness
surface. At X/D⫽9.98, the normal spread is under predicted by was predicted with less accuracy. The main reason is that compu-
about 50 percent. The predicted lateral spread ranges from 2.5 tations predicted more entrainment to the surface. The laterally-
⬍Z/D⬍7 above the surface to 3⬍Z/D⬍5.5 at the surface. In the averaged adiabatic effectiveness, however, does agree well with
experiment, the lateral spread was 2⬍Z/D⬍5.5, and was found measured values.
to be largest on the surface, which was not predicted. Thus, the As noted, this is a blind test. The qualitative and quantitative
computations under predict the normal spreading, over predict the comparisons given above show the computations to be physically
lateral spreading above the surface, and under predict the lateral reasonable.
spreading on the surface.
Figure 6 shows the computed and measured adiabatic effective- Discussion of Computed Results. With some confidence in
ness, which is the normalized temperature on the surface of an the computed results, this section discusses the flow field, tem-
adiabatic wall. Note that the computed results are presented in two perature distribution, and adiabatic effectiveness associated with
ways—one in which the value given at each point is the average the leading-edge film cooling problem shown in Fig. 2.
velocities is larger in the x-y plane though turning angles can be
larger in the y-z plane. These asymmetric separation bubbles in-
duce secondary flows in the cross stream. These secondary flows
were also observed by Leylek and Zerkle 关32兴. Further down-
stream along the holes 共about 5D to 6D兲, the flow does become
nearly axially symmetric. For this problem, the L/D ratio of the
film-cooling holes is 11.75 共L is length of hole兲, which is much
larger than 6. Thus, plenum-hole interactions are relatively unim-
portant here except in affecting the discharge coefficient. How-
ever, for holes with smaller L/D ratios, it can be important. For
such holes, their inlets can be shaped and/or angled to create the
desired symmetry or asymmetry in the flow that would bring
about the best cooling jet in terms of better adiabatic effective-
ness. At this point, it is noted that the mass flow rate through the
symmetry-plane 共1st row兲 holes is less than that through the
second-row 共2nd row兲 holes. This is because the pressure differ-
ence across the 1st row holes is smaller due to the stagnation zone
about the symmetry plane of the leading edge. The split between
the 1st and 2nd row holes is 48 to 52. Thus, the coolant-to-
mainstream mass-flux ratio is less than 2.0 for the 1st row holes
and greater than 2.0 for the 2nd row holes.
Once the cooling jets emerge from the 1st row holes, the fol-
Fig. 8 Velocity-vector field about the middle of a symmetry-
plane hole and the middle of a second-row hole in two x - y
lowing four interactions take place: jet-jet from 1st row holes, jet
planes and mainstream, spanwise acceleration due to favorable pressure
gradient induced by the curved geometry, and jets from 1st and
2nd row holes. Jet-jet interaction from 1st row holes along the
Flow Field. The flow field can be deduced from Figs. 8–11. z-direction causes an adverse pressure gradient region to form just
Figure 8 shows that there is a large separation bubble 共about upstream of each 1st row hole 共Fig. 9兲. These adverse pressure
D/sin(20deg) in length兲 at the inlets of the symmetry-plane and gradients produce small separation bubbles between holes 共P1 in
the second-row holes because of the vena contracta effect. The Fig. 9兲. It also causes jets from 1st row holes to spread along the
separation bubbles are not symmetric about the hole centerlines. X-direction. The effect of the mainstream on coolant from the 1st
They are largest in the x-y plane 共shown in Fig. 8兲 and smallest in row holes can be seen in Fig. 10 共P1, P2, and S1 to S6 for X/D
the y-z plane 共not shown兲. This is because the magnitude of the ⬍3兲. In planes P1 and P2 of Fig. 10, the 1st row hole jets can be
Fig. 9 Pressure contours on the cylindrical surface „X - Z plane at Y Ä0… and two Y - Z planes „P1: X Õ D Ä0, P3: X Õ D
Ä3.50; see Fig. 10…
Fig. 10 Projected streamlines at selected Y - Z and X - Y planes. P1, . . . , P6 correspond to X Õ D Ä0, 1.84, 3.50, 5.47, 8.47,
and 12.72, respectively. S1, . . . , S6 correspond to Z Õ D Ä1.10, 2.24, 3.37, 4.27, 5.73, and 7.64, respectively.
seen to cause the mainstream flow to acquire a velocity compo- exiting the holes 共see Figs. 8 and 10兲. This is because cooling jets
nent in the z-direction as it approaches the leading edge because from the 2nd row holes emerge with higher velocity than those
of viscous pumping. The z-component velocity acquired by the from 1st row holes. Reasons for the higher velocity include higher
mainstream flow coupled with the outward flow of the 1st row jets pressure difference between the plenum and the hole exits 共pres-
against the mainstream flow caused the formation of horseshoe- sure driven兲 and decreased effective hole-exit area because of
like vortices that eventually get underneath the 1st row jets. This flow distortion. Thus, effective mass-flux ratio from the 2nd row
brings hot gases in the mainstream next to the vane or blade holes is beyond the overall average of 2.0. Figure 5, planes P3,
surface 共Fig. 11兲. This indicates that coolant from 1st row holes P4, P5, and S2 in Fig. 10, and plane B in Fig. 11 show the strong
must completely overlap each other to prevent the formation of and asymmetric shear-layer interaction between the mainstream
hot spots. Figure 11 also shows the lateral spreading of the 1st and the coolant issuing from the 2nd row holes. This interaction
row jets in the X-direction, which is affected by flow acceleration causes the mainstream hot gas to be entrained by the cooling jet
along X and coolant issuing from the 2nd row holes. In that figure, and drawn to the surface, making the film-cooling jet from the
coolant from 1st row holes can be seen to flow towards the region second-row holes less effective.
between 2nd row holes. Some coolant from the 1st row holes can Figure 9 shows cooling jets from both the 1st and 2nd row
also be seen to flow over the jet issuing from the 2nd row holes holes to create considerable disturbances to the aerodynamics by
共see plane B in Fig. 11兲. inducing strong pressure gradients in the streamwise and spanwise
Cooling jets from the 2nd row holes separate and lift off once directions in the vicinity of the film-cooling holes and across the
Summary
Computations were performed by using the CFL3D code with
the SST turbulence model to study film cooling of a semi-
cylindrical leading edge. Comparison with experimentally mea-
sured temperature distribution and adiabatic effectiveness show
the computed results to be reasonable and physically meaningful.
The discrepancies were in coolant jet spreading rate and the
amount of hot gas entrainment. The computations under predict
normal spreading, over predict lateral spreading above the surface,
and under predict lateral spreading on the surface. These errors
can be attributed to the isotropic turbulence model used, which
cannot account for the Reynolds stress redistribution as eddies
flatten as they approach the wall.
Analysis of the computed results show the following:
• Plenum-hole interactions induce separation bubbles of con-
siderable length (D/sin(20deg)) in hole inlets because of the vena
contracta effect. These bubbles are asymmetric from asymmetry
in approach velocity and turning angles, and can produce second-
ary flows. These interactions are unimportant for the present prob-
lem since L/D is much larger than the separation bubble size.
• Coolant jets from the 1st row holes induce a velocity compo-
nent in the spanwise direction. This component coupled with the
outward flow of the cooling jets against the mainstream flow
cause the formation of horseshoe-like vortices that entrain hot
Fig. 11 Normalized temperature ⌰ at the symmetry plane and
the leading-edge surface gases and eventually get underneath the cooling jets.
• Velocity profiles at exits of 2nd row holes are distorted sig-
nificantly by the mainstream flow. The distortion further increases
the local coolant-to-mainstream mass flux ratio.
• Cooling jets from the 2nd row holes interact strongly with the
wakes of cooling jets. In the symmetry plane 共P1兲, Fig. 9 shows mainstream. These interactions entrain hot gases and cause lift off
the coolant from the 1st row holes to affect the pressure distribu- of the cooling jets.
tion up to about Y /D⫽1 from the leading-edge stagnation zone. • High acceleration induced by the curvature of the geometry
In the P3 plane of Fig. 9, the pressure disturbance induced by causes cooling jets from 2nd row holes to bend towards the
coolant jets can be seen to extend to Y /D of 2.5. streamwise direction rather abruptly despite the 20 deg angle from
which the jets emerge.
Temperature Distribution and Surface Adiabatic Effectiveness.
Figures 5 and 11 show normalized temperature ⌰ distributions.
From this figure, entrainment of hot gas into the cooling jet by Acknowledgment
mainstream-jet interaction can clearly be seen 共at X/D⫽4.86 in
Fig. 5 and planes B and C in Fig. 11兲. Basically, mainstream hot This work was supported by a grant from United Technologies
gas and coolant rolled together in layers. Cooling jets from the Pratt and Whitney in West Palm Beach, Florida. The computer
2nd row holes entrain more hot gas than do cooling jets from the time was provided by the NAS facility. The authors are grateful
symmetry-plane holes. This is because the relative velocity be- for this support. The second author is also grateful for the helpful
tween the mainstream flow and coolant jet issuing from 2nd row discussions with Martin Tabbita, Dean Johnson, Mark Stephens,
holes is higher. The entrainment of hot gas by cooling jets from and Fred Soechting of Pratt and Whitney and Dave Bogard of the
the 2nd row holes is highly asymmetric. Figures 5 and 11 show University of Texas at Austin.
the entrainment of hot gas by the vortex-like flow on the left side
of the coolant jet. This was also observed by Cruse et al. 关7兴. The Nomenclature
leading-edge surface becomes essentially unprotected by the cool-
ing jets after about 50 deg measured from the symmetry plane, but D ⫽ film-cooling hole diameter
there is a cool layer of gas just above it. This is because the DR ⫽ density ratio of coolant to mainstream ( ␳ c / ␳ ⬁ )
coolant jets were lifted off by the entrained hot gas. This indicates k ⫽ turbulent kinetic energy
the importance of minimizing entrainment induced by L ⫽ film-cooling hole length
mainstream-jet interaction. M ⫽ mass-flux ratio of coolant to mainstream ( ␳ c U c / ␳ ⬁ U ⬁ )
From Figs. 6 and 11, the following observations can be made T ⫽ temperature
about adiabatic effectiveness on the semi-cylindrical leading-edge Uc ⫽ average coolant velocity from both rows of holes
surface. The first observation is that there is a region between the U␶ ⫽ friction velocity
symmetry-plane holes, where adiabatic effectiveness reaches a U⬁ ⫽ mainstream velocity upstream of the leading edge
low value. This low adiabatic effectiveness came about because x-y-z ⫽ Cartesian coordinate system defined in Fig. 2
horseshoe-like vortices entrained hot gases from the mainstream y⫹ ⫽ normalized distance normal to surface ( ␳ U ␶ Y / ␮ )
and got underneath the cooling jets. The second observation is X ⫽ streamwise coordinate tangent to leading-edge surface
that adiabatic effectiveness is low between the second-row film Y ⫽ coordinate normal to surface
cooling holes. There are several reasons for this. First, cooling jets Z ⫽ lateral coordinate, normal to mainstream and tangent to
from the 1st row holes lift off when they approach the 2nd row leading-edge surface
holes due to the adverse pressure gradients induced by the cooling
jets from those holes. Second, large curvature in geometry causes Greek
high flow acceleration in the streamwise direction for both the ␩ ⫽ adiabatic effectiveness 关 (T aw⫺T ⬁ )/(T c ⫺T ⬁ ) 兴
mainstream flow and the coolant jets. Thus, the coolant jets from ␮ ⫽ dynamic viscosity
the 2nd row holes bend rather abruptly once they emerge 关3–32兴. ␳ ⫽ density
⌰ ⫽ normalized temperature 关 (T⫺T ⬁ )/(T c ⫺T ⬁ ) 兴关15兴 Chernobrovkin, A., and Lakshminarayana, B., 1998, ‘‘Numerical Simulation
␻ ⫽ dissipation rate per unit k and Aerothermal Physics of Leading Edge Film Cooling,’’ ASME Paper 98-
GT-504.
Subscripts and Superscripts 关16兴 Lin, Y.-L., Stephens, M. A., and Shih, T. I.-P., 1987, ‘‘Computations of
Leading-Edge Film Cooling With Injection Through Rows of Compound
aw ⫽ adiabatic wall Angle Holes,’’ ASME Paper 97-GT-298.
c ⫽ coolant 关17兴 Martin, C. A., and Thole, K. A., 1997, ‘‘A CFD Benchmark Study: Leading-
⬁ ⫽ mainstream Edge Film Cooling with Compound Angle Injection,’’ ASME 97-GT-297.
关18兴 Thakur, S., Wright, J., and Shyy, W., 1997, ‘‘Computation of a Leading-Edge
Film Cooling Flow over an Experimental Geometry,’’ ASME Paper 97-GT-
References 381.
关1兴 Simoneau, R. J., and Simon, F. F., 1993, ‘‘Progress Towards Understanding 关19兴 Menter, F. R., 1993, ‘‘Zonal Two-Equation k-␻ Turbulence Models for Aero-
and Predicting Heat Transfer in the Turbine Gas Path,’’ Int. J. Heat Fluid dynamic Flows,’’ AIAA Paper 93-2906.
Flow, 14, No. 2, pp. 106–128. 关20兴 Wilcox, D. C., 1993, Turbulence Modeling for CFD, DCW Industries, La
关2兴 Hanus, G. J., and L’Ecuyer, M. R., 1977, ‘‘Leading Edge Injection for Film Canada, California.
Cooling of Turbine Vanes,’’ J. Energy, 1, pp. 44–49. 关21兴 Kandula, M., and Wilcox, D. C., 1995, ‘‘An Examination of k-␻ Turbulence
关3兴 Karni, J., and Goldstein, R. J., 1990, ‘‘Surface Injection Effect on Mass Trans- Model for Boundary Layers, Free Shear Layers, and Separated Flows,’’ AIAA
fer from a Cylinder in Cross Flow,’’ ASME J. Turbomach., 112, pp. 477–487. Paper 95-2317.
关4兴 Camci, C., and Arts, T., 1991, ‘‘Effect of Incidence on Wall Heating Rates and 关22兴 Bardina, J. E., Huang, P. G., and Coakley, T. J., 1997, ‘‘Turbulence Modeling
Aerodynamics on a Film-Cooled Transonic Turbine Blade,’’ ASME J. Tur- Validation, Testing, and Development,’’ NASA TM 110446.
bomach., 113, pp. 493–501. 关23兴 Shih, T. I.-P., and Sultanian, B., 2001, ‘‘Computations of Internal and Film
关5兴 Mayle, R. E., and Anderson, A., 1991, ‘‘Velocity and Temperature Profiles for Cooling,’’ Heat Transfer in Gas Turbine Systems, B. Suden and M. Faghri,
Stagnation Film Cooling,’’ ASME J. Turbomach., 113, pp. 457–463. eds., WIT Press, Ashurst, Southhampton.
关6兴 Ou, S., and Han, J. C., 1992, ‘‘Influence of Mainstream Turbulence on Lead- 关24兴 Thomas, J. L., Krist, S. T., and Anderson, W. K., 1990, ‘‘Navier-Stokes Com-
ing Edge Film Cooling Heat Transfer Through Two Rows of Inclined Film putations of Vortical Flows Over Low-Aspect-Ratio Wings,’’ AIAA J., 28,
Slots,’’ ASME J. Turbomach., 114, pp. 724–733. No. 2, pp. 205–212.
关7兴 Salcudean, M., Gartshore, I., Zhang, K., and McLean, I., 1994, ‘‘An Experi- 关25兴 Rumsey, C. L., and Vatsa, V. N., 1993, ‘‘A Comparison of the Predictive
mental Study of Film Cooling Effectiveness Near the Leading Edge of a Tur- Capabilities of Several Turbulence Models Using Upwind and Central-
bine Blade,’’ ASME J. Turbomach., 116, pp. 71–79. Difference Computer Codes,’’ AIAA Paper 93-0192.
关8兴 Cruse, M. W., Yuki, U. M., and Bogard, D. G., 1997, ‘‘Investigation of Vari- 关26兴 Roe, P. L., 1981, ‘‘Approximate Riemann Solvers, Parameter Vector and Dif-
ous Parametric Influences on Leading Edge Film Cooling,’’ ASME Paper ference Schemes,’’ J. Comput. Phys., 43, pp. 357–372.
97-GT-296. 关27兴 Roe, P. L., 1986, ‘‘Characteristic Based Schemes for the Euler Equations,’’
关9兴 Ames, F. E., 1998, ‘‘Aspects of Vane Film Cooling with High Turbulence:
Annu. Rev. Fluid Mech., 18, pp. 337–365.
Part I—Heat Transfer,’’ ASME J. Turbomach., 120, pp. 768–776.
关28兴 Pulliam, W. R., and Chaussee, D. S., 1981, ‘‘A Diagonal Form of an Implicit
关10兴 Ames, F. E., 1998, ‘‘Aspects of Vane Film Cooling with High Turbulence:
Approximate Factorization Algorithm,’’ J. Comput. Phys., 39, pp. 347–363.
Part II—Adiabatic Effectiveness,’’ ASME J. Turbomach., 120, pp. 777–784.
关11兴 Garg, V. K., and Gaugler, R. E., 1996, ‘‘Leading Edge Film Cooling Effects 关29兴 Ni, R.-H., 1981, ‘‘A Multiple Grid Scheme for Solving the Euler Equations,’’
on Turbine Blade Heat Transfer,’’ Numer. Heat Transfer, Part A, 30, pp. AIAA Paper 81-1025.
165–187. 关30兴 Anderson, W. K., Thomas, J. L., and Whitfield, D. L., 1988, ‘‘Multigrid Ac-
关12兴 Heidmann, J., Rigby, D. L., and Ameri, A. A., ‘‘A Three-Dimensional celeration of the Flux-Split Euler Equations,’’ AIAA J., 26, No. 6, pp. 649–
Coupled Internal/External Simulation of a Film-Cooled Turbine Vane,’’ 654.
ASME Paper 99-GT-186. 关31兴 Stephens, M. A., Chyu, M. K., Shih, T. I-P., and Civinskas, K. C., 1996,
关13兴 Bohn, D. E., Becker, V. J., Kusterer, K. A., and Rungen, A. U., 1998, ‘‘Ex- ‘‘Calculations and Measurements of Heat Transfer in a Square Duct with
perimental and Numerical Conjugate Investigation of the Blowing Ratio Influ- Inclined Ribs,’’ AIAA Paper 96-3163.
ence on the Showerhead Cooling Efficiency,’’ ASME Paper 98-GT-85. 关32兴 Leylek, J. H., and Zerkle, R. D., 1994, ‘‘Discrete-Jet Film Cooling: A Com-
关14兴 Kercher, D. M., 2000, ‘‘Turbine Airfoil Leading Edge Film Cooling Bibliog- parison of Computational Results with Experiments,’’ ASME J. Turbomach.,
raphy: 1972–1998,’’ Int. J. Rotating Mach., 6, No. 5, pp. 313–319. 116, pp. 358–368.
Time-Resolved Thermal
Sean P. Kearney
Engineering Sciences Center, 9100,
Sandia National Laboratories,
Boundary-Layer Structure in a
P.O. Box 5800, MS 0834,
Albuquerque, NM 87185-0834
Pulsatile Reversing Channel Flow
e-mail: spkearn@sandia.gov
In this paper, the results of an experimental study of the time-resolved structure of a
Anthony M. Jacobi thermal boundary layer in a pulsating channel flow are presented. The developing lami-
Department of Mechanical nar regime is investigated. Two techniques were used for time-resolved temperature mea-
and Industrial Engineering, surements: a nonintrusive, pure-rotational CARS method and cold-wire anemometry. Re-
University of Illinois at Urbana-Champaign, sults are presented for differing degrees of flow reversal, and the data show that the
Urbana, IL 61801 primary impact of reversed flow is an increase in the instantaneous thermal boundary-
layer thickness and a period of decreased instantaneous Nusselt number. For the devel-
Robert P. Lucht oping laminar parameter space spanned by the experiments, time-averaged heat-transfer
Department of Mechanical Engineering, enhancements as high as a factor of two relative to steady flow are observed for nonre-
Texas A&M University, versing and partially reversed pulsating flows. It is concluded that reversal is not neces-
College Station, TX 77843 sarily a requirement for enhancement. 关DOI: 10.1115/1.1372317兴
Keywords: Boundary Layer, Convection, Enhancement, Heat Transfer, Periodic,

Pulsating
Introduction the convection process for the subset of the parameter space in-
Single-phase convective heat-transfer enhancement is important vestigated. If the confusion that exists regarding the heat-transfer
in a variety of residential and industrial end-use energy applica- impact of pulsatile flow is to be overcome, a fundamental under-
tions. Heat-transfer enhancement can be implemented using meth- standing of the unsteady convection process must be obtained.
ods that are active 共external energy input required兲 or passive 共no Such a detailed understanding requires knowledge of the time-
external energy input兲. The introduction of high-amplitude, peri- resolved flow, heat-transfer and temperature behavior over a
odic flow pulsation shows promise as an active enhancement broad range of the parameter space suggested in Eq. 1.
method. However, the literature provides conflicting reports, with The literature does, however, suggest a potential link between
some researchers reporting pulsation-induced heat-transfer en- significant heat-transfer enhancement and periodic, bulk flow re-
hancements 共Dec et al. 关1兴; Niida et al. 关2兴; Bayley et al. 关3兴; Gal- versal. In an early study by Bayley et al. 关3兴, the local, time-mean
itseyskiy et al. 关4兴; Fujita and Tsubouchi 关5兴兲 and others citing no heat-transfer coefficient was measured for pulsating flow over a
impact, or even a decrease in heat transfer 共Miller 关6兴; Feiler 关7兴; flat plate. The heat-transfer coefficient showed no frequency de-
Jackson and Purdy 关8兴兲. The reason for these conflicting reports is pendence, but increased with the applied pressure fluctuation am-
that the physics of the unsteady convection process over an ex- plitude if this amplitude exceeded a critical value. This critical
ceptionally broad parameter space are not fully understood. For
amplitude was postulated to be associated with flow reversal, but
example, a Buckingham ⌸ analysis shows that the local time-
no time-resolved velocity data were acquired to substantiate this
mean Nusselt number in a low-speed, wall-bounded laminar flow
may depend on as many as seven dimensionless groups: conclusion. Hanby 关9兴 and Keil and Baird 关10兴 observed both
nonreversing and reversing flows experimentally. In a study of a
Nux ⫽ f 共 ⌬u/Ū 兩 ⬁ ,⌬u/ ␻ L,a/ ␻ L,Wo,Rex ,Pr,GzT 兲 . (1) reacting, turbulent flow with 100-Hz resonant oscillations, Hanby
关9兴 observed a decrease in the time- and space-averaged heat-
Additional parameters arising from flow pulsation in Eq. 1 in- transfer coefficient for nonreversing flow and heat-transfer en-
clude: 共1兲 the Womersley number Wo, which is a ratio of the hancements upwards of 100 percent with flow reversal. In their
channel height to the Stokes or ‘‘acoustic’’ boundary-layer study of turbulent water flow in a shell-and-tube heat exchanger
thickness,1 共2兲 the centerline ‘‘amplitude ratio’’ ⌬u/Ū 兩 ⬁ , which with 0.4 to 1.1 Hz velocity pulsations, Keil and Baird 关10兴 ob-
is a measure of the amplitude of the velocity waveform and is served no heat-transfer impact for nonreversing flow and signifi-
greater than unity in the case of a bulk flow reversal, 共3兲 the cant enhancement when the flow was reversed.
parameter a/ ␻ L, which is a measure of the wavelength of the
Two experimental studies 共Fujita and Tsubouchi 关5兴 and Jack-
applied pressure oscillation to the relevant streamwise length
scale, and 共4兲 the grouping ⌬u/ ␻ L which represents the relative son and Purdy 关8兴兲 contradict the premise that heat transfer is
magnitudes of the Lagrangian and convective accelerations and enhanced in reversing flows only. Fujita and Tsubouchi 关5兴 mea-
dictates the importance of ‘‘acoustic streaming’’ 共as explained in sured local heat-transfer coefficients in a pulsed, nonreversing
Ref. 关8兴兲. laminar flow over a flat plate. They observed modest heat-transfer
The majority of reported experiments have focused on provid- enhancements of 10 to 20 percent for nonresonant forcing at fre-
ing empirical heat-transfer correlations for a specific geometry quencies between 2 and 23 Hz. the heat-transfer data presented by
and flow condition and have not provided measurements of tem- Fujita and Tsubouchi 关5兴 also suggest that an optimal forcing fre-
perature and velocity profiles that are needed to fully understand quency exists, in contrast to the earlier flat-plate measurements of
Bailey and co-workers, that suggested no frequency dependence.
1
The Stokes boundary layer is a measure of the penetration depth of unsteady In their study of an acoustically resonant, turbulent pipe flow,
vorticity fluctuations that are generated at the wall.
Jackson and Purdy 关8兴 observed no overall heat transfer impact
HEAT TRANSFER. Manuscript received by the Heat Transfer Division April 21, with pulsation, despite massive resonant velocity and pressure
2000; revision received February 5, 2001. Associate Editor: M. Hunt. fluctuations. Jackson and Purdy did observe both enhanced and
decreased time-mean heat-transfer coefficients at different local
measurement stations that resulted in a negligible surface-
averaged effect.
Three studies were found in which the time-resolved thermal
boundary-layer and wall heat-flux behavior were at least partially
addressed for pulsating flows with a nonzero mean mass flux.
Feiler and Yeager 关11兴 performed a qualitative flow visualization
study in a highly turbulent, flat-plate boundary layer using a
Schlieren method. Their results showed a dramatic thickening of
the thermal boundary layer during periodic flow reversal. In their
flat-plate study cited above, Fujita and Tsubouchi 关5兴 provided the
first reported time-resolved surface heat-transfer data. They found
significant phase differences between the Nusselt number and the Fig. 1 Schematic of the pulsatile flow facility. The bulk flow is
free-stream velocity oscillations. However, Fujita and Tsubouchi from left to right. Pulsations are imparted to the flow by a series
关5兴 presented time-resolved results for only a single operating of rotating vanes located upstream of the settling chamber.
condition and did not provide time-resolved boundary-layer tem-
perature data. The study of turbulent flow and heat transfer in a
pulsed combustor tailpipe by Dec and co-workers 关1,12–14兴 is the
most detailed set of time-resolved experiments to date. Dec and ployed in previous work, 关15–17兴 and provide sufficient velocity
co-workers presented nonintrusive point measurements of oscillation amplitude to reverse the flow over a limited range of
boundary-layer velocity 共using LDV兲 and temperature 共using forcing frequencies.
laser-induced fluorescence兲 as well as local wall heat flux results. The conditioned flow was accelerated through a 4.6-to-1, two-
Dec et al. 关1兴 observed a distinct Nusselt-number maximum of 2.7 dimensional contraction constructed using the matched-
times the accepted value for steady, turbulent pipe flow 共based on polynomial method of Morel 关18兴. At the exit plane of the con-
a Reynolds number evaluated with the time-mean velocity兲 during traction, the nozzle boundary-layer flow was removed through
the positive-to-negative reversal of the free-stream flow. This heat concentric suction slots of 10 mm minimum clearance, using an
transfer peak was followed by a period of decreasing heat flux approach suggested by Swearingen and Blackwelder 关19兴. The
during the negative-velocity portion of the cycle. Dec et al. 关1兴 suction slots were created by placing the first 25 mm of the test
postulated that the enhancement mechanism during reversal was section inside the nozzle and enclosing the nozzle/leading edge
an ejection of fluid from the near-wall region. Their analysis of assembly in a suction plenum where low pressures were created
the time-resolved data suggested that an increased turbulence using a high-head blower. Removal of the nozzle boundary layer
level, generated by elevated levels of shear in the Stokes boundary provided a well-defined origin for the thermal and momentum
layer, was the dominant mechanism for time-averaged heat- boundary layers, and allowed for transition between the flat nozzle
transfer enhancement, with the local maxima during flow reversal walls and the large-radius curvature introduced into the floor of
accounting for only 10 to 15 percent of the total enhancement. the test section as described below.
In this paper, new time-resolved temperature and heat flux data The test section was a rectangular channel composed of an
from a laminar, developing, pulsatile channel flow are presented. aluminum ceiling, a stainless-steel floor, and glass side windows.
The developing channel flow is chosen because it represents a To prevent leading-edge boundary-layer separation, both the ceil-
canonical problem for many heat-exchanger applications. The ing and the floor of the test section were machined to sharp lead-
thermal impact of nonreversed flows and flows that are partially ing edges, and knife-edged aluminum inserts were used to form
reversed 共reversal limited to a near-wall region only兲 or com- the side walls in the entrance region. The unheated ceiling of the
pletely reversed 共reversal across the entire channel height兲 is in- test section was flat and constructed from an aluminum plate that
vestigated using a mechanical, nonresonant flow-forcing scheme. was fitted with five removable blanks to allow for insertion of a
The literature cited above suggests that bulk flow reversal may hot/cold-wire probe at five streamwise locations. Uniform heat
lead to significant heat-transfer enhancement, but very little is input was applied to the backside of the wind tunnel floor using
understood about the interplay between the reversing flow and the Kapton resistance heaters controlled with a VARIAC. The test
time-dependent boundary-layer temperature and wall heat-flux section floor was fabricated from stainless steel, with a small
histories. Additionally, the impact of partial flow reversal, to our spanwise curvature machined into the floor so that the laser beams
knowledge, has not been previously addressed. Partial flow rever- used for the nonintrusive CARS temperature measurements were
sal is potentially important in heat-exchanger applications because not clipped as they entered the test section. The channel height
nonresonant, mechanical forcing schemes are likely to be pre- was 30 mm at the spanwise centerline, where all measurements
ferred to eliminate the high levels of noise and vibration associ- were made, and 35 mm at the side glass windows. The aspect ratio
ated with resonant, acoustic forcing. However, nonresonant forc- of the tunnel cross section was 5.3 when based on the centerline
ing will provide smaller velocity oscillations than resonant forcing channel height. Potential curvature effects were minimized, as
and may only generate partial or weak bulk flow reversals. care was taken to guarantee that the local radius of curvature of
the steel plate was always at least an order of magnitude greater
than the half-height of the channel, which represented the maxi-
Pulsatile Flow Facility mum attainable boundary-layer thickness.
A schematic of the pulsatile flow facility is provided in Fig. 1. The apparatus provided a pulsatile, developing channel flow
Airflow was supplied by a blower located upstream of the settling whose relative amplitude, ⌬u/Ū 兩 ⬁ , could be varied to provide
chamber. Pulsations were imparted to the flow by four co-rotating nonreversing, partially reversed, and completely reversed flow.
vanes located in the settling chamber, upstream of the flow- Here a partially reversed flow (⌬u/Ū 兩 ⬁ ⬍1) is a flow where re-
conditioning honeycomb and screens. The vanes rotated in phase versal is confined to some inner portion of the boundary layer,
and modulated the settling chamber pressure by providing a while the outer boundary-layer and the centerline or ‘‘core’’ flow
blockage that varied the flow area between an almost fully open do not reverse at any time during the cycle. A complete or ‘‘bulk’’
position and roughly 8 percent of the total settling chamber cross flow reversal (⌬u/Ū 兩 ⬁ ⬎1) is a flow that periodically reverses
section. Nominal forcing frequencies of 2.5 and 5.0 Hz were used everywhere across the channel height. In general, lower frequen-
so that, a/ ␻ LⰇ1, and the wavelength of the travelling pressure cies provided higher relative amplitudes. The channel flow appa-
wave was much longer than the length of the wind tunnel test ratus generated a developing laminar flow with a core velocity
section. Modulated blockage schemes of this type have been em- profile that was flat to within 3 percent, with a small 共typically
wire data from a non-reversing flow were extrapolated to a zero-
velocity condition at the wall. The uncertainty in the location of
the sensor relative to the wall was conservatively estimated at
⫾200 ␮m 共6.7 to 20 percent of the observed ␦ ␪ 兲 based on the
results of several fits in which the number of data points was
allowed to vary. The uncertainty in probe location was primarily
attributed to errors associated with the number of points chosen
for least-squares fitting and due to scatter in the hot-wire data. The
above stated probe-location uncertainties in both the cold-wire
and CARS measurements are purely bias errors, which shift the
temperature profiles by the stated amounts.
The response of the CCA to temperature fluctuations in the flow
was limited by the thermal inertia of the sensor. This thermal
inertia effect was compensated for by using the analytically based
transfer function derived by Wroblewski and Eibeck 关27兴, which
accounts for the thermal inertia of the active sensor element, plat-
ing, and support prongs:
H共 ␻ 兲⫽
1
␰w
再 1⫹
tanh关 ␰ 1/2
␰w lw
1/2
w lw兴 ␰w
␰c
冋
⫺1⫹
␰w ␰w
⫺
␰p ␰c
cosh关 ␰ 1/2
c lc兴
册冎 , (2)
Fig. 2 A comparison of mean square velocity profiles „pulsa-

tion kinetic energy… obtained from temperature-compensated where l k and ␶ k are the Bechtov cold length and the lumped-
hot-wire measurements to laminar, fully developed, pulsatile capacitance time constant of the indicated sensor elements, re-
flow theory from Currie †20‡ spectively, and ␰ k ⫽1⫹i ␻ ␶ k . The correction to the raw cold-wire
data was made in the frequency domain by dividing the discrete
Fourier transform of the ‘‘raw’’ time-series data by H( ␻ ) and
less than 5 percent variation in the cycle maximum and minimum then inverting the corrected transform. All cold-wire time series
velocity amplitudes兲 cycle-to-cycle variation in the velocity wave- were sampled for 100 periodic forcing cycles. The cold-wire
form. A comparison of temperature-compensated hot-wire mea- transfer function required knowledge of the lumped-capacitance
surements from a nonreversing, developing channel flow in our time constants, ␶ w and ␶ c , which depended on the local heat-
wind tunnel with fully developed laminar pulsatile flow theory transfer coefficient for the wire and support prongs. We used a
关20兴 is provided in Fig. 2. The flow exhibits the well-known time-mean heat-transfer coefficient calculated from a wire heat-
Stokes boundary-layer behavior 关21兴 characterized by an over- transfer correlation obtained in our laboratory 关23兴 to calculate ␶ w
shoot in the near-wall profile associated with the peak level of and ␶ c . This procedure neglects the cyclic variation in the wire
periodic velocity fluctuations and increased levels of mean shear heat-transfer coefficient, but it does allow for a tractable analysis
within the boundary layer. and an approximate correction to the cold-wire data.
Local time-mean velocity data were acquired from hot-wire
Measurement Instrumentation and Procedures measurements in the pulsatile boundary layer by switching the
single-wire probe to a constant-temperature bridge. Analysis of
Instrumentation. Time-resolved boundary-layer temperature the data required an iterative procedure, as the cold-wire tempera-
data were acquired using pure-rotational coherent anti-Stokes Ra- ture data were also used to temperature compensate the hot-wire
man scattering 共CARS兲 and cold-wire anemometry. CARS is a velocity data that were used to estimate the wire heat-transfer
nonlinear, spectroscopic technique that provides a coherent, laser- coefficient 共see Wroblewski and Eibeck 关27兴 or Kearney 关23兴 for
like signal whose spectral signature is similar to the rotational details兲. The process typically converged within one or two itera-
Raman spectrum of air, from which point measurements of tem- tions. The uncertainty in the cold-wire results was largely caused
perature may be obtained. The reader is referred to Kearney et al. by systematic errors associated with calibration, the estimated fre-
关22兴 and Kearney 关23兴 for a detailed explanation of the nonintru- quency compensation, and the intrusive nature of the sensor. The
sive pure-rotational CARS technique used here and an associated uncertainty relating to calibration was ⫾2 K 共2 to 2.5 percent of
uncertainty analysis. The CARS probe-volume provided for a spa- the temperature difference across the boundary layer兲, while the
tial resolution of 50 ␮m in the wall-normal and streamwise direc- uncertainty owed to frequency compensation and probe insertion
tions and for 1-mm resolution in the spanwise coordinate. The 10 was much more ambiguous and was difficult to quantify.
ns pulse width of the Nd:YAG laser provided for an essentially
infinite frequency response, but with a limited data-acquisition Phase-Averaging and Data Acquisition. A ‘‘triple-
rate so that ten temperature profiles per forcing cycle were ob- decomposition,’’ or phase-averaging, procedure was applied to all
tained. The uncertainty in the CARS temperature measurements of the time-series data obtained in this study. In applying the
was ⫾4 K 共3 to 3.4 percent of the temperature difference across triple-decomposition, the instantaneous value of a boundary-layer
the boundary layer兲, which is primarily a random 共precision兲 error time-series quantity, in this case temperature, is written as
associated with mode phase fluctuations in the broadband dye-
laser output 关24,25兴. The uncertainty in the location of the probe T 共 y,t 兲 ⫽T̄ 共 y 兲 ⫹T osc共 y,t 兲 ⫹T ⬘ 共 y,t 兲 , (3)
volume relative to the heat-transfer surface was ⫾50 ␮m 共1.6 to 5
percent of the observed ␦ ␪ 兲, or one probe-volume diameter.
where T̄ is the time-mean of the time-series data, T osc is the peri-
Additional time-resolved temperature measurements were per-
formed using a constant-current anemometer 共CCA兲 or ‘‘cold- odic component of the signal, and T ⬘ is a random contribution
wire’’ technique. The CCA supplied a 0.25 mA current through a resulting from small levels of turbulence and cycle-to-cycle varia-
commercially available, tungsten wire probe with copper plated tion in the flow. All time-series data to be presented here are in a
ends and dimensions d⫽4.5 ␮ m, l/d⫽726, and l /d⫽282. The phase-averaged form given by
*
location of the cold-/ hot-wire probe was determined using the
procedure of Quintana et al. 关26兴, in which boundary-layer hot- 具 T 典共 y,t 兲 ⫽T̄ 共 y 兲 ⫹T osc共 y,t 兲 . (4)
The phase average is dependent on the cycle time, 0⭐ ␻ t⭐2 ␲ , where H indicates a vertical integration across the channel height.
and is calculated by averaging all data within the cycle time bin, In Eq. 7, the absolute value of the velocity, which the hot-wire
␻ t to ␻ (t⫹⌬t), where ␻ is the fundamental circular frequency of instrument directly provides, has been used, as recommended by
the applied forcing. Zhang 关29兴, for calculation of T̄ m in reversing flows.
For the experiments reported here, the time bin for phase aver- Calculation of the Nusselt number from a wall-gradient mea-
aging was determined from the resolution of a shaft encoder that surement can be advantageous because it provides for a direct
was fixed to the lower-most rotating vane shaft and was used to measurement of the convective heat flux without any correction
synchronize and clock all data acquisition. All results are pre- for radiative heat transfer. Several groups of researchers have re-
sented in terms of a cycle time given in shaft-encoder crank-angle cently used wall-gradient information to obtain heat-flux results
degrees, or CAD. Because each complete rotation of the rotating 关26,30,31兴 and shear-stress data 关14,26兴. Only the cold-wire Nus-
vane assembly results in two flow-forcing cycles, a single period selt number results are presented here, as these data provide more
of the forcing is 180 CAD. Use of the shaft encoder output as a points per cycle and better dynamic range.
time base allowed for small levels of jitter in the rotational speed A procedure similar to the one recently outlined by Quintana
and the period of the rotating-vane assembly to be taken into et al. 关26兴 was used for calculation of the wall heat flux. The
account. temperature profiles obtained displayed a characteristic linear,
For the CARS experiments, the encoder clock pulse provided conduction-dominated region near the wall. A linear sublayer cri-
an external time base for a series of counter/timer chips that gen- terion of y⭐400 ␮ m was used in selection of the points used for
erated triggers to properly synchronize the firing of the Nd:YAG linear, least-squares fitting of the data. This criterion was justified
laser and the opening of the CCD detector’s mechanical shutter based on earlier CARS temperature measurements by Kearney
with the desired phase angle of the periodic flow forcing. At each et al. 关22兴 in a steady, tripped boundary layer flow conducted in
vertical location within the boundary layer, 100 single-pulse the same the facility used for the present measurements. These
CARS spectra were acquired at each of the 10 selected phase CARS data showed that the turbulent conduction sublayer could
angles of 2, 30, 50, 60, 70, 80, 100, 120, 140, and 178 CAD. The extend as far as y⫽800 ␮ m at Reynolds numbers similar to the
single-pulse averaging time of the CARS temperature measure- time-mean values employed in the present unsteady work. A 400
ments was 10 ns, or the duration of the Nd:YAG laser pulse. One
hundred single-pulse rotational CARS spectra were averaged in
software to yield an averaged spectrum for each vertical location
and phase angle. Use of 100-shot-averaged CARS spectra was
required to minimize the effects of dye-laser noise. Phase-
averaged temperatures were calculated from the averaged spectra
using the Sandia CARSFT code 关28兴 and the integrated-line-
intensity procedure described by Kearney et al. 关22兴. Using this
data-acquisition procedure, each CARS experimental run took
about eight hours in addition to the time required for the wind
tunnel floor to stabilize at temperatures between 100 to 150 K
above ambient.
For the cold-wire experiments, the encoder output was used to
clock the digitization of the CCA signal at a rate of one point per
CAD. This increase in data-acquisition rate relative to the CARS
technique provided for a more detailed picture of the unsteady
thermal-boundary-layer structure at the expense of an ambiguous
systematic error associated with the intrusive nature and thermal
inertia of the probe. The CCA output was sampled for 100 forcing
cycles, and phase-averaged temperatures were calculated after
correction of the raw CCA output using Eq. 2.
Wall Heat-Flux Calculation. Nusselt-number and
temperature-defect-thickness results were calculated from the
boundary-layer temperature data. The temperature-defect thick-
ness 关1兴,
具 ␦ ␪典 ⫽ 冕冉 0
⬁
1⫺
具 T 典共 y 兲 ⫺T w
T ⬁ ⫺T w 冊 dy, (5)
provides a robust and unambiguous integral measure of the cyclic

variation in thermal-boundary-layer thickness. The cyclically
varying Nusselt number was calculated from
具 q ⬙典 d h
具 Nu典 ⫽ , (6)
␬ ⬁ 共 T w ⫺T̄ m 兲
where 具 q ⬙ 典 is the wall heat flux calculated from the slope of the
near-wall, phase-averaged temperature profile, and T̄ m is the
mixing-cup temperature obtained by integrating phase-averaged
hot-wire velocity and cold-wire temperature data according to
T̄ m 共 t 兲 ⫽
1
ṁ ⬘ c p 冕
H
␳ 兩具 u 典兩 c p 具 T 典 dy, (7) Fig. 3 Time-resolved boundary-layer temperature and heat-
flux data from a nonreversing flow
Table 1 Experimental parameters for representative CARS
and cold-wire measurements „For all experimental runs a Õ ␻ L
š1 so that streamwise variation in the pressure time derivative
was unimportant. The parameter ⌬ u Õ ␻ L was less than 1 in all
experiments so that acoustic streaming effects were negli-
gible.…
Fig. 4 Sample near-wall temperature profile obtained from the

cold-wire data shown in the time series in Fig. 3„b…. The plot
shows that the nature of the deviation between the data and the
least-squares fits is correlated for different cycle times.
␮m cutoff was then justified given the ⫾200 ␮m uncertainty in

the location of the cold-wire probe. For the data reported in this tion of the cold-wire signal and from the intrusive nature of the
paper, six to eight temperature points were used in the linear fits. probe. This phase-lag is present to some extent in all of the re-
The uncertainty in the 具Nu典 data obtained from the cold-wire tem- ported data sets and is less severe for the 2.5-Hz results presented
perature results was determined from the statistical properties of below as should be expected.
the least-squares slope estimator 关23兴. The absolute Nusselt- The temperature data are nearly steady during the weak accel-
number uncertainty varied from ⫾10 percent to ⫾20 percent in eration at the crest of the velocity waveform and into the early
regions of high cyclic heat flux to as high as ⫾75 percent when stages of the fluid deceleration phase. The cold-wire results show
the phase-averaged heat flux was near zero. a cycle-minimum temperature between 75 and 85 CAD during the
Surface heat-transfer and thermal boundary-layer thickness re- later stages of the strong fluid deceleration phase. This tempera-
sults for a nonreversing flow are shown in Fig. 3共d兲. The high ture dip is significant because it is associated with the cycle-
level of negative correlation 共i.e., opposing trends兲 between the maximum Nusselt number. The magnitude of the temperature dip
Nusselt-number results and the robust, integral measurement of is on the order of the ⫾4 K random error in the CARS measure-
the temperature-defect thickness validates the gradient measurement, which combines with the coarse time grid to mask the tem-
ment of the wall heat flux. Furthermore, the scatter in the Nusselt- perature minimum in the CARS results.
number data is not nearly as large as the uncertainty estimates During the acceleration phase of the cycle, both the CARS and
quoted above would suggest. This is indicative of the large sys- cold-wire results exhibit a temperature rise on the order of 20
tematic component in the temperature-gradient error. To illustrate percent of the wall-to-air temperature difference. This temperature
the systematic nature of the error, a representative set of near-wall peak is a result of the advection of high-enthalpy fluid associated
cold-wire temperature data is shown in Fig. 4. The scatter in the with the increased fluid residence time and high levels of heat
temperature data about the least-squares fits is random for each transfer from the wall that occur at phase angles near the cycle-
individual fit, but the nature of the data scatter is correlated be- minimum velocity. The boundary-layer temperature decreases
tween the fits at different cycle times. Therefore, the scatter in the during the later stages of the fluid acceleration phase as the chan-
Nusselt number time-series results for a fixed number of points nel is ‘‘swept clean’’ of high-enthalpy fluid before the beginning
used in the fitting procedure is quite low. Because of the system- of the next forcing cycle. A temperature rise and decay of this
atic nature of the temperature-gradient uncertainty, the relative nature during the fluid acceleration portion of the cycle is ob-
error in the Nusselt number data is decreased, and trends in the served in the data for all three reversal regimes studied, and is
cyclic heat-flux variation are readily seen. believed to be caused by the same enthalpy-advection mechanism
in all cases. From a heat-transfer enhancement standpoint, it is
Results and Discussion important that the combination of convective velocities and forc-
Both CARS and cold-wire data were acquired under conditions ing frequencies be such that all of the high-enthalpy fluid flows
of nonreversing, partially reversed, and fully reversing conditions, through the channel before the start of the next forcing cycle.
with results obtained at several streamwise measurement stations. The results from the nonreversing flow suggest that the key
For brevity, only a representative sample of data from the most mechanism for the generation of periodic temperature fluctuations
upstream measurement location for each of the three flow regimes in this flow is the passive advection of enthalpy. Figures 3共b兲 and
is presented here in time-series form. The operating conditions for 3共c兲 show that the highest levels of periodic temperature fluctua-
the experiments are provided in Table 1. Some variation in the tion occur in the middle portion of the thermal boundary layer,
time-series data was observed with streamwise distance from the where the velocity oscillations within the Stokes layer are highest,
leading edge; however, the data were sufficiently similar to permit as shown in Fig. 2. The nonreversing flow data also show that the
a discussion of the convection process using the most upstream velocity oscillation and the wall heat transfer are out of phase,
data only. A more comprehensive presentation of the data is given while the Nusselt number is in phase with the bulk temperature
by Kearney 关23兴. fluctuation. This result strongly suggests that the velocity field
plays a more secondary role through the passive advection of
Nonreversing Flow. Phase-averaged time-series results from enthalpy and invalidates the simple quasi-steady models proposed
a nonreversing channel flow are shown in Fig. 3. A phase- by some researchers 共Hanby, 关9兴 and Keil and Baird, 关10兴兲, where
averaged hot-wire trace of the centerline flow is provided in Fig. the instantaneous Nusselt number is calculated from steady-flow
3共a兲 for phase reference. Both the CARS and cold-wire results correlations by using the instantaneous velocity.
show similar temperature time histories in the nonreversing flow.
However, the cold-wire results exhibit a phase lag of about 15–20 Partially Reversing Flow. Time-series data for a partially
CAD relative to the CARS data, a result that is likely caused by reversing boundary-layer flow at a forcing frequency of 2.5 Hz are
systematic errors arising from the imperfect frequency compensa- shown in Fig. 5. Both near-wall and centerline hot-wire traces are
difference, which occurs near the time of minimum centerline
velocity and the onset of flow reversal near the wall. This tem-
perature rise is followed by an equally strong decay in the
boundary-layer temperature. The rise and fall in temperature dur-
ing partial flow reversal is so rapid that the coarse time grid used
for the CARS measurements cannot resolve this feature. However,
the nonintrusive CARS results do capture the long-time-scale fea-
tures of the temperature oscillation. The cold-wire results from
this 2.5 Hz data set phase-lag the CARS measurements for this
case by about 10 CAD, or roughly half the observed lag for the
nonreversing flow at 5 Hz presented above.
The obvious thermal impact of flow reversal, seen both in the
partial reversal data and in the bulk-flow-reversal data presented
below, is a significant rise in the temperature history at cycle
times near flow reversal. Significant increases in 具␪典 near flow
reversal have been observed previously in the turbulent flow stud-
ies by Dec and Keller 关1兴 and by Feiler and Yeager 关11兴. Dec and
Keller 关1兴 postulated that this rise in 具␪典 is associated with an
ejection of fluid from the near-wall region during a flow reversal
event. The cold-wire and CARS temperature histories are also
suggestive of an ejection event during flow reversal, as close ex-
amination of the data reveals that the temperature changes near
the wall phase lead the changes in the outer portions of the bound-
ary layer.
In spite of the previous observations of both Dec and Keller 关1兴
and Feiler and Yeager 关11兴, an explanation of this thermal bound-
ary layer thickening has not been given. The energy equation
identifies four possible sources for the generation of temperature
fluctuations: 共1兲 advection of enthalpy, 共2兲 unsteady conduction,
共3兲 viscous dissipation, and 共4兲 compression heating. For the flows
investigated here, dissipation and compression heating are negli-
gible because the Eckert number is only of order 10⫺4 , and be-
cause the levels of the pressure time derivative and velocity are
moderate. Conduction at times of near-stagnant fluid velocities
could play a role in thickening the thermal boundary layer because
of the relatively high temperature differences used and the short
boundary-layer length scales involved. However, a scale analysis
of the unsteady conduction equation, using length and temperature
scales characteristic of our results, shows that the expected mag-
nitude of ⳵ T/ ⳵ t owed to conduction is an order of magnitude less
Fig. 5 Time-resolved boundary-layer temperature and heat-
than the experimentally observed gradients.
flux data from a partially reversed flow Based on the above scaling arguments and the behavior exhib-
ited in the data, passive advection of enthalpy must be the domi-
nant source of temperature fluctuation in this flow. Since our ex-
periments were conducted with constant heater input to the
shown in Fig. 5共a兲 for phase reference and to indicate the location backside of the test plate, it is possible that the reversal-induced
of the near-wall flow reversal event. When compared to the non- temperature rise is caused by streamwise advection of enthalpy
reversing flow discussed above, this partially reversed flow is from warmer downstream regions to cooler-surface-temperature,
characterized by the same levels of convective velocity at half the upstream locations. Analysis of the rectified portions of
forcing frequency, so that each CAD time unit is twice as large. temperature-compensated hot-wire traces, acquired in partially re-
The initiation of a local flow reversal event was determined to versing boundary layers at both 2.5 and 5.0 Hz forcing, showed
coincide with the rectification of the instantaneous hot-wire signal that the maximum upstream travel of a fluid particle was only
in the near-wall region. Identification of flow reversal events about three thermal boundary layer thicknesses. Based on this
based upon observed hot-wire signal rectification has been em- estimate of the upstream fluid particle displacement and the ob-
ployed in several previous studies of time-periodic flows includ- served magnitude of the temperature rise, the axial 共‘‘x’’兲 tem-
ing the work of Feiler 关7兴, Despard and Miller 关15兴, and Qiu and perature gradients present in the flow would necessarily be of
Simon 关32兴. The flow was characterized as partially reversed be- comparable order to the wall-normal 共‘‘y’’兲 gradients. Such a con-
cause hot-wire traces within the boundary layer displayed signal dition is inconsistent with the concept of a thin boundary layer
rectification in a limited region near the wall, but were not recti- and with the observed wall temperature distribution. Therefore,
fied in the outer regions of the boundary layer or in the inviscid the reversal-induced temperature rise must be caused by vertical
core flow—a condition which showed that the reversal event was transport of warm fluid. We believe that there are three potential
confined to a thin region near the wall. mechanisms for vertical fluid transport at cycle times near flow
Similar to the nonreversing boundary-layer flow, the boundary- reversal: 共1兲 a continuity-based ejection event, 共2兲 buoyancy in-
layer temperature history is nearly steady during the elongated duced flow, and 共3兲 ‘‘temporary’’ turbulence.
peak portion of the velocity waveform. A small temperature dip, An ejection event during periodic flow reversal can be ex-
similar to the one observed in the nonreversing flow, is observed plained in terms of continuity. Considering streamwise fluid mo-
in the cold-wire results during the steep deceleration phase of the tion at a fixed distance from the wall, the downstream flow will
cycle. This small temperature dip is followed by a distinct rise in respond more quickly to temporal changes in the pressure gradient
temperature of up to 50 percent of the wall-to-air temperature as a consequence of its lower momentum. A localized flow rever-
scale for the onset of free convection, ␶ ⫽ ␦ 2 / ␣ , where ␣ is the
thermal diffusivity of air, is calculated to be of order 1 sec, which
is much longer than the cycle time for which the velocities are
near zero during a flow reversal event.
A third potential source of bulk fluid mixing in the vertical
direction is the ‘‘temporary’’ turbulence that has often been re-
ported in oscillating and pulsatile flows. It has been well docu-
mented, 关34–37兴 that Stokes boundary layers are very receptive to
transition. Transition to turbulence is often temporary but explo-
sive, occurring during the deceleration phase of the cycle when
strong adverse pressure gradients and inflectional velocity profiles
destabilize the flow. This temporary turbulence is then relaminar-
ized by the strong, favorable pressure gradient encountered during
the cycle acceleration phase.
Boundary-layer hot-wire signals obtained in partially reversed
flows did not display increased levels of random velocity fluctua-
tions during the deceleration phase of the cycle where previous
measurements by Hino et al. 关34兴 suggest that transition will oc-
Fig. 6 Schematic of streamwise boundary-layer profiles dur- cur. For the case of a partial flow reversal, increased levels of
ing periodic flow reversal. The reversal initiates downstream at random fluctuation in the hot- and cold-wire signals were ob-
t Ä t 1 and rapidly propagates upstream. Continuity-based argu- served during the initial portion of the cycle acceleration phase,
ments show that there must be a vertical velocity event that from 85 to about 110 CAD, when the periodic temperature fluc-
accompanies flow reversal. tuations were highest. These random fluctuations were not explo-
sive in nature and transition to turbulence is not expected during a
strong acceleration, which often has a stabilizing effect. However,
sal will then initiate near the downstream end of the channel, as the increase in random fluctuations does indicate that some level
shown schematically for t⫽t 1 in Fig. 6. As reversed fluid begins of disturbance was present in the flow during the partial flow
to move upstream, it will immediately encounter stationary or reversal. The magnitude of this disturbance is moderate, and a
forward-moving fluid, and a secondary, transverse and/or span- good deal of the fluctuation in the anemometer output at times
wise flow must be established to preserve continuity. If we con- near flow reversal is caused by disturbed fluid that has passed over
sider a two-dimensional model and neglect the changes in gas the sensor and its supports for a second and third time, and by
density owed to temperature 共an assumption justified by similar imperfect temperature compensation of the dynamic hot-wire sig-
results from low-temperature-difference cold-wire tests兲, then the nal. Nevertheless, it is likely that this disturbance is due in part to
simplified continuity relation applies some level of transition, which contributes to the rapid tempera-
ture rise associated with reversal and acts in conjunction with the
⳵u ⳵v bulk ejection mechanism explained above to produce a rapid in-
⫹ ⫽0. (8)
⳵x ⳵y crease in thermal boundary-layer thickness during a flow reversal
event.
For a positive-to-negative, zero-velocity crossing, ⳵ u/ ⳵ x⬍0, so
that ⳵ v / ⳵ y⬎0. An ejection event must then occur almost simul- Complete Flow Reversal. Time-series data for the case of a
taneously with a local flow reversal to satisfy both continuity and complete flow reversal at a forcing frequency of 2.5 Hz are shown
the no-slip condition. A ‘‘reversal line’’ then quickly moves up- in Fig. 7. When compared to the partially reversed flow discussed
stream, as shown in Fig. 6, until: 共1兲 the near-wall flow is reversed above, this completely reversed flow is characterized by the same
everywhere, or 共2兲 the flow is accelerated in the positive direction frequency and by convective velocities that are reduced by more
again. Furthermore, any ejection must also be accompanied by a than a factor of two. Similar to the data sets presented above, the
sudden inrush in order to preserve continuity, which would ac- temperature history is essentially steady during the high-velocity
count for the subsequent dip in temperature following the rapid portion of the cycle, as high-enthalpy fluid resulting from heat
initial rise observed in the cold-wire results. This description and transfer and vertical transport of warm fluid during the previous
the necessary inflectional velocity profiles are similar in concept cycle has been advected past the measurement location before the
to steady separation, with the exception that the inviscid core flow next forcing cycle begins. Both the CARS and cold-wire results
still conforms to the shape of the wall. indicate a temperature rise associated with the initiation of bulk
Due to the high temperature differences employed in this study, flow reversal that is similar in nature to the partially reversed flow
buoyancy induced secondary flows in the form of thermal plumes results presented above. As in the case of a partial flow reversal,
could also be a source of vertical fluid motion at cycle times near the immediate impact of a positive-to-negative zero-velocity
flow reversal when the fluid velocity is nearly stagnant across the crossing is a rapid increase in thermal boundary-layer thickness
height of the channel. The impact of buoyancy was investigated associated with a fluid ejection event and a disturbed boundary
experimentally by acquiring temperature time-series data in a pul- layer. This increase in thermal boundary-layer thickness is accom-
sating channel flow with a cold wire. Channel floor temperatures panied by a cyclic heat-transfer minimum followed by a short-
of 4, 43, 60, 91, and 95 K above ambient were investigated. The lived thinning of the thermal boundary layer and a period of de-
cold-wire results, when normalized, showed that all of the tem- creased heat transfer caused by the advection of high-enthalpy
perature histories were both qualitatively and quantitatively of the fluid through the measurement volume.
same character 关23兴. This result is also consistent with the calcu- The duration of the period of near-cycle-minimum heat transfer
lated time scale for the generation and release of a buoyant plume is more than twice as long as observed for the case of partial flow
from a horizontal, heated wall in otherwise quiescent surround- reversal. The lower speed flow 共required to generate a bulk rever-
ings. Instability of the thermal boundary layer and release of a sal when using a mechanical forcing scheme兲 simply takes longer
buoyant plume occurs for Ra␦ ⬃1000 关33兴, where ␦ is the critical to sweep the channel clear of high-enthalpy fluid. In addition, high
thermal-boundary-layer thickness. For the conditions of our ex- levels of negative velocities can couple with the downstream wall
periments 共in air with ⌬T⬃100 K兲 this results in a length scale, ␦, boundary conditions to further influence surface heat transfer
of about 4 to 5 mm before the thermal boundary layer becomes through a streamwise advection mechanism that was unimportant
unstable and a plume is released. Using this length scale, the time for the small negative velocities observed in the partially reversed
Fig. 8 Time-mean heat transfer data plotted against the in-
verse of the thermal Graetz number. The data are compared to
accepted laminar heat transfer correlations obtained from Shah
and Bhatti †38‡ and show a heat-transfer enhancement with
flow pulsation in all cases. The error bars on the plot reflect a
Á15 percent uncertainty in time-mean Nusselt number for non-
reversing and partially reversed flow and a Á25 percent uncer-
tainty in the Nusselt number for the case of bulk flow reversal.
compared to a nonreversing, pulsatile flow. A similar decrease in

heat flux, relative to the positive-velocity cycle times, during the
negative-velocity portion of the cycle was also observed in the
100 Hz, turbulent flow results of Dec and Keller 关1兴. Relative to
the period of the applied forcing, the period of decreased heat
transfer is longer if the period of reversed flow is longer, so that
the lowest Nusselt numbers are obtained for a complete flow re-
versal.
At low GzT⫺1 共thin thermal boundary layers兲, the time-mean
Nusselt-number data for the nonreversing flow at 5 Hz and the
partially reversed flow at 2.5 Hz forcing differ by less than the
⫾15 percent uncertainty in the measurement, but the data are still
suggestive of higher heat-transfer rates for a nonreversing flow. At
higher GzT⫺1 共thicker thermal boundary layers兲, the Nusselt num-
bers for the partial flow reversal at 5 Hz forcing are higher or
Fig. 7 Time-resolved boundary-layer temperature and heat- equal within the measurement uncertainty to the data for bulk
flux data from a fully reversing channel flow reversal at 2.5 Hz. Previous authors 共Bayley et al. 关3兴; Keil and
Baird 关10兴; Hanby 关9兴兲 have postulated that flow reversal is a
requirement for heat-transfer enhancement. However, the time-
case. Estimates obtained from rectified, temperature-compensated mean heat transfer results presented here show that periodic flow
CTA traces acquired in the boundary layer suggest that warm fluid reversal does not necessarily lead to increased heat transfer rates
can travel as much as 40 percent of the channel length back to- relative to nonreversed pulsatile flows in the developing regime.
ward the leading edge during a bulk flow reversal. During the Furthermore, the heat-transfer results obtained in this study show
strong bulk flow reversal, high-enthalpy fluid was advected up- that the time-mean Nusselt numbers are generally enhanced rela-
stream to ‘‘reheat’’ the wall and produce a reversal of the wall tive to the accepted steady flow heat-transfer correlations for lami-
heat flux at this upstream measurement location for cycle times nar, two-dimensional, entry region flow provided by Shah and
from 70 to 110 CAD. No reversal in heat flux was observed with Bhatti 关38兴, regardless of reversal regime. This result indicates
bulk flow reversal at measurement locations further downstream. that pulsating flows show promise as a heat-transfer enhancement
This type of conjugate effect is suggestive of an increased sensi- technique in the developing laminar flows that are characteristic
tivity of pulsatile-flow heat transfer to the surface boundary con- of many single-phase heat-exchanger applications.
dition and is important in applications where surface temperature
distributions are often not isothermal.
Time-Averaged Heat Transfer. The time-mean Nusselt Conclusion
number results obtained from 12 sets of cold-wire experiments are The results of an experimental investigation of the time-
plotted against the inverse of the thermal Graetz number in Fig. 8. resolved thermal boundary-layer structure and surface heat flux in
In construction of Fig. 8, time-mean velocities have been used to a pulsatile, laminar channel flow have been reported. Two tech-
calculate the Graetz numbers for the unsteady data sets. The in- niques, pure-rotational CARS and cold-wire anemometry, have
verse of the Graetz number is a measure of the state of thermal been used to acquire phase-averaged temperature data. Pure-
boundary-layer development, and allows the effects of the thermal rotational CARS has been used for nonintrusive measurements of
entry length to be accounted for in a heat-transfer comparison of the long-time trends in the cyclic temperature oscillation and pro-
the data sets. vides measurements with an effectively infinite frequency re-
The phase-averaged heat transfer results shown in Figs. 3, 5, sponse. The increased data-acquisition rates offered by the inva-
and 7 indicate that the instantaneous impact of both partial and sive cold-wire technique provide a more detailed description of
complete flow reversals is a decrease in the Nusselt number when the temperature oscillation, at the expense of a small but finite
phase lag in the response of the probe resulting from imperfect x ⫽ streamwise coordinate 关m兴
frequency compensation and probe intrusion error. y ⫽ wall-normal coordinate 关m兴
The thermal impact of flow reversal has been systematically Greek Symbols
investigated. Time-resolved temperature data from nonreversing,
partially reversed, and complete flow reversals are presented. The ␦␪ ⫽ temperature defect thickness defined in Eq. 5 关m兴
cycle-resolved impact of flow reversal is a dramatic thickening of ␬ ⫽ thermal conductivity 关W/m-K兴
the thermal boundary layer, which appears to be caused by a ␯ ⫽ kinematic viscosity 关m2/s兴
vertical ejection of fluid from the near-wall region and by a lim- ␪ ⫽ normalized temperature ( ␪ ⫽T⫺T ⬁ /T w ⫺T ⬁ ) 关-兴
ited degree of turbulent transition. During the reversed-flow por- ␶ ⫽ lumped capacitance time constant, ( ␶ ⫽ ␳ c p d/4h) 关s兴
tion of the cycle, a cyclic heat-transfer minimum is attained. Fol- ⌬u ⫽ difference between min. and max. centerline velocities
lowing reversal, the wall heat transfer recovers as the accelerated 关m/s兴
boundary-layer flow sweeps the channel clear of high-enthalpy ␻ ⫽ circular frequency 关s⫺1兴
fluid. Subscripts
This time-resolved description is consistent with the observed
time-mean Nusselt-number results. The local time-mean Nusselt c ⫽ cold-wire plating property
number decreases as the duration of the reversal phase, and the osc ⫽ oscillating component
associated period of low heat-transfer rates, increases relative to p ⫽ cold-wire prong property
the cycle period. For the developing, laminar regime investigated, w ⫽ wall condition, cold-wire sensor property
the time-mean Nusselt number data suggest, within the bounds of x ⫽ evaluated at local streamwise location
the measurement uncertainty, that heat-transfer performance im- ⬁ ⫽ channel centerline condition
proves with lesser degrees of flow reversal. All of the pulsatile
flows studied exhibit enhancement relative to accepted Nusselt- References
number data for a developing, laminar channel flow, and enhance-
关1兴 Dec, J. E., Keller, J. O., and Arpaci, V. S., 1992, ‘‘Heat Transfer Enhancement
ments as high as a factor of two are observed for low values of in the Oscillating Turbulent Flow of a Pulsed Combustor Tail Pipe,’’ Int. J.
GzT⫺1 . These results show that enhancement can be achieved with Heat Mass Transf., 35, No. 9, pp. 2311–2325.
关2兴 Niida, T., Yoshida, T., Yamashita, R., and Nakayama, S., 1974, ‘‘The Influ-
low-amplitude 共e.g., low ⌬u/Ū 兩 ⬁ 兲, mechanical forcing schemes, ence of Pulsation on Laminar Heat Transfer in Pipes,’’ Heat Transfer-Jpn.
which will produce lower noise levels and require less energy Res., 3, No. 3, pp. 19–28.
input relative to acoustically resonant schemes that have been 关3兴 Bayley, F. J., Edwards, P. A., and Singh, P. P., 1961, ‘‘The Effect of Flow
used previously to produce large bulk flow reversals. Pulsations on Heat Transfer by Forced Convection from a Flat Plate,’’ Pro-
ceedings of the First International Heat Transfer Conference, Boulder, CO,
pp. 499–509.
关4兴 Galitseyskiy, B. M., Danilov, Y. I., Dreyster, G. A., Kalinin, E. K., and
Acknowledgment Koshkin, V. N., 1969, ‘‘Effect of Resonant Fluctuations of Coolant Pressure
The authors would like to thank the University of Illinois Air- on Convective Heat Transfer in Pipes,’’ Heat Transfer-Sov. Res., 1, No. 2, pp.
177–185.
Conditioning and Refrigeration Center 共ACRC兲 for partial fund- 关5兴 Fujita, N., and Tsubouchi, T., 1982, ‘‘An Experimental Study of Unsteady
ing of this work. The ACRC is a National Science Foundation Heat Transfer from a Flat Plate to an Oscillating Air Flow,’’ Heat Transfer-
Industry/University Cooperative Research Center 共IUCRC兲. The Jpn. Res., 11, pp. 31–43.
first author also wishes to thank the University of Illinois, Depart- 关6兴 Miller, J. A., 1969, ‘‘Heat Transfer in the Oscillating Turbulent Boundary
Layer,’’ J. Eng. Power, 91, pp. 239–244.
ment of Mechanical and Industrial Engineering for its financial 关7兴 Feiler, C. E., 1964, ‘‘Experimental Heat-Transfer and Boundary-Layer Behav-
support. The constructive reviews of Steven Trujillo, Tim O’Hern, ior with 100-CPS Oscillations,’’ NASA TN No. 2531, Washington, DC.
and Art Ratzel of Sandia are also gratefully acknowledged. 关8兴 Jackson, T. W., and Purdy, K. R., 1965, ‘‘Resonant Pulsating Flow and Con-
vective Heat Transfer,’’ ASME J. Heat Transfer, 87, pp. 507–512.
关9兴 Hanby, V. I., 1969, ‘‘Convective Heat Transfer in a Gas-Fired Pulsating Com-
Nomenclature2 bustor,’’ J. Eng. Power, 91, pp. 48–52.
关10兴 Keil, R. H., and Baird, M. H. I., 1971, ‘‘Enhancement of Heat Transfer by
Roman Symbols Flow Pulsation,’’ Ind. Eng. Chem. Process Des. Dev., 10„4…, pp. 473–478.
关11兴 Feiler, C. E., and Yeager, E. B., 1962, ‘‘Effect of Large Amplitude Oscilla-
a ⫽ sound speed 关m/s兴 tions on Heat Transfer,’’ NASA TR R-142, Washington, DC.
q⬙ ⫽ wall heat flux 关W/m2兴关12兴 Dec, J. E., and Keller, J. O., 1989, ‘‘Pulse Combustor Tail Pipe Heat Transfer
d ⫽ diameter 关m兴 Dependence on Frequency, Amplitude, and Mean Flow Rate,’’ Combust.
dh ⫽ hydraulic diameter 关m兴 Flame, 77, pp. 359–374.
关13兴 Dec, J. E., and Keller, J. O., 1990, ‘‘Time-Resolved Gas Temperatures in the
GzT ⫽ Graetz number (GzT ⫽RexPr/ 关 x/d h 兴 ) 关-兴 Oscillating Turbulent Flow of a Pulse Combustor Tail Pipe,’’ Combust. Flame,
h ⫽ heat-transfer coefficient 关W/m2-K兴 80, pp. 358–370.
H ⫽ channel half height 关m兴关14兴 Dec, J. E., Keller, J. O., and Hongo, I., 1991, ‘‘Time-Resolved Velocities and
L ⫽ streamwise length scale, channel length 关m兴 Turbulence in the Oscillating Flow of a Pulse Combustor Tail Pipe,’’ Com-
bust. Flame, 83, pp. 271–292.
l ⫽ Bechtov cold length, (l⫽ 冑␶␣ ), total length of cold- 关15兴 Despard, R. A., and Miller, J. A., 1971, ‘‘Separation in Oscillating Laminar
wire sensor 关m兴 Boundary Layer Flows,’’ J. Fluid Mech., 47, pp. 21–31.
l ⫽ active length of cold-wire sensor 关m兴关16兴 Charnay, G., and Mathieu, J., 1976, ‘‘Periodic Flow in a Wind Tunnel Pro-
* duced by Rotating Shutters,’’ ASME J. Fluids Eng., 98, pp. 278–283.
ṁ ⬘ ⫽ mass flux per unit depth in the spanwise direction 关kg/s 关17兴 Simpson, R. L., Sallas, J. J., and Nasburg, R. E., 1978, ‘‘Tailoring the Wave-
m兴 from of a Periodic Flow With a Programmable Damper,’’ ASME J. Fluids
Nux ⫽ local Nusselt number (Nux ⫽q x⬙ d h / ␬ ⬁ ⌬T) 关-兴 Eng., 100, pp. 287–290.
关18兴 Morel, T., 1977, ‘‘Design of Two-Dimensinal Wind-Tunnel Contractions,’’
Pr ⫽ Prandtl number (Pr⫽ v / ␣ ) 关-兴 ASME J. Fluids Eng., 99, No. 2, pp. 371–378.
Rex ⫽ Reynolds number (Rex⫽Ūx/v) 关-兴关19兴 Swearingen, J. D., and Blackwelder, R. F., 1987, ‘‘The Growth and Break-
T ⫽ temperature 关K兴 down of Streamwise Vortices in the Presence of a Wall,’’ J. Fluid Mech., 182,
t ⫽ time 关s兴 pp. 255–290.
关20兴 Currie, I. G., 1993, Fundamental Mechanics of Fluids, 2nd ed., McGraw-Hill,
U, u ⫽ streamwise velocity 关m/s兴 New York.
v ⫽ wall-normal velocity 关m/s兴关21兴 Stuart, J. T., 1963, ‘‘Unsteady Boundary Layers,’’ in Laminar Boundary Lay-
Wo ⫽ Womersley number based on channel half height (Wo ers, L. Rosenhead, ed., Clarendon Press, Oxford, UK.
⫽H 冑␻ / v ) 关-兴关22兴 Kearney, S. P., Lucht, R. P., and Jacobi, A. M., 1999, ‘‘Temperature Measure-
ments in Convective Heat Transfer Flows Using Dual-Broadband, Pure-
Rotational Coherent Anti-Stokes Raman Spectroscopy 共CARS兲,’’ Exp. Therm.
2
Note: All dimensionless groups are evaluated using the properties of air evalu- Fluid Sci., 19, pp. 13–26.
ated at the channel centerline. 关23兴 Kearney, S. P., 1999, ‘‘An Experimental Investigation of Thermal Boundary
Layer Structure and Heat Transfer in Oscillating Flows Using CARS Spectros- 关30兴 Qiu, S., Simon, T. W., and Volino, R. J., 1995, ‘‘Evaluation of Local Wall
copy and Cold-Wire Anemometry,’’ Ph.D. dissertation, The University of Il- Temperature, Heat Flux, and Convective Heat Transfer Coefficient from the
linois, Urbana, IL. Near Wall Temperature Profile,’’ ASME-HTD, 318, pp. 45–52.
关24兴 Kröll, S., and Sandell, D., 1988, ‘‘Influence of Laser Mode Statistics on Noise 关31兴 Volino, R. J., and Simon, T. W., 1997, ‘‘Velocity and Temperature Profiles in
in Nonlinear Optical Processes—Application to Single-Shot Broadband Co- Turbulent Boundary Layer Flows Experiencing Streamwise Pressure Gradi-
herent Anti-Stokes Raman Scattering Thermometry,’’ J. Opt. Soc. Am. B, 5,
ents,’’ ASME J. Heat Transfer, 119, No. 3, pp. 433–439.
No. 9, pp. 1910–1926.
关25兴 Aldén, M., Bengtsson, P. E., Edner, H., Kröll, S., and Nilsson, D., 1989, 关32兴 Qiu, S., and Simon, T. W., 1994, ‘‘Measurements of Heat Transfer and Fluid
‘‘Rotational CARS: A Comparison of Different Techniques with Emphasis on Mechanics within an Oscillatory Flow in a Pipe,’’ ASME-HTD, 285, pp. 45–
Accuracy in Temperature Determination,’’ Appl. Opt., 28, No. 15, pp. 3206– 52.
3219. 关33兴 Bejan, A., 1995, Convection Heat Transfer, Wiley, New York.
关26兴 Quintana, D. L., Amitay, M., Ortega, A., and Wygnanski, I. J., 1997, ‘‘Heat 关34兴 Hino, M., Sawamoto, M., and Takasu, S., 1976, ‘‘Experiments on Transition
Transfer in the Forced Laminar Wall Jet,’’ ASME J. Heat Transfer, 119, pp. to Turbulence in Oscillatory Pipe Flow,’’ J. Fluid Mech., 75, pp. 193–207.
451–459. 关35兴 Clamen, M., and Minton, P., 1977, ‘‘An Experimental Investigation of Flow in
关27兴 Wroblewski, D. E., and Eibeck, P. A., 1991, ‘‘A Frequency Response Com- an Oscillating Pipe,’’ J. Fluid Mech., 81, pp. 421–431.
pensation Technique for Cold Wires and Its Application to a Heat Flux 关36兴 Akhavan, R., Kamm, R. D., and Shapiro, A. H., 1991, ‘‘An Investigation of
Probe,’’ Exp. Therm. Fluid Sci., 4, pp. 452–463. Transition to Turbulence in Bounded Oscillatory Stokes Flows: Part 1 Experi-
关28兴 Palmer, R. E., 1989, ‘‘The CARSFT Computer Code for Calculating Coherent
ments,’’ J. Fluid Mech., 225, pp. 395–422.
Anti-Stokes Raman Spectra: User and Programmer Information,’’ Report No.
关37兴 Verzicco, R., and Vittori, G., 1996, ‘‘Direct Simulation of Transition in Stokes
SAND89-8206, Sandia National Laboratories, Albuquerque, NM and Liver-
more, CA. Boundary Layers,’’ Phys. Fluids, 8, pp. 1341–1343.
关29兴 Zhang, L. W., 1996, ‘‘A Numerical Study of Flow and Heat Transfer in 关38兴 Shah, R. K., and Bhatti, M. S., 1988, ‘‘Laminar Heat Transfer in Wall
Compact Heat Exchangers,’’ Ph.D. dissertation, The University of Illinois, Bounded Flows,’’ in Handbook of Single-Phase Convective Heat Transfer, R.
Urbana, IL. K. Shah, S. Kakac, and M. Yenir, eds., McGraw-Hill, New York.
Effect of Turbulence With Different
Vortical Structures on Stagnation
Aung N. Oo
Graduate Student
Faculty of Engineering and Applied Science
Region Heat Transfer
Memorial University of Newfoundland
St. John’s, Newfoundland, A1B 3X5, Canada The effect of freestream turbulence with different vortical structures on the stagnation
e-mail: aung@engr.mun.ca region heat transfer was experimentally studied. Reynolds numbers, based on leading
edge diameter of the heat transfer model with a cylindrical leading edge, ranged from
Chan Y. Ching 67,750 to 142,250. Turbulence generating grids of parallel rods were placed at several
Associate Professor positions upstream of the heat transfer model in orientations where the rods were per-
Dept. of Mechanical Engineering, pendicular and parallel to the stagnation line. The turbulence intensity and ratio of
McMaster University, integral length scale to leading edge diameter were in the range 3.93 to 11.78 percent
Hamilton, Ontario, L8S 4L7, Canada and 0.07 to 0.70, respectively. The grids with rods perpendicular to the stagnation line,
e-mail: chingcy@mcmaster.ca where the primary vortical structures are expected to be perpendicular to the stagnation
line, result in higher heat transfer than those with rods parallel to the stagnation line. The
measured heat transfer data and turbulence characteristics are compared with existing
correlation models. 关DOI: 10.1115/1.1375165兴
Keywords: Convection, Experimental, Heat Transfer, Turbulence, Vortex
Introduction oped through specific experiments are not applicable to sets of

data from other researchers. More importantly, most correlations
Heat transfer in the stagnation region is important in many en-
are for nearly isotropic turbulence generated using square-mesh
gineering applications. A critical application would be the heat
grids. However, turbulence in gas turbine engines is expected to
transfer from the combustion gases to the turbine blades in a gas be highly anisotropic 关15兴 and well laced with coherent vortical
turbine. In many instances, the temperature of combustion gases structures 关16兴.
can exceed the melting point of blade materials. Accurate predic- Heat transfer augmentation in the stagnation region is hypoth-
tion of turbine blade heat transfer is essential to improve the de- esized to be caused by vorticity amplification 关17–19兴. If a vorti-
sign of blade cooling systems. The estimation of heat transfer to cal filament, which is normal to the stagnation line and freestream
the first stage blades and vanes of a newly designed gas turbine flow direction, is considered, the filament is stretched and tilted as
can be in error by a factor of two or three under certain engine it is advected into the stagnation region due to divergence and
conditions, especially in the stagnation region 关1–3兴. acceleration around the bluff body 共see Fig. 1兲. The stretching
A complete understanding of the transport mechanisms of mo- causes the vorticity to be intensified through conservation of an-
mentum and heat in turbulent flows has not been achieved. It is, gular momentum. The vortical filament with intensified vorticity
however, well established that the coherent vortical structures in a interacts with the boundary layer and induces velocity gradients in
turbulent flow play an important role in momentum and heat the spanwise direction parallel to the stagnation line. The three-
transfer. A knowledge of the interaction of freestream vortical dimensional velocity gradients enhance the transport mechanism
structures with the stagnation region boundary layer should lead within the boundary layer resulting in higher heat transfer. On the
to improved models of stagnation region heat transfer. other hand, a vortical filament, which is parallel to the stagnation
There have been several studies on the effect of freestream line, is not stretched due to the lack of apparent velocity diver-
turbulence on stagnation region heat transfer 关4–14兴. Correlations gence in this direction as it approaches the stagnation region. Van-
between the stagnation region heat transfer and the characteristics Fossen et al. 关12兴 found that the correlation to estimate the stag-
of freestream turbulence such as turbulent intensity, integral nation line heat transfer from the characteristics of freestream
length scale and Reynolds number have been developed. Smith
and Kuethe 关4兴 suggested a semi-empirical theory for the augmen-
tation of heat transfer at the stagnation point of a circular cylinder.
They assumed a linear relation between the stagnation point Fr
(⬅Nu 冑ReD ) and the freestream turbulence level for a constant
Reynolds number, and their correlation was found to be valid up
to a turbulence intensity of about 6 percent. Kestin and Wood 关5兴
and Lowery and Vachon 关6兴 used the parameter Tu 冑ReD to cor-
relate the stagnation line heat transfer data. VanFossen et al. 关12兴
developed a correlation model for the stagnation point heat trans-
fer by incorporating the integral length scale in addition to Rey-
nolds number and turbulence intensity. Yeh et al. 关10兴 proposed a
correlation model for the heat transfer at the stagnation point of a
gas turbine blade based on the parameter developed by VanFossen
et al. 关12兴, but with different constants and exponents in order to
best fit their experimental data. In most cases, correlations devel-

HEAT TRANSFER. Manuscript received by the Heat Transfer Division June 30, 2000; Fig. 1 Vortex stretching and vorticity intensification around
revision received January 20, 2001. Associate Editor: M. Faghri. the leading edge
turbulence generated by square mesh grids under-predicted the
stagnation heat transfer rate using a grid of parallel rods. The
hypothesis of vortex stretching and heat transfer enhancement can
be experimentally investigated using a turbulent freestream with
different coherent vortical structures. This should also provide
further insight to the turbulent heat transfer mechanism in the
stagnation region. This experimental study was, therefore, con-
ducted to investigate the influence of freestream turbulence with
different coherent vortical structures on the stagnation region heat
transfer.
The objectives of the study are as follows:
1 to generate turbulence with different vortical structures 共one
with primary vortices susceptible to stretching and another
with primary vortices not susceptible to stretching as they
approach the stagnation region兲
2 to quantify the heat transfer enhancement in the stagnation
region by the two different turbulent flows in 共1兲 Fig. 3 Arrangements of grids
3 to examine the nature of heat transfer augmentation over the
stagnation region by different freestream coherent vortical
structures and a variable AC/DC transformer is used to supply power across
4 to investigate the difference in stagnation region heat trans- the foils to obtain a constant heat flux surface at the leading edge.
fer due to freestream turbulence with distinct vortical struc- Six calibrated, 36-gauge T-type copper-constant thermocouples
tures and that due to turbulence generated using square mesh are attached to the underside of every foil strip on one half 共0 deg
grids ⭐␪
⭐ 90 deg兲 of the cylindrical leading edge. The thermocouples are
Experimental Facilities and Test Procedures evenly distributed in the middle portion of the strip in such a way
that the distance between the edge of the strip and the nearest
The experiments were performed in an open circuit low speed
thermocouple is 3.25 cm as shown in Fig. 2. One instrumented
wind tunnel which has a test section of 1m⫻1m and is over 20 m
strip is also placed symmetrically opposite on the other half of the
long. The wind tunnel is driven by a 19 kW centrifugal blower.
leading edge in order to check the alignment of the model with the
The air passes through a screened diffuser and a large settling
mean flow direction. The thermocouples are threaded through
chamber with three single-piece precision screens before acceler-
small holes drilled on the leading edge. Additional thermocouples
ating into the test section through a 5:1 contraction The freestream
are attached on the inner wall of the leading edge to estimate the
turbulence intensity without the grids is less than 0.5 percent at all
conduction heat loss through the Plexiglass.
flow rates. The velocity in the test section is changed using mo-
Data from the thermocouples are acquired via two CYEXP 32
torized variable angle inlet vanes on the blower.
Multiplexors 共Cyber Research兲 connected in series. Analog out-
Figure 2 shows a schematic of the heat transfer model with a
puts from the Multiplexors are digitized using a DAS 1602 A/D
cylindrical leading edge to simulate the leading edge of a gas
converter 共Cyber Research兲 and data logging is controlled by the
turbine vane. The leading edge is attached to a flat body 60 cm
‘‘Labtech Buildtime’’ software package. Freestream velocity was
long. The after body is then streamlined with a long tapered tail in
measured by 8360-M-GB VelociCalc® Plus TSI air velocity
order to prevent the shift of the stagnation point due to vortex
meter, which has a resolution of 0.01 m/s. The voltage supply and
shedding at the end. The model is made of 0.635 cm thick Plexi-
current flow to the leading edge were recorded using a multimeter
glass. The cylindrical leading edge is 20.32 cm in diameter and
with a resolution of 0.01 volt and 0.01 ampere.
1m in height creating a flow blockage of 20 percent.
To generate freestream turbulence with well-defined primary
The middle portion of the leading edge is covered with a stain-
vortex lines, grids of parallel rods were used. The grids were
less steel surface, which is to be heated for the heat transfer test.
placed upstream of the model in different orientations as shown in
The stainless steel surface consists of nineteen strips of 0.005
Fig. 3. From hereafter, the grid with the rods perpendicular to the
cm-thick stainless steel foil, and the strips are evenly distributed
stagnation line will be called the grid in perpendicular orientation,
over the stagnation line at the center of the cylindrical leading
while the grid with the rods parallel to the stagnation line will be
edge. Each strip is 15.24 cm long, located vertically in Y direction
called the grid in parallel orientation. The rods in both grid orien-
over the leading edge surface, 1.5 cm wide and separated from
tations are perpendicular to the streamwise direction. Three grids
each other by a gap of 1 mm. The gaps between the strips are
with rod diameters of 2.86 cm 共1-1/8’’兲, 1.59 cm 共5/8’’兲 and 0.95
filled with super-glue and sanded until the surface is smooth and
cm 共3/8’’兲, and 50 percent open area were used. Each grid was
flush with the foil surface. The steel foils are connected in series,
placed at five different positions, 25d to 125d, upstream of the
stagnation line of the model. The wind tunnel was operated at
freestream velocities of 5, 8 and 10.5 m/s, corresponding to ReD
of 67,750, 108,350, and 142,250, for each grid position. The lead-
ing edge was heated to about 45°C for each heat transfer test. The
thermocouple readings were monitored every 15 minutes until
steady state conditions were reached. Three sets of temperature
distributions and voltage and current flow of the DC power supply
unit were recorded after steady state was attained. Each data set of
temperature distributions contains 180 data points 共data acquisi-
tion rate of 1.5 Hz for 2 minutes兲 for each thermocouple.
Single and X-wires were used to measure the freestream turbu-
lence characteristics at several positions downstream of the rod-
grids in perpendicular orientation in the absence of the heat trans-
fer model. Two sets of X-wire data, one for the simultaneous
Fig. 2 Schematic diagram of heat transfer model measurement of the velocity components in X and Y directions
and another in X and Z directions 共see Fig. 3兲, were obtained by then used to determine the total conduction heat losses taking into
rotating the X-wire probe 90 deg. The hot wires were operated by account the lateral end conduction heat losses. Correction factors
DANTEC 55M01 constant temperature bridges, and output volt- were obtained for different freestream Reynolds numbers to esti-
ages of the anemometers were digitized using a 16 channel 12 bit mate the conduction heat losses based on the average temperature
Keithley 570 System A/D converter, interfaced to a personal com- difference between the heated outer surface and inner wall of the
puter. The frequency response of the circuits was determined by leading edge. Heat transfer tests were also performed without any
standard square-wave tests, and found to be 30 kHz. The sampling grids to validate the heat transfer measurements. The stagnation
frequency of the study range from 20 to 30 kHz. line heat transfer Frossling numbers for the freestream with low
turbulence intensity 共0.5 percent for this study兲 ranged from 1.04
Data Reduction and Experimental Uncertainties to 1.06 for the Reynolds numbers of the study, and are in good
Electrical energy supplied to the series of stainless steel foils is agreement with the literature 关6,8兴.
lost to the freestream by convection, to the surrounding by radia- Streamwise turbulence intensity Tu(⫽u/U) and integral length
tion, and to the leading edge of the model by conduction. The scale (␭ x ) were calculated from the single hot wire data. The
Nusselt number was estimated as follows: method proposed by VanFossen et al. 关12兴 was used to estimate
␭ x by fitting the auto-correlation data with an exponential func-
Q covD tion. The rms fluctuating velocity components in spanwise direc-
Nu共 ␪ 兲 ⫽ (1) tions 共v and w兲 were calculated from the X-wire data. Experimen-
A 关 Tw 共 ␪ 兲 ⫺T ⬁ 兴 k
tal uncertainties in the v and w components and ␭ x , based on the
where uncertainty analysis of hot wire data by Yavuzkurt 关21兴, were
Q cov⫽Q in⫺Q rad⫺Q cond . estimated to be 3.09 percent and 10.95 percent, respectively. The
turbulence characteristics were found to be uniform in the span-
Energy input to the leading edge, Q in , is obtained from supply wise direction at all streamwise downstream locations, i.e., x/d of
voltage, V O , and current flow, I. An estimation for the radiation 25 to 125. This is consistent with the results of Roach 关22兴, where
heat loss, Q rad , was made by assuming gray body radiation to the turbulence was found to be homogeneous beyond ten mesh
black surroundings and emissivity of 0.17 for stainless steel foil. lengths, i.e., x/d⫽20 for this study.
Conduction heat losses, Q cond , through the leading edge wall was
computed from the measured temperature difference between the
outer and inner surface of the leading edge wall. Thermal conduc- Results and Discussion
tivity of the leading edge was 0.201 W/m-K as provided by the The streamwise turbulence intensity downstream of the three
manufacturer. In this experiment, radiation and conduction heat rod-grids are almost independent of Reynolds number 共Fig. 5兲,
losses were on the order of 2 percent and 20 percent, respectively. and well represented with the power law of Roach 关22兴.
An uncertainty analysis was conducted using the methods of Mof-
冉冊
5
fat 关20兴. The uncertainty in Fr is 3.92 percent and 3.67 percent for x ⫺7
the minimum and maximum freestream Reynolds number, Tu⫽C 1 . (2)
d
respectively.
The average of 180 data points was obtained for each thermo- The values of C 1 are 1.12, 1.24, and 1.20 for the rod-grids of
couple. The spanwise temperature distribution of the stainless 2.86 cm, 1.59 cm, and 0.95 cm, respectively. The turbulence in-
steel strips was found to be fairly constant and within the uncer- tensity decreases approximately from 12 percent to 4 percent as
tainty of the temperature measurements. The spanwise variation the grid-to-model distance increases from 25d to 125d. The
of the temperatures from the six thermocouples 共T1-T6兲 is about streamwise distributions of the ratio of integral length scale to the
two percent for all heated strips 共Fig. 4兲, where strip no. 1 is on diameter of the grid-rods (␭ x /d) for the sample ReD ⫽108,350 are
the stagnation line. Since the thermocouples are located in the given in Fig. 6. Unlike the turbulence intensity, ␭ x /d is dependent
middle portion of the stainless steel foils 共see Fig. 2兲, the end on Reynolds number, and the best-fit lines with the power law are
conduction heat losses from the heated strips do no significantly also presented in Fig. 6. The form of the power law used has the
influence the spanwise temperature distribution at the measured same growth rate of ␭ x proposed by Roach 关22兴.
locations on the strips. Therefore, an average value of the six
冉冊
1
thermocouple readings is taken as the temperature Tw( ␪ ) of a x 2
particular strip. ␭ x ⫽C 2 (3)
d
A number of test runs were performed with and without the
grids to estimate the conduction heat losses. During these test The values of C 2 for the different rod-grids and Reynolds num-
runs, the spanwise temperature distributions of the heated strips bers are presented in Table 1. The ␭ x /d increases by about 45
and the temperatures outside the strips and on the inner surfaces percent as x/d increases from 25 to 125, and is in reasonable
were recorded. A three-dimensional finite difference scheme was agreement with the results of Roach 关22兴. It should be noted that
Fig. 4 Spanwise temperature distributions of heated stainless Fig. 5 Streamwise turbulence intensity downstream of the
steel strips „䊊, strip no. 1; 䊐, 2; 䉭, 3; 〫, 4; *, 5; o, 6; ¿, 7; Ã, 8; grids „〫, —, 2.86 cm rod-grid; 䉭, - - -, 1.59 cm rod-grid; 䊊,
À, 9; 〫, 10… - — -, 0.95 cm rod-grid…
Fig. 6 Streamwise integral length scale downstream of the
grids for ReD Ä108,350 „〫, —, 2.86 cm rod-grid; 䉭, - - -, 1.59 cm Fig. 7 RMS fluctuating velocity components of grids in per-
rod-grid; 䊊, - — -, 0.95 cm rod-grid… pendicular orientation for ReD Ä108,350 „2.86 cm rod-grid; 〫,
u; 䉭, v; 䊊, w; 1.59 cm rod-grid; 〫, u; 䉭, v; 䊊, w; 0.95 cm
rod-grid: 〫, u; 䉭, v; 䊊 w…
both rod-grids in perpendicular and parallel orientations give the

same Tu and ␭ x /d since they are calculated based on the stream-
wise velocity fluctuations. Heat transfer at the stagnation line 共␪⫽0兲 with the rod-grids in
The rms of fluctuating velocity components, u, v , and w, down- perpendicular and parallel orientation are compared to investigate
stream of the rod-grids in perpendicular orientation for ReD the effect of different freestream vortical structures 共Fig. 9, 10,
⫽108,350 are presented in Fig. 7. At locations close to the grid, v and 11兲. The best-fit lines to the data are also presented in the
and u are approximately equal ( v /u⬇1), and higher than the w figures to highlight the trends. The detailed quantitative data of
component 共w/u is in the range 0.85 to 0.91兲. This is due to the these figures are given in Table 2. In all cases, heat transfer en-
fact that most turbulence eddies are aligned with the turbulence hancement by the grids in perpendicular orientation is larger than
generating rods and velocity fluctuation in the direction parallel to that by the grids in parallel orientation. It can be speculated that
the rods are not as intense as in other directions. As x/d increases, the larger heat transfer with the rod-grids in perpendicular orien-
the difference between the three components decreases. At x/d tation is due to the greater vortex stretching of the primary vorti-
⫽125, v /u, and w/u range from 0.99 to 0.9 and from 0.88 to 0.99, ces, which are parallel and align with the grid-rods. The increase
respectively. A plausible conclusion from Fig. 7 is that the distinct in heat transfer augmentation with the grids in perpendicular ori-
structures of turbulence due to the parallel array of rods become entation over grids in parallel orientation decreases with the rod
more homogeneous with distance from the grid. size. At x/d⫽25, the average increase in Frossling number from
Heat transfer curves for the three grids of 2.86 cm, 1.59 cm, and the perpendicular orientation over the parallel orientation is about
0.95 cm rods, where R and L denote rods in perpendicular and 7.39 percent, 4.63 percent, and 2.46 percent for the 2.86 cm, 1.59
parallel orientations, are presented in Fig. 8. The results for five cm, and 0.95 cm grid-rods, respectively. As evident in Fig. 9, 10,
grid locations for the sample ReD 共⫽108,350兲 are shown in the and 11, the difference between the two heat transfer curves for the
figures. The semi-theoretical solution of Frossling 关23兴 for a uni- two different grid orientations decreases with increasing grid-to-
form freestream is also provided for reference. Heat transfer in the model distance, and this is more pronounced for the bigger grid-
stagnation region decreases with increasing grid-to-model dis- rod. As x/d increases from 25 to 125, the average difference in
tance for a given Reynolds number. This is expected since the Frossling number between the perpendicular and parallel orienta-
turbulence intensity of the freestream decreases with downstream tions decreases from 7.39 percent to 1.06 percent for 2.86 cm
distance from the grid. Heat transfer augmentation at the stagna- rods, 4.63 percent to 0.81 percent for 1.59 cm rods and 2.46
tion line over the uniform freestream ranges from 37 percent to 75 percent to 0.89 percent for 0.95 cm rods.
percent for a grid-to-model distance (x/d) of 25. The augmenta- Since the literature on vortical structures behind grids of paral-
tion is reduced to the range 15 percent to 34 percent as x/d in- lel rods is very limited, the vortex dynamics of wakes behind
creases to 125. For a given x/d, the lowest heat transfer augmen- circular cylinders can be used to speculate on the freestream tur-
tation is obtained with the 2.86 cm rod-grid in parallel orientation bulent structure downstream of the grids used in this study. The
while the highest is obtained with the 0.95 rod-grid in perpendicu- Reynolds number based on the rod size, Red , in this study ranges
lar orientation. Heat transfer increases with increasing Reynolds from 3175 to 20,000. Hence, the wake is in the flow regime of
number for a given rod-grid and grid-to-model distance. For ex- shear layer transition 关24兴. Besides the primary vortices, which are
ample, the Frossling number at the stagnation line increases by parallel and aligned with the circular cylinder, intense shedding of
about 10 percent at x/d⫽125 as ReD increases from 67,750 to near wake (x/d⬍1) streamwise vortices are present in this re-
142,250. At x/d⫽25, the increase in Frossling number is about 15 gime. Furthermore, three-dimensional vortical structures with
percent for the same increase in ReD . For the same x/d and sizes ranging from the shear layer thickness to the Karman vorti-
ReD , a smaller rod-grid gives higher heat transfer. For example, ces are expected to develop in this flow regime 关25,26兴. The
at the stagnation line, Frossling number with the 0.95 cm rod-grid streamwise vortices, which evolve from the primary vortices in
is about 5 percent higher than that with the 2.86 cm the near wake region, deform into three-dimensional structures as
rod-grid. they travel further downstream. The primary vortices become dis-
located and cannot be precisely traced beyond x/d⬎50 in this
flow regime 关24兴. It is plausible to assume a similar vortex struc-
ture downstream of the grids in this study. At downstream dis-
Table 1 The constant C 2 of Eq. „3…
tances close to the grid (x/d⭐75), the two grid orientations gen-
erate freestream turbulence with primary vortices apparently in
different spanwise planes. The primary vortices from the perpen-
dicular grid orientation are primarily aligned normal to the stag-
nation line, and are more susceptible to stretching as they ap-
proach the stagnation region. This would explain the higher heat
transfer with the grids in perpendicular orientation. As the dis-
tance from the grids increases, the three-dimensionality of the
Fig. 8 Distribution of Frossling number in the stagnation region for ReD Ä108,350 grids „〫,
2.86 cm rod-grid; 䉭, 1.59 cm rod-grid; 䊊, 0.95 cm rod-grid; —— Frossling solution……
vortices increases, resulting in a smaller difference in heat transfer the sizes of primary vortices are relatively different for the three
with the perpendicular and parallel grid orientations. different rod-grids. It can be speculated that the greater difference
While the size of the primary vortices depends on the size of in heat transfer rates between perpendicular and parallel orienta-
the turbulence generating rod, the size of streamwise vortices are tions for the grid with larger rods is due to the more apparent
more dependent on the flow regime rather than the size of primary distinction between primary and secondary vortices with the
vortices 关26兴. Within a flow regime, there are relatively small larger rods.
changes in the character of vortex shedding over a large range of The difference in heat transfer augmentation with the two dif-
Red 关26兴. For the Red range of the present study, it seems likely ferent grid orientations over the entire stagnation region 共0 deg
that the size and strength of streamwise vortices, which promote ⬍␪⬍90 deg兲 are presented in Fig. 12 for all rod-grids at ReD
the three-dimensionality of the vortical structure downstream of ⫽108,350. For the 2.86 cm rod-grid, the difference in heat trans-
the grids, are somewhat comparable for all rod-grids. However, fer augmentation between the two grid-orientations is highest at
Fig. 9 Stagnation line Frossling number for 2.86 cm rod-grid
„〫, ReDÄ67,750 „R…; 〫, 67,750 „L…; 䉭, 108,350 „R…; 䉭, 108,350 Fig. 11 Stagnation line Frossling number for 0.95 cm rod-grid
„L…; 䊊, 142,250 „R…; o, 142,250 „L…… „〫, ReDÄ67,750 „R…; 〫, 67,750 „L…; 䉭, 108,350 „R…; 䉭, 108,350
„L…; 䊊, 142,250 „R…; o, 142,250 „L……
stagnation region. This distinct phenomena could be related to the

large scale secondary vortical structures, called double rollers in
the literature 关24,27兴, found in the far wake (x/d⬎100兲 of a
circular cylinder; however, a detailed explanation for this cannot
be offered at this point. For the 1.59 cm and 0.95 cm rod-grids,
the difference in heat transfer for the two grid orientations remains
relatively constant over the whole stagnation region for all
grid-to-model distances and ReD . The observed characteristics of
the off-stagnation region heat transfer indicate that the nature of
heat transfer enhancement in the stagnation region by different
Fig. 10 Stagnation line Frossling number for 1.59 cm rod-grid
vortical structures are dependent on the size of primary vortices,
„〫, ReDÄ67,750 „R…; 〫, 67,750 „L…; 䉭, 108,350 „R…; 䉭, 108,350 since the grid size is a determining factor of the size of primary
„L…; 䊊, 142,250 „R…; o, 142,250 „L…… vortices.
The data from the three grids are compared with the correlation
of VanFossen et al. 关12兴 for freestream turbulence from square bar
mesh grids in Fig. 13.
the stagnation line 共␪⫽0 deg兲, and decreases with streamwise dis-
tance from the stagnation line. The percentage difference in heat
transfer decreases by about three as ␪ increases from 0 deg to 80
deg for grid-to-model distances of 25d, 50d, 75d, and 100d. The
Fr⫽0.008 冑 TuReD 0.8 冉冊
␭x
D
⫺0.574
⫹0.939 (4)
heat transfer characteristics at grid-to-model distance of 125d are The discrepancy between the present experimental data and Eq.
anomalous, where the heat transfer with the parallel orientation is 共4兲 is significant for the 2.86 cm rod-grid, but the agreement im-
higher than with the perpendicular orientation in the off- proves with decreasing size of rod. For the 0.95 cm rod-grid, 87
Table 2 Stagnation line Frossling numbers
Fig. 12 Difference in heat transfer with grid in perpendicular over parallel orientation for
ReD Ä108,350 „〫, 2.86 cm rod-grid; 䉭, 1.59 cm rod-grid, 䊊, 0.95 cm rod-grid…
percent of data falls within ⫾4 percent of Eq. 共4兲. It is clear from form similar to that of VanFossen et al. 关12兴. The best fit obtained
Fig. 13 that the correlations developed for isotropic turbulence by the regression analysis for all data sets, i.e., for both orienta-
generated by square mesh grids should be corrected to predict the tions of rod-grids, is plotted in Fig. 14. The data fall within ⫾6
heat transfer due to the turbulence with distinct coherent vortical percent of the correlation given in Eq. 共5兲.
structures. The errors are more serious for turbulence with higher
integral length scales. Therefore, care should be taken in estimat-
ing heat transfer to gas turbine blades where the wake turbulence
from the preceding blades could have integral length scales com-
Fr⫽0.0018 冑 Tu 1.3463ReD 1.2053 冉冊
␭x
D
⫺0.2463
⫹0.939 (5)
parable to the leading edge diameter of the blades.

An attempt was made to highlight the distinct nature of stagna- Since the rod-grids in perpendicular orientation give higher heat
tion region heat transfer by freestream turbulence with different transfer in the stagnation region than those in parallel orientation,
vortical structures by developing a correlation using a functional correlations for each grid orientation were also developed 共Fig. 15
Fig. 13 Stagnation line Fr versus correlation parameter proposed by VanFossen et al. †12‡
„Orientation of rod-grids: 䊊, perpendicular; 䉭, parallel, Correlation lines: ——, van Fossen
et al.; - - -, ¿4%; — - —; À4%…
Fig. 14 Stagnation line Fr versus correlation parameter for Fig. 15 Stagnation line Fr versus correlation parameter for
both grid orientations „〫, Data, Correlation lines: ——, Eq. „5…; rod-grids in perpendicular orientation „〫, Data, Correlation
- - -, ¿6%; — - —; À6%… lines: ——, Eq. „6…; - - -, ¿4%; — - —; À4%…
5 The difference in heat transfer augmentation by grids in
perpendicular and parallel orientations decreases with the
size of grid-rods. This could be due to a more apparent dis-
tinction between primary and secondary vortices with the
bigger rods. At a grid-to-model distance of 25d, the average
difference in the Frossling number is 7.39 percent, 4.63 per-
cent, and 2.46 percent for the 2.86 cm, 1.59 cm, and 0.95 cm
rods, respectively.
6 The difference in heat transfer augmentation with grids in
perpendicular and parallel orientations decrease with increas-
ing grid-to-model distance. This is likely due to the greater
three-dimensionality of the freestream vortical structures as
Fig. 16 Stagnation line Fr versus correlation parameter for the distance from the grid increases.
rod-grids in parallel orientation „〫, Data, Correlation lines:
——, Eq. „7…; - - -, ¿4%; — - —; À4%…
7 For the 2.86 cm rod-grid, the difference in heat transfer
between perpendicular and parallel grid-orientations is high-
est at the stagnation line, and decreases with streamwise dis-
tance. However, this difference remains fairly constant over
and 16兲. The best-fit equations using the regression analysis are the whole stagnation region for the 1.59 cm and 0.95 cm
given in Eq. 共6兲 for the rod-grids in perpendicular orientation and rod-grids. This indicates that the heat transfer enhancement
in Eq. 共7兲 for those in parallel orientation. by different vortical structures are dependent on the size of
冑冉冊 ⫺0.1178 primary vortices.

␭x
Fr⫽0.0014 Tu 1.4858ReD 1.3058 ⫹0.939 (6) 8 The nature of heat transfer enhancement in the stagnation
D region by freestream turbulence with distinct coherent vorti-
冑冉冊 ⫺0.3123 cal structures is quite distinct from that by turbulence gener-

␭x ated using square-mesh grids. The heat transfer data of this
Fr⫽0.0013 Tu 1.2620ReD 1.2222 ⫹0.939 (7)
D study are poorly predicted using existing correlation models
The data for each grid orientation are within ⫾4 percent of the for turbulence generated from square-mesh grids. The dis-
correlation line. The different correlations that can be obtained to crepancies increase for the larger grid-rods. Any correlation
best fit a given data set highlights the difficulty of obtaining a model should take into account the distinct nature of the
correlation for gas turbine blade heat transfer. It is clear that coherent vortical structures of the turbulence to improve its
freestream turbulence with different orientations of coherent vor- predictive capability.
tical structures have different influences over the heat transfer in
the stagnation region.
Acknowledgments
Conclusions The support of the Natural Sciences and Engineering Research
An experimental study has been performed to investigate the Council 共NSERC兲 of Canada is gratefully acknowledged.
effect of freestream turbulence with different vortical structures
on stagnation region heat transfer. Freestream turbulence with dif-
ferent orientations of primary vortices were generated using grids
of 2.86 cm, 1.59 cm, and 0.95 cm diameter parallel rods in per- Nomenclature
pendicular and parallel orientations. Heat transfer measurements A ⫽ area of the heated portion of the leading edge (m2 )
were made in the stagnation region using a heat transfer model
d ⫽ diameter of a grid-rod 共m兲
with the grids 25d to 125d upstream of the model at three differ-
D ⫽ diameter of cylindrical leading edge 共m兲
ent ReD of 67,750, 108,350, and 142,250. The main conclusions
Fr ⫽ Frossling number 共Nu/ 冑ReD )
can be summarized as:
I ⫽ current 共ampere兲
1 Heat transfer in the stagnation region decreases with increas- k ⫽ thermal conductivity 共W/m-K兲
ing grid-to-model distance for a given Reynolds number. The Nu ⫽ Nusselt number
increase in the Frossling number at the stagnation line over a Q cond ⫽ conduction heat loss 共W兲
uniform freestream ranges from 37 percent to 75 percent at Q cov ⫽ convection heat transfer 共W兲
x/d⫽25, and 15 percent to 34 percent at x/d⫽125 for the Q in ⫽ heat input 共W兲
three different grids. Q rad ⫽ radiation heat loss 共W兲
2 Heat transfer increases with increasing ReD for a given rod- Red ⫽ Reynolds number based on d
grid and grid-to-model distance. The stagnation line ReD ⫽ Reynolds number based on D
Frossling number increases by 10 percent and 15 percent at Tu ⫽ streamwise turbulence intensity
x/d of 125 and 25 as ReD increases from 67,750 to 142,250. Tw( ␪ ) ⫽ temperature of the leading edge at angle ␪ 共K兲
3 For a given grid-to-model distance, x/d, and ReD , the grid T⬁ ⫽ temperature of freestream 共K兲
with smaller diameter rods gives higher heat transfer. At the u ⫽ rms of fluctuating velocity component in streamwise
stagnation line, the Frossling number with the 0.95 cm rod- direction 共m/s兲
grid is about 5 percent higher than that with the 2.86 cm U ⫽ mean freestream velocity 共m/s兲
rod-grid. This can be attributed to the smaller length scales v ⫽ rms of fluctuating velocity component in spanwise Y
associated with the smaller rod-grid. direction 共m/s兲
4 The heat transfer enhancement by the grids in perpendicular V0 ⫽ voltage 共V兲
orientation is larger than that by the grids in parallel orienta- w ⫽ rms of fluctuating velocity component in spanwise Z
tion. This can be speculated to be caused by greater vortex direction 共m/s兲
stretching, since the primary vortices generated by the per- x ⫽ distance downstream of the grid 共m兲
pendicular orientation are primarily aligned normal to the ␪ ⫽ angle measured from the stagnation point 共degree兲
stagnation line, and more susceptible to stretching as they ␯ ⫽ kinematic viscosity of freestream (m2/s)
approach the stagnation region. ␭x ⫽ streamwise integral length scale of turbulence 共m兲
Appendix References
关1兴 Maciejewski, P. K., and Moffat, R. J., 1992, ‘‘Heat Transfer with Very High
Analysis of Experimental Uncertainties. Following the un- Freestream Turbulence: Part I — Experimental Data; Part II — Analysis of
certainty analysis methods of Moffat 关20兴 the uncertainty in Fr Results,’’ ASME J. Heat Transfer 114, pp. 827–839.
can be expressed as: 关2兴 Larsson, J., 1997, ‘‘Turbine Blade Heat Transfer Calculations Using Two-
equation Turbulence Models,’’ Proc. Instn. Mech. Engrs. 221, Part A, pp.
共 UnFr兲 2 ⫽ 共 UnNu兲 2 ⫹0.25共 UnReD 兲 2 253–262.
关3兴 Guo, S. M., Jones, T. V., Lock, G. M., and Dancer, S. N., 1998, ‘‘Computa-
where UnFr⫽Uncertainty in Fr 共 % 兲 tional Prediction of Heat Transfer to Gas Turbine Nozzle Guide Vanes with
Roughened Surfaces,’’ ASME J. Turbomach. 120, pp. 343–350.
UnNu⫽Uncertainty in Nu 共%兲关4兴 Smith, M. C., and Kuethe, A. M., 1966, ‘‘Effects of Turbulence on Laminar
Skin Friction and Heat Transfer, ’’ Phys. Fluids 9, No. 12, pp. 2337–2344.
UnReD ⫽Uncertainty in ReD 共 % 兲关5兴 Kestin, J., and Wood, R. T., 1971, ‘‘The Influence of Turbulence on Mass
Transfer from Cylinders,’’ ASME J. Heat Transfer 93C, pp. 321–327.
The uncertainty in Fr and the influencing variables with 95 关6兴 Lowery, G. W., and Vacon, R. I., 1975, ‘‘Effect of Turbulence on Heat Trans-
percent confidence levels are given in Table 3 for both maximum fer from Heated Cylinders,’’ Int. J. Heat Mass Transf. 18, No. 11, pp. 1229–
1242.
and minimum freestream velocities. Uncertainties in temperature 关7兴 O’Brien, J. E., and VanFossen, G. J., 1985, ‘‘The Influence of Jet-Grid Tur-
were considered in estimating conduction heat loss, radiation heat bulence on Heat Transfer from the Stagnation Region of a Cylinder in Cross-
losses and convection heat transfer. Uncertainties associated with flow,’’ ASME Paper 85-HT-58.
leading edge diameter (D), thermal conductivity 共k) and area (A) 关8兴 Mehendale, A. B., Han, J. C., and Ou, S., 1991, ‘‘Influence of High Main-
are assumed to be negligible since these values are based on the stream Turbulence on Leading Edge Heat Transfer,’’ ASME J. Heat Transfer
113, pp. 843–850.
data specified by manufacturers. 关9兴 Han, J. C., Zhang, L., and Ou, S., 1993, ‘‘Influence of Unsteady Wake on Heat
The techniques proposed by Yavuzkurt 关21兴 were used to esti- Transfer Coefficients from a Gas Turbine Blade,’’ ASME J. Heat Transfer
mate the uncertainties in Tu and ␭ x . The uncertainties are given 115, pp. 904–911.
in Table 4. 关10兴 Yeh, F. C., Hippensteele, S. A., VanFossen, G. J., Poinsatte, P. E., and Ameri,
A., 1993, ‘‘High Reynolds Number and Turbulence Effects on Aerodynamics
and Heat Transfer in a Turbine Cascade,’’ AIAA-93-2252.
关11兴 Zhang, L., and Han, J. C., 1994, ‘‘Influence of Mainstream Turbulence on
Table 3 Experimental uncertainty in Fr „%… Heat Transfer Coefficients from a Gas Turbine Blade,’’ ASME J. Heat Trans-
fer 116, pp. 896–903.
关12兴 VanFossen, G. J., Simoneau, R. J., and Ching, C. Y., 1995, ‘‘Influence of
Turbulence Parameters, Reynolds Number and Body Shape on Stagnation Re-
gion Heat Transfer,’’ ASME J. Heat Transfer 117, pp. 597–603.
关13兴 Ahmaed, G. R., and Yovanovich, M. M., 1997, ‘‘Experimental Study of
Forced Convection from Isothermal Circular and Square Cylinders and Tor-
oids,’’ ASME J. Heat Transfer 119, pp. 70–79.
关14兴 Du, H., Ekkad, S., and Han, J. C., 1997, ‘‘Effect of Unsteady Wake with
Trailing Edge Coolant Ejection on Detailed Heat Transfer Coefficient Distri-
butions for a Gas Turbine Blade,’’ ASME J. Heat Transfer 119, 242–248.
关15兴 Johnston, J. P., 1974, ‘‘The Effects of Rotation on Boundary Layers in Tur-
bomachine Rotors,’’ NASA SP 304.
关16兴 Lakshminarayana B., 1996, Fluid Dynamics and Heat Transfer of Turboma-
chinery, John Wiley & Sons, Inc., New York.
关17兴 Sutera, S. P., Maeder, P. F., and Kestin, J., 1963, ‘‘On the Sensitivity of Heat
Transfer in the Stagnation-point Boundary Layer to Free-stream Vorticity,’’ J.
Fluid Mech. 16, part 3, pp. 497–520.
关18兴 Sutera, S. P., 1965, ‘‘Vorticity Amplification in Stagnation-point Flow and its
Effect on Heat Transfer,’’ J. Fluid Mech. 21, part 3, pp. 513–534.
关19兴 Morkovin, M. V., 1979 ‘‘On the Question of Instabilities Upstream of Cylin-
drical Bodies,’’ MASA CR-3231.
关20兴 Moffat, R. J., 1988, ‘‘Describing the Uncertainties in Experimental Results,’’
Exp. Therm. Fluid Sci. 1, pp. 3–17.
关21兴 Yavuzkurt, S., 1984, ‘‘A Guide to Uncertainty Analysis of Hot-Wire Data,’’
ASME J. Fluids Eng. 106, pp. 181–186.
Table 4 Experimental uncertainty in Tu and ␭ x „%… 关22兴 Roach, P. E., 1987, ‘‘The generation of Nearly Isotropic Turbulence by Means
of Grids,’’ Heat and Fluid Flow, 8, No. 2, pp. 82–92.
关23兴 Frossling, N., 1958, ‘‘Evaporating Heat Transfer and Velocity Distribution in
Two-Dimensional and Rotationally Symmetric Laminar Boundary Layer
Flow,’’ NACA TM-1432.
关24兴 Zdravkovich, M. M., 1997, Flow Around Circular Cylinders, Vol 1: Funda-
mentals, Oxford Science Publications, Oxford University Press; Oxford,
United Kingdom.
关25兴 Wei, T., and Smith, C. R., 1986, ‘‘Secondary Vortices in the Wake of Circular
Cylinders,’’ J. Fluid Mech. 169, pp. 513.
关26兴 Williamson, C. H. K., 1996, ‘‘Vortex Dynamics in the Cylinder Wake,’’
Annu. Rev. Fluid Mech. 28, pp. 477–539.
关27兴 Corke, T., Krull, J. D., and Ghassemi, M., 1992, ‘‘Three-dimensional Mode
Resonance in Far Wake,’’ J. Fluid Mech. 239, pp. 99–132.
Effects of Rib Arrangements on
Heat Transfer and Flow Behavior
in a Rectangular Rib-Roughened
Passage: Application to Cooling
Robert Kiml
Assistant Professor
E-mail: robert@mmlab.mech.tuat.ac.jp
of Gas Turbine Blade Trailing
Sadanari Mochizuki Edge
Professor, Mem. ASME
Experimentation was conducted to examine the heat transfer and pressure drop charac-
Akira Murata teristics in a rib-roughened rectangular passage with aspect ratio 2:1 for four rib con-
Professor figurations: 90 deg, 75 deg, 60 deg and 45 deg oblique ribs. The ribs were attached to two
opposing long side walls instead of short side walls. In this study the oblique ribs were
Tokyo University of Agriculture and Technology, intended to function as secondary flow inducers as well as turbulators to improve the heat
Department of Mechanical Systems Engineering, transfer of the bottom wall (one of the short side walls). The results revealed that, in order
Nakacho 2-24-16, Koganei-shi, Tokyo to enhance the heat transfer of the bottom wall, the oblique ribs should be arranged so
184-8588, Japan that the secondary flow along the ribs hits the top wall instead of the bottom wall. Flow
visualization test was performed to understand the heat transfer mechanisms. It was
confirmed that the heat transfer enhancement at the bottom wall was attributed to the
rib-induced secondary flow where the flow along the ribs hit the top wall, turned back and
carried cold air from the passage core region towards the bottom wall. The highest
average heat transfer was achieved for the 60 deg rib pattern due to the strong rotational
momentum of the secondary flow and higher heat transfer enhancement on the rib-
roughened walls in comparison to the other three rib patterns.
关DOI: 10.1115/1.1378019兴
Keywords: Cooling, Flow, Heat Transfer, Thermocouples, Visualization
Introduction the flow and heat transfer in a serpentine passage have been re-
ported by Cheng et al. 关19兴, Cheng and Shi 关20兴, Metzger et al.
With an increase in demands for the higher thermal efficiency,
关21兴, and Kiml et al. 关22兴.
components of a modern gas turbine engine, especially the first
However, with the advancement of gas turbine technology, it
moving stage blades, are exposed to the higher and higher turbine
became necessary to focus the attention on the regions exposed to
inlet gas temperature. Cooling techniques such as forced convec-
the most severe thermal conditions, leading and trailing edges, as
tion cooling, impingement cooling, film cooling and their combi-
shown in Fig. 1. In particular, the trailing edge, due to its signifi-
nations have been employed in order to maintain acceptable air-
cant shortage of cooling space for internal cooling, is becoming
foil temperature and secure blade life.
one of the most important problems to realize more effective gas
Internal forced convection cooling of a gas turbine rotor blade
turbine engines. In order to cool the trailing edge of a blade,
has been studied by many scientists. Johnson et al., 关1兴, Taslim
cooling-air ejection has been commonly used. However, ejection
关2兴, Rau et al. 关3兴, Mochizuki et al. 关4兴, Han 关5兴, Kiml 关6兴, and
of air degrades both the aerodynamic performance of the blade
Wagner 关7兴 studied the heat transfer and friction characteristics in
and thermal efficiency of the engine. Therefore, for future ad-
rib-roughened passages with different rib arrangements. They fo-
vanced engines, especially for the power plant use, cooling air
cused on the effects of the Reynolds number and rib geometry on
should be recovered from the blade after it passed through the
the heat transfer and pressure drop in the fully developed region
internal cooling passages. To accomplish this, installation of ob-
of a uniformly heated square and rectangular channels. All these
studies showed that angled ribs provide better heat transfer en- lique ribs in the internal passage near the trailing edge as
hancement than transverse ribs. Further studies Dutta et al. 关8兴, secondary-flow inducers as well as turbulators can be one solu-
Dutta et al. 关9兴, Dutta and Han 关10兴, El-Husayni and Han 关11兴, tion. Recently, study by Dutta et al. 关19兴, Zhang et al. 关23兴, Kiml
and Mochizuki et al. 关12兴 examined the effects of Coriolis forces, et al. 关24,25兴 and Hirota et al. 关26兴 showed the heat transfer data
centrifugal force and buoyancy force on the heat transfer charac- for the smooth and ribbed rectangular channels. However, the
teristics with different rib configurations. Studies by Taslim and authors investigated channels with different geometrical proper-
Kercher 关13兴, Hu and Shen 关14兴, Han et al. 关15,16兴, Ekkad and ties or they did not focus on the examination of the secondary
Han 关17兴, and Johnson et al. 关18兴 presented the heat transfer re- flow induced by the oblique rib patterns.
sults for the parallel, broken and V-shaped ribs mounted on two The present study focuses on the detailed examination of the
opposite walls. The effects of the 180 deg or 90 deg sharp turn on flow behavior and its effects on the local heat transfer distribution
in a rectangular passage with aspect ratio 2:1 where the ribs are
attached on the two opposing long side walls. The emphasis is
HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 27, placed on the investigation of rib-induced secondary flow and its
1999; revision received January 8, 2001. Associate Editor: J.-C. Han. effects on the heat transfer and pressure drop characteristics pro-
Fig. 3 Thermocouple layout and rib patterns tested in present
study
mixing device were insulated by an insulation material 共SAN

PERKA兲 with a heat conductivity of 0.0417 W/共mK兲. The insula-
Fig. 1 Schematic of internal and external cooling structures of tion thickness was 20 mm.
a gas turbine rotor blade Air from the turbo blower flows into the test section after pass-
ing through a straight entrance region, with length of 660 mm,
which has a rectangular flow cross section identical to the flow
duced by 4 different rib patterns: 45 deg, 60 deg, 75 deg, and 90 passage inside the test section. Air temperature at the inlet to the
deg ribs. The heat transfer and flow visualization experiments test section was measured by means of two K-type thermocouples
were performed only under stationary conditions. 共0.3 mm in diameter兲. The outlet bulk mean temperature was cal-
culated from temperature obtained from 12 thermocouples 共50 ␮m
in diameter兲 behind a flow mixing device which was installed at
Experimental Apparatus and Procedure the exit of the test section.
Heat Transfer Experiment. The heat transfer experiment The wall surface temperature was measured by means of 152
was performed in a rib-roughened passage with the aspect ratio K-type thermocouples 共50 ␮m in diameter兲 mounted on back side
2:1 共15 ⫻ 30 mm2, equivalent diameter d e ⫽20 mm兲 with the ribs of the conductive plastic heater, as can be also observed in Fig. 2.
attached on the two opposing long-side walls. From this point on, The thermocouples on all walls were located in three lines, as
the walls of the cooling channel will be called the ‘‘top’’, ‘‘bot- shown in Fig. 3. The ribs with a square cross section 共rib height
tom,’’ ‘‘left’’ and ‘‘right’’ walls, as presented in Fig. 2. The heat e⫽3 mm兲, machined from the bakelite, were attached to two op-
transfer test section was constructed from 5 mm thick Bakelite posing long side walls 共right and left walls兲 instead of short side
plates. This material was chosen for its low thermal conductivity, walls 共top and bottom walls兲. This material was chosen for its low
high mechanical strength and easy machining. The inside surfaces weight and high mechanical strength, which enables to perform
of the test section were covered with thin electric conductive plas- the heat transfer experiments under rotating conditions. Experi-
tic films, consisting of 180 ␮m thick polyethylene terephthalate ments were performed for 45 deg, 60 deg, 75 deg, and 90 deg rib
共PET兲 layer and a 20 ␮m thick conductive layer 共see Fig. 2兲. patterns which were intended to function as secondary flow induc-
Uniform wall heat flux conditions were achieved by passing an ers to improve the heat transfer of the bottom wall 共one of the
electric current through the conductive layer of the film. Its spe- short side walls兲. The Reynolds number was set to Re⫽15,000.
cific characteristics of highly uniform electric resistance fulfill the Flow Visualization Experiment. The flow visualization us-
requirement for uniform heat generation. The test section and ing fine particles as tracers was carried out in a 6 m long water
channel with a test section located 5 m downstream of the settling
Fig. 2 Schematic of rib-roughened passage in which heat

transfer experiment was performed Fig. 4 Schematic of flow visualization test section
chamber. The test section, made of transparent Plexiglas plates,
had a rectangular cross section 共150⫻75 mm2). Water was mixed
with the chemical compound NaSO4, with increased the specific
weight of the water so that the particles 共about 70 ␮m in diameter兲
had the same specific weight with that of the water. These par-
ticles were illuminated between the fifth and sixth ribs by an ar-
gon laser light-sheet 共3 mm thick兲 and photographed by a still
digital camera and video camera, as shown in Fig. 4. The camera
exposure time was set to 0.5 second to allow the particles to create
path lines. Top view patterns were photographed at 5 locations
x⫽0.66e 共near-bottom wall region兲, 2.66e, 5e 共center兲, 7.33e and
9.33e 共near-top wall region兲 from the bottom wall surface. Side
view patterns were obtained at y⫽0.33e 共near right wall region兲;
0.66e, 1e, 1.66e, and 2.5e 共central vertical plane兲 from the right
wall surface. The ribs, made of transparent Plexiglas with a square Fig. 6 Effects of rib configuration on friction coefficient, C f :
cross section 共15 ⫻ 15 mm2), were mounted on the left and right „a… flow pattern at x Ä0.66e „near the bottom wall…; „b… flow pat-
walls of the channel. Geometric proportions of the water channel, tern at x Ä5 e „channel center…; „c… flow pattern at x Ä9.33e
ribs, rib angle and rib pitch correspond to those used in heat „near the top wall…; „d… flow pattern at y Ä0.33e „close to the
right wall…; and „e… flow pattern at y Ä2.5e from right wall „chan-
transfer experiments. The Reynolds number was varied from Re
nel central vertical plane….
⫽10,000 to 20,000.
Data Reduction
The Reynolds number, wall heat flux, local heat transfer coef- considerably lower value on the top and bottom walls. In com-
ficient, friction coefficient and Nusselt numbers are calculated, parison to the 90 deg rib patterns, the oblique 关60 deg 共䊉兲, 75 deg
respectively, as follows: 共⽧兲 and 45 共䊏兲兴 rib patterns show a considerably lower wall Num
on the top wall and higher on the bottom, right and left walls.
Re⫽u m d e / ␯ (1)
Particularly, the 30 percent higher Num on the bottom wall in
q̇ w ⫽ṁc p 共 T out⫺T in兲 /A h (2) contrast to that on the top wall, where one might have expected
opposite situation due to the ribs’ inclination 共confirm in Fig. 2兲,
␣ ⫽q̇ w / 共 T w ⫺T b 兲 (3) may attract ones attention. It can be also seen from this figure that
the Num on the rib-roughened walls 共right and left walls兲 is the
Nu⫽ ␣ d e /k (4)
highest for 60 deg 共䊉兲 rib pattern, but on the bottom wall it is the
Nu⬁ ⫽0.022 Re 0.8 0.5
Pr (5) highest for the 45 deg 共䊏兲 rib pattern. The Num enhancement
caused by the use of ribs as secondary flow inducers is paid by a
C f ⫽⌬pd e / 共 2L ␳ u 2m 兲 (6) considerably larger pressure drop increase as shown in Fig. 6.
冒兺
n n The effects of all rib patterns on the passage mean Num p and
the friction coefficient in comparison to a Smooth passage 共䉺兲 are
Num ⫽ 兺
i⫺1
共 Nui A i 兲
i⫺1
Ai . (7) summarized in Table 1. The passage mean Num p in comparison to
the wall mean Num is evaluated from the Nu distribution on the
An uncertainty analysis using the ASME Performance Test left, right and bottom walls. The top wall is excluded because it is
Codes 共ANSI/ASME PTC 19.1-1985兲, was carried out for the heat not as important in this case due to the near location another
transfer experiment and it was estimated that the maximum uncer- cooling passage, as is shown in the blade internal structure in
tainties of the Nusselt number were estimated to be less than 9 Fig. 1.
percent. It can be seen from this table that every Num p enhancement
causes a large increase of C f . As mentioned above, the 90 deg rib
Experimental Results and Discussion pattern achieves the lowest Num p and C f of all rib patterns. The
most efficient rib patterns in terms of heat transfer are 60 deg and
Figure 5 shows the effects of rib configurations on the ratio of
45 deg parallel rib patterns. Unfortunately, while the heat transfer
the wall mean Nusselt number and Nusselt number for a fully
is only 2.67 or 2.60 times higher in comparison with the case of
developed flow in a smooth passage, Num /Nu⬁ , on the top, bot-
the Smooth passage, the pressure drop is 16.24 or 16.72 times
tom, right and left walls. It can be clearly observed from this
figure that the Num for 90 deg 共⽧兲 rib pattern achieves its maxi-
mum value on the rib-roughened walls 共left and right walls兲 and
Table 1 Passage mean Nusselt number and friction coeffi-
cient of four rib patterns compared to the smooth „䊊… surfaces
Fig. 5 Effect of rib configuration on average Nusselt number

on all four walls
Fig. 8 Schematic of secondary flow induced by 45 deg ribs
as presented in Figs. 5, 6, and Table 1, one has to focus on the

detailed Nu distribution and flow visualization results.
Flow Visualization. Figure 7 represents the flow visualiza-
tion results between the fifth and sixth ribs for the case of 60 deg
parallel ribs. The top view flow pattern, taken near the bottom
wall 共see Fig. 7共a兲兲, shows the creation of a separation bubble
behind the fifth rib and a flow reattachment between those two
ribs. From the flow reattachment the flow proceeds in the main
flow direction and due to the rib’s inclination is forced to move
along the rib towards the top wall 共see Fig. 7共d兲兲. Here, due to its
encounter with the top wall and a similar flow stream induced by
the ribs mounted on the opposite rib-roughened wall, the flow
sharply turns 180 deg and flows back towards the bottom wall, as
can be confirmed by Figs. 7共c兲 and 7共e兲. This flow behavior
causes a development of the secondary flow in form of two vor-
tices, which carry cold fluid from the central core region towards
the bottom wall, as schematically illustrated in Fig. 8. From the
Figs. 7共e兲 and 8 it can be also noted that the flow near the center
of the bottom wall creates a boundary layer with the flow proceed-
ing in the main flow direction, so that the secondary flow does not
impinge the bottom wall directly at this location, but in the near
rib-roughened wall regions.
Local Heat Transfer. The flow visualization results con-
firmed that the heat transfer enhancement at the bottom wall is
attributed to the rib-induced secondary flow where the flow is
forced to proceed along the ribs, hit the top wall, turn back and
carry cold and large momentum air from the passage core region
towards the bottom wall, as was schematically illustrated in Fig. 8
and as was shown in Figs. 7. This introduction of fresh cold air on
the bottom wall enhances the heat transfer appreciably. The big
difference between the Nu distributions on the top and bottom
walls for the 45 deg rib pattern is presented in detail in Figs. 9 and
10, where the Nu on the bottom wall 关共䊊兲, 共䉭兲, 共䊐兲 in Fig. 10兴 is
significantly higher than on the top wall 关共䊉兲, 共⽧兲, 共䊏兲 in Fig. 9兴.
It may be noted from these figures that the Nu appreciably en-
Fig. 7 Flow visualization results for 45 deg ribs, Re Ä20,000
higher. It may be also noted that 60 deg pattern shows very simi-
lar heat transfer with the case of the 45 deg rib patterns, but a
slightly lower pressure drop.
To explain these heat transfer and pressure drop characteristics, Fig. 9 Local Nu distribution on top wall for 45 deg ribs
Fig. 10 Local Nu distribution on bottom wall for 45 deg ribs Fig. 12 Local Nu distribution on left wall for 60 deg ribs
hances in the regions closer to the rib-roughened walls 关共⽧兲, 共䊏兲, idly deteriorates with the distance from the reattachment region
and 共䉭兲, 共䊐兲兴 in contrast to the central regions 关Top Center 共䊊兲 due to the boundary layer growth and the flow separation in front
and Bottom Center 共䊉兲兴. of the fifth or sixth ribs.
共For better understanding of this complicated Nu distribution Comparison between 45 Deg and 60 Deg Rib Patterns. As
and due to the different location of thermocouples between the was noticed above, the wall mean Num on the bottom wall was the
ribs Nu value plotted in brackets, ‘‘共兲,’’ substituted with the Nu highest for the 45 deg rib pattern and on the rib-roughened walls
value obtained by thermocouples located upstream or downstream it was the highest for the 60 deg rib pattern. It was also shown in
of the previous rib.兲 Table 1 that while the Num was very similar for both rib patterns,
In the case of the top wall, the Nu enhancement in the near the pressure drop was slightly higher for the 45 deg rib pattern.
rib-roughened walls regions takes place especially downstream The higher Num for the 60 deg rib pattern on the rib-roughened
and between the ribs’ connections with the top wall 共confirm in walls in comparison to that of the 45 deg rib pattern is caused by
Fig. 9兲. It is caused by 共1兲 the impingement of the local flow the strong Nu enhancement between the adjacent ribs, as can be
streams proceeding along the oblique ribs and impinging the top seen in Fig. 12. The Nu deterioration in the spanwise direction for
wall in this area 共see in Figs. 7共d兲 and 共8兲 and 共2兲 the high flow the 60 deg rib pattern is much smaller than in the case of the 45
velocity caused by the space contraction between the ribs. From deg rib pattern due to 共1兲 the smaller boundary layer growth be-
this local maximum, the Nu gradually deteriorates due to the tween the 60 deg ribs and 共2兲 the higher Nu enhancements, espe-
boundary layer growth. cially in the near top wall regions 关Left Top 共〫兲兴 compare Figs.
In contrast to the Nu distribution on the top wall, the Nu en- 12 and 11.
hancement in the near rib-roughened wall regions on the bottom The slightly higher mean Num on the bottom wall and higher
wall, 关Right 共䉭兲, Left 共䊐兲兴, is more peaked and can be clearly pressure drop for the 45 deg rib pattern in contrast to that of the 60
observed only in the regions near the ribs 共see Fig. 10兲 due to the deg rib pattern is considered to be caused by a stronger rotational
space contraction between the ribs which increases the flow ve- momentum of the secondary flow which impinges the bottom wall
locity appreciably. As the flow velocity quickly deteriorates from under greater angle and enhances the Nu appreciably, as can be
its peak within the space contraction, the Nu distribution de- confirmed by comparing Figs. 13 and 7.
creases likewise until it is very similar to the Nu in the central
region 关Bottom Center 共䊊兲兴. Comparison With the Other Researcher’s Results. Recently
The detailed Nu distribution on the left wall 共rib-roughened Hirota et al. 关26兴 examined the flow velocities and turbulent
wall兲 for 45 deg rib pattern between the fourth and fifth and fifth stresses in 90 deg rib-roughened passage with the aspect ration 2:1
and sixth ribs is presented in Fig. 11. The Nu in the area just where the channel width-to-height ratio was w/e⫽50. The authors
behind the fourth or fifth ribs is relatively low. This is because the showed secondary flow distribution near the rib-roughened walls,
area is covered by the separation bubble. In the area farther from however due to the different geometrical size 共channel width-to-
the rib, the Nu strongly enhances, particularly along the Left Bot- height ratio used in this study w/e⫽5兲 and only 90 deg rib ar-
tom 共䉭兲 and Left Center 共䉺兲 lines of measurement points due to rangement, the effects of the secondary flow induced by the rib
the existence of the flow reattachment. This Nu enhancement rap- inclination could not be compared.
Fig. 13 Flow pattern taken for 60 deg ribs at y Ä2.5e „central

vertical plane…. Comparison of secondary flow angle between
Fig. 11 Local Nu distribution on left wall for 45 deg ribs 45 deg and 60 deg ribs.
Zhang et al. 关23兴 examined the effects of compound turbulators V̇ ⫽ volumetric flow rate of air or water 关m3/s兴
on friction factors and heat transfer coefficients inside a high as- w ⫽ channel width
pect ratio rectangular channel 共aspect ratio 10:1兲 with two oppo-
z ⫽ passage length 关m兴
site ribbed-grooved walls. However, as in the previous study the
␣ ⫽ local heat transfer coefficient 关 W/m2 K兴 , Eq. 共3兲
authors investigated only 90 ribbed-grooved channel, where they
could not observe the flow behavior presented in this study. ␳ ⫽ air density 关kg/m3兴
To the authors knowledge there are no references describing a ␯ ⫽ air or water kinematic viscosity 关 m2/s兴
flow behavior and heat transfer distribution inside a similar rect-
angular passage with the aspect ratio 2:1, width-to-height ratio
w/e⫽5 and similar rib configurations, except Kiml et al. 关24,25兴
共study for the channel with the aspect ration 5:1兲.
References
关1兴 Johnson, B. V., Wagner, J. H., Steuber, G. D., and Yeh, F. C., 1993, ‘‘Heat
Conclusion Transfer in Rotating Serpentine Passages With Selected Model Orientations
The heat transfer experiments and flow visualization were perfor Smooth or Skewed Trip Walls,’’ ASME Paper 93-GT-305.
formed in a rectangular rib-roughened channel with aspect ratio 关2兴 Taslim, M. E., 1991, ‘‘An Experimental Investigation of Heat Transfer Coef-
ficients in a Spanwise Rotating Channel With Two Opposite Rib-Roughened
2:1. Ribs were attached to two opposing long side walls. Four rib Walls,’’ ASME J. Turbomach., 113, pp. 75–82.
patterns were tested: 90 deg, 75 deg, 60 deg, and 45 deg ribs. The 关3兴 Rau, G., Cakan, M., Moeller, D., and Arts, T., 1996, ‘‘The Effect of Periodic
following conclusions were drawn: Ribs on the Local Aerodynamic and Heat Transfer Performance of a Straight
Cooling Passage,’’ Proceedings, International Gas Turbine and Aeroengine
1 The flow visualization and heat transfer results revealed that Congress and Exhibition, Birmingham, UK, June 10–13, 1996.
the heat transfer enhancement at the bottom wall is attributed to 关4兴 Mochizuki, S., Murata, A., and Fukunaga, M., 1997, ‘‘Effects of Rib Arrange-
the rib-induced secondary flow which conveys the cold air from ment on Pressure Drop and Heat Transfer in a Rib-roughened Channel with a
Sharp 180° Turn,’’ ASME J. Turbomach., 119, pp. 610–616.
the passage core region towards the bottom wall. This introduc- 关5兴 Han, J. C., 1985, ‘‘Heat Transfer Enhancement in Channels With Turbulence
tion of fresh cold air on the surface of the bottom wall enhances Promoters,’’ ASME J. Eng. Gas Turbines Power, 106, pp. 774–781.
the heat transfer appreciably. 关6兴 Kiml, R., 2000, ‘‘Internal Cooling of Gas Turbine Blades,’’ Doctoral thesis,
2 The highest mean Num on the bottom wall and the highest Tokyo University of Agriculture and Technology, pp. 1–190.
friction coefficient were achieved for the 45 deg parallel rib pat- 关7兴 Wagner, J. H., 1992, ‘‘Heat Transfer in Rotating Serpentine Passages With
Trips Normal to the Flow,’’ ASME J. Turbomach., 114, pp. 847–457.
tern due to the stronger rotational momentum of the secondary 关8兴 Dutta, S., Han, J.-C., and Zhang, Y.-M., 1995, ‘‘Influence of Rotation on Heat
flow which impinges the bottom wall under larger angle than in Transfer from a Two-Pass Channel with Periodically Placed Turbulence and
the other cases. Secondary Flow Promoters,’’ Int. J. Rotating Mach., 1, pp. 129–144.
3 The highest mean Num on the rib-roughened walls was 关9兴 Dutta, S., Andrews, J. M., and Han, J.-C., 1996, ‘‘Prediction of Turbulent
achieved for the 60 deg rib pattern. It was considered to be caused Flow and Heat Transfer in Rotating Square and Rectangular Smooth Chan-
nels,’’ ASME Paper 96-GT-234.
by 共a兲 the stronger Nu enhancement between the adjacent ribs 关10兴 Dutta, S., and Han, J.-C., 1996, ‘‘Local Heat Transfer in Rotating Smooth and
than for the other rib patterns and 共b兲 the smaller Nu deterioration Ribbed Two-Pass Square Channels with Three Channel Orientations,’’ ASME
in the spanwise direction. J. Turbomach., 118, pp. 587–584.
关11兴 El-Husayni, H. A., Taslim, M. E., and Kercher, D. M., 1994, ‘‘Experimental
Heat Transfer Investigation of Stationary and Orthogonally Rotating Asym-
Acknowledgments metric and Symmetric Heated Smooth and Turbulated Channels,’’ ASME J.
Turbomach., 116, pp. 124–132.
The authors would like to express their gratitude to Mr. M. 关12兴 Mochizuki, S., Beier, M., and Murata, A., 1996, ‘‘Detailed Measurement of
Yagi and Mr. T. Yokoyama who were very helpful in carrying out Convective Heat Trans. In Rotating Two-Pass Rib-Roughened Coolant Chan-
the experiments. nels,’’ ASME Paper 96-TA-6.
关13兴 Taslim, M. E., Li, T., and Kercher, D. M., 1994, ‘‘Experimental Heat Transfer
and Friction in Channels Roughened with Angled V shaped and Discrete Ribs
Nomenclature on Two Opposite Walls,’’ Paper 94-GT-163.
关14兴 Hu, Z., and Shen, J., 1996, ‘‘Secondary Flow and It’s Contribution to Heat
a ⫽ passage width 关m兴 Transfer Enhancement in a Blade Cooling Passage with Discrete Ribs,’’ Pro-
Ai ⫽ cross sectional area of channel 关 m2兴 ceedings, International Gas Turbine and Aeroengine Congress & Exhibition
Ah ⫽ heat transfer surface area 关 m2兴 Birmingham, UK, June 10–13, 1996.
b ⫽ passage side length 关m兴关15兴 Han, J. C., and Zhang, Y. M., 1992, ‘‘High Performance Heat Transfer Ducts
with Parallel, Broken, and V-shaped ribs,’’ Int. J. Heat Mass Transf., 35, pp.
cp ⫽ specific heat of air 关J/kgK兴 513–523.
Cf ⫽ friction coefficient ( f /4) 关16兴 Han, J. C., Zhang, Y. M., and Lee, C. P., 1991, ‘‘Augmented Heat Transfer in
de ⫽ equivalent diameter (4ab/ 关 2(a⫹b) 兴 ) Square Channels With Parallel, Crossed, and V-Shaped Angled Ribs,’’ ASME
e ⫽ rib height 关m兴 J. Heat Transfer, 113, pp. 590–596.
k ⫽ air thermal conductivity 关W/mK兴关17兴 Ekkad, S. V., and Han, J.-C., 1997, ‘‘Detailed Heat Transfer Distribution in
Two-Pass Square Channels with Rib Turbulators,’’ Int. J. Heat Mass Transf.,
L ⫽ passage length 关m兴 40, No. 11, pp. 2525–2537.
ṁ ⫽ air mass flow rate 关kg/s兴关18兴 Johnson, B. V., Wagner, J. H., Steuber, G. D., and Yeh, F. C., 1994, ‘‘Heat
Nu ⫽ Nusselt number, Eq. 共4兲 Transfer in Rotating Serpentine Passages With Trips Skewed to the Flow,’’
Nu⬁ ⫽ Nusselt number for fully developed flow in smooth ASME J. Turbomach., 116, pp. 113–123.
passage, Eq. 共5兲关19兴 Cheng, K. C., Shi, L., and Kurokawa, M., 1992, ‘‘Visualization of Flow Pat-
terns in a 180-Degree Sharp Turn of a Square Duct,’’ Proceedings, 4th Inter-
Num ⫽ wall mean Nusselt number, Eq. 共7兲 national Symposium on Transport Phenomena and Dynamics of Rotating Ma-
Nump ⫽ passage mean Nusselt number chinery, USA, Hawaii, A, pp. 124–132.
Num0 ⫽ mean Nusselt number for the smooth passage 关20兴 Cheng, K. C., and Shi, L., 1996, ‘‘Visualization of Flow Patterns in a 90-
q̇ w ⫽ wall heat flux 关 W/m2兴 , Eq. 共2兲 Degree Sharp Turn of a Square Duct,’’ Proceedings, 6th International Sympo-
sium on Transport Phenomena and Dynamics of Rotating Machinery, USA,
⌬p ⫽ pressure loss 关Pa兴 Hawaii, 2, pp. 651–660.
Pr ⫽ Prandtl number of air 关21兴 Metzger, D. E., Plevich, C. W., and Fan, C. S., 1984, ‘‘Pressure Loss Through
Re ⫽ Reynolds number, Eq. 共1兲 Sharp 180 Degree Turn in Smooth Rectangular Channels,’’ Trans. ASME J.
Tb ⫽ local bulk temperature 关K兴, (T in⫹(z/L)(T out⫺T in)) Eng. Gas Turbines Power, 106, pp. 677–681.
T in ⫽ inlet bulk temperatures 关K兴关22兴 Kiml, R., Mochizuki, S., and Murata, A., 1998, ‘‘Influence of 180 Degree
Sharp Turn on the Heat Transfer and Flow Behavior in a Smooth Square Cross
T out ⫽ outlet bulk temperatures 关K兴 Sectional Serpentine Channel,’’ 8th International Symposium on Flow Visual-
Tw ⫽ wall temperature 关K兴 ization, CD Rom Proc. of 8th International Symposium on Flow Visualization,
um ⫽ air or water mean velocity 关m/s兴 Sorrento, Italy 共Paper No. 067兲, pp. 1–8.
关23兴 Zhang, Y. M., Gu, W. Z., and Han, J. C., 1994, ‘‘Heat Transfer and Friction in lators and Secondary Flow Inducers,’’ 9th International Symposium on Flow
Rectangular Channels Ribbed and Ribbed-Grooved Walls,’’ ASME J. Heat Visualization, CD Rom Proc. of 9th International Symposium on Flow Visu-
Transf., 116, pp. 58–65. alization, Edinburgh, United Kingdom 共Paper File No. 164兲, pp. 1–10.
关24兴 Kiml, R., Mochizuki, S., and Murata, A., 2000, ‘‘Influence of the Gap Size 关26兴 Hirota, M., Fujita, H., Yokosawa, H., and Tanaka, Y., 1996, ‘‘Characteristics
between Side Walls and Ribs on the Heat Transfer in a Stationary and Rotating of Turbulent Flow in a Rectangular Duct With Rib-Roughened Long-Side
Straight Rib-Roughened Duct,’’ Int. J. Rotating Mach. 6, No. 4, pp. 253–263. Walls,’’ Proc. of the 9th International Symposium on Transport Phenomena in
关25兴 Kiml, R., Mochizuki, S., and Murata, A., 2000, ‘‘Function of Ribs as Turbu- Thermal-Fluids Engineering, Singapore, June 25–28, 1996.
Energy Separation and Acoustic
R. J. Goldstein Interaction in Flow Across
University of Minnesota,
Department of Mechanical Engineering,
111 Church Street, S.E.,
a Circular Cylinder
Minneapolis, MN 55455
Energy separation in a flow around an adiabatic circular cylinder is investigated using a
Boyong He surface-mounted thermocouple. Energy separation mechanisms in different regions
510 Maloney Road, around the cylinder are discussed. Velocity measurements near the rear stagnation point
Poughkeepsie, NY 12603 and acoustic measurements indicate that shedding vortices and the wind tunnel intrinsic
resonant acoustics can strengthen each other when their frequencies match producing
strong energy separation. 关DOI: 10.1115/1.1378020兴
Keywords: Heat Transfer, Measurement Techniques, Vortex, Wakes
Introduction r⫽ 冑Pr (6)

Fluid in motion without any heat sources or heat sinks can or
separate spontaneously into high energy 共total temperature or en-
thalpy兲 regions and neighboring low energy regions. This phe- S l ⫽ 冑Pr⫺1,
nomena is called energy separation. Energy separation occurs in
while for a turbulent boundary layer,
various flow situations, including laminar and turbulent boundary
layers, and vortex flows. The intensity of energy separation can be r⬵ 3冑Pr (7)
quantitatively measured with the energy separation factor. Energy
separation associated with vortex flows is a much stronger process or
S l ⬵ 3冑Pr⫺1.
compared with that associated with boundary layer flows. The
mechanisms of energy separation have been a topic of discussion
for many years and continue to be in dispute. Here r is the recovery factor and S l is the energy separation
factor based on the free stream velocity immediately outside the
Steady Boundary Layer Flow. The unsteady energy equa- boundary layer.
tion assuming constant specific heat and no body force can be
written from Eckert and Drake 关1兴 as Unsteady Flow Without Friction and Conduction. For un-
冉冊
steady flow without friction and conduction, Eq. 共1兲 reduces to
DT t ⍀ P ⳵ ⳵Ts ⳵
␳Cp ⫽ ⫹ k ⫹ 共 u ␶ 兲, (1) DT t ⳵ p
Dt ⳵t ⳵xi ⳵xi ⳵x j i ij ␳Cp ⫽ . (8)
Dt ⳵t
where
Equation 共8兲 indicates that the total energy of a fluid particle
D ⳵ ⳵ will change along path-line due to the pressure fluctuations at a
⫽ ⫹u i (2)
Dt ⳵ t ⳵xi fixed location in the low field. Strong pressure fluctuations, for
example in vortex type flows, can cause a large energy separation.
and for Newtonian fluid, Energy separation due to pressure fluctuations caused by vor-
␶i j⫽␮ 冉 ⳵ui ⳵u j
⫹
⳵x j ⳵xi 冊
⫹␦i j␮⬘
⳵uk
⳵xk
. (3)
texes has been studied for flow in vortex tubes 共Ranque 关2兴,
Hilsch 关3兴, Hartnett and Eckert 关4兴, Marshall 关5兴, and Kurosaka
关6兴兲, in resonant tubes 共Sprenger 关7兴兲, in a free shear layer
For steady state two-dimensional boundary layer flow, 共O’Callaghan and Kurosaka 关8兴兲, and in jet flows 共Goldstein et al.
关9兴, Goldstein et al. 关10兴, Fox et al. 关11兴 and Seol 关12兴兲.
⳵ ⳵ ⳵ Eckert and Weise 关13兴 observed energy separation in the flow
Ⰷ , (4)
⳵y ⳵x ⳵z around a circular cylinder; and an energy separation factor of
⫺1.1 was reached at the rear stagnation line. This means that the
and Eq. 共1兲 reduces to cylinder surface recovery temperature at the rear stagnation line is
␳Cp u冉 ⳵Tt
⳵x
⫹v
⳵Tt
⳵y
⫽
⳵
⳵y 冊冉冊冉冊
k
⳵Ts
⳵y
⫹
⳵
⳵y
u␮
⳵u
⳵y
, (5)
even lower than the static temperature of the upstream fluid. Ryan
关14兴 and Thomann 关15兴 confirmed Ecker’s experimental results.
Eckert 关16兴 presented a model to analyze cross energy transport
where x is the streamwise direction, y in the direction normal to through a pathline in a single vortex which can also be used in the
the wall, and z is the direction orthogonal to x and z. analysis of the energy separation in a vortex street. He assumed
The total energy 共temperature兲 can vary due to the imbalance the vortex has a viscous core and inviscid outer layer, and the
between the energy transferred by conduction and shear stress. maximum and minimum total temperature T t are located at a
Within a boundary layer, energy separation is a relatively weak place close to the border between the viscous and inviscid part of
process dependent on the Prandtl number. For a flow over a flat the vortex. Energy separation starts as the vortex is generated and
plate laminar boundary layer has a recovery factor, decreases as the strength of the vortex decreases. Kurosaka and
Gertz 关17兴 also investigated energy separation phenomena around
a circular cylinder in the Mach number range of 0.2⭐M a⭐0.5.
HEAT TRANSFER. Manuscript received by the Heat Transfer Division April 18, They found that the pressure fluctuations due to the convective
2000; revision received February 12, 2001. Associate Editor: B. Chung. vortex separate the flow field instantaneously into hot spots and
cold spots and resonance has very strong effects on the energy grammable Stepper Motor Controller 共Velmex Inc.兲 connected to
separation. Ng et al. 关18兴 used an aspirating probe to measure the a computer through an IO Tech Serial-to-IEEE 488 Converter and
transient total pressure and total temperature simultaneously in the an IEEE 488 cable.
wake of a circular cylinder at free stream Mach number of 0.4 and
The Splitter Plates. A splitter plate installed immediately
Reynolds number of 2.3⫻105 . From the frequency domain of the
behind the cylinder can eliminate the shedding vortex in the cyl-
measured transient total temperature and total pressure signals, it
inder wake. Two different splitter plates with thicknesses of 3.0
was confirmed that both signals have a dominant frequency which
mm and 9.4 mm are used. The splitter plates have a contoured
is the same as the vortex shedding frequency.
face toward the cylinder with curvature equal to that of the cylin-
It can be seen from the above discussion that energy separation
der and 1.0 mm away from it.
across a circular cylinder and its relation with the shedding vortex
and the acoustic sound has been studied since 1940 and it is gen- Surface Hot Wire and Thermocouple Mounting. A TSI
erally agreed that the pressure fluctuation caused by the shedding T1.0 tungsten wire with a 2.54 ␮m diameter is soldered on the
vortex will intensify the energy separation around the cylinder cylinder surface as shown in Fig. 1. The wire is approximate 120
rear stagnation point and acoustic resonance will have strong ef- ␮m above the cylinder surface and can be used as both hot wire
fect on the energy separation. It is paper’s intention to answer and cold wire. When used as a hot wire, it is calibrated against the
if the acoustic sound will intensify the vortex shedding thus potential flow and used to measure the fluid velocity immediately
the energy separation through temperature, flow and acoustic outside the boundary layer, U( ␪ ). The velocity in the potential
measurement. flow region 共outside the cylinder boundary layer兲 before the sepa-
ration can be expressed as follows:
Experimental Apparatus
The Wind Tunnel. The suction-type subsonic tunnel used U 共 ␪ 兲 ⫽2U ⬁ sin共 ␪ 兲 . (9)
has an entrance section with a bell-mount inlet, a flow straightener
and filter, three turbulence reducing screens and two stages of area
contraction to the test section. The total area contraction ratio is Calculating the boundary layer thickness from the similarity
15:1. The test section constructed of 19 mm thick Plexiglas has a solution of Wang 关19兴, the hot wire is outside the boundary layer
120 mm 共width兲⫻103 mm 共height兲 cross section and 1345 mm for the potential flow at least to ␪⫽30 deg. This provides a veloc-
overall length. With this design, the velocity in the wind tunnel ity reference for an approximate calibration of the surface
test section without a cylinder inside can reach 120 m/s. Measured mounted hot wire against the potential flow velocity 共Eq. 9兲, while
by a TSI 1218-T1.5 hot wire, the velocity across the wind tunnel the incoming fluid velocity, U ⬁ , is measured with a pitot tube.
test section with no cylinder present is uniform and the turbulence A calibrated T-type 30 gauge thermocouple, flush mounted on
intensity is 0.3 percent. the cylinder surface with an insulator as shown in Fig. 1, measures
The Test Cylinder and Its Stepper Motor Control. The the cylinder surface recovery temperature T r . Before mounting
test cylinder, made of phenolic and shown in Fig. 1 has an OD of the surface thermocouple, a groove in the cylinder’s longitude
19.05 mm and an ID of 10.16 mm. Phenolic has a thermal con- direction and an angled hole were cut as shown in Fig. 1. Then the
ductivity of 0.3344W/m•K and a thermal diffusivity of 1.6 thermocouple was laid down on the groove and both thermo-
⫻10⫺7 m2 /s. The hot-wire probe leads and a thermocouple leads couple wires can be pulled out of the hollowed cylinder center
come out of the hollow cylinder center. The test cylinder, by from the angled hole. The thermocouple insulation cement was
connection through a 6.4 mm diameter circular extension shaft, then used to glue thermocouple wire and the thermocouple junc-
can be rotated by a Superior Electric M063-LS09 model stepper tion in position. During this process, one took care that the ther-
motor with a resolution of 400 steps per revolution 共0.9 deg per mocouple junction is just slightly above the cylinder surface. The
step兲. The stepper motor is controlled by a Unislide NF-90-3 Pro- cylinder was put back to the original shape by machining away
the extra insulation cement. A sand paper with very fine grain size
was then used to carefully sand the cylinder and the thermocouple
junction so that the cylinder surface is smooth and the thermo-
couple junction is flush with the cylinder surface. Four T-type 30
gauge thermocouples mounted on the entrance section of the wind
tunnel monitor the room temperature, which closely approximates
the total temperature of the incoming fluid, T ⬁ .
Microphone Acoustic Measurement. One objective of this
test is to investigate acoustic effects. A Radio Shack 270-092B
microphone mounted on the top wall of the test section three
diameters 共D兲 downstream of the cylinder monitors the dominant
acoustic frequency. A hole in the top wall enables flush mounting
of the microphone to the inside of the wind tunnel.
Flow Visualization. In order to investigate the relationship
between the moving vortices and energy separation around a cir-
cular cylinder, the flow pattern around the cylinder was studied.
The near surface flow was visualized using a mixture of oil and
black powder painted on the cylinder.
Uncertainty Analysis. The uncertainty in energy separation
factor S ⬁ depends on the uncertainties in incoming fluid total
temperature T t,⬁ , recovery temperature T r and dynamic tempera-
ture T d , according to the definition of S ⬁ . From the uncertainty
analysis method 共Coleman and Steele 关20兴兲, it is estimated that the
Fig. 1 Schematic drawing of cylinder and surface mounted uncertainty in energy separation factor S ⬁ is 5.23 percent at the
hot wire cylinder rear stagnation point with U ⬁ ⫽84.25 m/s 共He 关21兴兲.
Results and Discussion
Energy Separation Without a Splitter Plate. Figure 2
shows the variation of energy separation with ␪ along with flow
visualization when U ⬁ ⬇84 m/s. The energy separation factor,
S ⬁ , decreases continuously from a value of zero at the front stag-
nation point 共recovery factor r⫽1.0) to a value of ⫺1.2 at the rear
stagnation point 共recovery factor r⫽⫺0.2). The negative recov-
ery factor at the rear stagnation point indicates that the recovery
temperature is even lower than the incoming fluid static tempera-
ture. For U ⬁ ⬇84 m/s case, the energy separation is surprisingly
strong. During the recovery temperature measurements at this free
stream velocity, an extremely strong resonant acoustic sound was
Fig. 3 Proposed traveling vortex pattern around a circular
heard and it was suspected that the surprisingly strong energy cylinder
separation phenomena 共low recovery temperature兲 in the region
near the rear stagnation point is related to the resonant acoustic
sound.
The distribution of S ⬁ , local velocity U( ␪ ) and turbulence in- factor S ⬁ does not change significantly. It also can be seen from
tensity T u around the cylinder can be separated into five regions the U( ␪ ) distribution that the velocity decreases dramatically in
共Fig. 3兲 which correspond to the five regions from the flow visu- this region.
alization shown in Fig. 2. 3 Region 3 extends from about 90 deg to about 110 deg. S ⬁
decreases dramatically from ⫺0.4 to ⫺0.75. It is expected from
1 Region 1 extends from the front stagnation point 共␪⫽0 deg兲 the flow visualization picture that the negative pressure gradient in
to about 72 deg. In this region, S ⬁ continuously decreases from ␪ direction due to the laminar separation creates a clockwise vor-
zero to ⫺0.4 at a moderate slope. Both the U( ␪ ) distribution and tex in this region as shown in Fig. 3. As this clockwise vortex
the flow visualization indicate a laminar boundary layer flow in moves away from the cylinder surface in region 3, main stream
this region and separation point at 72 deg, somewhat lower than fluid is drawn close to the cylinder surface and the cylinder sur-
often observed. In region 1, the energy separation is dominated by face recovery temperature drops dramatically.
laminar boundary layer flow and is relatively weak. 4 Region 4 extends from about 110 deg to 150 deg. S ⬁ de-
2 Region 2 extends from about 72 deg to about 90 deg. In this creases from ⫺0.75 to ⫺1.2 at a moderate rate but still greater
region, the energy separation factor changes very little. It is sus- slope than that in region 1. The moving clockwise vortex creates
pected from the flow visualization that this region has a near stag- a strong pressure fluctuation which dominates the energy separa-
nant flow—a very weak flow exists so that the energy separation tion in this region, but the cylinder surface recovery temperature
does not drop as sharply as that in region 3.
5 Region 5 extends from 150 deg to 180 deg. In this region, the
clockwise vortices from the upper part of the cylinder and the
counter-clockwise vortex from the bottom part of the cylinder
alternatively affect the flow near the cylinder surface. Pressure
fluctuations due to the alternately shedding vortex produce the
lowest S ⬁ in this region.
The energy separation factor distribution around the cylinder
for five different velocities are shown in Fig. 4. S ⬁ distributions
for all velocity cases generally follow the five regions discussed
above. At U ⬁ ⫽84 m/s case, the acoustic sound is the strongest;
the sound levels for U ⬁ ⫽72 m/s, 93 m/s, and 100 m/s cases are
weaker than that when U ⬁ ⫽84.25 m/s. No acoustic sound is
heard when U ⬁ ⫽43 m/s. Up to the laminar separation points, the
S ⬁ distributions for all velocity cases are virtually the same and
match Eckert and Weise 关13兴 very well. The resonant acoustics
have no effects on the energy separation inside the laminar bound-
Fig. 2 Comparison of energy separation factor, S ⴥ , and flow

visualization around the cylinder for U ⴥ É84 mÕs Fig. 4 Energy separation factor, S ⴥ
Fig. 5 Energy separation factor, S l , distributions for 5 differ- Fig. 7 Acoustic measurement for U ⴥ Ä100 mÕs
ent velocities
tion point has a dominant frequency of 1640 Hz, so the single

ary layer and this is also demonstrated in Fig. 5 which shows that vortex shedding frequency is 820 Hz, which is virtually the same
the S l distributions for all five cases inside the laminar boundary as the first dominant acoustic frequency. The calculated Strouhal
layer before the separation point are virtually the same. Both Fig. number, S t , from the 820 Hz single shedding vortex frequency is
2 and Fig. 5 indicate that before flow separation, S l which is based 0.186 which closely matches the value found by Ericsson and
on the local velocity, closely matches the recovery factor ( 冑Pr) Reding 关22兴. This suggests that there is an intrinsic resonant
for a laminar boundary layer. But it is very evident in Fig. 4 that acoustic frequency around 810 Hz. Taking the intrinsic acoustic
S ⬁ at a free stream velocity of U ⬁ ⫽84 m/s decreases from 90 deg wavelength, ␭, to be 2 times the wind tunnel height 共203 mm兲 and
to 180 deg at larger slope than at other free stream velocities. the sound speed (c⫽ 冑␯ RT⫽344.8 m/s兲, then the frequency for a
It is interesting to compare the S ⬁ distributions in region 4 and standing wave in the wind tunnel can be calculated as f ⫽c/␭
5 for different free stream velocities. S ⬁ has lowest value in re- ⫽850 Hz. This frequency is very close to the measured first domi-
gion 4 and 5 when U ⬁ ⫽84 m/s. The greater the difference in U ⬁ nant acoustic frequency and the shedding vortex frequency when
from U ⬁ ⫽84 m/s, the larger S ⬁ is. If there is a resonance in the U ⬁ ⬇84 m/s.
wind tunnel used at U ⬁ ⫽84 m/s, then the energy separation effect When there is a shedding vortex at the same frequency as the
in the area after the laminar separation point will be weaker if the wind tunnel intrinsic resonant acoustic frequency, the resonant
incoming fluid velocity is further away from this resonant velocity acoustic will be excited to a stronger level, but at the same time,
(U ⬁ ⫽84 m/s兲. this strong acoustic sound also strengthens the shedding vortex
Acoustic Effects on Energy Separation. It has been found after the separation point around the cylinder. This is why S ⬁
by Ryan 关14兴 and confirmed by Kurosaka and Gertz 关17兴 that decreases with a large slope after separation and has the low value
acoustic sound has strong effect on energy separation around a of ⫺1.2 at the rear stagnation point for the case of U ⬁ ⫽84 m/s.
circular cylinder. By using a surface mounted hot wire an flush With no cylinder present there is also no dominant frequency
mounted microphone, the effect of acoustic on the flow, vortex inside the acoustic signal for U ⬁ ⫽100 m/s 共c.f. Fig. 7兲, but with a
shedding can be related to energy separation. cylinder present, there are two dominant frequencies whose am-
Figure 6 indicates that there is no dominant frequency inside plitudes are substantially higher than the others. The first one is
the acoustic signal when U ⬁ ⬇84 m/s without a cylinder present. f a1 ⫽820 Hz, which is related to the wind tunnel intrinsic resonant
But once a circular cylinder is present, certain dominant frequen- frequency. The second one is f a2 ⫽1015 Hz (S t ⫽0.193) which is
cies appear in the acoustic signal. The first dominant frequency is essentially the shedding vortex frequency for this velocity. But the
f a1 ⫽810 Hz, the nth dominant frequency is f an ⫽n⫻ f a1 , but the acoustic signal level at both frequencies is substantially lower
amplitude of the first dominant frequency is much higher than the than the acoustic signal level at 810 Hz for U ⬁ ⫽84 m/s case. The
others. The hot wire signal measured at the cylinder rear stagna- hot wire signal amplitude at the dominant shedding vortex fre-
quency is also lower than at U ⬁ ⫽84 m/s. As the shedding vortex
frequency gets close to the wind tunnel intrinsic resonant fre-
quency, both the shedding vortex strength and the resonant acous-
tic get quite strong and the energy separation gets larger in the
region after the laminar separation point where the energy sepa-
ration is dominated by the shedding vortex, as occurs at U ⬁ ⫽84
m/s. When the shedding vortex frequency is away 共but not too far
away兲 from the wind tunnel intrinsic resonant frequency 共cases for
U ⬁ ⫽100 m/s, 93 m/s, and 72 m/s兲, both the resonant acoustic at
th wind tunnel intrinsic resonant frequency 共810 Hz兲 and the
acoustic at the shedding vortex frequency will be excited by the
shedding vortex, but at a weaker level than when U ⬁ ⫽84 m/s as
there is a weaker interaction between the resonant acoustic and the
shedding vortex. The weaker shedding vortex results in smaller
energy separation in the region after the laminar separation point.
When U ⬁ ⫽43 m/s case, there is very little interaction between the
shedding vortex and the resonant acoustic since the shedding vor-
tex frequency is away from the wind tunnel intrinsic resonant
frequency, resulting in a relatively weak energy separation in the
Fig. 6 Acoustic measurement for U ⴥ Ä84 mÕs rear of the cylinder 共Fig. 4兲.
Fig. 10 Energy separation factor, S ⴥ , comparison among
cases with 3.0 mm splitter plate for U ⴥ Ä72 mÕs, 84 mÕs, 94 mÕs,
100 mÕs
virtually the same for all three cases, but there is a profound
difference after 90 deg. S ⬁ remains constant at ⫺0.4 with either
splitter plate. The energy separation is much weaker with a splitter
plate. During the recovery temperature measurements with splitter
plates, no acoustic sound is heard. A comparison between the
energy separation factor, distributions and the flow visualization
around the cylinder with a 3.0 mm splitter plate in Fig. 8 shows
that the flow still separates at ␪ ⬇72 deg even with a splitter plate
present and laminar boundary layer flow still dominates the region
from 0 deg to 72 deg. This is why the S ⬁ distributions for all three
Fig. 8 Comparison between energy separation factor, S ⴥ , cases in Fig. 9 are the same up to the laminar separation point.
distributions and flow visualization around the cylinder for After the separation, flow visualization in Fig. 8 shows that the
U ⴥ Ä84 mÕs with 3.0 mm splitter plate flow reattaches at ␪ ⬇91 deg and a separation bubble is present in
the region between ␪ ⬇72 deg and ␪ ⬇91 deg. So with a splitter
plate, there is no moving convective vortex after the separation;
Energy Separation With a Splitter Plate. It was found 共see the flow in the cylinder wake region after the separation is not
previous section兲 that the wind tunnel’s intrinsic acoustic and- dominated by vortex type flow, but rather by reattached boundary
shedding vortex will have strong interaction only when the shed- layer type flow. That is why the energy separation is so weak with
ding vortex frequency matches to wind tunnel’s intrinsic acoustic a splitter plate present. The flow pattern differences in the region
frequency. In this case, the energy separation around the cylin- after the laminar separation point with and without a splitter plate
der’s rear stagnation point is greatly intensified. Thus, it is ex- causes profound differences in energy separation in this region.
pected the energy separation around the rear stagnation point of Figure 10 shows S ⬁ distributions with a 3.0 mm splitter plate
the cylinder will be much weaker if a splitter plate is placed at the present for four velocities. Unlike the distribution without a split-
rear stagnation point to eliminate the shedding vortex. ter plate, the S ⬁ distributions in Fig. 10 are similar for all four
Figure 9 shows energy separation distributions around the cyl- velocities all around the cylinder even in the region after the lami-
inder with 3.0 mm and 9.4 mm splitter plates and without a split- nar separation point at ␪ ⬇72 deg. The S ⬁ distributions with a 9.4
ter plate at U ⬁ ⬇84 m/s. Up to 72 deg, the S ⬁ distributions are mm splitter plate present at same velocities as 3.0 mm splitter
plate are exactly the same as that in Fig. 10. The presence of a
splitter plate has eliminated the shedding vortex after the separa-
tion and energy separation is totally independent of splitter plate
and Reynolds number in the region studied.
Conclusions
In this paper, the energy separation effect around a circular
cylinder is measured using a surface mounted thermocouple. To
reveal the direct relationship between the acoustic and energy
separation, a surface mounted hot wire is used to measure the
shedding vortex frequency and a flush mounted microphone was
used to measure the acoustic signal inside the wind tunnel. Thus,
the relationship between the strength of shedding vortex and the
acoustic sound can be found. Then from the energy separation
measurement around the cylinder at the same flow conditions, a
relationship between the energy separation and the acoustic is
obtained.
Around a cylinder in cross flow, the flow in the region from
Fig. 9 Energy separation factor, S ⴥ , comparison among ␪ ⫽0 deg to ␪ ⫽72 deg is laminar; the energy separation is due to
cases with 3.0 mm, 9.4 mm and without splitter plate for the imbalance between conduction heat transfer and shear stress
U ⴥ É84 mÕs work and is relatively weak. After flow separation, the flow
around the cylinder is dominated by shedding vortices and the xi ⫽ coordiante in Cartesian coordinate system
energy separation in this region is due to the strong pressure fluc- x ⫽ coordiante along flow direction
tuations due to the shedding vortices. y ⫽ coordiante perpendicular to flow direction
Shedding vortices can interact with the wind tunnel intrinsic z ⫽ coordiante orthogonal to x and y
acoustics. If the shedding vortex frequency is close to the tunnel ␦ij ⫽ Kronecker delta
intrinsic resonant acoustic frequency, both the shedding vortex ␪ ⫽ angle around cylinder measured from forward stagna-
and the resonant acoustic will be strengthened by each other and tion point
the energy separation in the rear of the cylinder is strong due to ␭ ⫽ acoustic wavelength
the strong shedding vortex. The energy separation gets weaker ␮ ⫽ fluid kinetic viscosity
because of a weaker interaction between the shedding vortex and ␮⬘ ⫽ viscosity
the intrinsic resonant acoustic as the shedding vortex frequency ␯ ⫽ specific heat ratio
shifts further from the wind tunnel intrinsic resonant acoustic fre- ␳ ⫽ fluid density
quency. Interaction between the shedding vortices and the reso- ␶ij ⫽ shear stress tensor
nant acoustic has little effects on energy separation in the region
before flow separates.
With a splitter plate present, the shedding vortices are not References
present. The flow still separates at ␪ ⬇72 deg, but reattaches at 关1兴 Eckert, E. R. G., and Drake, R. M. J., 1972, Analysis of Heat and Mass
␪ ⬇90 deg. Energy separation in this case is dominated by bound- Transfer, McGraw-Hill, New York, pp. 265–270.
关2兴 Ranque, G., 1933, ‘‘Expériences sur la Détente Giratoire avec Productions
ary layer type flow. The incoming flow velocity and the splitter Simultanées d’un échappement d’air chaud et d’un échappement d’air Froid,’’
plate thickness have little effect on energy separation. J. Phys. Radium, 4, pp. 112–114.
关3兴 Hilsch, R., 1947, ‘‘The Use of Expansion of Gases in a Centrifugal Field as a
Cooling Process,’’ Rev. Sci. Instrum., 18, pp. 108–113.
Acknowledgment 关4兴 Hartnett, J. P., and Eckert, E. R. G., 1957, ‘‘Experimental Study of the Veloc-
This work was supported by the Engineering Research Program ity and Temperature Distribution in a High-Velocity Vortex-Type Flow,’’
Trans. ASME, 79, pp. 751–758.
of the Office of Basic Energy Sciences at the Department of En- 关5兴 Marshall, J., 1977, ‘‘Effect of Operating Conditions, Physical Size and Fluid
ergy. We also wish to thank Dr. Bumsoo Han for his helpful Characteristics on the Gas Separation Performance on the Linderstrom-Lang
suggestions. Vortex Tube,’’ Int. J. Heat Mass Transf., 20, pp. 227–231.
关6兴 Kurosaka, M., 1982, ‘‘Acoustic Streaming in Swirling Flow and the Ranque-
Hilsch 共Vortex Tube兲 Effect,’’ J. Fluid Mech., 124, pp. 139–172.
Nomenclature 关7兴 Sprenger, H., 1954, ‘‘Über Thermische Effecte in Resonanzrohren,’’ Mittei-
c ⫽ speed of sound lungen aus dem Institut für Aerodynamik, Eidgen. Tech. Hochschule Zürich,
21, pp. 18.
cp ⫽ specific heat at constant pressure 关8兴 O’Callaghan, J. J., and Kurosaka, M., 1993, ‘‘Vortex-Induced Energy Separa-
D ⫽ cylinder diameter tion in Shear Flow,’’ AIAA J., 13, pp. 1157–1159.
f ⫽ frequency 关9兴 Goldstein, R. J., Behbahani, A. I., and Heppelmann, K. K., 1986, ‘‘Streamwise
f a1 ⫽ first dominant frequency inside acoustic signal Distribution of the Recovery Factor and the Local Heat Transfer Coefficient in
an Impinging Air Jet,’’ Int. J. Heat Mass Transf., 29, pp. 1227–1235.
f an ⫽ n th dominant signal inside acoustic signal 关10兴 Goldstein, R. J., Sobolik, K. A., and Seol, W., 1990, ‘‘Efefct of Entrainment
fs ⫽ vortex shedding frequency on the Heat Transfer to a Heated Circular Air Jet Impinging on a Flat Sur-
k ⫽ fluid thermal conductivity face,’’ , 112, pp. 608–611.
关11兴 Fox, M., Kurosaka, M., and Hirano, K., 1990, ‘‘Total Temperature Separation
P ⫽ pressure in Jets,’’ AIAA Paper 90-1621.
Pr ⫽ prandtl number 关12兴 Seol, W., 1993, ‘‘Energy Separation in a Jet Flow,’’ Ph.D. thesis, Department
r ⫽ recovery factor, r⫽1⫹S ⬁ of Mechanical Engineering, University of Minnesota, Minneapolis, MN.
R ⫽ universal gas constant 关13兴 Eckert, E. R. G., and Weise, W., 1940, ‘‘Messungn der Temperaturverteilung
Re ⫽ Reynolds number, R e ⫽ ␳ U ⬁ D/ ␮ auf der Oberfläche Schnell Angeströmter umbeheizter Körper,’’ Jahrbuch
1940 der Deutschen Luftfahrtforschung, 2, pp. 25–31.
S⬁ ⫽ energy separation factor based on incoming fluid ve- 关14兴 Ryan, L. F., 1951, ‘‘Experiments on Aerodynamic Cooling,’’ Ph.D. thesis,
locity, S ⬁ ⫽T r ⫺T t,⬁ /T d Eidgen. Tech. Hochschule, Zürich.
Sl ⫽ energy separation factor based on local fluid velocity 关15兴 Thomann, H., 1959, ‘‘FFA Report 84,’’ Stockholm, Sweden.
关16兴 Eckert, E. R. G., 1987, ‘‘Cross Transport of Energy in Fluid Stream,’’ Wärme
immediately outside the boundary layer, S l ⫽T r und Stoffübertragung, 21, pp. 73–81.
⫺T t,⬁ /0.5U( ␪ ) 2 /C p 关17兴 Kurosaka, M., and Gertz, J. B., 1987, ‘‘Energy Separation in a Vortex Street,’’
St ⫽ Strouhal number, St⫽ f s D/U ⬁ J. Fluid Mech., 178, pp. 1–29.
t ⫽ time 关18兴 Ng, W. F., Chakroun, M., and Kurosaka, M., 1990, ‘‘Time-Resolved Measure-
ment of Total Temperature and Pressure in the Vortex Stress Behind a Cylin-
T t,⬁ ⫽ incoming fluid total temperature der,’’ Phys. Fluids A, 12, pp. 971–978.
T s,⬁ ⫽ incoming fluid static temperature 关19兴 Wang, H. P., 1997, ‘‘Local Mass Transfer from a Turbine Blade: Influence of
Tr ⫽ recovery temperature High Turbulence with Large Length Scale on Heat/Mass Transfer,’’ Ph.D.
Td ⫽ dynamic temperature, T d ⫽T t,⬁ ⫺T s,⬁ ⫽U 2⬁ /2C p thesis, Department of Mechanical Engineering, University of Minnesota, Min-
neapolis, MN.
u ⫽ fluid velocity in flow direction inside a flat plate 关20兴 Coleman, H. W., and Steele, W. G. S. J., 1989, Experimentation and Uncer-
boundary layer tainty Analysis for Engineers, Wiley, New York, pp. 75–116.
U⬁ ⫽ incoming fluid velocity 关21兴 He, B., 1997, ‘‘Energy Separation for High Speed Flow Across a Circular
U共␪兲 ⫽ potential flow velocity around circular cylinder at Cylinder,’’ Master’s thesis, Department of Mechanical Engineering, Univer-
sity of Minnesota, Minneapolis, MN.
angle ␪ 关22兴 Ericsson, L. E., and Reding, J. P., 1979, ‘‘Criterion for Vortex Periodicity in
ui ⫽ veloicty component in Cartesian coordinate system Cylinder Wakes,’’ AIAA J., 17, pp. 1013.
Experimental and Numerical
Study of Conjugate Heat Transfer
in a Horizontal Channel Heated
Wilson K. S. Chiu From Below
Cristy J. Richards Conjugate heat transfer has significant relevance to a number of thermal systems and
techniques which demand stringent temperature control, such as electronic cooling and
Yogesh Jaluria chemical vapor deposition. A detailed experimental and numerical study is carried out to
Fellow ASME investigate conjugate heat transfer in a common configuration consisting of a horizontal
channel with a heated section. Experimental data obtained from this study provides physi-
Department of Mechanical cal insight into conjugate heat transfer effects and facilitates validation of numerical
and Aerospace Engineering, conjugate heat transfer models. The basic characteristics of the flow and the associated
Rutgers, The State University of New Jersey, thermal transport are studied. The numerical model is used to carry out a parametric
New Brunswick, NJ 08903 study of operating conditions and design variables, thus allowing for the characterization
of the conjugate heat transfer effects. It is found that the numerically predicted flow field
and heat transfer results validate well to experimental observations. Conjugate heat
transfer is shown to significantly affect the temperature level and uniformity at the heated
section’s surface, channel walls and the gas phase, thus impacting the rate of heat trans-
fer. This study provides guidelines and fundamental insight into temperature control
during the combined modes of heat transfer, with implications to various thermal manu-
facturing methods. 关DOI: 10.1115/1.1372316兴
Keywords: Conjugate, Experimental, Heat Transfer, Manufacturing, Mixed Convection
Introduction effects are considered. Fotiadis et al. 关10兴 further showed that dif-
ferent wall materials in a vertical channel altered the isotherms,
Many investigators have demonstrated the importance of con- but did not significantly affect the flow field.
jugate heat transfer in a number of important technological areas, The preceding literature review indicates the importance of
where the coupling of conduction, convection and/or radiation conjugate heat transfer effects. Several researchers have numeri-
heat transfer significantly affects the temperature distribution and cally investigated conjugate transport, but a need exists for experi-
rate of heat transfer. Examples where conjugate heat transfer mental data, satisfactory model validation, and characterizing of
analysis has important and significant impact include electronic the conjugate effects for a wide range of operational parameters.
cooling 关1–3兴, solar collectors 关4兴, chemical vapor deposition This paper presents an experimental and numerical study of con-
共CVD兲关5,6兴, and extrusion of materials 关7兴. Conjugate heat trans- jugate heat transfer of air flow in a horizontal channel with con-
fer analysis is critical in predicting the system’s temperature dis- ducting walls, as motivated by CVD systems. Conjugate heat
transfer effects are considered for different wall materials and
tribution and heat transfer rate. However, there is a need to further
wall thicknesses. The flow field is visualized using a light sheet
understand how the different modes of heat transfer affect the with smoke injection at the inlet, while wall and gas phase tem-
thermal characteristics of the system 共i.e., temperature distribu- perature distributions are measured using thermocouple probes.
tion, heat transfer rate兲, and provide experimental data for conju- Experimental results obtained at different flow rates, heating rates
gate model validation. and heated region materials are compared to numerical predic-
A limited amount of work on mixed-mode heat transfer studies tions. Good agreement between the two is observed.
exist in the literature. In CVD, for example, Chinoy et al. 关8兴
numerically modeled thermal radiation exchange using a uniform Experimental Apparatus
temperature heated region coupled with conservation equations.
The experimental system shown in Fig. 1 consists of a plexi-
The study demonstrated that a two-band model is able to satisfac-
glass channel, flow regulation system with a smoke generator, DC
torily predict GaAs growth rates in a horizontal CVD reactor. power supply, light source, and temperature measurement probes
Durst et al. 关9兴 and Fotiadis et al. 关10兴 investigated radiation heat connected to a data acquisition system. The 91.4 cm long plexi-
transfer coupled with heat conduction through quartz walls and glass channel measures 5.08 cm high by 25.4 cm wide from the
ambient convection heat loss. Durst et al. 关9兴 examined conjugate inside. The channel is insulated on all sides with 5.08 cm thick
effects using an electrically heated region in a channel reactor, fiberglass insulation to minimize heat loss.
while Fotiadis et al. 关10兴 considered a vertical channel with induc- Air is pressure-regulated and introduced into the channel using
tion heating and radiation heating by heat lamps. Both investiga- a variable area float-type flowmeter. A 1.89⫻10⫺2 m3 stagnation
tors stressed the importance of wall conduction, and found signifi- chamber is used to provide a flow at constant temperature and to
cant differences in heated surface temperatures when conjugate dampen pressure fluctuations associated with compressed air sup-
plied by the house line. The stagnation chamber is also used to
premix smoke for flow visualization. Smoke is generated using a
HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 16, commercial smoke generating device 共ROSCO Laboratories
1999; revision received February 1, 2001. Associate Editor: D. A. Kaminski. 1600兲 which vaporizes heated aqueous glycol solution, producing
toshop and Premiere兲 is used for image post-processing. A data
acquisition system 共National Instruments SCXI and LabView兲 is
used to record outputs from the thermocouples. Graphical presen-
tation of the results are carried out on workstations connected to
the data acquisition system.
Convective Heat Flux Calculation. Since the bottom wall is
not perfectly insulated, heat loss from the bottom of the experi-
mental arrangement is accounted for by using the following ex-
pression:
q ref⫽q ⍀ ⫺q k , (1)
where the amount of thermal energy input to the heated region
(q ref) is the difference between the heat flux due to resistance
Fig. 1 Diagram of the experimental system heating (q ⍀ ) and heat flux due to conduction loss through the
bottom. Since temperature measurements show small temperature
differences between the heated bottom surface and the bottom
wall, radiative heat loss through the bottom is neglected in Eq. 共1兲.
glycol particles in suspension. The gas mixture is injected into the Assuming uniform heat flux q ⍀ and q k through the bottom wall,
channel, where 10.16 cm of high porosity polyurethane is used as q k is calculated using a one-dimensional conduction model with a
a diffuser and flow straightener to give rise to a uniform laminar 12.7 mm thick air gap, 6.4 mm thick G-10 Garolite and 50.8 mm
entrance flow. thick fiberglass insulation. This correction is incorporated in all
The heated section consists of a 16.51 cm long by 21.22 cm cases presented in this study. Heat flux into the heated surface
wide by 6.35 mm thick heated region plate 共susceptor兲 mounted in (q ref) is used to calculate numerical parameters for the model.
a G-10 Garolite fiberglass tub, as shown in Fig. 2. G-10 Garolite There is a minimal feedback effect from numerically corrected
is chosen because of its high thermal resistance and excellent q ref used for validation studies since q ref is determined using a
machinability. The plate is mounted flush to the top surface using separate one-dimensional conduction model, and contribution
ceramic spacers. Heating pads are constructed of 0.13 mm thick from the conduction component is small, where q k /q ⍀
by 2.03 mm wide inconel wire sandwiched between two 0.06 mm ⭐5 percent.
thick kapton sheets 共Technical Heaters, CA兲. The heating pads are In calculating the total convective heat flux 共q兲 at the heated
secured to the back of the heated region plate using a high tem- region, one must determine the amount of conduction and radia-
perature acrylic adhesive to ensure good thermal contact. Heating tion contribution to the overall heat flux. The heat flux across the
power is supplied by a DC power supply. Power input is calcu- heated region q is written as the uniform heat input q ref less con-
lated by measuring the heater resistance and the amount of voltage
duction loss q ⬘k to the walls and radiation loss q r in the channel:
supplied to the heaters. Heater resistance deviates less than 0.5
percent under the temperature range considered. q⫽q ref⫺q ⬘k ⫺q r . (2)
Heated section temperature measurements are obtained by em-
bedding 0.25 mm diameter Type K thermocouples using OME- Conduction losses are calculated using the current conjugate heat
GABOND high thermal conductivity cement. A total of five ther- transfer model. Radiative heat flux is calculated using a net radia-
mocouples are centered in the plate, placed 3.31 cm apart along tive enclosure model 关11兴. Calculations show that radiative effects
the longitudinal centerline and at half the plate thickness. Two contribute between 2.5 percent to 7.5 percent of q ref , with the
more thermocouples are centered along the plate’s transverse cen- largest q r contribution occuring at the highest heated section tem-
terline, spaced 5.08 cm apart, to measure transverse temperature perature. Furthermore, q r remains fairly constant 共⬍5 percent de-
distributions within the plate. Since the embedded thermocouples viation兲 over the range of temperatures considered, and can be
are placed at half the plate thickness, a one-dimensional conduc- easily factored out as a constant in the results. Since these results
tion calculation is performed to obtain the surface temperature. show that convective and conductive heat transfer components
Two upstream and three downstream thermocouples are mounted dominate heat transfer, radiative effects are not included in this
on the bottom plexiglass wall with similar uniform spacing. A study.
thermocouple probe consisting of eight thermocouples mounted
Uncertainty Analysis. Many experiments were repeated to
on a rack is used to measure gas phase temperatures. On the rack,
ensure that the results and imposed conditions were accurate and
the bottom four thermocouples are spaced 3.2 mm apart with the
repeatable. An uncertainty in repeatability of less than 10 percent
bottom thermocouple 6.4 mm from the bottom wall. The next
was found for the results presented here. The uncertainty in the
three thermocouples are spaced 6.4 mm apart, and the top ther-
data is also estimated to be of this order based on the inaccuracies
mocouple is 6.4 mm from the top wall.
in the measuring instrumentation.
A vertical light sheet is passed through the top of the channel
Other uncertainties arising in the experimental study include the
for flow visualization. A side view is obtained by centering the
measurement of dimensional values (H, L, X sep), heat fluxes,
sheet along the longitudinal direction, giving a two-dimensional
flow rates and temperature values. Embedded thermocouples fluc-
view of flow traveling down the channel. The image is captured
tuate less than ⫾0.1°C, while gas phase temperature readings are
using a CCD camera, and video editing software 共ADOBE Pho-
within ⫾0.25°C. All thermocouples are calibrated to within
⫾0.1°C using a constant water bath at temperatures bounding the
range encountered in this study 共300 K–400 K兲. Thermocouples
mounted in the spanwise direction recorded less than ⫾0.1°C
from the centerline temperature, revealing negligible heat loss in
the spanwise direction. Spanwise flow visualization revealed neg-
ligible flow variation across the heated plate when measurements
are taken at least one channel height away from the sidewalls. It is
estimated that temperature and other measurements account for
less than 10 percent uncertainty in experimental measurements.
Fig. 2 Cross sectional view of the heated region „susceptor… The majority of uncertainty arises from material property evalua-
assembly tion, where uncertainties up to 15 percent have been observed for
some materials used in this study. These uncertainties will have a wall covered with fiberglass insulation and convective losses from
significant impact on experimental parameter calculations 共Re, Gr, the insulation surface to the ambient 共Fig. 3共b兲兲, resulting in a
Pr, K兲 and on consequent comparisons with numerical predictions. convective boundary condition of the form
Top Wall:
⳵␪
⳵y
冏 top
⫽⫺Bitop␪ top . (7)
Numerical Model
The value of Bitop is calculated by considering the total thermal
The full elliptic governing equations for a two-dimensional
resistance arising from a 9.5 mm thick plexiglass wall (k
steady-state laminar flow in a horizontal channel with constant
⫽0.25 W/mK), 50.8 mm thick fiberglass insulation (k
properties are considered. The two-dimensional assumption is
⫽0.04 W/mK), and natural convection heat transfer from the in-
valid if the channel aspect ratio is large and buoyancy effects do
sulation surface into the environment approximated by h
not dominate 共Gr/Re2 small兲, as verified by observing the corre-
⫽5.0 W/m2K 关14兴. The resulting total heat transfer coefficient at
sponding two-dimensional flow in the experimental study. In the
present study, buoyancy effects are considered using the Bouss- the top boundary is calculated as h top⫽0.66 W/m2K, correspond-
inesq approximations 关12兴. The dimensionless governing equa- ing to Bitop⫽1.27.
tions for continuity, momentum, and energy are The lower boundary is a no-slip, impermeable solid material of
finite thickness (H s ). It is composed of three sections as shown in
⳵u ⳵v Fig. 3共a兲. To replicate experimental conditions, thermal contact
⫹ ⫽0 (3) resistance between each section is prescribed a 1 mm air gap (K
⳵x ⳵y
⫽1.0). The heated region’s 共susceptor兲 surface is electrically
⳵x
⫹
⳵y
⫽ 冋
⳵ 共 u ␾ 兲 ⳵ 共v ␾ 兲 1 ⳵ 2 ␾ ⳵ 2 ␾
⌫ ⳵x2 ⳵y 册
⫹ 2 ⫹S 共 x,y 兲 , (4)
heated from the bottom by resistance heaters, and approximated
by a constant source of uniform heat flux (q ref) as shown in Fig.
3共c兲. The insulation sections to the left and right of the heated
where the general dependent variable ␾, the effective diffusion region are solved using the conjugate model, while heat loss
coefficient 1/⌫, and the source term S(x,y) for momentum and through the remaining insulation material at y⬍0, composing of
冋册
energy conservation are defined as plexiglass and fiberglass 共Fig. 3共d兲兲, is approximated using a total
heat transfer coefficient. At the y⫽0 boundary, convective heat
⳵P loss into the ambient occurs through a 28.6 mm thick plexiglass
冋册冋册
⫺
u Re ⳵x wall (k⫽0.25 W/mK), 50.8 mm thick fiberglass insulation (k
⫽0.04 W/mK), and convectively from the outer insulation sur-
␾⫽ v , ⌫⫽ Re , S 共 x,y 兲 ⫽ ⳵P Gr . (5) face, as approximated by h⫽5.0 W/m2K representing natural con-
␪ ⫺ ⫹ ␪
RePr ⳵ y Re2 vection to the ambient 关14兴. The total heat transfer coefficient at
0 the bottom boundary is calculated as h insul⫽0.63 W/m2K, corre-
sponding to Biinsul⫽1.22. In summary, the thermal boundary con-
Nondimensional variables are scaled using the channel height H, ditions for the bottom wall are
冏
free stream velocity U ⬁ , reference heat flux q ref , and properties
evaluated at the reference temperature T ref . Thermal buoyancy ⳵␪
Heated Region: ⫽⫺1/K sus (8)
effects are represented by the mixed convection parameter, ⳵y sus
Gr/Re2, appearing in the source term of the second momentum
equation, Eq. 共5兲. In the energy conservation equation, an order of
magnitude analysis 关12兴 shows that viscous dissipation and pres-
Insulation:
⳵␪
⳵y
冏 insul
⫽
Biinsul
␪
K insul bottom
, (9)
sure work are negligible, and are, therefore, not included in this
study. where Biinsul⫽h insulH/k insul , and ␪ bottom is the temperature at the
bottom wall (y⫽0). The ends of the insulation section are ex-
Conjugate Modeling. Conjugate heat transfer couples heat posed to the ambient, hence are assumed to be at the ambient
transfer in the solid and fluid regions. The present problem con- temperature ( ␪ ⫽0).
siders heat transfer through the gas and bottom wall of thickness The flow exiting the channel is quite complex due to flow re-
H s . Conjugate effects due to fiberglass insulation and plexiglass circulation and entrainment of ambient gas from the exit. In such
walls are accounted for using convective boundary conditions cases, gas exiting the channel is assumed to be dominated by
with total heat transfer coefficients. Energy transfer in the fluid is convection heat transfer, hence the fully developed thermal and
governed by Eq. 共4兲. The dimensionless energy equation for the velocity boundary condition is prescribed. If gas is entrained into
solid region is given as the channel at the exit (u⭐0), it enters at the ambient temperature
冋册
under fully developed flow conditions 关15兴. In summary, the exit
K ⳵ 2␪ ⳵ 2␪
⫹ ⫽0. (6) boundary conditions are
RePr ⳵ x 2 ⳵ y 2
⳵u
Interface conditions are derived by performing an energy balance ⫽ v ⫽0
⳵x
across the solid-fluid interface. Using these equations, energy
transport in the entire region, including solid and gas regions, is ␪ ⫽0 if u⭐0 共 inflow兲 (10)
solved simultaneously.
In the non-conjugate model, only the fluid region is considered ⳵␪
共i.e., H s ⫽0兲. Conjugate boundary conditions 共Eqs. 7 to 9兲 are still ⫽0 if u⬎0 共 outflow兲 .
⳵x
used, but surface heating input goes directly into convectively
heating the gas flow, as achieved by specifying K sus⫽K insul⫽1. The inflow temperature was experimentally measured as the am-
bient temperature. This technique is verified to ensure overall
Boundary Conditions. The velocity of the flow entering the mass conservation in the computational domain. Pressure bound-
channel is prescribed as the analytically obtained parabolic profile ary conditions are not needed since the SIMPLER algorithm 关16兴
u⫽6(y⫺y 2 ) 关13兴. The flow enters at the ambient temperature calculates pressure values directly from the velocity field. Since
T ⬁ ( ␪ ⫽0). The inlet flow conditions are experimentally verified pressure differences are calculated in this study, the ambient pres-
by velocity and temperature measurements at the inlet. The upper sure is specified at one point at the inlet to eliminate drifting of the
boundary is a no-slip, impermeable wall consisting of a plexiglass absolute pressure.
Fig. 3 Boundary conditions and parameters used in the numerical model. The heated region is
labeled as the susceptor.
Numerical Method. The numerical model solves the coupled by verifying mass and energy conservation to less than 1 percent
non-linear partial differential equations governing continuity, mo- discrepancy. The independence of other user prescribed variables,
mentum, and energy conservation. A finite volume technique is such as initial guess, relaxation factors, outflow boundary condi-
used with a uniform staggered grid arrangement, while different tions, and channel length are varied for all cases to ensure that the
grid sizing of uniform spacing is used in the fluid and solid re- results are independent of values chosen.
gions. The algorithm solves for primitive variables in velocity
(u, v ), pressure 共p兲, and temperature 共␪兲. Since momentum and Results and Discussion
heat transfer are strongly coupled, the continuity 共pressure兲, mo-
mentum, and energy conservation equations are solved simulta- Experimental and numerical results are presented for a horizon-
neously. This coupling arises through buoyancy effects and con- tal channel heated from below. The channel is 5.08 cm high by 28
jugate boundary conditions. The resulting non-linearly coupled cm long with a 12 cm long entrance length defined as insulation in
system is solved using Successive Over Relaxation 共SOR兲 for the Fig. 3. Consider a system where air enters the chamber at a
pressure equation and the Tridiagonal Matrix Algorithm 共TDMA兲 freestream velocity of U ⬁ ⫽1.24 cm/s. The bottom wall is 3.18
for the remaining equations. Under-relaxation values of 0.1–0.4 mm thick, while the heat region material is ceramic (k
are used during iteration of the linearized equations 关17兴. ⫽1.26 W/mK) and the insulation material is composed of plexi-
Typical cases use a grid size of 280⫻40 in the fluid region and glass (k⫽0.25 W/mK). The heated region is heated uniformly at
10 grid points per H s ⫽0.10 thickness in the solid region. The grid q ref⫽12.7 W/m2, yielding T ref⫽25.3 K. Transport properties are
size is chosen based on solution accuracy and computational ef- evaluated at 300 K as ␯ ref⫽15.89⫻10⫺6 m2/s, k ref⫽26.3
fort. Grid density is increased until the converged solution differs ⫻10⫺3 W/mK, and ␣ ref⫽22.5⫻10⫺6 m2/s. Using these values,
by less then 1 percent between sizes. This procedure ensures the nondimensional parameters for the reference case are calculated
independence of results on grid size. Iterative convergence is as Re⫽39.8, Pr⫽0.7, Gr⫽4.3⫻105 , K sus⫽47.9, K insul⫽9.5, and
achieved when two consecutive comparisons of the calculated val- H s ⫽0.125.
ues do not indicate significant differences (⬍10⫺6 ). Convergence The flow field is studied by comparing the experimentally ob-
is ensured by comparing intermediate flow and thermal fields and served flow field to numerical predictions. The study then pro-
Fig. 4 Comparison between experimental observations and numerical predictions of
streamlines for a ceramic heated region at „a… ReÄ9.48 and „b… ReÄ29.7
ceeds to compare numerically calculated and experimentally mea- of flow separation兲, as given by X sep . Figure 5 shows predicted
sured temperature distributions in the channel for different heat X sep compared to experimentally measured values for a range of
region materials and flow rates. A parametric study of several mixed-convection parameter 2.71⫻102 ⭐Gr/Re2⭐6.79⫻104 .
conjugate heat transfer parameters, namely K sus , K insul , and H s , With the exception of the numerical data point at Gr/Re2⫽2.71
are performed to quantify their impact on temperature uniformity ⫻102 , there is good agreement between numerical and experi-
and heat transfer at the heated region’s surface, yielding better mental results. Discrepancy between numerical and experimental
physical insight and provide quantitative information for the ther-
X sep at Gr/Re2⫽2.71⫻102 is due to inherent difficulty in match-
mal design of thermal manufacturing systems.
ing exit boundary conditions to experimental conditions. When
Flow Field. Predicted streamlines are compared to experi- the plume observed in Fig. 4 reaches near the exit, flow across the
mental observations for flow of air over a region heated by a outlet deviates from the fully developed conditions given by Eq.
uniform heat flux source. The comparison is presented for the 共10兲. Other numerical outflow boundary conditions, such as zero-
reference case with two flow rates given by Re⫽9.48 and 29.7, as diffusion and different channel lengths, have been investigated.
shown in Fig. 4. As the gas travels over the heated region (6.0 The zero-diffusion outflow boundary condition is a less stringent
⭐x⭐9.25), gas heating causes buoyancy effects to arise. In the form of Eq. 共10兲, where a zero second gradient is defined for u, v ,
case of low Re (Re⫽9.48), a plume develops above the heated and ␪. This choice of boundary condition assumes negligible dif-
region. This flow pattern generates two transverse rolls, with their fusion across the boundary, but does not demand fully developed
axis of rotation perpendicular to the flow direction. The upstream flow at the exit. Both methods have been investigated without
roll produces a recirculation region, while the downstream roll success. Since it was experimentally observed that the plume in-
entrains flow from outside. These rolls can have significant alter- tensifies across the channel exit under these conditions, a possible
cations to the surface temperature and heat transfer rates. In solution is to extend to numerical domain to include flow into the
chemical vapor deposition 共CVD兲, for example, deposition rate environment. This option was not pursued in this study due to
and film uniformity can be affected by the convective displace- computational limitations.
ment of reactants. As Re is increased 共Fig. 4共b兲兲, the plume shifts Parameters in the conjugate model characterizing heat transfer
downstream due to bulk gas flow. Consequently, the two trans- in the solid material, namely K sus , K insul , and H s , are systemati-
verse rolls become smaller and flow entrainment from the outside cally varied in order to investigate the generation of transverse
is reduced. The appearances of oscillatory flow, turbulent flow, rolls. For each case, the separation distance relative to the non-
transverse and longitudinal rolls are observed to be dependent on conjugate model X sep /X sep,nc is numerically calculated and plotted
the channel cross-sectional aspect ratio, flow rate and heating rate. in Fig. 6. Reference conditions are prescribed to the remaining
These findings are reported in Chiu et al. 关18兴. operating parameters. Results show that a variation in the heated
The plume location is measured by calculating the distance region’s thermal conductivity (K sus) does not significantly alter
from the heated region’s leading edge to the plume’s center 共point X sep . This is expected since a thin solid material yields low ther-
Temperature Field. Numerically predicted temperature
fields shown in Fig. 7 are compared to experimental measure-
ments. Figure 7共a兲 shows streamlines overlaying isotherms for the
reference case with Re⫽9.48 and an aluminum heated region
(K sus⫽6730). It demonstrates the coupling of strong mixed con-
vection effects with conjugate heat transfer modeling to predict a
highly isothermal aluminum heated region in the wall and flow
field comparable to experimental observations. Further informa-
tion may be obtained by examining horizontal temperature pro-
files in the channel.
Corresponding temperature distributions at various y locations
are obtained and plotted with experimental data in Fig. 7共b兲. At
temperature distributions above the surface (y⬎0.0) in Fig. 7共b兲,
the thermal plume produces a peak in the temperature profile near
x⬇8.25. This trend is also observed in the experimental data,
reinforcing the validity of numerical model in gas phase tempera-
ture predictions. Horizontal asymmetry in temperature profiles is
caused by bulk gas flow through the channel. The temperature
field shows that the heated region remains at high temperature
relative to the insulation material, resulting in the generation of a
thermal plume above the heated region surface. Due to alumi-
num’s high thermal conductivity, the heated region surface tem-
Fig. 5 Thermal plume generation as characterized by flow perature (6.0⭐x⭐9.25) is highly uniform, as indicated by the last
separation at various GrÕRe2 temperature distribution (y⫽0.0) in Fig. 7共b兲. Comparison of ex-
perimental measurements and numerical predictions show good
agreement on the wall and gas phase temperature measurements,
mal resistance, thereby allowing convection to be the dominant with less than 5 percent deviation at the heated region’s surface.
mode of heat transfer. As the insulation thermal conductivity is When low thermal conductivity materials such as ceramics are
increased, the energy is redirected into the surrounding insulation. used for the heated region, the temperature characteristics differ
Small X sep deviations occur at K insul⬍2.5 because gas heating significantly from the aluminum case, as shown in Fig. 8. In this
primarily occurs when the fluid flows over the heated region. At comparison, the surface temperature distribution (y⫽0.0) for a
K insul⬎2.5, significant heat loss to the insulation results in reduced ceramic heated region (K sus⫽47.9) at two different flow rates
thermal buoyancy, hence increasing the separation distance. At (Re⫽9.48 and 39.8兲 is shown, with less than 12 percent deviation
very large K insul , heat transfer is limited to the heated region’s between numerical and experimental values. A low thermal con-
conduction toward the insulation, resulting in a leveling of the ductivity heated region material reduces the surface temperature
curve. For small solid material thickness (H s →0), the conjugate uniformity. This is due to a number of factors, such as increased
model approaches the non-conjugate case, as expected. When the heat transfer to the insulation and poor thermal diffusivity charac-
solid material thickness is increased, enhanced heat transfer teristics of low K sus materials. Discussion and explanation of this
through the insulation reduces heat transfer into the gas flow, phenomenon will be presented later. At increased flow rates, the
thereby reducing buoyancy effects and delaying plume separation, flow field 共Fig. 4兲 shows a gradual transition where the two trans-
as indicated by an increase in X sep . verse rolls created by the plume decrease in size as the plume
shifts downstream. In consequence, enhanced cooling will de-
crease the heated region’s temperature, as shown in Fig. 8. Dis-
crepancies at x⬎9.25 are again due to difficulties in matching the
exit boundary conditions to experimental conditions, as discussed
earlier. The numerical and experimental cases compare well at
other flow rates as well.
Parametric Study. Operational parameters describing the
heated region and surrounding materials, namely K sus , K insul and
H s have a direct impact on conjugate heat transfer. This section
presents a systematic numerical investigation of these three pa-
rameters in detail. The parameters are varied in the range 1.0
⭐K sus , K insul⭐6730, and 0.0⭐H s ⭐1.0, where H s ⫽0.0 repre-
sents the nonconjugate case. The effect of each parameter on the
heated region’s surface temperature and uniformity is character-
ized by the average heated region surface temperature ␪ avg and
heated region surface temperature deviation of the mean ␪ ␴ , re-
spectively. Besides K sus , K insul , and H s , other operational param-
eters are kept at the reference values.
The significance and importance of conjugate heat transfer
modeling is demonstrated in Fig. 9, where the wall surface tem-
perature predicted using a non-conjugate model is compared to
temperature predictions using conjugate heat transfer modeling
under similar conditions. The conjugate parameters used in this
comparison are H s ⫽0.125, K insul⫽9.5 共plexiglass兲 and various
heated region thermal conductivities. Heated region conductivities
Fig. 6 Flow separation dependence on conjugate parameters. of K sus⫽9.5, 47.9, 100, 500, and 6730 are typical values for plexi-
Flow separation predicted by the non-conjugate model is glass, ceramics, stainless steel and aluminum, respectively. When
X sep,ncÄ4.67. heated by a uniform heat flux source, which is typical with radia-
Fig. 7 „a… Streamline and temperature contours and „b… experimental temperature data
comparison „symbols… with numerically predicted results for an aluminum heated region
tion or electrical heating, the non-conjugate model predicts a gion’s bottom, the non-conjugate model directs all available ther-
highly nonuniform wall temperature profile. The temperature im- mal energy into the gas phase. Conjugate modeling demonstrates
mediately upstream of the heated region section is heated above that thermal energy also travels through the solid material, even-
the ambient temperature ( ␪ ⬎0) due to axial diffusion of thermal tually being lost to the ambient or reintroduced into the channel at
energy. A slight dip in the wall temperature at x⬇11.0 is due to locations other than the heated region’s surface. Since these pro-
the presence of a plume, as discussed previously. When conjugate cesses are highly temperature sensitive, these issues must be re-
effects are considered, thermal energy transport besides convec- solved in order to improve temperature and heat transfer rate pre-
tion heat transfer into the flowing gas becomes increasingly im- dictions. In CVD processing, these quantities are strongly related
portant. As the heated region conductivity (K sus) is increased, not to film quality, uniformity and productivity.
only does thermal energy redistribute itself within the heated re- In order to evaluate the importance of heated region surface
gion to produce a highly uniform temperature distribution, ther- temperature predictions using conjugate heat transfer modeling,
mal energy loss to the insulation also increase. In consequence, the average heated region temperature ␪ avg predicted at various
paths of thermal energy transfer are manifold compared to the K sus , K insul , and H s are compared to the non-conjugate prediction
non-conjugate model. Under uniform heating of the heated re- ( ␪ avg,nc) by using the ratio ␪ avg / ␪ avg,nc . Using this ratio, the effect
Fig. 8 Surface temperature comparisons for a ceramic heated
region
Fig. 10 Deviation of average heated region temperature from
the non-conjugate case „ ␪ avg,ncÄ0.229…
of increasing K sus is shown in Fig. 10. The figure shows a result-
ant increase in the average temperature due to enhanced heat
transfer of the heated region material coupled with limited con- solid material thickness is increased incrementally from H s ⫽0.0
vective heat transfer to the gas flow. Furthermore, Fig. 11 shows a 共non-conjugate兲 to a thickness of one channel height (H s ⫽1.0),
dramatic drop in temperature deviation ␪ ␴ across the heated reheat loss through the insulation is the major contributor to a large
gion’s surface when K sus is increased. A decease in ␪ ␴ is indica- drop in ␪ avg and decreasing heated region temperature deviation
tive of a high thermal conductivity material’s ability to redistrib- ( ␪ ␴ ), as indicated in Figs. 10 and 11. These aspects are very
ute thermal energy, thus producing a nearly isothermal surface. important when a process is temperature sensitive, and when pre-
Low K sus materials exhibit highly non-isothermal behavior, result- dictable and controllable temperatures are desirable.
ing in large temperature deviations across the heated region’s sur- A further understanding of conjugate effects is achieved by
face. quantifying the amount of energy transfer into the gas phase under
An increase in insulation thermal conductivity (K insul) allows conjugate conditions for a variety of parameters K sus , K insul , and
for thermal energy loss to the insulation, resulting in the reduction H s . Figure 12 compares total convective heat transfer across the
of ␪ avg 共Fig. 10兲 and increased ␪ ␴ 共Fig. 11兲. In consequence, a heated region surface predicted using conjugate modeling and
temperature drop occurs at locations where the heated region is in non-conjugate modeling. Comparison to non-conjugate predic-
contact with the insulation; a phenomena known as an ‘‘edge tions is achieved by scaling the total convective heat transfer to
effect.’’ This phenomenon, although important particularly near
the heated region’s edges, is fairly insensitive to variations in
K insul , as indicated by small variations in ␪ ␴ in Fig. 11. When the
Fig. 11 Heated region temperature uniformity dependence on

Fig. 9 Effect of heated region „susceptor… material on surface solid material thickness „ H s …, heated region thermal conductiv-
temperature distribution ity „ K sus… and insulation thermal conductivity „ K insul…
In the experimental study, the gas flow field, temperature distri-
bution and paths of heat transfer are examined for different con-
ditions. The ranges of operational parameters are further extended
to numerically examine conjugate effects by considering a variety
of materials and wall thicknesses and studying their effects on
temperature uniformity and heat transfer.
Experimental data obtained in this study provides valuable in-
sight into the physical phenomenon which governs conjugate heat
transfer. The flow and temperature field in a heated horizontal
channel with thermally conducting walls of different materials can
be quite complicated. This is demonstrated experimentally by the
appearance of a two-dimensional thermal plume and transverse
roll structures, and by large deviations in experimentally mea-
sured temperature distributions for different heated region materi-
als. The aluminum heated region exhibited nearly isothermal con-
ditions, while a ceramic plate yielded highly non-uniform
temperature along its surface, illustrating the importance of con-
jugate effects in system material selection. Comparisons of ex-
perimental results with numerical predictions show good
agreement.
A parametric study of operational parameters related to the con-
jugate heat transfer 共K sus , K insul , and H s 兲 revealed that the addi-
tion of conjugate heat transfer significantly affects the temperature
and heat transfer rates at the heated region’s surface. In conse-
Fig. 12 Fraction of thermal energy across the heated region quence, channel wall design and modeling of thermal processes
surface as compared to non-conjugate modeling
with high temperature sensitivity require proper treatment of con-
jugate heat transfer effects. Situations where non-conjugate mod-
eling is satisfactory are also discussed in detail.
the non-conjugate value q/q nc . For example, since the non-
conjugate model has a zero-thickness wall, heat input goes di-
rectly into the gas flow, i.e., q/q nc⫽1.0. When conjugate effects Acknowledgments
are considered, alternate routes of thermal energy transfer become Computational resources are provided by the National Partner-
significant, and q/q nc becomes less than unity. Increasing K sus or ship for Advanced Computational Infrastructure 共NPACI兲 under
K insul reduces convection heat transfer across the heated region’s Grant No. RUT206 and the High Performance Computing Project
surface, and thermal energy reroutes into the insulation. In the 共HPCP兲 of the Information Sciences Strategic Planning Commit-
range of K sus considered, convective heat transfer into the channel tee at Rutgers University. The authors thank Professor N. G. Glu-
(q/q nc) drops as much as 12 percent with highest thermal losses mac for many helpful discussions. The authors also wish to ac-
occurring when an aluminum heated region (K sus⫽6730) pro- knowledge partial support provided by the Center for
duces a hot 共high ␪ avg兲 isothermal heated region. Heat losses of 8 Computational Design 共CCD兲 at Rutgers University. The help
percent occured at K insul⫽1, and has slightly increased at K insul provided by Troy Kim in some of the experiments is also ac-
⫽6730, indicating that K insul has a fairly weak effect on heat loss knowledged.
through the insulation. However, the inclusion of insulation con-
jugate heat transfer is necessary to account for a significant
amount of thermal energy loss 共up to 9 percent兲 in the solid ma- Nomenclature
terial and resultant edge effects on the heated region. When the Bi ⫽ Biot number (⫽hH/k)
solid material thickness is increased up to H s ⫽1.0, thermal en-
ergy loss through the insulation becomes very significant, with as
Gr ⫽ Grashof number (⫽g ␤ T refH 3 / v 2f )
much as 28 percent of the thermal energy being lost to the insu- Gr/Re2 ⫽ mixed convection parameter (⫽g ␤ T refH/(U ⬁ ) 2 )
lation at H s ⫽1.0. H ⫽ channel height
These observations indicate that the inclusion of conjugate heat h ⫽ heat transfer coefficient
transfer effects is necessary, as evident from a significant 共up to K ⫽ thermal conductivity ratio (⫽k/k f )
12 percent兲 drop in convective heat transfer due to material prop- k ⫽ thermal conductivity
erty variations in the heated region and the insulation L ⫽ channel length
(K sus ,K insul), and as much as 28 percent of the thermal energy P ⫽ local pressure in the fluid
input is lost through the insulation due to H s variations. Numeri- Pr ⫽ Prandtl number (⫽ v f / ␣ f )
cal models incorporating conjugate heat transfer are needed to q ⫽ heat flux
capture these effects. For certain cases, especially where the walls Re ⫽ Reynolds number (⫽U ⬁ H/ v f )
are very thin, and low K insul material is used to contain thermal T ⫽ temperature
energy in the heated region, conjugate modeling may not be criti- T ref ⫽reference temperature difference (⫽q refH/k ref)
cal. In such instances, a high K sus material may be approximated U⬁ ⫽ free stream velocity
as an isothermal surface, while low K sus material may be modeled u, v ⫽ velocity components along the x and y axes, respec-
using a prescribed heat flux condition. Care must be taken in tively
choosing the appropriate model, and a detailed investigation in- X sep ⫽ flow separation distance from the heated region’s
corporating conjugate heat transfer modeling may be necessary in leading edge
order to identify the appropriate design variables. x, y ⫽ dimensionless coordinates 共scaled by H兲
Greek Letters
Conclusions ␣ ⫽ thermal diffusivity (⫽k/ ␳ c p )
A detailed experimental and numerical study is carried out to ␤ ⫽ coefficient of thermal expansion (⫽⫺1/␳⳵␳ / ⳵ T 兩 p )
investigate conjugate heat transfer in horizontal channels with a ⌫ ⫽ effective diffusion coefficient
heated section, and its impact on thermal processing techniques. ␯ ⫽ kinematic viscosity
␪ ⫽ dimensionless temperature (⫽(T⫺T ⬁ )/T ref) 关6兴 Mahajan, R. L., 1996, ‘‘Transport Phenomena in Chemical Vapor-Deposition
␾ ⫽ generalized dependent variable Systems,’’ Adv. Heat Transfer, 28, pp. 339–425.
关7兴 Lin, P., and Jaluria, Y., 1998, ‘‘Conjugate Thermal Transport in the Channel
Subscripts of an Extruder for Non-Newtonian Fluids,’’ Int. J. Heat Mass Transf., 41, pp.
3239–3253.
⬁ ⫽ ambient or free stream 关8兴 Chinoy, P. B., Kaminski, D. A., and Ghandhi, S. K., 1991, ‘‘Effects of Ther-
avg ⫽ arithmetic mean mal Radiation on Momentum, Heat, and Mass Transfer in a Horizontal Chemi-
insul ⫽ insulation cal Vapor Deposition Reactor,’’ Numer. Heat Transfer, Part A, 19, pp. 85–
nc ⫽ obtained from non-conjugate model 100.
关9兴 Durst, F., Kadinski, L., and Schäfer, M., 1995, ‘‘A Multigrid Solver for Fluid
ref ⫽ reference value Flow and Mass Transfer Coupled with Grey-Body Surface Radiation for the
s ⫽ solid Numerical Simulation of Chemical Vapor Deposition Processes,’’ J. Cryst.
sus ⫽ susceptor 共heated region兲 Growth, 146, pp. 202–208.
top ⫽ top wall 关10兴 Fotiadis, D. I., Kieda, S., and Jensen, K. F., 1990, ‘‘Transport Phenomena in
Vertical Reactors for Metalorganic Vapor Phase Epitaxy,’’ J. Cryst. Growth,
102, pp. 441–470.
References 关11兴 Siegel, R., and Howell, J. R., 1992, Thermal Radiation Heat Transfer, Taylor
关1兴 Fedorov, A. G., and Viskanta, R., 2000, ‘‘Three-Dimensioanl Conjugate Heat & Francis, Philadelphia, PA.
Transfer in the Microchannel Heat Sink for Electronic Packaging,’’ Int. J. Heat 关12兴 Gebhart, B., Jaluria, Y., Mahajan, R. L., and Sammakia, B., 1988, Buoyancy-
Mass Transf., 43, pp. 399–415. Induced Flows and Transport, Taylor & Francis, Philadelphia, PA.
关2兴 Heindel, T. J., Ramadhyani, S., and Incropera, F. P., 1995, ‘‘Conjugate Natural 关13兴 Panton, R. L., 1984, Incompressible Flow, John Wiley & Sons, Inc., NY.
Convection from an Array of Discrete Heat Sources: 1. Two- and Three- 关14兴 Incropera, F. P., and DeWitt, D. P., 1990, Fundamentals of Heat and Mass
Dimensional Model Validation,’’ Int. J. Heat Mass Transf., 16, pp. 501–510. Transfer, John Wiley & Sons, Inc., NY.
关3兴 Sugavanam, R. A., Ortega, A., and Choi, C. Y., 1995, ‘‘A Numerical Investi- 关15兴 Ku, A. C., Doria, M. L., and Lloyd, J. R., 1976, ‘‘Numerical Modeling of
gation of Conjugate Heat Transfer from a Flush Heat-Source on a Conductive Buoyant Flows Generated by Fire in a Corridor,’’ Proc. 16th Symp. 共Int.兲 on
Board in Laminar Channel Flow,’’ Int. J. Heat Mass Transf., 38, pp. 2969– Combustion, Combustion Institute, pp. 1372–1384.
2984. 关16兴 Patankar, S. V., 1980, Numerical Heat Transfer and Fluid Flow, Taylor &
关4兴 Aronov, B., and Zvirin, Y., 1999, ‘‘A Novel Algorithm to Investigate Conju- Francis, Philadelphia, PA.
gate Heat Transfer in Transparent Insulation: Application to Solar Collectors,’’ 关17兴 Jaluria, Y. and Torrance, K. E., 1986, Computational Heat Transfer, Hemi-
Numer. Heat Transfer, Part A, 35, pp. 757–777. sphere Publishing Corporation, USA.
关5兴 Chiu, W. K. S., Glumac, N. G., and Jaluria, Y., 2000, ‘‘Numerical Simulation 关18兴 Chiu, W. K. S., Richards, C. J., and Jaluria, Y., 2000, ‘‘Flow Structure and
of Chemical Vapor Deposition Processes Under Variable and Constant Prop- Heat Transfer in a Horizontal Converging Channel Heated from Below,’’
erty Approximations,’’ Numer. Heat Transfer, Part A, 37, pp. 113–132. Phys. Fluids, 37, pp. 2128–2136.
Conjugate Mixed Convection With
C. Gururaja Rao
Research Scholar and Lecturer
Surface Radiation From a Vertical
Department of Mechanical Engineering,
Regional Engineering College,
Warangal—506 004 (AP), India
Plate With a Discrete Heat Source
The results of a numerical study of the problem of two-dimensional, steady, incompress-
C. Balaji ible, conjugate, laminar, mixed convection with surface radiation from a vertical plate
Assistant Professor with a flush-mounted discrete heat source are reported. The governing equations, written
in vorticity-stream function form, are solved using a finite-volume based finite difference
S. P. Venkateshan method. A hybrid grid system has been employed for discretization of the computational
Professor domain. The effects of (i) the magnitude and location of the heat source, (ii) the material
e-mail: spv46@usa.net and surface properties of the plate, and (iii) the free-stream velocity on both heat transfer
and fluid flow have been studied. Based on a large set of (more than 550) numerical data,
Heat Transfer and Thermal Power Laboratory, correlations have been developed for maximum and average non-dimensional plate tem-
Department of Mechanical Engineering, peratures and mean friction coefficient. A method for evaluating the forced convection
Indian Institute of Technology Madras, mean friction coefficient component, which may be used in estimating the power input
Chennai-600 036, India required for maintaining the flow, has been proposed. 关DOI: 10.1115/1.1373654兴
Keywords: Conjugate, Finite Difference, Heat Transfer, Mixed Convection, Radiation
Introduction vorticity-stream function 共␻-␺兲 form and are then normalized. As

the present problem involves ‘‘conjugate’’ heat transfer, an ‘‘ob-
Interaction of different modes of heat transfer continues to be a
vious’’ reference temperature difference does not exist. Hence, a
topic of interest because of its application in several areas, like the
‘‘modified’’ reference temperature difference is introduced as
cooling of electronic equipment. Vertical board mounted elec-
⌬T ref⫽(q v L h t/k s ). The normalized governing equations are
tronic components are cooled by the removal of the heat generated
in the components, with air as one of the promising cooling me-
dia. Zinnes 关1兴 presented both numerical and experimental results
of the problem of interaction of conduction with natural convec-
U
d␻
⳵X
⫹V
⳵␻
⳵Y
⫽⫺RiL*
⳵␪
⫹
1 ⳵ 2␻ ⳵ 2␻
⫹
⳵ Y ReL ⳵ X 2 ⳵ Y 2 冋册 (1)
tion from a vertical plate with arbitrary surface heating, and ⳵ 2␺ ⳵ 2␺

showed that the degree of coupling between plate conduction and ⫹ ⫽⫺ ␻ (2)
⳵X2 ⳵Y 2
natural convection in the fluid is influenced by the plate-fluid ther-
mal conductivity ratio. Gorski and Plumb 关2兴 numerically inves-
tigated the problem of conjugate heat transfer from a single dis-
crete heat source, flush-mounted in a flat plate. Here, the problem
U
⳵␪
⳵X
⫹V
⳵␪
⫽
1 ⳵ 2␪ ⳵ 2␪
⫹ 冋
⳵ Y PeL ⳵ X 2 ⳵ Y 2
. 册 (3)
was solved using the well-known Blasius velocity profile for lami- Computational Domain
nar forced convection as the input, and a correlation relating the
average Nusselt number to Peclet number, the heat source size Based on an earlier work of the present authors, Gururaja Rao
and the length ratio was developed. Hossain and Takhar 关3兴 nu- et al. 关6兴, the computational domain in this case is extended be-
merically investigated the effect of radiation on mixed convection yond the trailing edge of the plate by a length equal to that of the
from a heated vertical plate with uniform free-stream and surface plate 共L兲, while the width 共W兲 of the domain is taken equal to the
temperatures. Cole 关4兴 addressed, numerically, the problem of plate length.
electronic cooling, from the perspective of scaling, applied to a
steady viscous flow over a heated strip on a plate. However, the Boundary Conditions
results for the problem of conjugate laminar mixed convection At the bottom, as the fluid enters with a uniform velocity, u ⬁ ,
with surface radiation from a vertical plate with a flush-mounted and temperature, T ⬁ , ( ⳵ ␺ / ⳵ Y )⫽1, ␻ ⫽0 and ␪ ⫽0. Along the
discrete heat source are not available in the literature. Hence, a plate, ␺ ⫽0 and ␻ ⫽⫺ ⳵ 2 ␺ / ⳵ Y 2 . Energy balance on the plate el-
detailed numerical analysis of this problem is reported here. ement, which is shown, enlarged, in the inset to Fig. 1, yields the
following equation for the plate temperature distribution, in the
Mathematical Formulation non-dimensional form:
冉冊冋冉冊册
The governing equations for two-dimensional, steady, incom-
⳵ 2␪ ⳵␪ T 4
pressible, laminar, mixed convection from a vertical plate are ⫹␥ ⫹A r 1 A r 2 ⫺␧ ␥ N RF ⫺1 ⫽0. (4)
available in a number of references, e.g., Bejan 关5兴. The schematic ⳵X2 ⳵Y Y ⫽0
T⬁
of the problem geometry, consisting of a vertical plate with a
Equation 共4兲 is valid in the region that contains the heat source. In
flush-mounted discrete heat source, is shown in Fig. 1. The left,
the region outside the heat source, the term A r 1 A r 2 ⫽0. Energy
top and bottom faces of the plate are insulated, while heat transfer
occurs from the right face, by convection 共free and forced兲 and balance on the bottom and top insulated ends will give the appro-
surface radiation. The governing equations are first converted into priate equations, depending on whether or not these two ends are
part of the heat source. With regard to the extended length of the
left boundary, because of symmetry, V⫽0, which means that
HEAT TRANSFER. Manuscript received by the Heat Transfer Division May 30, 2000; ( ⳵ ␺ / ⳵ X)⫽0, implying that ␺ is a constant. Since ␺ ⫽0 has been
revision received November 25, 2000. Associate Editor: R. Skocypec. taken along the plate, the same is used along the extended length
Fig. 1 Schematic of the problem geometry „inset showing an
enlarged plate element…
of the left boundary also. The vorticity, ␻ ⫽0 here. Because there

is no heat transfer across the extended length, ( ⳵ ␪ / ⳵ Y )⫽0. On the Fig. 2 Grid pattern used for a typical case along with bound-
top, the fully developed condition, ( ⳵ ␺ / ⳵ X)⫽0, has been used for ary conditions
␺. Since the domain is extended, ␻ ⫽0. When U is positive, the
fully developed condition ( ⳵ ␪ / ⳵ X)⫽0 is used for ␪, and when U
is negative, ␪ ⫽0. On the right, a mixed condition, ( ⳵ 2 ␺ / ⳵ X ⳵ Y ) Validation
⫽0, is imposed on ␺, while ␻ and ␪ are each taken to be zero. The fundamental difference between the boundary conditions
for laminar mixed convection from an isothermal vertical plate,
Method of Solution e.g., Gururaja Rao et al. 关6兴 and those used in the present work is
The governing Eq. 共1兲–共3兲 are transformed into finite difference the temperature variation along the plate. In the former, the tem-
equations using a finite-volume based finite difference method of perature was considered uniform along the plate, while here, the
Gosman et al. 关7兴 and are then solved using the Gauss-Seidel temperature varies along the plate, provided with a heat source,
iterative procedure. The details of the solution procedure are due to multi-mode heat transfer. Thus, the validation of results for
available in Gururaja Rao et al. 关6兴. A hybrid grid system is used the entire mixed convection regime, provided in Gururaja Rao
for discretizing the computational domain, keeping in mind the et al. 关6兴 with reference to both numerical and experimental re-
fact that the temperature distribution along the plate depends on sults available in the literature, will ensure the accuracy of the
the height and the position of the heat source. Based on a grid present results.
sensitivity test to be discussed in the ensuing section a grid size of
111⫻111 is chosen. The grid pattern used for a typical case is
shown in Fig. 2, which also shows all the boundary conditions. Table 1 Range of parameters †L Ä0.10 m, t Ä0.0015 m, L h
All the calculations are done for air (Pr⫽0.71) and the range of Ä0.0125 m‡
parameters used in the present work is listed in Table 1.

Grid Sensitivity Analysis. To study the effect of grid size
(M ⫻N) on the solution, a case with q v ⫽5⫻105 W/m3, A 1
⫽0.4375, k s ⫽0.25 W/m K, ␧⫽0.45, ReL⫽1275, and RiL* ⫽2 is
considered, and the results are shown in Table 2. The analysis is
made in two stages-first with M fixed and then with N fixed. The
results of the former show that the difference in ␪ max between grid
sizes 111⫻111 and 111⫻131 is 0.06 percent, while the difference
in C̄ f between the same grid sizes is 0.62 percent. The results of
the grid sensitivity with N fixed show that the differences in ␪ max
and C̄ f between grid sizes 111⫻111 and 131⫻111 are 0.07 per-
cent and 0.54 percent, respectively. In view of the above, M and N
have both been fixed as 111. The nodal number 111 is considered
as the basis after some initial studies with various other numbers.
Table 2 Grid sensitivity analysis †L Ä0.10 m, L h Ä0.0125 m, t ␧⫽0.85, ␪ max increases by as much as 51 percent as the heat
Ä0.0015 m, L l Ä0.04375 m, q v Ä5Ã105 WÕm3, k s Ä0.25 WÕm K, source shifts from A 1 ⫽0 to A 1 ⫽0.875. Further, ␪ max decreases
␧ Ä0.45, k f Ä0.0291 WÕm K, ReLÄ1275 and RiL* Ä2‡ with increasing ␧ for a given RiL* . The effect of ␧ on ␪ max is more
pronounced in the free convection dominant regime than in the
forced convection dominant regime, which is consistent with the
findings of earlier studies on natural convection-surface radiation
interaction in enclosures, e.g., Dehghan and Behnia 关8兴.
Heat Transfer Characteristics

Figure 4 shows the contributions of convection and radiation,
plotted against A 1 , for two values of RiL* 共25 and 0.1兲 and two
values of ␧ 共0.05 and 0.85兲, for the case with q v ⫽106 W/m3, k s
⫽0.25 W/m K. A notable feature is the behavior of the two curves
The results of the present problem are also tested for mass and drawn for RiL* ⫽25 and ␧⫽0.85. It is seen that the contribution of
energy balance. For this, a typical case with q v ⫽106 W/m3, k s convection decreases from 45 percent for A 1 ⫽0 to about 41.5
percent for A 1 ⫽0.05, but from then increases monotonically to a
⫽0.25 W/m K, ␧⫽0.85, ReL⫽2500, and RiL* ⫽1 is taken up. As
maximum of 65 percent for A 1 ⫽0.875. An exact ‘‘mirror image’’
many as 13 different positions for the discrete heat source (0
variation is noticed for radiation. The two curves cross each other
⭐A 1 ⭐0.875) are considered. The mass and the energy balance
for A 1 ⫽0.625, implying equal contributions from the two modes.
are found satisfactory within ⫾ 0.007 percent and ⫾ 4.02 percent,
The contribution from radiation decreases continuously from the
respectively.
free convection limit (RiL* ⫽25) to the forced convection limit
Variation of Maximum Plate Temperature „ ␪ max… With (RiL* ⫽0.1), for a given ␧. For RiL* ⫽25 and ␧⫽0.85, radiation is
significant with maximum and minimum contributions of 58.5
Other Parameters percent and 36 percent, respectively, depending on the position of
Figure 3 shows the variation of ␪ max with non-dimensional heat the heat source. Even for RiL* ⫽0.1 共and the same ␧⫽0.85兲, radia-
source position (A 1 ), for two values of u ⬁ 共RiL* ⫽25 and 0.1兲 and tion plays a key role with maximum and minimum contributions
two values of surface emissivity 共␧⫽0.05 and 0.85兲, for the case of 30.5 percent and 18.3 percent, respectively. Convection is the
with q v ⫽106 W/m3 and k s ⫽0.25 W/m K. It is clear that there is a dominant mode of heat transfer for a good reflecting surface (␧
sharp increase in ␪ max from A 1 ⫽0 to A 1 ⫽0.125 and again from ⫽0.05). For the case of the heat source located at the center of the
A 1 ⫽0.825 to A 1 ⫽0.875, with a considerably slower increase for plate, for RiL* ⫽0.1 and ␧⫽0.05, convection takes away as much
A 1 ⫽0.125 to 0.825. As RiL* decreases from 25 to 0.1, the amount as 97.6 percent of the heat, while the radiation contribution is only
by which ␪ max increases between the positions A 1 ⫽0.125 and 2.4 percent. The contribution from radiation increases with ␧, with
A 1 ⫽0.825 also increases. In the case considered here, the increase a proportionate decrease in convection. For RiL* ⫽25 and A 1 ⫽0,
in ␪ max , between A 1 ⫽0.125 and A 1 ⫽0.825, for RiL* ⫽25 and ␧ radiation contributes about 6.1 percent for ␧⫽0.05, while, for ␧
⫽0.85, is only 2.8 percent, while that for RiL* ⫽0.1 is 14.4 per- ⫽0.85, there is more than nine fold increase to 55.1 percent.
cent. The ␪ max decreases with increasing ReL 共or decreasing RiL* 兲,
for a given A 1 . The figure shows that the best position for the heat
source is A 1 ⫽0 共leading edge of the plate兲, while the least advis-
able position is A 1 ⫽0.875 共heat source ending on the trailing
edge兲 for all values of RiL* and ␧. For example, for RiL* ⫽0.1 and
Fig. 3 Maximum non-dimensional plate temperature with Fig. 4 Percentage heat transfer due to convection and radia-
modified Richardson number and surface emissivity for vari- tion with modified Richardson number and surface emissivity
ous positions of the discrete heat source for various positions of the discrete heat source
Effect of Thermal Conductivity of the Plate
The thermal conductivity of the plate (k s ) is present in ⌬T ref ,
which, in turn, is present in GrL* , RiL* , and N RF . Thus, k s alone
cannot be varied by setting all the three dimensionless variables,
GrL* , RiL* , and NRF , for a given q v . However, keeping all the
primary variables fixed and letting k s alone vary 共0.25, 0.5, and 1
W/m K兲, the dimensional maximum plate temperature (T max) is
calculated for a representative case with q v ⫽106 W/m3, u ⬁
⫽0.25 m/s, T ⬁ ⫽25°C 共298 K兲, ␧⫽0.45, and A 1 ⫽0.4375. It has
been noticed that T max decreases by 18.9 percent as k s increases
form 0.25 to 1 W/m K, in this particular case. This demonstrates
the importance of k s in the present problem.
Correlations
A correlation, having a correlation coefficient of 0.99 and an
error band of ⫾5.8 percent, is evolved for ␪ max , based on a large
set of 514 data, as
␪ max⫽51.5181共 1⫺A 1 兲 ⫺0.14␥ ⫺0.69 冉 N RF

1⫹N RF 冊 0.07
共 1⫹␧ 兲 ⫺0.51
⫻ 共 1⫹RiL* 兲 ⫺0.139 ReL⫺0.323 . (5) Fig. 5 Percentage contributions to mean friction coefficient
A correlation, having a correlation coefficient of 0.986 and an from forced and free convection components
error band of ⫾6.1 percent, is developed for ␪ av as
␪ av⫽54.4132共 1⫺A 1 兲 0.25␥ ⫺0.89 冉 N RF

1⫹N RF 冊 ⫺1.91
共 1⫹␧ 兲 ⫺0.78 convection falling in the range 1⭐RiL* ⭐10. In view of this, it has
been decided to check whether 0.1⭐RiL* ⭐1 would serve as the
⫻ 共 1⫹RiL* 兲 ⫺0.179 ReL⫺0.415 . (6) forced convection asymptote. To do this, 26 randomly chosen
data, covering the entire range (0.1⭐RiL* ⭐25), are generated
Two separate correlations are generated for C̄ f for two different
with ␤ ⫽0 共no free convection兲. The values of C̄ f 共forced兲, thus
ranges of RiL* , viz., 0.1⭐RiL* ⭐1 and 1⬍RiL* ⭐25. The C̄ f for
‘‘low’’ modified Richardson number range (0.1⭐RiL* ⭐1), based obtained, are compared with the values of C̄ f obtained by letting
on 245 data, having a correlation coefficient of 0.992 and an error RiL* ⫽0 in Eq. 共7兲, and a very good agreement, with an error band
band of ⫾4.6 percent, correlated as of ⫾6.3 percent, is noticed. Therefore, C̄ f 共forced兲 for the whole
C̄ f ⫽6.8719共 1⫺A 1 兲 0.03␥ ⫺0.08 冉 N RF

1⫹N RF 冊 ⫺0.8
共 1⫹␧ 兲 ⫺0.05
range of RiL* (0.1⭐RiL* ⭐25) may be obtained from Eq. 共7兲, by
simply setting RiL* ⫽0, as
⫻ 共 1⫹RiL* 兲 0.239 ReL⫺0.672 . (7) C̄ f 共 forced兲 ⫽6.8719共 1⫺A 1 兲 0.03␥ ⫺0.08 冉 N RF

1⫹N RF 冊 ⫺0.8
The correlation for C̄ f for ‘‘high’’ modified Richardson number

range (1⬍RiL* ⭐25), based on 196 data, having a correlation co- ⫻ 共 1⫹␧ 兲 ⫺0.05 ReL⫺0.672 . (9)
efficient of 0.99 and an error band of ⫾5.4 percent, came out as From the C̄ f 共forced兲, the wall shear, the drag force and thus the
C̄ f ⫽38.155共 1⫺A 1 兲 0.39␥ ⫺0.47 冉 N RF

1⫹N RF 冊 ⫺0.25
共 1⫹␧ 兲 ⫺0.54
fan power input may be calculated.
Conclusions
⫻ 共 1⫹RiL* 兲 0.493 ReL⫺0.792 . (8) The best position for the heat source is the leading edge of the
plate and the least preferable position is the trailing edge, as the
Calculation of Forced Convection Friction Coefficient latter position increases ␪ max by 35–50 percent for the entire range
Component of RiL* . The ␪ max , for the given values of q v and k s , decreases
Equations 共7兲 and 共8兲 give the sum of the free and forced con- with ␧ for any given ReL or RiL* . However, the degree of decrease
vection components of the mean friction coefficient. Of these, of ␪ max with ␧ decreases, as one moves from the free convection
only the forced convection component requires the power input limit (RiL* ⫽25) to the forced convection limit (RiL* ⫽0.1). Radia-
共pumping power兲, through a fan or blower, for maintaining the tion is found to play a significant role in the present problem. For
flow. The free convection related flow, results from buoyancy, and example, for ␧⫽0.85, for RiL* ⫽25, radiation is found to contrib-
hence does not need any power input. Now, the point of interest is ute 35–60 percent of heat transfer, depending on the position of
to obtain C̄ f 共forced兲 from the correlations for the total C̄ f evolved the heat source. While, for the same ␧⫽0.85, for RiL* ⫽1 and 0.1,
above. For this, the present problem is solved for two typical the radiation contributions have been found to be 28–45 percent
cases, for as many as eight values of RiL* covering the entire range and 18–30 percent, respectively. For a given RiL* 共say for RiL*
0.1⭐RiL* ⭐25 in each case. The problem is solved first as such ⬃1兲, convection is dominant, taking away as much as 90–95
and later with ␤ ⫽0 共no free convection兲, which, respectively, percent of the heat generated, in the case of a good reflecting
gave the values of C̄ f 共total兲 and C̄ f 共forced兲. From these, the surface (␧⫽0.05), with radiation being insignificant. However,
values of C̄ f 共free兲 are obtained. The percentages of the two com- the radiation contribution increases with ␧ for the same RiL* and it
ponents are plotted against RiL* , for both the cases, as shown in may be as high as 45–50 percent, in the case of a good emitting
Fig. 5, which reveals that RiL* ⫽0.1 and 25 could be taken as the surface (␧⫽0.85). Correlations are developed for ␪ max and ␪ av
forced and free convection limits, respectively, with pure mixed along with two separate correlations for C̄ f for two different
ranges of RiL* . A method for evaluating the forced convection ⌬T ref ⫽ modified reference temperature difference,
friction coefficient component is provided, which helps in the es- q v L h t/k s ,K
timation of the fan or blower power input required for maintaining ⌬x ⫽ height of the plate element, m
the flow. ⌬X ⫽ non-dimensional height of the plate element, ⌬x/L
␧ ⫽ emissivity of the plate surface
Nomenclature ␦ ⫽ convergence criterion in percentage form, 兩 ( ␨ new
A 1 ⫽ non-dimensional position of the heat source, ⫺ ␨ old )/ ␨ new兩 ⫻100 percent
(L 1 /L) ␥ ⫽ non-dimensional thermal conductivity parameter,
A r 1 ,A r 2 ⫽ aspect ratios pertaining to problem geometry, (L/t), k f L/k s t
␯ ⫽ kinematic viscosity of the fluid, m2/s
(L/L h ), respectively ␻⬘ ⫽ vorticity, s⫺1
C̄ f ⫽ mean friction coefficient, (2/ReL)兰10(⳵U/⳵Y)Y⫽0dX ␻ ⫽ non-dimensional vorticity, ␻ ⬘ L/u ⬁
GrL* ⫽ modified Grashof number based on L, ␳ ⫽ density of the fluid, kg/m3
g ␤ ⌬T refL 3 / ␯ 2 ␴ ⫽ Stefan-Boltzmann constant, 5.6697⫻10⫺8 W/m2 K4
g ⫽ acceleration due to gravity, 9.81 m/s2 ␺⬘ ⫽ stream function, m2/s
H, W ⫽ height and width of the computational domain, re- ␺ ⫽ non-dimensional stream function, ␺ ⬘ /u ⬁ L
spectively, m ␪ ⫽ non-dimensional temperature, (T⫺T ⬁ )/⌬T ref
k ⫽ thermal conductivity, W/m K ␨ ⫽ dependent variable 共␺, ␻, or ␪兲 over which conver-
L, t ⫽ length and thickness of the vertical plate, respec- gence test is applied
tively, m
L h , L 1 ⫽ height and starting length of the discrete heat Subscripts
source, respectively, m av, max, p ⫽ average, maximum and local values of the plate
M, N ⫽ number of grid points in horizontal and vertical temperature, respectively
directions, respectively f, s ⫽ fluid and plate material, respectively
N RF ⫽ radiation-flow interaction parameter, new, old ⫽ values of the dependent variable from present
␴ T 4⬁ ⌬T ref/(k f /L) and previous iterations, respectively
P ⫽ pressure at any location, Pa
PeL ⫽ Peclet number based on L, u ⬁ L/ ␣
q v ⫽ volumetric heat generation rate in the discrete heat
source, W/m3 References
Q ⫽ heat transfer rate, W
关1兴 Zinnes, A. E., 1970, ‘‘The Coupling of Conduction with Laminar Natural
ReL ⫽ Reynolds number based on L, u ⬁ L/ ␯ Convection From a Vertical Flat Plate With Arbitrary Surface Heating,’’
RiL* ⫽ modified Richardson number based on L, (GrL* /ReL2 ) ASME Journal of Heat Transfer, 92, pp. 528–534.
关2兴 Gorski, M. A., and Plumb, O. A., 1992, ‘‘Conjugate Heat Transfer from an
or g ␤ ⌬T refL/u 2⬁ Isolated Heat Source in a Plane Wall,’’ Proc. Winter Annual Meeting of The
T ⫽ temperature at any location, K American Society of Mechanical Engineers, ASME HTD-210, pp. 99–105.
u ⬁ ⫽ free-stream velocity of the fluid, m/s 关3兴 Hossain, M. A., and Takhar, H. S., 1996, ‘‘Radiation Effect on Mixed Con-
u ⫽ vertical velocity, m/s vection Along a Vertical Plate with Uniform Surface Temperature,’’ Heat and
Mass Transfer/Warme-und Stoffubertragung, 31, pp. 243–248.
U ⫽ non-dimensional vertical velocity, u/u ⬁ or ⳵ ␺ / ⳵ Y 关4兴 Cole, K. D., 1997, ‘‘Conjugate Heat Transfer from a Small Heated Strip,’’ Int.
v ⫽ horizontal velocity, m/s J. Heat Mass Transf., 40, pp. 2709–2719.
V ⫽ non-dimensional horizontal velocity, v /u ⬁ or 关5兴 Bejan, A., 1984, Convection Heat Transfer, Wiley, New York, pp. 112–114.
⫺⳵␺/⳵X 关6兴 Gururaja Rao, C., Balaji, C., and Venkateshan, S. P., 2000, ‘‘Numerical Study
of Laminar Mixed Convection from a Vertical Plate,’’ I. J. Trans. Phenomena,
x, y ⫽ vertical and horizontal distances, respectively, m 2, pp. 143–157.
X, Y ⫽ non-dimensional vertical and horizontal distances, 关7兴 Gosman, A. D., Pun, W. M., Runchal, A. K., Spalding, D. B., and Wolfshtein,
x/L,y/L, respectively M., 1969, Heat and Mass Transfer in Recirculating Flows, Academic Press,
␣ ⫽ thermal diffusivity of the fluid, m2/s London, pp. 89–137.
关8兴 Dehghan, A. A., and Behnia, M., 1996, ‘‘Combined Natural Convection—
␤ ⫽ isobaric cubic expansivity of the fluid, ⫺(1/␳ ) Conduction and Radiation Heat Transfer in a Discretely Heated Open Cavity,’’
⫻( ⳵␳ / ⳵ T) P ,K ⫺1 ASME Journal of Heat Transfer, 118, pp. 56–64.
Minoru Takahashi
e-mail: mtakahas@nr.titech.ac.jp
Research Laboratory for Nuclear Reactors,
Tokyo Institute of Technology,
Heat Transfer in Direct Contact
2-12-1 O-okayama, Meguro-ku,
Tokyo 152-8550, Japan Condensation of Steam
Arun Kumar Nayak
to Subcooled Water Spray
e-mail: redths@apsara.barc.ernet.in
Reactor Engineering Division, The condensation heat transfer of saturated steam to a hollow-cone spray of subcooled
Bhabha Atomic Research Centre, water was investigated experimentally and analytically. The spray water temperature rose
Trombay, Mumbai 400085, India more steeply in flow direction than those in the previous studies, because of the use of
smaller thermocouple which was capable of measuring the temperature in a thin water
sheet and water droplets more accurately. The result of the condensation heat transfer
Shin-ichi Kitagawa coefficient suggested the breakup of the water sheet into droplets. A pure conduction
model underpredicted the heat transfer in the sheet region significantly, which was better
Hiroyuki Murakoso predicted by considering turbulence in the sheet. The heat transfer in the droplet region
was well estimated by considering internal circulation and mixing inside the droplets.
Department of Nuclear Engineering, 关DOI: 10.1115/1.1370510兴
Tokyo Institute of Technology,
2-12-1 O-okayama, Meguro-ku, Keywords: Condensation, Cooling, Droplet, Heat Transfer, Sprays
Tokyo 152-8550, Japan
1 Introduction havior at the interface of steam and spray. That is because it has
been technically difficult to measure the temperature profile across
The study of direct contact condensation of steam on spray of
the sheet whose thickness may vary from few hundred micrometer
water has been of great interest to researchers because of its nu- to few micrometer depending on the structure of spray cone. The
merous applications in nuclear industries, e.g., the pressure con- previous researchers have not mentioned about the sizes of ther-
trol in the pressurizers of the Pressurized Water Reactors 共PWR兲, mocouples used for measuring the spray water temperature. If
the emergency core cooling in the Boiling Water Reactors they have employed relatively large thermocouples to measure the
共BWR兲, the containment cooling during the loss of coolant acci- temperature of the sheet, their measurements may provide much
dent 共LOCA兲 of the Light Water Reactors 共LWR兲, and the bleed different spray temperatures from each other. For example, they
condensers of the Indian Pressurized Heavy Water Reactors are a spatially averaged temperature in a certain volume having
共PHWR兲. The other applications include direct contact condensers the water sheet and a steam flow, and not the average temperature
used in chemical industries, geothermal power plants, etc. across the sheet which has higher temperature in thermal bound-
Figure 1 depicts the structure of a spray flow issuing from a ary layers adjacent to the sheet surfaces but a local low bulk
pressure-swirl hollow-cone nozzle. The liquid moves through spi- temperature at one point inside the sheet. The measurement of
ral like channels inside the nozzle, which discharges it tangen-
tially into a vortex or swirl chamber. The liquid thus emerging
from the nozzle forms a conical sheet at the nozzle outlet due to
its tangential, radial and axial momentum forces. After some dis-
tance, the sheet breaks into ligaments and finally into droplets due
to the destabilizing forces. Therefore, the spray flow can be
broadly classified into two distinct regions, viz., a continuous liq-
uid sheet region followed by a droplet region.
Many experimental and analytical studies have been carried out
in the past to investigate the hydrodynamics and the heat transfer
mechanism in the above two regions. Studies on vapor condensing
on single moving drops have been carried out by Brown 关1兴, Lekic
and Ford 关2兴, Sundararayan and Ayyaswamy 关3,4兴, Chung and
Chang 关5兴 and Celata et al. 关6兴 and on liquid sheet formed out of
a spray by Lee and Tankin 关7兴. Hasson et al. 关8兴 studied the heat
transfer behavior during direct contact condensation of steam on a
water sheet analytically. Weinberg 关9兴 studied experimentally and
analytically the heat transfer behavior considering both sheet and
droplet regions. Mayinger and Chavez 关10兴 carried out experi-
ments to reveal the heat transfer behavior in the sheet and droplet
regions by using pulsed laser holography.
In spite of all these works, there is a greater need for further
research in this area in order to generate more reliable experimen-
tal data which are scanty and to understand the heat transfer be-

HEAT TRANSFER. Manuscript received by the Heat Transfer Division July 10, 2000;
revision received January 30, 2001. Associate Editor: H. H. Bau. Fig. 1 Hollow-cone spray
temperature profiles across the sheet thickness require the use of
thermocouples which are smaller than the sheet thickness and
have short response time, and a fast data acquisition system. The
temperature profile across the sheet thickness can be measured by
traversing the thermocouples normal to the spray flow path, which
can be averaged to obtain the local average temperature.
Further, there are many analytical models for the heat transfer
behavior in the sheet and droplet regions with different assump-
tions in literature. However, the models have not been adequately
tested especially for steam condensing on a hollow-cone spray of
water. In order to clarify the heat transfer mechanism in the sheet
and droplet regions, it is required to predict the heat transfer be-
havior using these analytical models and compare the result with
the experimental data.
For this purpose, experiments were carried out to study the heat
transfer behavior of steam condensing on a hollow-cone spray.
The temperature profiles across the sheet and droplet regions were
measured using very small thermocouples and a fast data acquisi-
tion system. The effect of thermocouple response time on the
spray temperature was studied by varying the size of thermo-
couples in the experiment. The temperature rise in the flow direc-
tion in the sheet and droplet regions were predicted using analyti-
cal models. The heat transfer mechanism in the sheet and droplet
regions are discussed from the comparison of the test data with
the predictions. Fig. 3 Test section
2 Experiment
ter to the test vessel, and the temperature in the test vessel were
2.1 Experimental Apparatus and Procedure. Figure 2 measured with sheathed thermocouples 1 mm in diameter. The
shows a schematic diagram of the experimental apparatus which vessel pressure was measured with a semi-conductor type pressure
consists of a water circulation loop and a steam supply system. transducer. In order to prevent the steam from condensing inside
Water was circulated by the pump from the tank below the test the transducer, a tube introduced into the transducer was filled
vessel through the subcooler, the orifice, the flow control valve to with silicon oil and electrically heated to the saturation tempera-
the test vessel. Saturated steam was generated in the boiler and ture of steam, while the tube near the transducer was water-cooled
supplied into the test vessel maintained at near atmospheric pres- to keep the sensor at the room temperature. During the measure-
sure. In order to keep the water level in the test vessel and boilers ment, the outlet valve was closed, and the pressure in the test
constant, the amount of the water accumulated in the vessel due to vessel was controlled to be the one corresponding to the saturation
condensation was returned to the boiler continuously using a temperature in the boiler.
pump, and thermally steady condition was maintained throughout A spray condensation was realized in a cylindrical test vessel
the operation. The inlet temperatures of steam and subcooled wa- made of stainless steel shown in Fig. 3. The hollow-cone spray
Fig. 2 Experimental apparatus
Table 1 Experimental conditions
confirmed that the sizes of the beads were nearly the same of the
wire diameters. The hot junction was directed toward the up-
Fig. 4 Hollow-cone spray nozzle „B1Õ4GG-SS3030HC of Spray- stream of the spray in order not to disturb the water sheet. The
ing Systems Japan Co.… thin thermocouple wires were connected to the thick wires with a
lead-tin alloy solder. The surface of the wires except for the hot
junction were covered with the silicon glue for electrical insula-
nozzle: Spraying Systems Japan Co., B1/4GG-SS3030HC, was tion and also to support the thin thermocouple wires. In order to
inserted from the top into the vessel, and a subcooled water was avoid splashing of the water at the surface of the thermocouple
discharged from the nozzle into the vessel, which resulted into a assembly, which may affect the measured temperature signal, the
hollow-cone spray. The inside diameter of the nozzle is 3.6 mm. glue cover was made as small as possible. The thermocouple was
The configuration of the nozzle is shown in Fig. 4, and the depen- inserted through the side wall, and traversed both radially and
dences of the flow rate and the spray angle on the pressure drop axially to measure the temperature distributions in the region of
are shown in Fig. 5. It is found that the spray angle is nearly the spray sheet and droplets.
constant with the flow rate except for the conditions of low flow The response time of the 0.025 mm thermocouple was esti-
rate. Thus, the present experiment was not performed with the mated to be approximately 0.03 ms if the unsteady heat conduc-
parameter of the spray angle. tion is considered in a cylindrical thermocouple wire. This was
The outside of the entrance pipe from the top of the vessel to estimated from the convective heat transfer coefficient of steam
the nozzle was thermally insulated with a silicon rubber sheet 3 and water on the surface of the thermocouple wire and found to be
mm in thickness. The temperature rises from the measuring posi- about 0.05–0.15 ms for water flowing with velocity of 1–14 m/s,
tion of the inlet water temperatures to the nozzle outlet were es- respectively, and about 9–28 ms for steam flowing with velocity
timated to be 0.3°C–0.5°C, which was 1.3 percent of T sat⫺T i and of 1–14 m/s, respectively. The data of the fluctuating temperature
negligibly small compared with the temperature rise in the spray. in the saturated steam and subcooled water sheet or droplets were
Thus, it is valid to consider the measured inlet temperature as the stored using a fast data acquisition system with the data sampling
nozzle outlet temperature. frequency of 50 kHz which was high enough compared with the
Non-condensable gas was removed from the vessel by blowing temperature fluctuation frequency less than 1 kHz which could
the steam through the steam outlet valve for about 30 minutes be estimated from the characteristic period of the data shown in
before each operation and also for about 15 minutes before every Fig. 7.
measurement. The experimental conditions are listed in Table 1, As shown in Fig. 7, the fluctuating temperature in the saturated
where the superficial nozzle outlet velocity indicates the cross- steam and subcooled water sheet or droplets was successfully de-
sectional average velocity estimated from the cross-sectional area tected with fast response. It is found that the practical response
of the nozzle and the volumetric flow rate without considering an time is much shorter than that estimated from the heat transfer in
air core that exists inside the present centrifugal nozzle. a steam flow. If it is assumed that the axial velocity is equal to the
To measure the temperature of the spray with the steam con- superficial velocity listed in Table 1 and does not change in the
densation, two types of themocouples 共K-type兲 with bead diam- flow direction, the sheet thickness is estimated to be 108 ␮m–540
eters of 0.025 mm and 0.05 mm shown in Fig. 6 were used in ␮m depending on the radius of the spray sheet r at the water mass
order to investigate the influence of the thermocouple size on the flow rate of 0.081 kg/s as shown in Table 2. Practically these
spray temperature measurement. Chromel and alumel wires with estimated values of the thickness may be valid in the order of the
the diameters of 0.025 mm and 0.05 mm were spot-welded to magnitude. Inside the sheet with the thickness, there are two thin
make the hot junctions. From the microscope observation, it was thermal boundary layers adjacent to the both sheet surfaces in
which temperature is higher than in the bulk of the sheet. It may
Fig. 5 Performance of hollow-cone spray nozzle „B1Õ4GG- Fig. 6 Thermocouple used for spray temperature measure-
SS3030HC of Spraying Systems Japan Co.… ment
Fig. 7 Temperature fluctuation Fig. 8 Radial temperature profiles at various axial distance
from the nozzle outlet
Table 2 Estimated sheet thickness at the water mass flow rate

of 0.081 kgÕs
This is because, the sheet is very stable near the nozzle outlet.
However, as the spray moves downstream, it becomes unstable
due to the dynamic shear between the water sheet and the ambient
steam and the sheet vibrates at any position. After some distance
from the nozzle outlet, the sheet breaks into ligaments and finally
drops. In the sheet region, even though the thickness of sheet
be difficult to detect the higher temperature in the thin thermal reduces with distance from nozzle, the sheet vibrates away from
boundary layers using thermocouples larger than the thickness of nozzle outlet due to instability which makes as if a thicker sheet.
the sheet. Since the mixing of the hot and cold water inside the The phenomena is a thinner sheet having a wider radial movement
sheet cannot be expected when the sheet contacts the thermo- away from the nozzle. So the thermocouple measures a wider
couples, only the lower temperature inside the bulk of the sheet temperature profile at any downstream position. In the droplet
may be detected with the large thermocouple. With the small ther- region, various sizes of droplets are formed from the breakup of
mocouple with short response time the size of which is compa- sheet. All the drops do not move in the same direction. In fact the
rable with the thickness of the thermal boundary layers, the higher smaller droplets move towards the center of spray as they move
temperature in the boundary layers can be detected. downstream. So at any axial position, the thermocouple finds the
At the interface of steam and sheet of the spray, the thermo- droplets over a wider region.
couple output signal sometimes contain the steam temperature due It is also found that the temperature profiles for T i ⫽65°C are
to the instability of the sheet and hence the thermocouple touches slightly narrower than that of T i ⫽75°C at any vertical positions,
the steam. Similarly in the droplet region, since the droplet flow is that is, the sheet is thinner and droplet region is narrower for
discontinuous and dispersed at any axial and radial location, the higher nozzle outlet water subcooling. This may be possibly be-
thermocouple output contains the mixture of steam and droplet cause the higher subcooling increases the heat flux, and then the
temperature at any position. Hence, to estimate the spray tempera- steam inflows to the sheet and droplets from both sides, and as a
ture at any position, the steam temperature signal is removed from result reduces the radial expanding motions or instability of the
the thermocouple output signal so that the measured temperature sheet and droplets flow. Furthermore, the spray angle is smaller
is only the spray temperature as shown in Fig. 7. Then, the water for the higher subcooling condition. This is possibly because of
temperature was time-averaged to obtain mean local water tem- the lower pressure inside the hollow cone due to the higher vapor
perature in the sheet and droplets of the spray. condensation rate at the higher subcooling condition.
The errors of the measured radial distance was within ⫾ 0.05 The average water temperature T̄ at each elevation was calcu-
mm when the thermocouple was traversed horizontally. The mea- lated by integrating the radial temperature profile as
冕
sured temperature signal had some electrical noise, but it was r2
reduced to be less than 1 K in the noise peak-to-peak values. T̄⫽ 2T 共 r 兲 rdr/ 共 r 22 ⫺r 21 兲 . (1)
Thus, the error of the measured temperature was within ⫾0.5 K. r1
The error of the measured T i was also within ⫾0.5 K.
The local average heat transfer coefficient was estimated from the
2.2 Experimental Results and Discussion. The tempera- local average heat flux between two measuring positions and the
ture profiles in the spray sheet and droplets for different spray temperature rise across them. The local average heat flux q̄ be-
inlet temperatures are shown in Fig. 8, where r and x are the radial tween two positions was calculated from the axial enthalpy rise as
and axial coordinates respectively. It is found that the temperature
profiles are steep near the nozzle outlet. However, in the down- 1 ⌬T
q̄⫽ṁC p , (2)
stream region, the temperature profiles are broader and higher. 4 ␲ r̄ ⌬x/cos ␾
Fig. 9 Variation of heat transfer coefficient in the flow direc-
tion
where ṁ is the spray mass flow rate, ⌬T is the spray temperature Fig. 10 Variation of condensation efficiency in the flow direc-
rise between two positions for which the vertical distance is ⌬x, ␾ tion
is the spray angle and r̄ is the spray radius, and the over-bar
means the arithmetical average of variables at two positions. The
local average heat transfer coefficient h̄ is given by
significant heat transfer takes place in the sheet region as com-
q̄⫽h̄ 共 T sat⫺T̄ 兲 . (3) pared to that in the droplet region, even though the length of the
sheet region is very small.
The magnitude of the error in the estimate of h̄ is smaller as the The axial rise in the water average temperature, or the conden-
axial temperature rise ⌬T is larger and the average spray tempera- sation efficiency is shown in Fig. 10, where T⫺T i is the water
ture T̄ is lower because of the smaller errors in the estimate of ⌬T temperature rise from the nozzle outlet to the measuring position,
and the heat flux q̄ in Eq. 共2兲 and the smaller error in the estimate and T sat⫺T i is the difference between the saturation temperature
of the heat transfer coefficient h̄ in Eq. 共3兲. The errors of h̄ are and the water temperature at the nozzle outlet. The axial distance
estimated to be ⫾8 percent at x/d⫽0.69, ⫾23 percent at x/d from the nozzle x is normalized by the nozzle diameter, d. The
⫽2.08, ⫾32 percent at x/d⫽3.47, ⫾48 percent at x/d⫽4.86, error in the estimate of the condensation efficiency is within
⫾70 percent at x/d⫽6.25, and ⫾105 percent at x/d ⫾0.04 due to the errors of T̄ and T i mentioned above. The tem-
⫽7.64– 10.4. perature measured with the thermocouple having the diameter of
The experimental results for the local average heat transfer co- 0.05 mm is somewhat lower than that measured with the thermo-
efficient h̄ is shown in Fig. 9. There appears two sheets and drop- couple having the diameter of 0.025 mm in the region where
lets regions in the temperature measurement plane, that is, the x/d⬍10. This is because it is difficult to detect the higher tem-
region of z⬍0 and that of z⬎0 as shown in Fig. 3. The heat perature in the thin thermal boundary layers using thermocouples
transfer coefficients obtained in the regions of z⬍0 and z⬎0 are larger than the thickness of the sheet. While only the lower tem-
separately shown in Fig. 9, although appreciable differences are perature inside the bulk of the sheet may be detected with the
not observed between these results. It is found that the heat trans- large thermocouple, the higher temperature in the thermal bound-
fer coefficient is high near the nozzle outlet, and it decreases in ary layers can be detected using the small thermocouple with short
the x-direction. The breakup point indicated in Fig. 9 is deter- response time the size of which is comparable with the thickness
mined from the temperature fluctuation data, assuming that when of the thermal boundary layers, and as a result the average tem-
measured temperature reaches the saturation temperature periodi- perature appears to be higher with the small thermocouples having
cally, the sheet has broken up into droplets. It is assumed that if shorter response time as shown from the comparison of the results
the thermocouple measures continuously temperature less than the with the thermocouples with the sizes of 0.025 mm and 0.05 mm
steam temperature as it penetrates radially at any axial position in Fig. 10.
along the spray, it is the sheet region of spray. On the other hand, In addition, the present data are found to be higher than the data
if it measures sometimes low temperature and sometimes satura- obtained by Weinberg 关9兴 for a nozzle with the inner diameter of
tion temperature as it penetrates along the spray at any axial po- 7.62 mm. It means that with smaller thermocouples the response
sition, it measures the droplet temperature. This is because of the time is short enough compared with the time scale of temperature
discontinuity of the flow of droplets at any axial and radial posi- fluctuation. Hence, such small thermocouples are required to mea-
tion unlike that of a liquid sheet which is continuous along its sure reliable spray water temperatures. However, the analytical
thickness. The breakup point obtained from the temperature pro- results by Lee and Tankin 关7兴, who assumed a laminar flow in the
file measurement occurs at 5.5x/d which is almost the same smooth spray sheet, are much lower than the present experimental
as that predicted from the empirical correlation of Lee and result and Weinberg’s data. This suggests that the condensation
Tankin 关7兴. heat transfer may be influenced greatly by disturbed interfaces, a
It is obvious that the heat transfer coefficient in the sheet region turbulent flow inside the water sheet and recirculating flows in
is comparatively higher than that in the droplet region, as a result droplets.
3 Analysis L tb ⫽1.875⫻10⫺4 Vd 2 / ␯ ⫹0.11, (12)
which takes into account 共i兲 the effect of radial velocity profile
3.1 Analytical Model development along the jet axis; 共ii兲 the decrease of the turbulence
caused by the lack of solid walls; 共iii兲 the edge effects at the
3.1.1 Summary. As mentioned before, the structure of the nozzle exit due to the surface tension and the viscosity; 共iv兲 the
spray can be characterized by two distinct regions, viz., 共a兲 con- effect of surface instability; 共v兲 the thermal effect due to the set-
tinuous liquid sheet region and 共b兲 droplet region. There are many tling of the radial temperature profile; 共vi兲 the effect of the inter-
analytical models in the literature for the heat transfer behavior in face shear stress between the condensate and the liquid jet. The
these regions. Depending on the assumptions made in these mod- K tot obtained by the model is substituted into Eqs. 共5兲 or 共6兲 to
els, the heat transfer mechanism in the sheet and droplet regions
calculate ¯␪ . An empirical K tot similar to the above cannot be
can be broadly classified into two modes, i.e., 共i兲 heat transfer
easily found out in the spray condensation case. That is because
taking place by purely conduction with negligible resistance of
the propagation of turbulence generated inside the swirl chamber
condensate film, and 共ii兲 consideration of a turbulent thermal con-
of a hollow-cone nozzle is quite different from that of a jet issuing
ductivity in the sheet region or internal circulation and mixing in
from a pipe. Besides, the sheet thickness and the turbulence inten-
the droplet region. The models considered in the present analysis
sity reduce with distance from the nozzle outlet, which is a com-
are discussed below.
plete different phenomena unlike that observed from the cylindri-
3.1.2 Modeling of Heat Transfer in the Sheet Region. If con- cal jets. Further, the sheet is never stable and the droplet flow is
duction is assumed to be the only mode of heat transfer from the highly dispersed. So it is impossible to estimate accurately the
surface of the sheet to the center without considering the growth heat flux from the measured temperature in the surface of the
of condensate film on the surface, the model due to Hasson et al. sheet which is unstable and the highly dispersed flow of droplets.
关8兴 developed for a fan spray sheet can be directly applied. A The present temperature measurements are only good enough for
numerical solution similar to that carried out by Lee and Tankin the average temperature of the sheet or the droplet region at any
关7兴 for the sheet region is possible, but the Hasson et al. model 关8兴 axial position.
is easier to apply provided the sheet thickness is known at any
3.1.3 Modeling of Heat Transfer in the Droplet Region. The
axial distance from the nozzle. The present analysis considers this
heat transfer in the droplet regions depends on the size of the
model and the sheet thickness is estimated from the equation of
droplets. Since droplets of different sizes are formed due to the
continuity
breakup of the liquid sheet, many analytical models consider a
m droplet size distribution parameter. However, it is required to de-
s⫽ , (4) termine the maximum droplet diameter and other distribution pa-
2␲r␳V
rameters from experiments. In the absence of that, it is possible to
where V is the velocity in the flow direction, m is the mass flow evaluate the heat transfer behavior by considering a mean diam-
rate of spray, and r is the radius of spray cone at a distance x from eter of the droplets. Weinberg 关9兴 carried out experiments using
the nozzle outlet. various types of spray nozzles with water as fluid. The size of
The mean temperature rise in the sheet can be expressed as nozzle varied from a minimum of 0.07 in 共1.78 mm兲 to a maxi-
mum of 3.75 in 共95.25 mm兲 which covered the present nozzle size
8
¯␪ ⫽ exp共 ⫺4 ␲ 2 /3Gz 兲 , (5) of 3.6 mm. Weinberg developed the correlation which is appli-
␲2 cable for flow parameter F up to 500 and pressure up to 20 psi and
for small Graetz numbers Gz; and given by
8 D⫽0.0047⫻10⫺6 共 F 3 /⌬ p 兲 1/7, (13)

¯␪ ⫽1⫺ , (6)
冑3 ␲ Gz where F⫽Q/ 冑⌬p, Q is the flow rate in gph and ⌬p is the pres-
for large Graetz numbers. sure drop across the nozzle in psi. Since in the present case, the
However, the spray discharged from the nozzle possesses cer- maximum value of F was 33, and the pressure was atmospheric,
tain amount of turbulence which damps along the flow path of the which was in the range of the Weinberg’s ones, Eq. 共13兲 was used
spray. To the authors’ knowledge, no theoretical model has ever for the prediction of the droplet diameter.
considered this phenomenon in the sheet region of the spray. One If we assume the droplet to be a sphere and the mode of heat
complex problem with spray condensation phenomena is model- transfer is by conduction from the surface to the center, the appli-
ing the turbulence generated inside the swirl chamber of the cable differential equation is 共Ford and Lekic 关14兴兲
冉冊
nozzle, and how it propagates with the spray after it issues from
⳵␪ ⳵ 2␪ 2 ⳵ ␪
the nozzle while condensation occurring on it. However, some ⫽␣ ⫹ (14)
analytical studies have considered this behavior for circular jets ⳵t ⳵y2 y ⳵y
共Kutateladze 关11兴; Celata et al. 关12兴; Celata et al. 关13兴兲 issuing
from nozzles. Hence, an attempt has been made here to consider where ␪ ⫽(T⫺T i )/(T sat⫺T i ) with the following boundary condi-
the turbulent thermal conductivity model of Celata et al. 关13兴 on tions: 共i兲 droplet initially at a uniform temperature: ␪ (y,0)⫽0; 共ii兲
the heat transfer behavior in the sheet region. This model is given droplet surface immediately reaches saturation temperature:
by ␪ (R,t)⫽1; 共iii兲 symmetry: ( ⳵ ␪ / ⳵ y) y⫽0 ⫽0; and 共iv兲 heat balance
at the interface: K(T sat⫺T i )( ⳵ ␪ / ⳵ y) y⫽R ⫽␭ ␳ dR/dt. The tem-
K tot⫽K 共 1⫹A 共 B⫹C 兲兲 (7) perature distribution, considering 共as an approximation necessary
for the solution兲 that the heat flux at y⫽R is the same for a sphere
where
with constant radius, is given by
0.12
A⫽ 共 H/d 兲共 Pe兲共 d/H 兲
冉冊冉冊
, (8) ⬁
2R 共 ⫺1 兲 n y ␲ 2n 2t
C⫽3.6⫻10 ⫺4
共 T sat⫺T 兲 / 共 T sat⫺T i 兲 , (9) ␪ 共 y,t 兲 ⫽1⫺ 兺
␲ y n⫽1 n
sin n ␲
R
exp ⫺ ␣
R2
.
⫺3
B⫽1.36⫻10 共 L tb /x 兲共 ␣ L /d 兲 exp关 ⫺ 共 L tb /x 兲共 H/d 兲兴 ,
1.8 0.3 (15)
(10)
The rate of droplet growth due to condensation can be expressed
␣ L ⫽ 冑␥ /g⌬ ␳ (11) as
D
Di 冋
⫽1⫹ ␸ 1⫺exp ⫺
D 2i 冉
4 ␲ 2␣ t
冊册 1/2
;
冋
␸ ⫽ 1⫹
C p 共 T sat⫺T i 兲
␭ 册 1/3
⫺1. (16)
The motion of the droplet in the vapor environment can be ex-

pressed as 共Weinberg 关9兴兲
␲ 3 dV ␲ 3 1
D ␳ ⫽ D 共 ␳ ⫺ ␳ g 兲 g⫺ C D ␳ g ␲ D 2 V 2 , (17)
6 dt 6 8
or
dV
dt
⫽g 1⫺ 冉冊
␳g
␳
3
⫺ CD
4
␳g V2
␳ D
. (18)
The diameter and velocity of the droplet at any time can be esti-
mated from Eqs. 共16兲 and 共18兲. The average non-dimensional tem-
perature of the droplet at any time can be written as
冉冊
⬁
6 1 4 ␲ 2␣ t
兺
Fig. 11 Variation of condensation efficiency in the flow direc-
␪ m ⫽1⫺ 2 2 exp ⫺n
2
. (19) tion „comparison of experiment and analysis…
␲ n⫽1 n D 2i
Many researchers have considered the effects of both internal
circulation and mixing inside the droplet 共Ohba et al. 关15兴; Rose rise is more than 60 percent of the total ⌬T. This may be due to
and Kinter 关16兴; Skelland and Wellek 关17兴兲. Adding both internal the fact that the turbulence intensity is the largest at the nozzle
circulation and mixing inside the droplet to the pure conduction outlet and decreases in the flow direction. Previous experimental
hypothesis, the mean temperature the mean temperature of the studies of Weinberg 关9兴 and Mayinger and Chavez 关10兴 have
droplet can be expressed as 共Celata 关18兴兲 shown similar behavior very close to the nozzle outlet. Lee and
Tankin 关7兴 have attributed the high heat transfer in the experimen-
冉冊
⬁
6 1 4 ␲ 2␣ t tal data of Weinberg 关9兴 to the inaccuracy of temperature mea-
␪ m ⫽1⫺ 兺
␲ n⫽1 n
2 2 exp ⫺n C
2
D 2i
, (20) surement. Whereas, it is due to the turbulence generated inside the
swirl chamber that enhances the heat transfer in the sheet region,
where C is a factor which takes into account the effects of turbu- which needs to be considered in the analytical model. Further, the
lence inside the droplets. present calculations show that the turbulent thermal conductivity
model due to Celata et al. 关13兴 is not adequate to reveal the heat
3.2 Analytical Results and Discussion. Figure 11 shows a transfer in the sheet region.
comparison of the predictions with the experimental data for spray If we consider the internal circulation inside the droplet, there is
velocity of 0.057 kg/s with its inlet temperature at 75°C. The an increase in the heat transfer over that of a simple conduction
region I and region II shown in the figure are the estimated sheet model which is very close to the experimental data in the region
region and droplet regions respectively. The breakup length was II.
estimated using the empirical correlation of Lee and Tankin 关7兴
which was obtained from the photographs of a spray, since the 4 Conclusions
present breakup length which agreed well with the predicted from
the above correlation was only demarcated from the measured The spray condensation heat transfer were investigated experi-
temperature profile of the spray with just an approximate method. mentally and analytically. The following conclusions are obtained
The diameter of the droplet is estimated to be about 0.295 mm from the above study:
for the condition by using Eq. 共13兲. The droplet Reynolds number 1 Reliable temperature in the spray can be obtained for both
Re⬅VD/v is estimated to be 107 if the velocity of the droplet V is sheet and droplet regions by using smaller size thermocouples
approximately equal to the maximum superficial nozzle velocity with fast response.
of 8.2 m/s, and the steam kinematic viscosity v is 2.26⫻10 2 The experimental results indicate that the spray water tem-
⫺7 m2/s. This value is much lower than the oscillation criteria of perature rises more steeply in flow direction than those obtained
200. Thus, the effect of the oscillations of the droplets on heat by previous investigators. Further, maximum heat transfer is
transfer has been neglected. achieved in the sheet region as compared to that in the droplet
It can be observed that when pure conduction is assumed in region.
both the sheet and droplet regions, the mean temperature of the 3 The measured temperature rise in water spray is much higher
spray is very much under-predicted. However, if the turbulence is than those predicted by the pure heat conduction models both in
considered, the heat transfer increases significantly especially in spray sheet and droplets. The temperature prediction is much im-
the sheet region. Even though, the predictions are lower than the proved by considering turbulence in a spray sheet and internal
experimental data, still it clarifies the mechanism of heat transfer circulation and mixing inside droplets, which clarifies the heat
in the sheet region which has never been considered before. The transfer mechanism during direct contact condensation of steam
previous analytical studies 共Lee and Tankin 关7兴; Hasson et al. 关8兴兲 on subcooled water spray.
have considered only heat conduction in the sheet, but the present
results show that this assumption is not true. In fact, the turbu- Nomenclature
lence generated inside the swirl chamber of a hollow-cone nozzle
greatly augments the heat transfer in the sheet region eventhough C ⫽ constant, C⫽a(Pe⬘ ) b where a⫽0.153 and b⫽0.454
the sheet is laminar 共nearly 90 percent of thermal efficiency is 关6兴
achieved in this region兲. C D ⫽ drag coefficient
In addition, it is also found from the experimental results that C p ⫽ specific heat 共J/kg-K兲
even at a distance of 5 mm from the nozzle outlet, the temperature D ⫽ droplet diameter 共m兲
d ⫽ nozzle diameter 共m兲 m ⫽ mean
g ⫽ acceleration due to gravity 共m/s2兲 sat ⫽ saturation
Gz ⫽ Graetz number (Gz⫽4s2 V/ ␣ ) Superscripts
H ⫽ length of nozzle 共m兲 ⫺
h ⫽ heat transfer coefficient 共W/m2K兲 ⫽ average
h̄ ⫽ local average heat transfer coefficient 共W/m2K兲
K ⫽ thermal conductivity 共W/mK兲 References
L ⫽ axial distance 共m兲
关1兴 Brown, G., 1951, ‘‘Heat Transmission by Condensation of Steam on a Spray
ṁ ⫽ mass flow rate 共kg/s兲 of Water Drops,’’ Proc. of General Discussion on Heat Transfer, Institution of
⌬p ⫽ pressure drop in nozzle 共psi兲 Mechanical Engineers.
Pe ⫽ Peclet number (dV/ ␣ ) 关2兴 Lekic, A., and Ford, J. D., 1980, ‘‘Direct Contact Condensation on a Spray of
Pe⬘ ⫽ modified Peclet number Subcooled Liquid Droplets,’’ Int. J. Heat Mass Transf., 23, pp. 1531–1537.
关3兴 Sundararayan, T., and Ayyaswamy, P. S., 1984, ‘‘Heat and Mass Transfer
(Pe⬘ ⫽(dV/ ␣ )( ␮ g /( ␮ g ⫹ ␮ ))) Associated with Condensation of a Moving Drop: Solutions for Intermediate
q ⫽ heat flux 共W/m2兲 Reynolds Numbers by a Boundary Layer Formulation,’’ ASME J. Heat Trans-
fer , 107, pp. 409–415.
q̄ ⫽ axially average heat flux 共W/m2兲关4兴 Sundararayan, T., and Ayyaswamy, P. S., 1985, ‘‘Hydrodynamics and Heat
Q ⫽ flow rate 共gph兲 Transfer Associated with Condensation on a Moving Drop: Solution for Inter-
r ⫽ radius 共m兲 mediate Reynolds Numbers,’’ J. Fluid Mech., 149, pp. 33–58.
r̄ ⫽ spray radius 共m兲关5兴 Chung, J. N., and Chang, T. H., 1984, ‘‘A Mathematical Model of Condensa-
tion Heat and Mass Transfer to a Moving Droplet in its Own Vapor,’’ , 106,
R ⫽ radius of droplet 共m兲 pp. 417–424.
s ⫽ sheet thickness 共m兲关6兴 Celata, G. P., Cumo, M., D’Annibale, F., and Farello, G. E., 1991, ‘‘Direct
t ⫽ time 共s兲 Contact Condensation of Steam on Droplets,’’ Int. J. Multiphase Flow, 17, pp.
T ⫽ temperature 共K兲 191–211.
关7兴 Lee, S. Y., and Tankin, R. S., 1984, ‘‘Study of Liquid Spray 共Water兲 in a
T̄ ⫽ integral average temperature across sheet and droplet Condensable Environment 共Steam兲,’’ Int. J. Heat Mass Transf., 27, pp. 363–
region 共K兲 374.
V ⫽ velocity in flow direction 共m/s兲关8兴 Hasson, D., Luss, D., and Peck, R., 1964, ‘‘Theoretical Analyses of Vapor
x ⫽ vertical downward distance from nozzle outlet 共m兲 Condensation on Laminar Liquid Jets,’’ Int. J. Heat Mass Transf., 7, pp. 969–
981.
y ⫽ local radius of droplet 共m兲关9兴 Weinberg, S., 1952, ‘‘Heat Transfer to Low Pressure Sprays of Water in a
z ⫽ horizontal coordinate with z⫽0 on the center axis in Steam Atmosphere,’’ Proc. Inst. Mech. Engrs, London, Vol. 1B, No. 6, p.
the temperature measurement plane 共m兲 240–252.
关10兴 Mayinger, F., and Chavez, A., 1992, ‘‘Measurement of Direct Contact Con-
Greek Symbols densation of Pure Saturated Vapor on an Injection Spray by Applying Pulsed
Laser Holography,’’ Int. J. Heat Mass Transf., 35, pp. 691–702.
␣ ⫽ Thermal diffusivity 共m2/s兲关11兴 Kutateladze, S. S., 1952, ‘‘Heat Transfer in Condensation and Boiling,’’ 2nd
␾ ⫽ Spray angle 共rad兲 ed. Chpt. 7, Moscow-Leningrad, English translation by U.S. Atomic Energy
␭ ⫽ latent heat of vaporization 共J/kg兲 Commission, AEC-tr-3770.
␮ ⫽ dynamic viscosity 共N-s/m2兲关12兴 Celata, G. P., Cumo, M., Farello, G. E., and Focardi, G., 1986, ‘‘Direct Con-
tact Condensation of Steam on Slowly Moving Water,’’ Nucl. Eng. Des., 96,
␯ ⫽ kinematic viscosity 共m2/s兲 pp. 21–31.
␪ ⫽ non-dimensional local temperature inside droplet, ␪ 关13兴 Celata, G. P., Cumo, M., Farello, G. E., and Focardi, G., 1989, ‘‘A Compre-
⫽(T⫺T i )/(T sat⫺T i ) hensive Analysis of Direct Contact Condensation of Saturated Steam on Sub-
¯␪ ⫽ non-dimensional mean temperature, ¯␪ ⫽(T sat⫺T̄)/
cooled Liquid Jets,’’ Int. J. Heat Mass Transf., 32, pp. 639–654.
关14兴 Ford, J. D., and Lekic, A., 1973, ‘‘Rate of Growth of Drops During Conden-
(T sat⫺T i ) sation,’’ Int. J. Heat Mass Transf., 16, pp. 61–64.
␪m ⫽ condensation efficiency in Eq. 共19兲, ␪ m ⫽(T̄⫺T i )/ 关15兴 Ohba, K., Kitada, H., and Nishiguchi, A., 1982, ‘‘Direct Contact Condensation
on a High Spray Jet of Subcooled Water,’’ Heat Transfer in Nuclear Reactor
(T sat⫺T i ) Safety, S. G. Bankoff and N. H. Afgan, eds., Hemisphere Publishing Corpo-
␳ ⫽ density 共kg/m3兲 ration.
␴ ⫽ surface tension 共N/m兲关16兴 Rose, P. M., and Kintner, R. C., 1966, ‘‘Mass Transfer from Large Osillalting
Drops,’’ AIChE J., 12, pp. 530–534.
Subscripts 关17兴 Skelland, A. H. P., and Wellek, R. M., 1964, ‘‘Resistance to Mass Transfer
Inside Droplets,’’ AIChE J., 10, pp. 491–496.
d ⫽ droplet 关18兴 Celata, G. P., 1991, ‘‘Direct Contact Condensation of Steam on Subcooled
g ⫽ Vapor Water,’’ Invited Lecture in Phase-Interface Phenomena in Multiphase Flow, G.
i ⫽ initial or nozzle outlet F. Hewitt, F. Mayinger, and J. R. Riznic, eds., pp. 345–371.
Experimental and Theoretical
Study of Heating/Drying of Moist
J. Seyed-Yagoobi
Paper Sheet With a Gas-Fired
A. N. Husain1 Infrared Emitter
Drying Research Center, Infrared (IR) radiation is used for heating/drying of moist paper in the paper making
Department of Mechanical Engineering, process owing to its penetration characteristic inside porous media and a high absorp-
Texas A&M University, tivity of water in the IR wavelength range. However, there is a lack of fundamental
College Station, TX 77843-3123 understanding of the transport phenomena occurring within moist paper during IR
heating/drying. A series of experiments has been conducted along with the development of
a theoretical model to fundamentally understand IR heating/drying of moist paper. The
experimental results confirm the IR radiation energy penetration into the paper sheet and
the theoretical analysis indicates that the penetration depth decreases with increasing
sheet moisture content. 关DOI: 10.1115/1.1372324兴
Keywords: Drying, Emitting, Heat Transfer, Moisture, Porous Media, Radiation
Introduction Though it has been used for a long time, insufficient fundamen-
tal information is available on paper drying with IR heaters.
Heating of media and surfaces is achieved using various modes
Riikonen et al. 关4兴 developed a computer simulation, which com-
of heat transfer namely conduction, convection, and radiation. Pa-
bined IR and convective drying of pulp by adding the radiation
per manufacturing is an energy intensive process that involves
contribution to a previously developed model for convective and
drying of paper primarily with steam-heated cylinders. A method
conductive heat transfer. The IR energy was considered to be
that has been used to increase drying efficiency is infrared 共IR兲
absorbed into the top-most layer of the pulp sheet. It was found
radiation heating/drying. Though IR energy use in the US paper
that the observed experimental drying rates were smaller com-
industry accounted for less than 3 percent in 1980 关1兴, the usage
pared to the model predictions.
has increased significantly in the past decade, and a survey con-
Kuang et al. 关5兴 developed a numerical model to study the dry-
ducted in 1996 for US and Canadian paper industries indicated
ing of paper using a gas-fired IR dryer, taking into account the
that 27 percent of the paper machines were using IR energy for
three modes of heat transfer. The model took into account the
drying purposes 关2兴.
radiation heating as a volumetric absorption that attenuated pro-
The wide application of IR radiation for paper drying purposes
gressively in paper thickness direction. The absorption coefficient
is because of the fact that IR radiation is not solely a surface-
was calculated by the volumetric weighted average of the absorp-
heating source, rather the energy penetrates inside the medium,
tion coefficients of water and solid fiber. Local scattering and
imparting volumetric heating. IR heating also offers the priceless
emission within the medium were neglected. The model devel-
benefit of heating/drying of media in a non-contact mode that
oped demonstrated a high sensitivity to water and paper absorp-
makes its application very attractive for the drying of coated pa-
tion coefficients. The simulated moisture content profiles were in
per. Two conventional types of IR heaters are being commercially
agreement with the experimental results. However, no data were
used namely electric heaters and gas-fired heaters. Electric IR
available for temperature profiles within the paper thickness direc-
heaters emit radiation by passing an electric current through a
tion for comparison with the simulation results.
resistance, which raises its temperature. These are operated up to
Ojala and Lampinen 关6兴 studied the efficiency of gas and elec-
a temperature of 2200°C, and emit near IR radiation 共0.7 to 1.4
tric IR emitters using radiation exchange calculations. Various
␮m兲 which does not correspond to a high absorptivity in moist
dryer configurations were studied. Optical properties of paper
paper, and a portion of the energy is either transmitted through the
were evaluated over discrete wavelength intervals 共0.4 to 20 ␮m兲
paper thickness or reflected 关3兴. In many applications, the trans-
and used in the radiation exchange calculations. A comparison of
mitted energy is redirected by placing back reflectors on the op-
the model with a pilot unit was made but it was vague owing to
posite side of the web thus increasing the efficiency of the electric
the differences in operating conditions and geometry of the pilot
IR heaters. Gas-fired IR heaters produce combustion on the burner
unit and those used in the theoretical model. The optical proper-
surface by ignition of a premixed air and fuel stream. The com-
ties determined for paper 共with varying moisture content兲 by Ojala
bustion raises the surface temperature to ranges of 800–1100°C.
and Lampinen have been used extensively for modeling the IR
The heater surface, which can be a ceramic plate with many holes,
radiation absorption in the paper thickness direction.
a matrix from metallic or ceramic fibers, or porous ceramic foam,
Fernandez and Howell 关7兴 developed a one-dimensional model
incandesces at high temperatures to emit radiation. The radiation
that took into account the three modes of heat transfer with mass
emitted at these temperatures is mainly in the medium wavelength
transfer. The radiation was assumed to be a volumetric phenom-
range 共1.6 to 10 ␮m兲 of the spectrum, which is absorbed well by
enon, with the material internally absorbing, scattering, and emit-
moist paper due to presence of water in the web 共see Fig. 1兲.
ting energy. The absorption and scattering coefficients were spec-
1
trally dependent. Water was assumed to exist in three phases;
Current Address: Carrier Corporation, Carrier Parkway, Syracuse, NY 13221
bound liquid, free liquid, and vapor. The model accounted for
HEAT TRANSFER. Manuscript received by the Heat Transfer Division July 24, 2000; spectral dependence of radiative properties, and the radiation heat
revision received January 22, 2001. Associate Editor: F. B. Cheung. flux was calculated for two different spectral bands. Model simu-
Fig. 2 Schematic of experimental set-up
chamber, is a 16.50 cm circular sample port provided for position-

Fig. 1 Spectral absorption coefficient of water †16‡ ing the test samples parallel to the IR emitter. The sample port is
flush with the upper plate of the heat exchanger, which forms the
floor of the enclosure. The paper samples are placed on a mesh
lations were not compared with any experimental data to validate stand 共0.0083 m2 open area versus a total area of 0.0188 m2兲
the model and the values used for the optical properties. which in turn is placed on a digital weighing balance. The digital
Pettersson and Stenstrom 关8兴 proposed a model based on multi- balance has a resolution of one milligram and a signal output
layered idealization of a paper sheet as proposed earlier by Scallan frequency of 2.5 Hz. The balance is calibrated to account for the
and Borch 关9兴. Their argument, based on microscopic analysis of mass reading indicated due to air flow over the sample. The data
paper structure, was that a beam of radiation had to pass a number acquisition system used for collecting the sample transient tem-
of parallel layers within the paper depth to propagate through the perature and the heat flux data has a data output frequency of 5
medium. The reflectivity and the transmissivity of a single layer Hz.
were fitted to the experimental literature data for the reflectivity The sample port is covered with a radiation insulating shield
and transmissivity of paper at different wavelengths. The model prior to the drying experiment. A 0.127 mm diameter T type
developed predicts the IR energy penetration inside the paper 共copper-constantan junction兲 thermocouple is mounted on the
thickness without accounting for other transport mechanisms that outer periphery of the sample for signaling the starting time in
are present during heating/drying of moist paper. It is therefore heating/drying process when the sample port is exposed to the IR
not possible to compare the model with any drying data. radiation. For surface heat flux measurements, a circular foil heat-
As presented above, limited work has been carried out in the flux gage 共Vatell Corporation, model number 1000-1兲 with a sur-
past to analyze the drying process of moist paper with IR radiation face total hemispherical emissivity of 0.94 is used. The IR emitter
energy. Specifically, there is a lack of fundamental understanding surface temperature measurements are made with an IR camera
of the transport phenomena occurring within the moist sheet dur- 共Mikron, model 6T62兲. These measurements are based on the IR
ing IR heating/drying. The intent of this paper is to provide such emitter surface emissivity of 0.99 determined with the IR camera.
information through experimental study and theoretical analysis.
Paper Samples Construction
Experimental Set-Up The paper samples were made in accordance with the Technical
Details regarding the experimental facility can be found in Association of Pulp and Paper Industries 共TAPPI兲 standard 205.
Wirtz 关10兴 and Husain 关11兴. The facility, shown in Fig. 2, consists The paper samples were all fabricated from linerboard fibers. The
of a single gas-fired IR surface burner in a look down configura- experimental program consisted of two different sets of samples.
tion, suspended in the middle of an aluminum enclosure 共214 cm In the first set, the initial dry basis moisture content 共DBMC兲 of
high, 154 cm wide, and 123 cm deep兲. The purpose of the enclo- the samples was kept constant at 1.5. Fibers with 605 Canadian
sure is to minimize the effect of outside air entrainment. The Standard Freeness 共CSF兲 were used in preparation of samples.
burner consists of a porous ceramic matrix, which is 60 cm wide CSF is a measure of drainability of free water from the stock. The
and 40 cm deep. Combustion air and natural gas are fed into a basis weights 共BW兲 of the samples were selected to be 100, 200,
mixing tube located upstream of the IR emitter. The premixed and 300 g/m2. In the second set, fibers with 680 CSF were used
stream passes through the ceramic matrix and ignites at the sur- and the basis weight of all the samples was 200 g/m2 with initial
face of the porous structure. The ignition produces a very uniform DBMC of 1.5, 1.0, and 0.5.
flame at the exit face of the ceramic that protrudes about 1 mm Single and multi layer samples were tested for both of the stud-
from the surface. The combustion heat is partially transferred to ies. Multi layer samples were used to obtain temperature profiles
the ceramic surface, which incandesces to produce the desired IR within the thickness direction. The multi layer samples consisted
radiation. Aluminum cooling air shroud is provided around the IR of four layers of equal weight with the total weight equal to the
emitter with a 2 mm gap between the edges of the ceramic face required basis weight of the sample. Three T type micro thermo-
and the cooling shroud. Cooling air is introduced through four couples of 0.08 mm diameter were inserted in between the adja-
independent ports and passes out through the gap between the cent layers prior to adhering them together with a press roll ac-
shroud and the ceramic face. cording to TAPPI-205 procedure. Table 1 provides the
On the floor of the rig, parallel to the firing face, is a large information on the location of the thermocouples inserted within
water-cooled heat exchanger, for absorbing incident IR radiation the samples prior to heating/drying process. For each sample, the
energy. The top surface of the heat exchanger, facing the IR emit- initial thickness was measured at four locations 共25 mm in from
ter, is painted black to increase its absorptivity. Located in the the edge on 90 degree intervals兲. These measurements were per-
center of the water-cooled heat exchanger on the floor of the test formed with a dual-plate dial micrometer. Using the assumption
Table 1 Initial sample thickness and thermocouple locations where the effect of water vapor and air components on the aver-
measured from sample surface exposed to IR emitter; initial age specific heat of the sheet has been considered negligible. The
DBMCÈ1.5; Samples: linerboard with 605 CSF. conduction heat flux is given by the Fourier’s law
⳵ T sh
q ⬙ ⫽⫺k eff• (3)
⳵z
and water flux is given by the Darcy’s law
J w ⫽⫺ 冉冊
K ⳵ P ca
␷w
•
⳵z
. (4)
that each target sample was uniform, the depth of the thermo- The vapor flux is given by Fick’s law
冋册
couples from the sample top surface was determined. The maxi-
mum standard deviation for the data given in Table 1 was 0.020 D a p •M W v ⳵ C v
J v ⫽⫺ . (5)
mm. 共 1⫺y v 兲 ⳵z
Details for determining parameters such as P ca , D a p , K, C v , and
Error Analysis ␧ are given in Asensio and Seyed-Yagoobi 关13兴. Similar to the
average specific heat of the sheet 共see Eq. 共2兲兲, the effective ther-
For each individual measurement, the propagation of uncer- mal conductivity of the sheet is also calculated assuming a parallel
tainty was calculated using the method proposed by Kline and heat flow model where fluid and solid phases are parallel, cross-
McClintock 关12兴. The uncertainty associated with each measure- sectional areas are proportional to void and solid fractions, and
ment term was comprised of error induced due to uncertainty phase lengths are equal to the sheet thickness
associated with the measurement device itself and the error in-
duced due to the variance within the experimental data 共the k eff⫽ 共 1⫺␧ 兲 k f ⫹S␧k w ⫹ 共 1⫺S 兲 ␧k a , (6)
t-value corresponding to the confidence level of 90 percent was
used兲. The uncertainty values for the calculated DBMC of 100 where the effect of water vapor on the effective thermal conduc-
g/m2 samples were less than 20 percent while for 200 and 300 tivity has been considered negligible.
g/m2 samples they were below 15 percent. The high uncertainty The two relevant boundary conditions for the present case are
values corresponded to the low moisture contents. For tempera- for the paper bottom surface in contact with the unheated mesh on
ture measurements, the uncertainty values with initial DBMC of which the samples are placed, and for the paper top surface ex-
1.5 were below 15 percent for the top, middle, and bottom ther- posed to the IR radiation heat flux. The surface nodes are modeled
mocouples. The total uncertainty associated with the heat flux as half thickness control volumes in the energy and mass balance
measurement on the target surface was 6.7 percent. As described equations. For the bottom control volume, the mass balance equa-
above, this uncertainty was comprised of uncertainty due to the tion is given as
sensitivity of the heat flux meter 共⫾6 kW/m2兲 and the uncertainty
⳵M 1
in the measured data. Further details in the error analyses are ⫽ 共 ⫺J w ⫺J v ⫹J v ,o 兲 (7)
given in Husain 关11兴. ⳵t ⌬z
and the energy conservation equation is
Theoretical Analysis ⳵ T sh
The theoretical model predicts the temperature and moisture 关 c w ␳ w ␧S⫹c f ␳ f 共 1⫺␧ 兲兴
⳵t
冉冊
distributions inside paper with IR as a heat source and is based on
transport equations developed by Asensio and Seyed-Yagoobi ⳵ T sh
关13兴 for simulation of conventional steam heated cylinder drying 1 k eff ⫺J w H w ⫺J v H v ⫹J v ,o H v
⫽ ⳵z .
of paper. A volumetric heat source term to account for radiation ⌬z
penetration is added to this basic model. Boundary conditions are ⫹h botttom共 T amb⫺T sh 兲 ⫹q r⬙ 兩 z⫽⌬z ⫺q r⬙ 兩 z⫽0
also modified to represent the experimental conditions of this (8)
study. Note that the sheet shrinkage and porosity variation that
occur due to moisture and temperature changes are also accounted For the top control volume exposed to IR radiation, the mass
for in the original model by Asensio and Seyed-Yagoobi. The balance equation is given as
following provides a brief description of the theoretical model.
The local moisture content and temperature of the sheet are ⳵M 1
⫽ 共 ⫹J w ⫹J v ,i ⫺J v ,o 兲 , (9)
assumed to be function of z 共see Fig. 2兲 and IR exposure time. The ⳵t ⌬z
conservation of mass for water is given by
and the energy conservation equation includes the radiation and
⳵M ⳵Jw ⳵Jv convection 共promoted mainly by the IR emitter兲 heat flux terms
⫽⫺ ⫺ . (1)
⳵t ⳵z ⳵z ⳵ T sh
关 c w ␳ w S␧⫹c f ␳ f 共 1⫺␧ 兲兴
The actual mass of vapor and air within the paper, especially in ⳵t
冉冊
the early stages of drying 共this is the case here兲, will be much
smaller than the liquid mass. Thus, under the assumption of no T sh
1 ⫺k eff ⫹J w H w ⫹J v ,i H v ⫺J v ,o H v
vapor or air accumulation, only a single continuity equation needs
⫽ ⳵z
to be written. ⌬z
The conservation of energy is as follows: ⬙ ⫹q r⬙ 兩 z⫽Z ⫺q r⬙ 兩 z⫽Z⫺⌬z
⫹q conv
⳵ T sh ⳵ q ⬙ ⳵ 共 H w J w ⫹H v J v 兲 (10)
关 c w ␳ w S␧⫹c f ␳ f 共 1⫺␧ 兲兴 ⫽⫺ ⫺
⳵t ⳵z ⳵z q ⬙conv represents the convective contribution of heating on the top
surface of the sample. This value is determined by deducting the
⳵ q r⬙ 兩 z calculated incident radiative heat flux from the measured total
⫹ , (2)
⳵z heat flux. This is discussed at length in Husain 关11兴. For the
evaporation at the paper-air interface, J v o , the mass transfer co- Table 2 Parameters used for theoretical analysis
efficient is determined from the Chilton-Colburn form of the Rey-
nold’s analogy as described in Asensio 关14兴.
The radiation heat flux term in Eqs. 共2兲, 共8兲, and 共10兲 represents
the amount of energy that penetrates into the paper in the thick-
ness direction and is given by Bouguer’s law
q r⬙ 兩 z ⫽q r⬙ 兩 Z exp 冉冕冊
z
Z
a•dz , (11)
where q r⬙ 兩 Z represents the radiation heat flux incident on the sur-

face of paper exposed to the IR emitter. Note that the z-direction
is defined from the paper sample bottom surface toward the top
共IR exposed兲 surface 共see Fig. 2兲. The amount of energy that
penetrates up to a certain thickness decays exponentially as a
function of the penetration thickness and absorption coefficient
‘‘a.’’ The radiation penetration given by Eq. 共11兲 does not take
the scattering of radiation within the paper into account. Thus, this
analysis is not precise, however, it allows for significant simplifi-
cation of the problem especially that insufficient data are available
in the literature for optical properties of the paper fibers.
To determine the net radiation heat flux incident at the top
surface of paper 共i.e., q r⬙ 兩 Z 兲, a classical radiation exchange calcu-
lation was performed among the IR emitter surface, paper sample
top surface, heat exchanger top surface, and the surroundings 关11兴.
The gas within the enclosure was assumed to be nonparticipating.
The magnitude of q r⬙ 兩 Z was mainly a function of the IR emitter
surface temperature and the distance between the emitter and the
paper sample.
The absorption coefficient in Eq. 共11兲 has been determined by
taking a weighted average of absorption coefficient for dry fiber
and water Experimental Results
a⫽a f 共 1⫺␧ 兲 ⫹␧Sa w . (12) The experimental results discussed in this section are provided
The above equation for calculating the absorption coefficient is an in terms of the average moisture content and temperature distri-
approximation. The porosity and saturation in Eq. 共12兲 at various bution of the sample sets specified earlier. All the experiments
moisture content levels are determined from the paper structure presented here are carried out at an emitter fuel consumption rate
model given by Asensio 关14兴. The absorption coefficient for paper corresponding to 50.1 kW 共209 kW/m2, based on emitter surface
fiber is determined from the following equation based on Bougu- area兲 and an emitter to paper sample surface spacing of 11.5 cm.
er’s law, in terms of the total hemispherical absorptivity and re- At this operating condition, the ceramic surface temperature of the
flectivity of fiber. IR emitter is measured 共with an IR camera兲 to be 934°C 共flame
temperature of 1500°C兲 resulting in 66.0 kW/m2 共determined from
1
a f ⫽ ln 1⫺
L 冉 ␣f
共 1⫺R f 兲
. 冊 (13)
classical radiation calculations兲 of incident radiation heat flux on
the paper sample surface. Note that this incident radiation heat
flux is much lower than the heat flux generated at the emitter
The total hemispherical absorptivity of fiber, ␣ f , was obtained on surface 共based on the fuel heating value兲 since only a portion of
the basis of spectral absorptivity from 0.4 to 20.0 ␮m for a 41 the heat generated at the emitter results in radiation emission from
g/m2 basis weight paper with a moisture content of 6 percent, the IR ceramic surface. More details and additional experimental
measured by Ojala and Lampinen 关6兴. Potential differences in the data for different operating conditions can be found in Husain
fiber types used in this paper and the ones studied by Ojala and 关11兴.
Lampinen contribute towards errors in the theoretical predictions. Figure 3 shows the effect of varying sample basis weight on IR
However, there is very limited information available in the litera- drying. Higher basis weight samples take a longer duration to
ture about optical properties of paper. The value for the total achieve a certain DBMC due to large amount of water present in
hemispherical reflectivity of fiber, R f , was obtained from Kuang the sample. A brief period of approximately one second of con-
关15兴. The value for absorption coefficient of water, a w , has been stant DBMC 共no moisture loss兲 is visible for the 300 g/m2 basis
taken from Kuang et al. 关5兴 for a temperature that corresponded to weight samples which represents the initial sensible heating. The
the operating temperature of the IR heater in the current experi- lower basis weights of 100 g/m2 and 200 g/m2 do not indicate a
ments. Table 2 gives the values of these properties along with the constant DBMC period, rather, rapid moisture reduction is evident
other parameters utilized in this study. in both cases. This is primarily due to the fact that the thermal
The theoretical model presented above was solved numerically mass 共water and fiber兲 associated with lower basis weight samples
using finite difference techniques. A marching technique 共fully at identical initial DBMC of 1.5 is small resulting in a very short
explicit兲 in the time direction was employed. The paper thickness sensible heating period under similar IR radiation energy level.
was divided into twelve equally spaced nodal points. The step size Note that all the samples at initial DBMC of 1.5 are saturated with
in the time direction was 1⫻10⫺4 s. There was no observable water. This shorter duration of sensible heating for lower basis
difference in the moisture or temperature values when more nodal weight samples has not been captured in Fig. 3 owing to relatively
points in the z-direction or smaller steps in time direction were low frequency of data acquisition system for the weight
used. Numerical convergence with respect to the temperature and measurement.
moisture content 共weight of water divided by weight of dry fiber兲 Figure 4 illustrates the effect of initial DBMC on IR drying for
in the z-direction corresponded to 10⫺3 K and 10⫺3 , respectively. 200 g/m2 basis weight samples 共680 CSF兲. The drying time is
More details on the numerical solutions are given in 关11兴 and 关14兴. shorter for lower initial DBMC samples due to smaller amount of
Fig. 5 Temperature profile for 100 gÕm2 basis weight sample;
initial DBMCÈ1.5; samples: linerboard with 605 CSF.
Fig. 3 Effect of basis weight on IR drying; initial DBMCÈ1.5;
samples: linerboard with 605 CSF.
drying. However, initially the sheet thickness, thus, the location of
the thermocouples remain unchanged due to the lack of significant
water content within the sample. A sensible heating period of moisture loss in the early stage of the sample exposure to the IR
approximately 1.0 s and 0.7 s is visible for samples with initial energy.
DBMC of 0.5 and 1.0, respectively. The sensible heating period Figure 5 illustrates the temperature profile of 100 g/m2 共605
was not captured for samples with initial DBMC of 1.5. The CSF兲 basis weight samples with an initial DBMC of 1.5. The
longer duration of sensible heating for samples with initial DBMC temperature response for all the layers in Fig. 5 can be divided
of 0.5 can be related primarily to the unavailability of water on the into three distinct regimes. The process starts with sensible heat-
sample surface. For samples with the initial DBMC of 1.5, the ing indicated by an increase in temperature. The initial sensible
surface is saturated with water and with the interception of the IR heating is followed by evaporation at constant temperature, to-
energy, the water rapidly evaporates. wards the end of which, the process of moisture removal is essen-
The discussion on the temperature distribution in the paper tially completed. At the completion of constant temperature
thickness direction is based on the temperature data obtained from evaporation, a second sensible heating follows. From Fig. 5, a
the thermocouples embedded within the layers of paper, as men- sharp temperature rise is evident for all the three layers. However,
tioned earlier. The initial calculations indicated no need for the the top layer demonstrates the most rapid temperature rise, owing
radiative corrections on the thermocouple outputs. The top, to its closer location to the surface exposed to IR emitter. The
middle, and bottom layers represent the temperature measured by energy that reaches the top, middle, and bottom layers can be due
the thermocouples between the first and second, second and third, to IR radiation penetrating to this depth, as well as the energy that
and third and fourth layers, respectively. See Table 1 for the initial propagates by conduction and convection through the thickness.
location of the embedded thermocouples. Note that the location of The response of the middle and bottom layers demonstrates that
the thermocouples changes due to the shrinkage of samples with the magnitude of initial temperature rise decreases with increasing
distance from the surface exposed to the IR emitter.
As shown in Fig. 5, the duration of constant temperature evapo-
ration stage for the top layer is shorter compared to the middle and
bottom layers. This is representative of moisture transport mecha-
nism within paper during drying process. Transport of water vapor
from the top layer occurs in either direction with evaporation oc-
curring from the surface exposed to the IR heating as well as
vapor transfer to the lower layers. The water vapor reaching the
lower layers condenses back to liquid phase first followed by
evaporation period. This was observed experimentally by Wirtz
关10兴 through weight measurement of the individual layers at dif-
ferent times. In addition to the above, the evaporation time is
longer for lower layers because the amount of energy reaching the
lower layers decreases with distance. In Fig. 5, the evaporation
temperature is below 100°C corresponding to partial saturation
pressure of the water vapor. As shown in Fig. 5, the middle layer
temperature approaches the top layer temperature following 7 s of
IR exposure time because the top layer starts to discolor, crack,
and eventually ignite.
Effect of an increased basis weight on the temperature profile is
illustrated in Fig. 6, which shows the temperature response of 300
g/m2 samples 共605 CSF兲. Distinct regimes of sensible heating,
evaporation, and final sensible heating are visible. A delay in re-
sponse time, especially of the bottom layer temperature, in con-
Fig. 4 Effect of initial DBMC on IR drying; basis weight trast to the top layer immediate response, is indicative of the fact
Ä200 gÕm2 ; samples: linerboard with 680 CSF. that the IR energy initially does not penetrate this far 共1.600 mm
Table 3 Calculated initial temperature rise „based on pure
conduction heat transfer analysis… versus measured initial tem-
perature rise „from Fig. 6…; basis weightÄ300 gÕm2, initial
DBMCÈ1.5; samples: linerboard with 605 CSF
magnitude of the depth of penetration is dependent on the operat-

Fig. 6 Temperature profile for 300 gÕm2 basis weight sample; ing conditions. The data presented here, however, prove the fact
initial DBMCÈ1.5; samples: linerboard with 605 CSF. that the IR radiation energy should be treated as a volumetric
phenomenon rather than a surface phenomenon with respect to a
porous paper sheet.
and beyond from Table 1兲. The depth of penetration of the IR
energy into the sheet is dependent on various parameters includ- Comparison of Numerical and Experimental Results
ing IR emitter operating conditions and paper sheet moisture The numerical solutions presented here are based on the theo-
level. retical model presented earlier. Transient moisture content and
To estimate the initial penetration depth of the IR energy under local temperature profiles in the paper sheet are simulated and
the operating conditions corresponding to Fig. 6, a classical con- compared with the experimental data. The values used for the
duction heat transfer analysis was carried out to determine the parameters in the theoretical model are given in Table 2. Param-
temperature rise in the locations of the first, second, and third eters, such as water properties, not listed in this table are readily
thermocouples that would have resulted if the incident radiation available in the literature. As shown in the theoretical model, the
heat flux on the sample upper surface was only a surface phenom- properties are saturation dependent and their values do change
enon. Since the sample at DBMC of 1.5 is completely saturated, during the drying process. The net IR energy incident on paper
the heat transfer due to mass transport within the sheet is initially sheet surface, q r⬙ 兩 z⫽Z , is a calculated value as described earlier in
absent. Assuming that the IR energy incident on the sample sur-
the paper. The convection heat flux on the top surface of the paper
face does not penetrate into the sheet, the only mode of transfer of
heat into the sample will be by conduction. A heat conduction sample, q ⬙conv , is determined by subtracting q r⬙ 兩 z⫽Z from the total
analysis for a semi-infinite body under a constant surface heat flux measured heat flux using the heat flux sensor.
共sum of the radiation and convection heat fluxes兲 was performed Figure 7 compares the theoretical and experimental DBMC as a
and the results are summarized in Table 3. In the conduction function of IR exposure time for 100, 200, and 300 g/m2 samples.
analysis, it was assumed that the saturated sheet had properties of Results are provided up to an IR exposure time of 3.0 s to avoid
water, thus exaggerating the calculated temperature values. Also
included in this table are the corresponding measured tempera-
tures. The micro thermocouples used in this study have a response
time of approximately 0.05 s. According to Table 3, the contribu-
tion of the pure conduction heat transfer to temperature rise is
negligible during the first 0.2 s of exposure to IR energy. On the
other hand, the measured temperature rise is significant even at
0.1 s of exposure to IR energy at 0.401 mm from the surface. For
0.803 mm from the surface, there is a small 0.3°C temperature rise
at 0.1 s indicating a negligible portion of the IR radiation energy
penetrating this far into the sample. It must be mentioned that with
the IR energy penetrating into the sheet, the conduction heat trans-
fer 共in the early stage of exposure to IR energy兲 from the IR
penetrated region to the region unaffected by the IR energy is
promoted due to the increase in the temperature of the IR affected
region. A similar analysis was carried out for a 100 g/m2 sample
based on the data given in Fig. 5, and it was concluded that the IR
energy did penetrate to the third thermocouple location 共i.e. 0.382
mm, see Table 1兲. In the case of 200 g/m2 samples, the experi-
mental data 共not given in this paper, see Husain 关11兴兲 revealed that
the IR energy reached the middle thermocouple location 共0.441
mm兲 but barely penetrated up to the third thermocouple location
共0.667 mm兲. From the above discussion, one can assume that
under the corresponding operating conditions, the IR radiation Fig. 7 Comparison of theoretical predictions with experimen-
energy penetrated into the saturated sheet up to approximately tal data; DBMC versus IR exposure time for three different ba-
0.667 mm from the surface. As previously mentioned, the exact sis weights; samples: linerboard with 605 CSF.
Fig. 8 Comparison of theoretical predictions with experimen- Fig. 9 Theoretical penetration depth for 1 percent of incident
tal data; temperature versus IR exposure time for BW radiation energy versus DBMC „distributed uniformly…;
Ä100 gÕm2, initial DBMCÈ1.5; samples: linerboard with 605 samples: linerboard with 605 CSF.
CSF.
the exposure to the IR energy, is very small compared to the

theoretical predictions in the hygroscopic region. This is because radiative heat source term 共the last term on the right hand side of
the theoretical model does not take into account the effect of Eq. 共2兲兲. Therefore, one can conclude that in the beginning 共e.g.,
bound water transport that becomes pronounced at low moisture within the IR exposure time of less than 1.0 s兲, the dominant terms
content levels. According to Fig. 7, the theory demonstrates a in the energy balance 共i.e. Eq. 共2兲兲 are the transient and radiative
trend that is similar to the experimental data. For basis weight of heat source terms, indicating that the IR energy initially raises the
300 g/m2, the agreement between the theory and the experimental sheet temperature over time as illustrated in Figs. 7 and 8. Fol-
data is very good, however, the agreement becomes worse at lowing the initial stage of the exposure to IR energy, all the terms
lower basis weights where the samples dry quickly. This is pri- given in Eq. 共2兲, contribute to the local energy balance within the
marily due to 共1兲 limitation of the theoretical model, especially at sheet. Further details on the transport mechanisms taking place
low moisture content levels 共e.g., exclusion of bound water ef- within a moist sheet during the drying process in the absence of
fects兲, 共2兲 limitation of the IR energy penetration model due to exposure to the IR energy are given in 关13兴.
neglecting internal scattering within the paper, and 共3兲 lack of data Using Eqs. 共11兲, 共12兲, and 共13兲, the initial depth of IR energy
for optical properties of paper fibers. The optical properties of penetration 共defined based on 1 percent of the radiation energy
paper fibers significantly affect the IR radiation energy penetration incident at the surface兲 is presented as a function of initial mois-
into the sheet. Their effect on drying of a thin paper 共low BW兲 ture content in Fig. 9. The radiation penetration depth decreases
becomes more pronounced since the depth of IR energy penetra- with increasing moisture content. This is because of high absorp-
tion becomes comparable to the sheet thickness. As presented tivity of water versus fiber within the corresponding wavelength
later in this section, the depth of IR energy penetration increases range of the IR radiation energy. According to Fig. 9, the depth of
with a decrease in the moisture content indicating the importance penetration at DBMC of 1.5 is approximately 0.24 mm. This is
of the values of the optical properties of paper fibers especially at much lower than the experimentally predicted value. However, it
the later stage of drying. must be realized that the experimentally determined depth of pen-
Figure 8 compares the predicted temperature profiles with the etration was determined primarily based on the temperature values
experimental data for a basis weight of 100 g/m2 and initial corresponding to 0.1 s of IR exposed time. As shown in Fig. 3, the
DBMC of 1.5. The trend demonstrated by the simulation is very DBMC of the sample, especially with BW of 100 g/m2, decreases
similar to the experimental data. Sensible heating is predicted for within this time period. This drop in the moisture content is solely
all the three layers with the most rapid temperature rise demon- from the top layer giving a lower DBMC for this layer, which
strated for the top layer followed by the middle and bottom layers. should result in a slightly higher theoretical depth of penetration
Quantitatively, the theoretical predictions compare well with the than 0.24 mm. Nevertheless, the theoretically predicted value is
experimental data for the top layer, but the deviation becomes much smaller than the experimentally determined value. The dis-
larger with the progressive layers. The deviation becomes worse crepancy between the theoretical and experimental results is pri-
for the 300 g/m2 sample, which shows a qualitative, rather than a marily attributed to the errors associated with the values assumed
quantitative agreement with the experimental data. The theoretical for the optical properties. The depth of penetration is strongly
predictions also deviate from the experimental data for the top dependent on the absorption coefficient 共see Eq. 共11兲兲. Other fac-
layer once this layer reaches the second sensible heating region. tors discussed earlier in this section also contribute to this
The shrinkage and degradation of the top layer, because of the discrepancy.
over exposure to the IR emitter, contribute to this deviation.
In the early stage of the sample exposure to the IR emitter, due Conclusion
to the lack of significant moisture loss, the contribution of the Experimental and theoretical results of drying of paper using IR
convective liquid phase and vapor phase energy transfer 共the sec- radiation energy as a heating source have been presented. The
ond term on the right hand side of Eq. 共2兲兲 to the total energy measurement of temperature distribution inside the paper thick-
balance within the moist sheet is negligible. Furthermore, as dis- ness and transient DBMC provide a good basis for understanding
cussed earlier in this paper, the conduction term 共the first term on of the IR paper drying process. The theoretical model gives quali-
the right hand side of Eq. 共2兲兲, especially during the early stage of tatively comparable trends to experimental data. The experimental
data indicate a higher IR energy depth of penetration into the Subscripts
paper sample than the numerical prediction. Absorption coeffi- a ⫽ air
cient theoretically determines the level of IR penetration within a amb ⫽ ambient
medium and it is therefore a critical parameter in the theoretical ap ⫽ apparent
model. Thus, there is a great need for accurate measurement of bd ⫽ bone dry
optical properties of various types of paper fibers. ca ⫽ capillary
conv ⫽ convection
Acknowledgment e ⫽ emitter
This work was partially supported by the Drying Research Cen- eff ⫽ effective
ter at Texas A&M University. The gas-fired IR burner was pro- f ⫽ fiber
vided by Measurex-Devron. ins ⫽ instantaneous
r ⫽ radiation
Nomenclature sh ⫽ paper sheet
v ⫽ vapor
A ⫽ area of the sample 共m2兲 v ,i ⫽ vapor from interior
a ⫽ absorption coefficient 共1/m兲 v ,o ⫽ vapor to exterior
BW ⫽ basis weight; 共weight of air dry sheet/sheet area兲, w ⫽ water
共g/m2兲
C ⫽ molar concentration 共kmol/m3兲 References
c ⫽ specific heat 共J/kg K兲关1兴 McConell, R. R., 1980, ‘‘A Literature Review of Drying Research in the Pulp
CSF ⫽ Canadian Standard Freeness and Paper Industry,’’ Drying, Hemisphere Publishing Corp., Washington,
D ⫽ diffusivity 共m2/s兲 D.C., 2, p. 330.
DBMC ⫽ dry basis moisture content, DBMC⫽W ins⫺W bd /W bd 关2兴 Bean, C. E., and Cocagne, J. M., 1996, ‘‘Assessment of Gas-fired Infrared
Heaters in the Paper Industry,’’ Topical Report, GRI-96/0087, Gas Research
h bottom ⫽ heat transfer coefficient for bottom paper surface ex- Institute, Chicago, IL, p. 3.
posed to ambient conditions 共W/m2K兲关3兴 Pettersson, M., and Strenstrom, S., 1998, ‘‘Evaluation of Gas-fired and Elec-
H ⫽ enthalpy 共J/kg兲 trically Heated Industrial Infrared Paper Dryers,’’ Proceedings of International
J ⫽ mass flux 共kg/m2 s兲 Gas Research Conference, San Diego, CA, p. 213.
关4兴 Riikonen, J., Jarkonen, E., and Palosaari, S., 1987, ‘‘Modeling of Infrared
K ⫽ permeability 共m2兲 Drying of Pulp,’’ Drying ’87, Hemisphere Publishing Corp., New York, NY,
k ⫽ thermal conductivity 共W/m K兲 pp. 18–23.
k m ⫽ mass transfer coefficient 共kg/m2 s兲关5兴 Kuang, H., Thibault, J., Grandjean, B., and Chen, R., 1994, ‘‘Study of Heat
L ⫽ thickness of bone dry sheet with 41 g/m2 basis and Mass Transfer during Drying of Paper,’’ Drying Technol., 3, pp. 545–575.
关6兴 Ojala, K. T., and Lampinen, M. J., 1995, ‘‘Modeling, Measurements, and
weight Efficiencies of Infrared Dryers for Paper Drying,’’ Handbook of Industrial
M ⫽ mass of water in sheet 共kg/m3兲 Drying, 2, A. S. Mujumdar, ed., ISBN no. 0-8247-9644-6, pp. 931–976.
MW ⫽ molecular weight 共kg/kmol兲关7兴 Fernandez, M. L., and Howell, J. R., 1997, ‘‘Radiative Drying Model of Po-
P ⫽ pressure 共Pa兲 rous Materials,’’ Drying Technol., 15, No. 10, pp. 2377–2339.
关8兴 Pettersson, M., and Strenstrom, S., 1998, ‘‘Absorption of Infrared Radiation
q ⫽ heat transfer 共W兲 and Radiation Transfer Mechanism in Paper,’’ J. Pulp Pap. Sci., 24, No. 11,
q ⬙ ⫽ heat flux 共W/m2兲 pp. 349–355.
R ⫽ total hemispherical reflectivity 关9兴 Scallan, A. M., and Borch, J., 1972, ‘‘An Interpretation of Paper Reflectance
Based upon Morphology: I. Initial Considerations,’’ Tappi J., 55, No. 7, pp.
S ⫽ saturation; volume of liquid/volume of voids 583–588.
t ⫽ instantaneous drying time 共s兲关10兴 Wirtz, J. W., 1999, ‘‘An Experimental Study of Gas-Fired Infrared Drying of
T ⫽ temperature 共K兲 Paper,’’ M.S. thesis, Texas A&M University, College Station, TX.
W ⫽ weight of sample 共kg兲关11兴 Husain, A. N., 2000, ‘‘Measurement and Analysis of Heating of Paper with
Gas-Fired Infrared Burner,’’ M.S. thesis, Texas A&M University, College
y ⫽ mole fraction Station, TX.
z ⫽ paper thickness direction 共m兲关12兴 Kline, S., and McClintock, F., 1953, ‘‘Describing Uncertainities in Single
Z ⫽ paper sample thickness 共m兲 Sample Experiments,’’ Mech. Eng. 共Am. Soc. Mech. Eng.兲, 75, pp. 3–8.
关13兴 Asensio, M. C., and Seyed-Yagoobi, J., 1993, ‘‘Simulation of Paper-Drying
Greek Symbols Systems With Incorporation of an Experimental Drum/Paper Thermal Contact
␣ ⫽ total hemispherical absorptivity Conductance Relationship,’’ ASME J. Energy Resour. Technol., 115, No. 4,
pp. 291–300.
⌬z ⫽ step size in z direction 共m兲关14兴 Asensio, M. C., 1992, ‘‘Simulation of Paper Drying with Incorporation of an
⑀ ⫽ emissivity Experimental Drum/Paper Thermal Contact Conductance Relationship,’’ M.S.
␧ ⫽ porosity; volume of voids/total volume thesis, Texas A&M University, College Station, TX.
关15兴 Kuang, H., 1994, ‘‘Le Sechage du Papier Par Infrarouge,’’ Ph.D. dissertation,
␭ ⫽ wavelength 共␮m兲 Universite Laval, Quebec, Canada.
␷ ⫽ kinematic viscosity 共m2/s兲关16兴 Siegel, R., and Howell, J. R., 1992, ‘‘Thermal Radiation Heat Transfer,’’ Third
␳ ⫽ density 共kg/m3兲 Edition, Hemisphere Publishing Corp., Washington, D.C., pp. 175–176.
The Effect of Dissolving Gases or
Solids in Water Droplets Boiling
on a Hot Surface
Qiang Cui We conducted experiments on the effect of dissolving either a gas (carbon dioxide) or a
solid salt (sodium carbonate or sodium bicarbonate) in water droplets boiling on a hot
Sanjeev Chandra stainless steel surface. Substrate temperatures were varied from 100°C to 300°C. We
Mem. ASME recorded the boiling of droplets with a video system, and photographed droplet impact
e-mail: chandra@mie.utoronto.ca using short-duration flash photography. At surface temperatures that were too low to
initiate nucleate boiling, dissolved salts were found to reduce the evaporation rate since
Susan McCahan they lower the vapor pressure of water. Dissolved gas had the opposite effect: it came out
of solution and formed bubbles in the liquid, enhancing evaporation. In the nucleate
Department of Mechanical and boiling regime dissolved carbon dioxide enhanced heat transfer by a small amount. How-
Industrial Engineering, ever, sodium carbonate prevented coalescence of vapor bubbles and produced foaming in
University of Toronto, the droplet, greatly enhancing heat transfer and reducing the droplet lifetime to approxi-
Toronto, Ontario M5S 3G8, Canada mately half that of a pure water drop. Sodium bicarbonate, which decomposes to give
carbon dioxide and sodium carbonate when heated, produced an even larger enhance-
ment of heat transfer. When the surface temperature was raised above the Leidenfrost
temperature of water, droplets went into film boiling and bounced off the surface follow-
ing impact. Dissolved carbon dioxide was found to suppress heterogeneous bubble for-
mation in the droplet during impact. However, dissolved salts promoted bubble formation
and led to droplet break-up during impact. 关DOI: 10.1115/1.1376394兴
Keywords: Additives, Boiling, Droplet, Enhancement, Heat Transfer, Sprays
Introduction attractive fire extinguishing agent available; unfortunately it is not

efficient enough to use on board aircraft and vehicles, where the
Spray cooling is frequently used in industrial processes to dis-
weight of liquid that can be carried is severely restricted. Much
sipate extremely high heat fluxes. A solid surface loses heat much
recent research has been directed at finding additives that can
more rapidly to small droplets of liquid impinging on it than it enhance the fire suppression capabilities of water. Finnerty 关6兴
does to the same liquid flowing continuously over it. High heat dissolved an assortment of inorganic and organic salts in water
transfer rates are beneficial because they allow the size, cost, and sprays used to extinguish a liquid pool fire of standard size. A
complexity of heat exchange equipment to be reduced. Practical pure water spray took 31.9 s to extinguish the fire. Sprays in
applications of water sprays include quenching of metals in which a salt was dissolved took from 3.2 s to 42.7 s depending on
foundries, fire fighting, and emergency core cooling in nuclear the additive used. The reason for this large variation was not clear.
plants. King et al. 关7兴 studied the effect of some of the salts that Finnerty
To design spray cooling equipment we need to know how the
heat transfer rate from the surface varies with the size, velocity
and physical properties of impacting spray droplets. To provide
such information many experimental studies have examined water
droplets evaporating on a hot surface 共e.g., 关1–3兴兲. The water used
in all these experiments was carefully degassed and purified to
eliminate any impurities that could influence boiling. However,
water sprays used in industrial applications typically contain sig-
nificant amounts of dissolved salts and gases. We need to under-
stand the effect of these contaminants on boiling heat transfer to
accurately predict cooling rates. Even more usefully, if some im-
purities are found to significantly enhance heat transfer, we can
deliberately add controlled amounts of them to the water to in-
crease cooling efficiencies. For example, Qiao and Chandra 关4,5兴
have demonstrated that adding traces of a surfactant to droplets or
sprays impinging on a heated surface can increase nucleate boil-
ing heat transfer by up to 300 percent.
The use of additives to enhance spray cooling has taken on
added importance in the light of efforts to find replacements for
Halon 1301, which is a very effective fire suppression agent, but
whose manufacture has been banned because it depletes atmo-
spheric ozone. Environmental considerations make water the most

HEAT TRANSFER. Manuscript received by the Heat Transfer Division July 31, 2000;
revision received January 11, 2001. Associate Editor: F. B. Cheung. Fig. 1 Schematic diagram of the experimental apparatus
Fig. 2 Evaporation of droplets on a hot surface at 100°C: „a… pure water; „b… water containing dissolved CO2 ;
„c… 1 percent by weight solution of Na2CO3 ; „d… 1 percent by weight solution of NaHCO3
关6兴 had found to have a large effect, by placing droplets of salt transfer coefficients. The magnitude of heat transfer enhancement
solution on a hot surface 共at temperatures below the boiling point increased with gas solubility. Torikai et al. 关11兴 observed that the
of water兲 and photographing their evaporation. They found that wall superheat necessary to initiate pool boiling of water de-
dissolved salts decreased droplet evaporation rates. creased with increasing air content. You et al. 关12兴 did an exten-
In the absence of any information on the effect of solid or sive experimental investigation on the enhancement of pool boil-
gaseous impurities in droplets boiling on a hot surface, the most ing heat transfer using highly wetting dielectric fluids 共FC series兲.
relevant information is available from pool boiling experiments. Dissolved gases were found to influence boiling initiation only
Jamialahmadi and Müller-Steinhagen 关8兴 and Najibi et al. 关9兴 in- when gas content was high. Jeschar et al. 关13兴 measured the cool-
vestigated the effect of dissolved salts on pool boiling of water. ing rate of a hot sphere dropped into a water bath in which carbon
They found that, in general, dissolved salts reduce heat transfer. dioxide and various salts were dissolved. They reported that high
This effect was most pronounced for salts with low solubility that concentration of dissolved gas yielded a more stable vapor film on
precipitated on to the heater surface and formed a thermal barrier. the heater surface and reduced the temperature at which film boil-
Dissolved gases also affect pool boiling heat transfer. Muller- ing was initiated.
Steinhagen et al. 关10兴 studied sub-cooled boiling of heptane and Studies on the effect of dissolving gases and salts in boiling
water and established that dissolved gases always increased heat water have not considered the case of droplets placed on a hot
surface. Boiling of droplets is distinct from pool boiling in that they did not significantly change surface tension. Previous mea-
heat transfer is to a thin liquid film, so that vapor bubbles pro- surements of the effects of these additives on the surface tension
duced at the hot surface cannot rise away but tend to coalesce. of water 关17,18兴 confirm that they have little influence at low
This large vapor mass distorts the droplet shape and alters the concentrations.
liquid-solid contact area. Therefore, additives that can influence Evaporation of droplets deposited on the hot surface was re-
bubble nucleation and coalescence are of interest in the study of corded using a video camera, and evaporation time for each drop-
droplet boiling. let measured using a video timer. The resolution of our measure-
We deposited water droplets, in which either a solid salt or gas ments was 33 ms, corresponding to the 30 frames per second
was dissolved, on a heated stainless steel surface and photo- recording rate of the video camera. Droplet diameter and liquid-
graphed them as they boiled. Based on information found in the solid contact angle during evaporation was measured by importing
pool boiling literature and some preliminary experiments, we ex- video images into a computer-based image analysis package. The
pected sodium carbonate (Na2CO3) to have significant influence resolution of our measurement was restricted to the size of a
on droplet boiling and selected it as the first salt to study. The gas single pixel of the video image, which, at the magnification used
used was carbon dioxide (CO2), because it is highly soluble in in our experiments, was ⫾ 0.02 mm. Impacting droplets were
water. Finally we experimented with dissolving sodium bicarbon- photographed by using either a 35 mm camera or a digital still
ate (NaHCO3), which decomposes when heated to yield Na2CO3 camera to take a single flash photograph of an impacting droplet.
and CO2, thus combining their individual effects. Substrate tem- By varying the time between the droplet landing on the surface
perature was varied in our experiments from 100°C to 300°C, and triggering the flash, different stages of droplet impact could
encompassing the entire range of heat transfer regimes from be recorded. The impact process was repeatable enough that we
nucleate to film boiling.
Experimental Method
The water used in our experiments was successively distilled,
deionized and degassed. Distilled water was deionized in a
NANOpure Bioresearch Deionization System 共Model #D4754,
Barnstead Thermolyne Corporation兲 to an electrical resistivity of
18.3 M⍀-cm. A stainless steel chamber 共88.9 mm diameter and
152.4 mm height兲 was filled with deionized, distilled water and
placed in a Dewar flask filled with dry ice. Once the water froze
the space above it was evacuated for over an hour. The water was
then allowed to melt, after which the freezing and evacuation
process was repeated two more times.
Salt solutions were prepared by dissolving 1 percent by weight
of powdered Na2CO3 or NaHCO3 共99.7 percent pure, ACP
Chemical Inc.兲 in purified water at room temperature, and then
allowing the solution to stand for about 12 hours to ensure com-
plete mixing. The salt concentrations were well below their solu-
bility limits in water, which are 18 percent by weight for Na2CO3
and 9 percent by weight for NaHCO3 at room temperature 关14兴.
Carbon dioxide was dissolved directly in water contained in the
stainless steel vessel used for degassing water by pressurizing it
with gas at 1 bar from a CO2 tank for two days. The container was
shaken occasionally to speed up the diffusion rate of carbon diox- Fig. 3 Variation of liquid-solid contact diameter with time dur-
ide into the water. A Natelson microgasometer 共Model #650, Sci- ing evaporation of droplets on a stainless steel surface at
entific Industries Inc.兲 that adapts the classical Van Slyke 关15兴 100°C
manometeric method, was used to determine the gas content. The
carbon dioxide content of water measured in the tests was typi-
cally 0.74 mm3/mm3 at STP, which was close to the saturated
value 共0.75 mm3/mm3 at STP兲 reported in the literature 关16兴.
The apparatus shown in Fig. 1 was used to record the impact
and boiling of a single droplet on a hot surface. It consisted of a
droplet generation and delivery system, a heated test surface on
which droplets were placed, and a camera. The test surface, 50.8
mm square and 6.35 mm thick, was made of stainless steel and
finished with 600 grit emery cloth and metal polish. During ex-
periments it was cleaned with a cotton swab soaked in distilled
water each time a droplet was deposited on it. The test plate was
bolted to a copper block in which two cartridges heaters were
embedded. Chromel-Alumel thermocuples were used to measure
surface temperature, which was held constant within ⫾0.5°C by a
temperature controller before depositing a droplet on it.
Droplets were formed by allowing water to flow from the res-
ervoir though a needle valve and detach under its own weight
from the tip of a 33-gauge stainless steel needle held 50 mm
above the test surface. Droplet diameter was measured from pho-
tographs of spherical droplets falling after detachment and found
to be 2.0⫾0.02 mm in all our experiments. There was no measur-
able difference between the diameter of droplets in which any of Fig. 4 Variation of droplet-air interface area during evapora-
the three additives was dissolved, from which we concluded that tion of droplets on a stainless steel surface at 100°C
Fig. 5 Boiling of droplets on a hot surface at 130°C: „a… pure water; „b… water containing dissolved CO2 ; „c…
1 percent by weight solution of Na2CO3
Fig. 6 Boiling of 1 percent by weight solution of NaHCO3 droplets on a hot surface at 130°C
than that in a drop of pure water. The bubble produced little
change in the contact diameter 共Fig. 3兲, though it did increase the
droplet surface area exposed to air 共see Fig. 4, 0⬍t⬍5 s兲. The
increased area for evaporation produced a small decrease in drop-
let lifetime. The effect is not very significant because, as shown
by computer simulations 关1,2兴, most heat transfer to an evaporat-
ing droplet occurs around its edge where the substrate is hottest
and the liquid film is thinnest. Therefore the length of its periph-
ery, rather than surface area, principally determines droplet evapo-
ration time.
Addition of Na2CO3 or NaHCO3 to water droplets placed on a
surface at 100°C led to an increase in their lifetime 共Fig. 2共c兲 and
共d兲兲, in agreement with the findings of King et al. 关7兴. Dissolving
a salt in water is known to reduce its vapor pressure. The magni-
tude of this effect can be calculated using the method described by
Cisternas et al. 关19兴. We estimated that the reduction of vapor
pressure is small for dilute solutions: dissolving 1 percent by
weight of NaHCO3 reduces the vapor pressure to 100 kPa at
100°C, only slightly lower than that of pure water 共101 kPa兲.
However, as the droplet evaporated and its salt concentration in-
creased, the lowering in vapor pressure would become more pro-
Fig. 7 Variation of droplet lifetime for droplets deposited on a nounced. The decrease of vapor pressure would diminish the rate
stainless steel surface at temperatures ranging from 100°C to of mass transfer from the droplet surface and increase droplet
210°C evaporation time.
As the droplets of salt solution became smaller salt crystals
precipitated along their edges, where the evaporation rate was
could reconstruct droplet impact dynamics from a sequence of highest 共see Fig. 2共c兲 and 共d兲 at t⫽31.55 s兲. The ring of salt pre-
such photographs. Qiao and Chandra 关3,4兴 have described the vented inward movement of liquid at the edges of the droplets
method in greater detail. 共see Fig 2共c兲 and 共d兲, t⫽31.55 s兲. In fact droplets spread out
slightly towards the end of their life, as is evident in both photo-
Results and Discussion graphs 共see Fig. 2共c兲 and 共d兲 at t⫽26.55 s兲 and our measurements
When water droplets were deposited on a stainless steel plate at of droplet diameter. Figure 3 shows that droplet contact diameter
a temperature below 120°C, the surface temperature was not high remained constant for droplets that had NaHCO3 dissolved in
enough to trigger bubble nucleation and droplets evaporated with- them, whereas droplets of pure water and those containing CO2
out initiating nucleate boiling. Figure 2 shows photographs of shrank rapidly towards the end of their lifetime (t⬎20 s). The salt
water droplets, and those containing the three different additives stain left after all the water had evaporated was in the form of a
used in our experiments, evaporating on a surface at 100°C. Each circle along which there were regularly spaced fingers, perhaps
column in Fig. 2 shows successive stages during the droplet life- created by a surface tension driven instability 关20兴.
time. Each row of photographs shows droplets photographed at Increasing the surface temperature to 130°C led to the forma-
the same time 共t兲, measured from the instant of droplet impact on tion of vapor bubbles in droplets deposited on the surface and the
the surface. The time t in s is indicated to the left of each row. onset of nucleate boiling. Figure 5 shows photographs of pure
A small air bubble was trapped at the center of a water droplet water droplets 共Fig. 5共a兲兲 and droplets containing dissolved CO2
when it landed on the surface 共see Fig. 2共a兲, t⫽0.75 s兲. The droplets 共Fig. 5共b兲兲 and Na2CO3 droplets 共Fig. 5共c兲兲 on a surface
bubble grew larger as the droplet evaporated. Measurements from at 130°C. By examining the video record of droplet evaporation it
photographs showed that as the volume of liquid diminished, the was evident that a number of small bubbles nucleated inside the
liquid-solid contact angle decreased while the contact area be- droplets immediately after impact, formed in cavities on the heater
tween the droplet and substrate remained constant. Eventually, surface. However, for both pure water droplets and those with
when the contact angle reached a lower limit, equal to the reced- dissolved CO2 these bubbles rapidly coalesced, so that by t
ing contact angle 共measured by Qiao and Chandra 关4兴 to be ap- ⫽0.35 s 共the first frame in Fig. 5兲 only one large bubble was
proximately 10 deg兲, the contact diameter began to decrease rap- evident. This bubble grew larger as the droplet evaporated, until it
idly until the droplet disappeared. We measured the evolution of finally broke through the droplet surface. No more bubbles were
contact diameter during droplet evaporation with the results produced in the droplet; heat loss to the liquid cooled the stainless
shown in Fig. 3. The contact diameter of pure water droplets was steel substrate too much to support further nucleation. Dissolving
constant for t⬍20 s, after which it decreased rapidly. We also CO2 in the droplet increased the number of bubbles nucleated
calculated the surface area of the droplet exposed to air 共from immediately after impact, and they coalesced to produce a large
which liquid evaporated兲 by measuring the droplet height and di- bubble. The bubble grew as CO2 came out of solution and in-
ameter and assuming it to be a segment of a sphere. The variation creased the droplet diameter. Droplet lifetime was therefore
of droplet surface area is shown in Fig. 4, and a steady decrease in slightly reduced.
area is evident as the droplet evaporates. The rate at which the Dissolved Na2CO3 had two significant effects on nucleate boil-
surface area diminished increased towards the end of the droplet ing of a water droplet 共Figure 5共c兲兲. First, the bubbles that formed
lifetime (t⬎20 s) when the contact diameter began to shrink. immediately after the droplet was deposited did not coalesce 共see
Dissolving CO2 in a water droplet placed on a surface at 100°C Figure 5共c兲, t⫽0.35 s兲. Second, at approximately t⫽3 s, a large
slightly reduced its evaporation time 共see Fig. 2共b兲兲. Inspection of number of new vapor bubbles nucleated. These also did not com-
videotapes of evaporating droplets showed that several vapor bine, but grew larger, greatly expanding the droplet diameter. The
bubbles nucleated within the drop immediately after it was depos- bubbles burst, reducing the droplet lifetime by almost half com-
ited on the surface. These rapidly merged with each other, so that pared to that for pure water.
by t⫽0.75 s 共the first frame in Fig. 2兲 there remained a single The ability of dissolved salts to prevent coalescence of vapor
large bubble. The solubility of CO2 decreases with rising tempera- bubbles has been well documented 共e.g., 关21–24兴兲. The merging
ture and gas escaping from solution made the bubble much larger of two bubbles is inhibited if the liquid film between them is
Fig. 8 Impact of droplets on a surface at 220°C: „a… pure water; „b… water containing dissolved CO2 ; and „c…
1 percent by weight solution of NaHCO3
The boiling of a droplet of 1 percent by weight solution of
NaHCO3 is shown in Fig. 6. A larger number of bubbles were
visible in the droplet soon after impact 共Fig. 6, t⫽0.35 s兲, prob-
ably because of CO2 released by decomposition of NaHCO3.
These did not coalesce, but grew larger. The abrupt nucleation of
more bubbles was again observed at approximately t⫽1.2 s, much
earlier than for droplets in which Na2CO3 was dissolved. The
bubbles disrupted the droplet so that it evaporated in approxi-
mately 3 s. As expected, NaHCO3 reduced the droplet lifetime
more than either CO2 or Na2CO3 alone.
Our measurements of droplet lifetime on surfaces at tempera-
tures (T w ) ranging from 100°C to 210°C are shown in Fig. 7.
Each data point represents the average of five measured values
with error bars marking the highest and lowest measured values.
To summarize our findings, at a low surface temperature (T w
⫽100°C), CO2 reduced the droplet lifetime slightly, while the
two salts increased it. At higher temperatures (T w ⬎120°C) both
salts produced a large reduction in droplet lifetime, NaHCO3 hav-
ing a larger effect than Na2CO3.
Droplets dropped on a surface at a temperature higher than
210°C went into a state of transition boiling. Water droplets
Fig. 9 Variation of droplet lifetime for droplets deposited on a evaporated so rapidly that the pressure of vapor generated at the
stainless steel surface at temperatures ranging from 210°C to liquid-solid interface was sufficient to lift them off the surface.
300°C Figure 8 shows three sequences of photographs of the impact and
recoil of droplets of water 共Fig. 8共a兲兲, water with dissolved CO2
共Fig. 8共b兲兲, and NaHCO3 solution 共Fig. 8共c兲兲, on a surface at
220°C. Since Na2CO3 had an effect similar to that of NaHCO3, a
stabilized. Surfactants achieve this by altering the surface tension separate set of photographs for Na2CO3 solution droplets is not
of the liquid. Qiao and Chandra 关4兴 demonstrated that adding a shown. Dissolved CO2 had no significant effect on droplet behav-
surfactant to water droplets boiling on a hot surface prevents ior. However NaHCO3 solution droplets had a visibly larger num-
bubble coalescence, making the liquid foam and greatly reducing ber of bubbles in them 共see Fig. 8, t⫽1.1 ms兲 whose coalescence
droplet lifetime. It is unlikely that inorganic salts change surface was inhibited by the salt. Bubbles bursting out of the liquid made
tension enough to produce foaming 关21兴. Using published data for the droplet disintegrate (t⫽5.3 ms).
the surface tension and viscosity of Na2CO3 and NaHCO3 solu- To record the evaporation time of droplets on surfaces with
tions we concluded that adding 1 percent by weight of solute T w ⬎210°C, they were dropped on a stainless steel plate ma-
would produce a negligible change in both surface tension and chined to make the upper surface slightly concave, which pre-
viscosity. The ability of salts to prevent coalescence is linked to vented droplets from rolling off. Measured droplet lifetimes for
their ionic strength: electric charge accumulated on the surfaces of 210°C⭐T w ⭐300°C are shown in Fig. 9. Data for the two salt
bubbles produces a repulsive force, preventing them from ap- solutions is not shown, since the droplets shattered during impact,
proaching each other 关24兴. There is also a more complex effect, in and it was not possible to define a meaningful droplet lifetime.
which dissolved electrolytes strengthen the electrostatic bonds be- There was little difference in measurement for pure water droplets
tween water molecules and make the liquid film separating and those with dissolved CO2. Droplet lifetimes increased with
bubbles stronger 关23兴. substrate temperature for 210°C⭐T w ⭐240°C. The surface tem-
The sudden nucleation of bubbles in the droplets of Na2CO3 perature corresponding to the maximum droplet lifetime is known
solution fragmented them and produced a large decrease in drop- as the ‘‘Leidenfrost temperature’’ 关3,4兴. Above this droplets are
let lifetime 共see Fig. 5共c兲 at t⫽3.55 s兲. We believe that this was fully levitated on a thin film of their own vapor, and are in a state
caused by salt particles precipitating from the evaporating droplet, of stable film boiling. Droplet lifetimes then decrease with further
which serve as nuclei to trigger bubble formation. The particles increases of temperature.
were too small to observe directly, but to test our hypothesis, we Figure 10 shows photographs of droplets landing on a surface at
sprinkled a few salt particles on the hot surface before a droplet 300°C. Droplets containing CO2 had far fewer bubbles than those
was deposited on it. These produced instantaneous nucleation of of pure water 共see Fig. 10, t⫽1.5 ms兲. Vapor bubbles are formed
bubbles in water droplets dropped on them. by heterogeneous nucleation when liquid touches the solid sur-
The boiling rate of water droplets in nucleate boiling was in- face. When a droplet approached the heated plate, dissolved CO2
creased by a small amount by dissolving CO2 in them, and by a quickly escapes from the bulk liquid since the solubility of CO2
much larger amount by dissolving Na2CO3. Though dissolved decreases sharply with increasing water temperature. Gas emerg-
CO2 increased the number of vapor bubbles nucleated, they ing from under the droplet would form a film that prevents liquid
quickly merged. Dissolved Na2CO3, which prevented bubble coa- from contacting the hot substrate, suppressing nucleation. Jeschar
lescence, had a much larger effect on droplet lifetime. We there- et al. 关13兴 have also noted that dissolved gas makes the vapor film
fore expected that a combination of the two additives, in which formed around a heated sphere immersed in water more stable. On
CO2 enhanced bubble production and Na2CO3 prevented their the other hand NaHCO3 promoted bubble nucleation, which was
merger, would reduce droplet lifetimes even further. To achieve vigorous enough to fragment the droplet 共Fig. 10共c兲, t⫽7.5 ms兲.
this combined effect we selected sodium bicarbonate (NaHCO3)
as an additive, because it decomposes into Na2CO3 and CO2 at
high temperatures: 2NaHCO3→Na2CO3⫹H2O⫹CO2. We esti- Conclusions
mated the decomposition rate of NaHCO3 using kinetic data from We observed the effects of dissolving either a gas (CO2) or a
the literature 关25兴, and calculated that approximately 10 percent of salt 共Na2CO3 or NaHCO3兲, in water droplets impinging on a
the NaHCO3 in a 1 percent by weight aqueous solution decom- heated surface. At surface temperatures too low to initiate nucle-
poses during the droplet lifetime. This would produce about 10 ate boiling it was found that droplet lifetime was determined by
percent of the mass of CO2 that was dissolved in pure water dur- heat and mass transfer around the periphery of the droplet. Dis-
ing our previous tests, i.e., 0.07 mm3 of CO2 /mm3 water at STP. solved NaHCO3 and Na2CO3 both reduced droplet evaporation
Fig. 10 Impact of droplets on a surface at 300°C: „a… pure water; „b… water containing dissolved CO2 ; and „c…
1 percent by weight solution of NaHCO3
rates since they lower the vapor pressure of water. Dissolved CO2 from below the drop and prevented it touching the substrate, sup-
enhanced the evaporation rate slightly, because the gas bubble pressing heterogeneous nucleation. Conversely, both salts pro-
trapped in the droplet made it swell and increased its circumfer- moted bubble nucleation and foaming so vigorously that droplets
ence. In the nucleate boiling regime, Na2CO3 prevented coales- disintegrated upon impact.
cence of bubbles in the liquid and promoted foaming, reducing
droplet lifetimes. Salt particles precipitated from the solution dur-
ing evaporation also triggered vapor nucleation. Dissolved References
NaHCO3 reduced droplet lifetimes even more than Na2CO3, be-
关1兴 di Marzo, M., Evans, D. D., 1989, ‘‘Evaporation of Water Droplet Deposited
cause it decomposed when heated to produce CO2, further aug- on a Hot High Thermal Conductivity Solid Surface,’’ ASME J. Heat Transfer,
menting bubble formation. When a droplet was deposited on a 111, pp. 210–213.
surface above the Leidenfrost temperature, dissolved CO2 escaped 关2兴 Chandra, S., di Marzo, M., Qiao, Y. M., and Tartarini, P., 1996, ‘‘Effect of
Liquid-Solid Contact Angle on Droplet Evaporation,’’ Fire Saf. J., 27, pp. Compounds,’’ Binary Systems, Part 1, 1 Pergamon Press, New York, pp. 115–
141–158. 116.
关3兴 Qiao, Y. M., and Chandra, S., 1996, ‘‘Boiling of Droplets on a Hot Surface in 关15兴 Van Slyke, D. D., and Neill, J. M., 1924, ‘‘The Determination of Gases in
Low Gravity,’’ Int. J. Heat Mass Transf., 39, No. 7, pp. 1379–1393. Blood and Other Solutions by Vacuum Extraction and Manometric
关4兴 Qiao, Y. M., and Chandra, S., 1997, ‘‘Experiments on Adding a Surfactant to Measurement-I,’’ J. Biol. Chem., 61, pp. 523–573.
Water Drops Boiling on a Hot Surface,’’ Proc. Phys. Soc., London, Sect. A, 关16兴 Fogg, P. G. T., and Gerrand, W., 1991, ‘‘Solubility of Gases in Liquids: A
453, pp. 673–689. Critical Evaluation of Gas/Liquid Systems in Theory and Practice,’’ Chiches-
关5兴 Qiao, Y. M., and Chandra, S., 1998, ‘‘Spray Cooling Enhancement by Addi- ter, Wiley, New York.
tion of a Surfactant,’’ ASME J. Heat Transfer, 120, pp. 92–98. 关17兴 Jho, C., Nealon, D., Shogbola, S., and King, Jr., A. D., 1977, ‘‘Effect of
关6兴 Finnerty, A. E., 1995, ‘‘Water-Based Fire-Extinguishing Agents,’’ Proceed- Pressure on the Surface Tension and Water: Adsorption of Hydrocarbon Gases
ings of Halon Options Technical Working Conference, pp. 461–471. and Carbon Dioxide on Water at Temperatures Between 0 and 50°C,’’ J.
关7兴 King, M. D., Yang, J. C., Chien, W. S., and Grosshandler, W. L., 1997, Colloid Interface Sci., 65, No. 1, pp. 141–154.
‘‘Evaporation of a Small Water Droplet Containing an Additive,’’ Proceedings 关18兴 International Critical Table, 1933, Vol. II and V, McGraw-Hill, New York.
of the 22nd National Heat Transfer Conference, 4, Baltimore, MD, pp. 45–57. 关19兴 Cisternas, L. A., and Lam, E. J., 1991, ‘‘An Analytic Correlation for the Vapor
关8兴 Jamialahmadi, M., and Muller-Steinhagen, H., 1990, ‘‘Pool Boiling Heat Pressure of Aqueous and Non-Aqueous Solutions of Single and Mixed Elec-
Transfer to Electrolyte Solutions,’’ Chem. Eng. Process., 28, pp. 79–88. trolytes, Part II. Application and Extension Fluid Phase Equilibria,’’ 62, pp.
关9兴 Najibi, S. H., Muller-Steinhagen, H., and Jamialahmadi, M., 1996, ‘‘Boiling
11–27.
and Non-Boiling Heat Transfer to Electrolyte Solutions,’’ Heat Transfer Eng.,
关20兴 Zhang, N., Wang, B. X., and Xu, Y., 1987, ‘‘Thermal Instability of Evaporat-
17, No. 40, pp. 46–63.
ing Droplets on a Flat Plate and Its Effects on Evaporation Rate,’’ Int. J. Heat
关10兴 Muller-Steinhagen, H., Epstein, N., and Watkinson, A. P., 1987, ‘‘Subcooled
Boiling of Heptane and of Water with Various Dissolved Gases,’’ Proceedings Mass Transf., 30, No. 3, pp. 469–478.
of 1987 ASME-JSME Thermal Engineering Joint Conference, pp. 125–131. 关21兴 Marrucci, G., and Nicodemo, L., 1967, ‘‘Coalescence of Gas Bubbles in Aque-
关11兴 Torikai, K., Shimamune, H., and Fujishiro, T., 1970, ‘‘The Effects of Dis- ous Solution of Inorganic Electrolytes,’’ Chem. Eng. Sci., 22, pp. 1257–1265.
solved Gas Content Upon Incipient Boiling Superheats,’’ Proceedings of 4th 关22兴 Lessard, R. R., and Zieminski, S. A., 1971, ‘‘Bubble Coalescence and Gas
International Heat Transfer Conference., V, pp. B2.11. Transfer in Aqueous Electrolytic Solutions,’’ Ind. Eng. Chem. Fundam., 10,
关12兴 You, S. M., Simon, T. W., Bar-Cohen, A., and Hong, Y. S., 1995, ‘‘Effects of No. 2, pp. 260–269.
Dissolved Gas Content on Pool Boiling of a Highly Wetting Fluid,’’ ASME J. 关23兴 Zieminski, S. A., and Whittemore, R. C., 1971, ‘‘Behavior of Gas Bubbles in
Heat Transfer, 117, pp. 687–692. Aqueous Electrolyte Solutions,’’ Chem. Eng. Sci., 26, pp. 509–520.
关13兴 Jeschar, R., Kraushaar, H., and Griebel, H., 1996, ‘‘Influence of Gases Dis- 关24兴 Keitel, G., and Onken, U., 1982, ‘‘Inhibition of Bubble Coalescence by Sol-
solved in Cooling Water on Heat Transfer During Stable Film Boiling,’’ Steel utes in Air/Water Dispersions,’’ Chem. Eng. Sci., 17, pp. 1635–1638.
Res., 67, No. 6, pp. 227–234. 关25兴 Hu, W., Smith, J. M., Dogu, T., Dogu, G., 1986, ‘‘Kinetics of Sodium Bicar-
关14兴 Stephen, H., and Stephen, T., 1963, ‘‘Solubilities of Inorganic and Organic bonate Decomposition,’’ AIChE J., 32, No. 9, pp. 1483–1490.
Interface Shape and
Thermally-Driven Convection in
Vertical Bridgman Growth of
Gallium Selenide: A
Semiconductor With Anisotropic
Solid-Phase Thermal Conductivity
Hanjie Lee
Postdoctoral Research Associate For vertical Bridgman growth of thermally anisotropic semiconductors, we present a
e-mail: hanjie@ajphp1.me.uiuc.edu detailed model accounting for heat transfer, flow driven by thermal buoyancy and solidi-
fication shrinkage, and interface deformation. The model allows for anisotropic solid-
Arne J. Pearlstein phase thermal conductivity, characteristic of nonlinear optical materials, as well as con-
Professor of Mechanical Engineering duction in the ampoule wall, and conduction and convection in the liquid. The interface
e-mail: ajp@uiuc.edu shape is determined as part of the solution of a moving boundary problem. For the
Mem. ASME nonlinear optical material gallium selenide and a range of growth conditions of practical
interest, we present steady axisymmetric computations of the isotherms, flow, and inter-
Department of Mechanical and Industrial face shape. For ampoule-wall temperature profiles typical of three-zone Bridgman fur-
Engineering, naces, the strength of the flow and deflection of the interface increase considerably with
University of Illinois at Urbana-Champaign, increasing growth rate, while the temperature distribution is relatively insensitive, except
1206 West Green Street, near the interface. Interface deflection decreases as the maximum ampoule-wall tempera-
Urbana, IL 61801 ture gradient increases. The flow depends significantly on whether the melting tempera-
ture is ‘‘centered’’ between the high and low temperatures. The 23°C uncertainty in the
melting temperature of GaSe is shown to have little effect on the flow and interface shape
over the entire range of growth conditions. We show that properly accounting for thermal
anisotropy is critical to predicting the flow and interface shape, both of which are rela-
tively insensitive to the temperature dependence of the viscosity and thermal conductivi-
ties. We also show that localized heating along the ampoule wall can both reverse the
direction of flow along the interface, which is expected to significantly influence distribu-
tion of dopants or impurities in the solid phase, as well as reduce interfacial curvature.
When GaSe is grown under zero gravity conditions, the only flow is due to solidification
shrinkage, and is essentially normal to the interface, whose shape is similar to those
computed at normal gravity. Comparison of results for GaSe to previous work for ben-
zene, a surrogate for organic nonlinear optical materials, shows that the qualitatively
different results are associated with differences in the anisotropy of the thermal
conductivity. 关DOI: 10.1115/1.1372194兴
Keywords: Crystal Growth, Microgravity, Solidification
1 Introduction tical, magnetic, and other materials has been discussed elsewhere
关5兴. Reference 关5兴 also reviews previous work on directional so-
Crystals of nonlinear optical materials have anisotropic optical
lidification in which thermal anisotropy has been considered, in-
and electro-optical properties. In many cases, the solid-phase ther-
cluding experiments by Gau and Viskanta 关6兴 in horizontal plane
mal conductivity and thermal expansivity are also anisotropic, and
layers of apparently quiescent fluid heated from below, and com-
can have important effects on crystal growth. First, anisotropy of
putations by Weaver and Viskanta 关7兴 in which heat transfer oc-
the solid-phase conductivity can significantly affect heat transfer
curs only by conduction. 共Conduction calculations by Huang, El-
from the liquid to solid, as well as removal of latent heat liberated
well, and Feigelson 关8兴 for CdGeAs2, a nonlinear optical material
at the liquid/solid interface, thereby affecting interface shape and
whose solid-phase thermal conductivity is thought to be aniso-
liquid flow. Second, anisotropic thermal expansion 共or contraction
tropic on the basis of its crystal structure, do not account for flow,
关1兴兲 can lead to development of either cracks that degrade crystal
latent heat, or crystal growth.兲
quality 关1–3兴, or gaps between the growing boule and the am-
We have recently reported 关5,9兴 the first numerical simulations
poule wall into which impurity- or dopant-enriched liquid can
of vertical Bridgman growth of benzene, a thermally anisotropic
infiltrate and solidify 关4兴.
material, with thermophysical properties similar to those of most
The importance of vertical Bridgman growth of anisotropic op-
organic nonlinear optical materials for which property data is
available. The results of those calculations are consistent with the
HEAT TRANSFER. Manuscript received by the Heat Transfer Division August 29, work of Huang et al. 关8兴 in that they show that solid-phase con-
2000; revision received February 2, 2001. Associate Editor: C. Beckermann. ductivity anisotropy plays an important role in determining inter-
Table 1 Dimensionless property value ratios Finally, the larger kinematic viscosity of GaSe relative to other
molten semiconductors gives rise to a much larger Schmidt num-
ber, Sc⫽ v /D, and considerably thinner concentration boundary
layers, critical to distribution of dopants or impurities in the liq-
uid. Thus, accurate computation of the flow and the temperature
distribution that drives it will be important in predicting distribu-
tions of dopants 共e.g., 关11,34–39兴兲 and impurities 关40–42兴 in
Bridgman-grown GaSe, as well as in doped crystals of other
III–VI semiconductors, including InSe 关43兴.
To date, the only computational work on melt growth of GaSe
is that described in papers by Singh and co-workers 关39,40兴. The
first 关39兴 presented axial temperature distributions ‘‘near the cen-
ter’’ of the ampoule at five instants based on a ‘‘pure conduction’’
model apparently accounting for conduction in the GaSe, am-
poule, and boron oxide encapsulant, but not for convection in the
liquid. It is unclear whether the growth rate was nonzero, since no
mention was made of it or the interfacial position or shape. Ex-
clusion of convection was justified by the statement that ‘‘the
Prandtl number of GaSe is very low,’’ although no Prandtl num-
ber was cited. 共As shown in Table 1, the Prandtl number for GaSe
face shape. Depending on growth conditions, qualitatively differ- at 1000°C is about 2.8, compared to 0.01 for Ge 关18兴 and 0.11 for
ent interface shapes are found, including shapes for which neither Hg0.8Cd0.2Te 关22兴.兲 The second paper 关40兴 presented the ‘‘pure
the liquid nor solid region is convex1. conduction’’ axial temperature profile at one additional time, and
Gallium monoselenide 共GaSe兲 is a nonlinear optical material for two times showed an apparently axisymmetric interface,
whose large nonlinear susceptibility and birefringence, transpar- whose shape cannot be discerned. The partial differential equa-
ency over the wavelength range 0.65–18 ␮m, and relatively high tions shown included a momentum equation with thermal and
thermal conductivity perpendicular to the preferred optical axis solutal Rayleigh numbers in the buoyancy terms, which terms did
make it attractive in a number of frequency conversion applica- not depend on temperature or composition. No physical property
tions 关11–12兴. Reviews of the properties, growth, and character- values were given in either paper.
ization of GaSe are available 关13–16兴. Vertical Bridgman growth Here, we focus on temperature distributions, interface shapes,
is the dominant 关14兴 approach used to produce single crystals of and flow in vertical Bridgman growth of GaSe in quartz ampoules
GaSe. of circular cross-section, used in almost all previous experimental
Gallium selenide differs significantly in several key respects work. The temperature distribution is critical to predicting stress
from benzene 关5兴, and from semiconductors 关17–27兴 for which distributions and dislocation formation 共cf. 关33,44兴兲. The interface
Bridgman growth has previously been simulated. shape is shown to be important in understanding and avoiding
First, all known GaSe polytopes consist of hexagonally sym- secondary nucleation and growth in GaSe 关14兴 and other crystals
metric layers, leading to isotropy about the c axis 关28–29兴. Thus, 关45兴, while the flow is important in dopant transport 关9,18,19兴
there are exactly two distinct thermal conductivities, both of
which have been measured 共see §2兲. For benzene, the crystal
structure is orthorhombic bipyramidal 关30兴, and one might expect
2 Model and Numerical Approach
the thermal conductivity to assume three different values along We consider vertical Bridgman growth in a quartz
the principal axes, even though only two have been measured 关5兴. 关11,31,38,46–51兴 ampoule of inner radius r i , moving downward
In all previous simulations of Bridgman solidification of semicon- at velocity ⫺Uez on a vertical axis past a fixed temperature dis-
ductors, the material has been 共or has been taken to be兲 isotropic. tribution T b (z), which we take to obtain on the outer wall of the
Table 1 shows that the solid-phase thermal conductivity of GaSe ampoule, as shown in Fig. 1. Except in §3.4, we consider the
is highly anisotropic, with the conductivity along the c axis 共par- hyperbolic tangent profile
allel to the sixfold symmetry axis兲 being about one-eighth the
1
conductivity along the orthogonal a and b axes 关16,31兴. In our T b共 z 兲 ⫽ 关 T ⫹T h ⫹ 共 T h ⫺T c 兲 tanh ␭z 兴 , (1)
work for benzene 关5兴, the ratio of the axial conductivity to the 2 c
lateral conductivity in the solid was 2.2, whereas for the c axis of where the temperatures in the liquid and solid far from the inter-
GaSe aligned with the growth direction, the corresponding ratio is face, T h and T c , respectively, bracket the melting temperature
about 0.12. Thus, we see that radial conduction in the solid will be T M . The ampoule-wall temperature profile 共1兲 is an excellent ap-
a much more efficient means to remove heat from the interface proximation to those in Bridgman growth experiments 共cf. refer-
共relative to axial conduction兲 in GaSe than in benzene. Anisotropy ences cited in 关5,9兴 as well as in other experimental and compu-
and liquid/solid conductivity ratios are expected to strongly influ- tational 关52,53兴 work兲, in which an ‘‘adiabatic zone’’ lies between
ence interface shape 共cf. 关8兴兲, which in turn is known to affect hot and cold constant-temperature zones. Here, ␭
twinning and other determinants of crystal quality in Bridgman- ⫽2(dT b (0)/dz)/(T h ⫺T c ), where dT b (0)/dz is the maximum
grown materials 关32–33兴. axial derivative of the ampoule-wall temperature profile 共1兲. As in
Second, beyond its anisotropic solid-phase thermal conductiv- previous simulations of Bridgman growth of electronic materials
ity, GaSe has liquid- and solid-phase thermophysical properties 共cf. 关23,27,54兴兲, we neglect radiative transfer within the liquid and
quite different from those of other semiconductors and benzene, solid. Assessment of the effect of this approximation on growth of
as shown in dimensionless terms in Table 1. The Prandtl number GaSe, which is essentially transparent over most of the infrared, is
for GaSe 共⬃2.8 at the melting temperature兲 is smaller than that for beyond the scope of the present work.
benzene by a factor of three, but is more than two orders of We consider GaSe growing as a single crystal, with its c axis
magnitude higher than for Ge. parallel to the ampoule axis, as in a number of experimental in-
vestigations 关11,39–41,46兴. 共Crystals of GaSe can also be grown
perpendicular to the c axis 关41,46,55兴.兲 In that case, the conduc-
1
In what follows, we use ‘‘convex-solid’’ and ‘‘concave-solid’’ as adjectives to
unambiguously describe interfaces bounding convex and concave solid regions, re-
spectively. Use of ‘‘convex’’ and ‘‘concave’’ to modify regions rather than interfaces
tivity can be written as ks ⫽k t er er ⫹k t e␾ e␾ ⫹k c ez ez , where k c and
is consistent with standard topological nomenclature 关10兴, is shorter than ‘‘convex to k t are the conductivities parallel and transverse to the c axis, re-
the solid,’’ etc., and avoids confusion. spectively. We consider the situation in which the temperature
sor accounting for the temperature dependence of the solid-phase
thermal conductivity. The subscript M denotes a property evalu-
ated at the melting temperature.
As in Ref. 关5兴, the temperature dependence of the viscosity is
accounted for using an Arrhenius fit to the data. Results presented
in §3.3 show that these variable-property effects are small. Aside
from the viscosity, liquid- and solid-phase conductivities, and liq-
uid density 共vide supra兲, all other thermophysical properties are
taken as constant. We account for shrinkage flow due to the dif-
ference between the liquid and solid densities. The heat flux sin-
gularity associated with the discontinuous thermal conductivity
across the liquid/solid interface, and the moving contact singular-
ity associated with shrinkage, both at the liquid/solid/ampoule
wall three-phase junction, are discussed in 关9兴.
The Grashof number and liquid-phase Péclet numbers are de-
fined by Gr⫽ ␣ gr 3i ⌬T/ ␯ 2M and Pe lM ⫽Ur i / ␬ lM , respectively.
Here, ␣ , g, and U are the thermal expansivity of the liquid, mag-
nitude of the gravitational acceleration, and growth rate, respec-
tively, while ␬ lM ⫽k lM /( ␳ lM c p,l ), ␬ tM ⫽k tM /( ␳ s c p,s ), and ␬ a
⫽k a /( ␳ a c p,a ) are the thermal diffusivities in the liquid, solid
共transverse to the c axis兲, and ampoule wall.
We assume that on the liquid/solid interface, the deflection
␩ I ( ␰ ) is a single-valued function of the radial coordinate, and that
the dimensionless interfacial boundary conditions are
␪ l 共 ␰ , ␩ I 共 ␰ 兲兲 ⫽ ␪ s 共 ␰ , ␩ I 共 ␰ 兲兲 ⫽ ␪ M (5a)
k cM ⳵ ␪ s 共 ␰ , ␩ I 共 ␰ 兲兲 ⳵ ␪ s 共 ␰ , ␩ I 共 ␰ 兲兲
⫺ ␩ I⬘ 共 ␰ 兲
k tM ⳵␩ ⳵␰
Fig. 1 Schematic of vertical Bridgman growth with imposed

⫺
k tM 冋
k lM ⳵ ␪ l 共 ␰ , ␩ I 共 ␰ 兲兲
⳵␩
⫺ ␩ I⬘ 共 ␰ 兲
⳵␰ 册
⳵ ␪ l 共 ␰ , ␩ I 共 ␰ 兲兲 ␬ lM
⫽ Pe St
␬ tM lM
ampoule-wall temperature distribution T b „ z … (5b)
Pe lM 1 ␳s
en "u共 ␰ , ␩ I 共 ␰ 兲兲 ⫽⫺ , (5c)
Pr 冑1⫹ ␩ I⬘ 共 ␰ 兲 2 ␳ lM
distribution, interface shape, and flow are axisymmetric 关27,54兴,
and steady in an interface-fixed reference frame in which the solid Pe lM ␩ I⬘ 共 ␰ 兲
and ampoule translate downward at ⫺Uez . The ampoule is taken es "u共 ␰ , ␩ I 共 ␰ 兲兲 ⫽⫺ , (5d)
to be unbounded in the vertical direction, corresponding to a long Pr 冑1⫹ ␩ I⬘ 共 ␰ 兲 2
boule grown under steady conditions. The relevance of this
growth regime to the production of axially uniform boules has corresponding to an isothermal interface at T M , conservation of
been discussed elsewhere 关9,52兴. energy, conservation of mass, and no-slip, respectively. Here, en
With these assumptions, the dimensionless governing equations and es are unit vectors normal and tangential to the interface,
for conservation of energy, mass, and momentum are respectively, St⫽⌬H f /(c p,s ⌬T) is the Stefan number, and ⌬H f
is the heat of fusion.
␬ lM ⳵ ␪ s On the inner wall of the ampoule, we have
⫺ Pe lM ⫽ⵜ"共 bs ⵜ ␪ s 兲 , (2a)
␬ tM ⳵ ␩ ␪ l 共 1,␩ 兲 ⫽ ␪ a 共 1,␩ 兲 (6a)
Pr M u"ⵜ ␪ l ⫽ⵜ"关 f k,l 共 ␪ l 兲 ⵜ ␪ l 兴 , (2b) ⳵ ␪ l 共 1,␩ 兲 k a ⳵ ␪ a 共 1,␩ 兲
⫽ (6b)
⳵␰ k lM ⳵␰
␬ lM ⳵ ␪ a
⫺ Pe lM ⫽ⵜ 2 ␪ a , (2c)
␬a ⳵␩ u⫽⫺ 共 Pe lM / Pr M 兲 e␩ (6c)
ⵜ"u⫽0, (3) above the liquid/solid/ampoule junction ( ␩ ⬎ ␩ I (1)), and

␪ s 共 1,␩ 兲 ⫽ ␪ a 共 1,␩ 兲 (6d)
u"ⵜu⫽⫺ⵜ⌸⫹ⵜ"兵 f v 共 ␪ l 兲关 ⵜu⫹ 共 ⵜu兲 T 兴其 ⫹Gre␩ , (4)
where the temperature ( ␪ l , ␪ s , and ␪ a , in the liquid, solid, and ⳵ ␪ s 共 1,␩ 兲 k a ⳵ ␪ a 共 1,␩ 兲
⫽ (6e)
ampoule wall, respectively兲, relative velocity u, and cylindrical ⳵␰ k tM ⳵␰
coordinates 共axial ␩ and radial ␰兲 have been scaled with ⌬T
below that junction ( ␩ ⬍ ␩ I (1)).
⫽T h ⫺T c , ␯ M /r i , and r i , respectively 关5兴. We take the specific
On the axis, symmetry and boundedness of the velocity and
heats (c p,l , c p,s , and c p,a ), ampoule conductivity (k a ), and den-
temperature require
sities of the solid ( ␳ s ) and ampoule ( ␳ a ) to be constant. The liquid
density is to taken to be constant in the inertial term and continu- e␰ "u⫽0 (7a)
ity equation, and to vary linearly with temperature 共as amply jus-
tified by the data for GaSe 关56兴兲 in the body force. The functions e␩ "u finite (7b)
f v ⫽ ␯ (T)/ ␯ M and f k,l ⫽k l (T)/k lM account for the temperature de-
pendence of the kinematic viscosity and liquid-phase thermal con- ⳵ ␪ l 共 0,␩ 兲
⫽0 (7c)
ductivity, respectively, while bs ⬅ks /k tM is a dimensionless ten- ⳵␰
Table 2 Parameters in terms of which results are presented braic equation system, is as described for benzene 关5兴, with the
exception that a longer computational domain is required, as de-
scribed below. Compared to benzene, the relatively more vigorous
flows in GaSe growth give rise to primary toroidal vortex cells
that penetrate considerably farther upward into the liquid, as dis-
cussed in §§3 and 4. This requires that the far-field computational
boundary for the liquid (L 1 in dimensionless terms 关5兴兲 be farther
from the interface. This question, and the effects of domain size
on convergence, are discussed in §3.
For r⫺z coordinates, two distinct streamfunction definitions
satisfy the continuity equation. We use here the one for which the
standard two-dimensional interpretation of ‘‘no flow across a
⳵ ␪ s 共 0,␩ 兲 streamline’’ pertains. As in our earlier work for benzene 关5,9兴, all
⫽0. (7d)
⳵␰ streamlines shown are referred to a frame translating with the
In the liquid, solid, and ampoule, the velocity and temperature ampoule and solid.
should approach finite limits far from the interface. These values For ease of reference, we define in Table 2 the parameters in
can be obtained by seeking ␩-independent solutions of the gov- terms of which results are presented. These quantities and the
erning equations 共2兲–共4兲 subject to the radial boundary conditions symbols denoting them are identical to those used for benzene 关5兴.
共6兲, 共7兲 The property values used are listed in Table 3, along with the
sources of data 共Refs. 关14,31,56–67兴兲.
lim u⫽⫺ 共 Pe lM / Pr M 兲关 1⫹2 共 ␳ s / ␳ lM ⫺1 兲共 1⫺ ␰ 2 兲兴 e␩ (8a)
␩ →⬁ 3 Results
lim ␪ l ⫽1 (8b) We have computed temperature distributions, interface shapes,
␩ →⬁ and the corresponding thermally-driven flows for several sets of
growth conditions.
lim ␪ s ⫽0 (8c) In most previous simulations of vertical Bridgman growth 共cf.
␩ →⫺⬁
关5,18,19兴兲, the high and low temperatures were chosen so that T M
lim ␪ a ⫽1 (8d) lies midway between. Except where otherwise stated 共in §3.2兲, we
␩ →⬁ also choose the temperatures of the hot and cold zones to ‘‘cen-
ter’’ T M . We use ⌬T⫽T h ⫺T c ⫽150°C, corresponding to the dif-
lim ␪ a ⫽0. (8e) ference between hot- and cold-zone temperatures reported by
␩ →⫺⬁
Singh and co-workers for vertical Bridgman growth of GaSe
For ␳ s ⫽ ␳ lM , the far-field flow 共8a兲 corresponds to superposition 关39,40兴. For this ⌬T, temperatures that ‘‘center’’ T M ⫽937°C are
of the growth velocity and a Poiseuille flow driven by solidifica- T h ⫽1012°C, and T c ⫽862°C. We refer to these temperatures and
tion shrinkage. a maximum ampoule-wall temperature gradient of dT b (0)/dz
The model determines the interface shape as part of the solution ⫽30°C cm⫺1 关39,40,69,70兴 as the ‘‘standard’’ thermal conditions.
关9,54兴, and accounts for latent heat release at the interface, con- The fused quartz ampoule was taken to have an inner diameter
vection in the liquid, conduction in the liquid, solid, and ampoule 2r i ⫽1 cm 关11,36,41,48,55,70–75兴 and outer wall thickness of 1
wall, and anisotropy of the solid-phase thermal conductivity, a mm.
ubiquitous property of nonlinear optical materials 关5,9兴. Resolution convergence was established by increasing the num-
The computational approach, involving domain decomposition, ber of radial and axial expansion functions. For L 1 ⫽30 and L 2
Chebyshev collocation, and iterative solution of a nonlinear alge- ⫽15 共where L 1 and ⫺L 2 are the locations of the upper and lower
computational boundaries, respectively 关5,9兴兲, Table 4 shows that
for the ‘‘standard case’’ and U⫽1.0 ␮m sec⫺1, the streamfunction
Table 3 Thermophysical properties of GaSe and temperature ␪ l ⫽ 关 T l ⫺T c 兴 /⌬T at six points 共axial positions in
the primary and secondary vortices; radial positions near the cen-
terline and ampoule wall, and midway between兲, and the overall
interface deflection ⌬z⫽ 关 z(r i )⫺z(0) 兴 /r i , are well converged for
M ⫽20 radial and N⫽60 axial expansion functions.
For M ⫽20 and N⫽60, we assess domain convergence using
domains of varying dimensionless axial extent. We consider the
Table 4 Resolution convergence for dT b „0…Õ dz Ä30°C cmÀ1,

U Ä1.0 ␮m secÀ1, L 1 Ä30, L 2 Ä15
Table 5 Domain convergence for dT b „0…Õ dz Ä30°C cmÀ1,
U Ä1.0 ␮m secÀ1, M Ä20, N Ä60, L 2 Ä L 1 Õ2
Fig. 3 Interface shapes for dT b „0…Õ dz Ä30°C cmÀ1. U Ä0.25

␮m secÀ1, ⌬ z Õ r i Ä0.120; U Ä0.50 ␮m secÀ1, ⌬ z Õ r i Ä0.152; U
standard case with U⫽1.0 ␮m sec⫺1 and L 2 ⫽L 1 /2 for 3⭐L 1 Ä1.0 ␮m secÀ1, ⌬ z Õ r i Ä0.215; U Ä3.0 ␮m secÀ1, ⌬ z Õ r i Ä0.437.
⭐30 共corresponding to 4.5⭐L 1 ⫹L 2 ⭐45). This range contrasts to
L 1 ⫹L 2 ⫽4 关18兴 and L 1 ⬇6 关19兴 used for Ge, and L 1 ⫹L 2 ⭓8 关5兴
used for benzene. For these growth conditions, Table 5 shows that quires a larger domain than for z/r i ⫽0.5. At both axial locations,
the interface deflection and liquid temperature are all well con- however, the streamfunction is well converged for the two largest
verged even for the smallest computational domain. Not surpris- domains (L i ⫽15 and 30兲.
ingly, convergence of the streamfunction values at z/r i ⫽3 re- Together, these results demonstrate convergence of the com-
puted solutions.
3.1 Effect of Growth Rate and Maximum Temperature
Gradient. Vertical Bridgman growth of GaSe has been reported
at rates between 0.1 and 12 mm hr⫺1 共0.028 ␮m sec⫺1⭐U⭐3.33
␮m sec⫺1兲, for 10°C cm⫺1⭐dT b (0)/dz⭐80°C cm⫺1 关37–
39,41,46–49,55,69,70,72–74,76,77兴. After first considering
growth rates covering most of the U range for the standard case,
we then consider additional values of dT b (0)/dz covering most of
the range studied in previous experiments.
Figures 2共a–c兲 show isotherms and streamlines for U⫽0.25,
1.0, and 3.0 ␮m sec⫺1. For these growth rates and U⫽0.5
␮m sec⫺1, Fig. 3 shows the corresponding interface shapes. Fig-
ure 2共a兲 shows that at the lowest U, the thermally-driven buoyant
flow consists of two counter-rotating toroidal cells. The upper
‘‘primary’’ vortex extends many diameters into the liquid, while
the smaller ‘‘secondary’’ vortex is confined to a region adjacent to
the interface. 共This nomenclature was adopted in our work on
benzene 关5兴, in which sometimes only a single large vortex ex-
tending far into the liquid was found.兲 The thermal Péclet number
in the liquid Pe lM ⫽Ur i ␳ lM c p,l /k lM is so small (4.5⫻10⫺3 ,
based on the growth rate U⫽0.25 ␮m sec⫺1兲 that the liquid-phase
isotherms 共Fig. 2共a兲兲 are almost horizontal except near the inter-
face and wall. 共In this and all other contour plots, the dashed and
solid streamlines shown are at equal increments between ␺ min and
0, and between 0 and ␺ max , respectively. The isotherms shown are
at increments of ⌬T/30.) Slight isotherm curvature near the wall
reflects thermal boundary layers associated with locally upward
and downward flows of the primary and secondary vortices, re-
spectively. The primary vortex 共dashed streamlines, correspond-
ing to negative streamfunction values兲 transports hot fluid down
toward the interface along the centerline, while heat released at
the interface is conducted down into the solid. Isotherms in the
solid are essentially horizontal, i.e., perpendicular to the growth
direction. The aspect ratio (L 1 ⫹L 2 ) of the domain required to
adequately capture the flow is larger than in our benzene compu-
tations, since 关 dT b (0)/dz 兴 / 关 (T h ⫺T c )/r i 兴 is relatively smaller
共0.1 for dT b (0)/dz⫽30°C cm⫺1, ⌬T⫽150°C, and 2r i ⫽1 cm,
with a range of 0.05–0.4 for the GaSe cases considered, compared
Fig. 2 Isotherms and streamlines for dT b „0…Õ dz Ä30°C cmÀ1. to a range of 0.375–3 for benzene 关5兴兲. As a consequence, the
„a… U Ä0.25 ␮m secÀ1: ␺ minÄÀ0.0261, ␺ maxÄ0.0432; „b… U Ä1.0 primary vortex extends much farther up into the liquid. Figure 3
␮m secÀ1: ␺ minÄÀ0.0253, ␺ maxÄ0.0708; and „c… U Ä3.0 shows that at U⫽0.25 ␮m sec⫺1, the interface shape is concave-
␮m secÀ1: ␺ minÄÀ0.0233, ␺ maxÄ0.116. solid, with maximum deflection ⌬z/r i ⫽0.120 on the centerline.
Fig. 5 Interface shapes for dT b „0…Õ dz Ä60°C cmÀ1 . U Ä0.25
Ä1.0 ␮m secÀ1, ⌬ z Õ r i Ä0.150; U Ä3.0 ␮m secÀ1, ⌬ z Õ r i Ä0.275.
30°C cm⫺1 case兲, while there is little change in the corresponding

value for the primary vortex. However, the primary vortex is sig-
nificantly shortened.
For dT b (0)/dz⫽60°C cm⫺1, Fig. 5 shows that the interfacial
curvature increases significantly with growth rate, as for
30°C cm⫺1, but with considerably smaller deflections at
60°C cm⫺1. The ratio of the interfacial deflection at 60°C cm⫺1 to
that at 30°C cm⫺1 decreases from 0.111/0.120⫽0.92 at
Fig. 4 Isotherms and streamlines for dT b „0…Õ dz Ä60°C cmÀ1 , 0.25␮m sec⫺1 to 0.275/0.437⫽0.63 at 3.0 ␮m sec⫺1.
with L 1 Ä30 and L 2 Ä8. „a… U Ä0.25 ␮m secÀ1: ␺ minÄÀ0.0608,
␺ maxÄ0.0490; „b… U Ä1.0 ␮m secÀ1: ␺ minÄÀ0.0598, ␺ max
For dT b (0)/dz⫽15°C cm⫺1 关14兴, results 共not shown兲 demon-
Ä0.0704; and „c… U Ä3.0 ␮m secÀ1: ␺ minÄÀ0.0572, ␺ max strate that the strengths of the flow in the primary vortex and
Ä0.117. especially the secondary vortex are reduced, while the vertical
extent of the former and the deflection of the concave-solid inter-
face are both considerably greater than for the standard
30°C cm⫺1 case.
For U⫽0.5 ␮m sec⫺1, the thermal Péclet number based on U is 3.2 Effect of Hot- and Cold-Zone Temperatures and Melt-
still so small (9⫻10⫺3 ) that the temperature distribution is domi- ing Temperature Uncertainty. As stated above, most previous
nated by conduction. Except near the interface, the isotherms 共not simulations of vertical Bridgman growth have considered the case
shown兲 are indistinguishable from those for U⫽0.25 ␮m sec⫺1 where T M lies midway between T h and T c . There is, however,
共Fig. 2共a兲兲. Hence, the thermally-driven primary vortex is nearly evidence suggesting that ‘‘off-centered’’ temperature distributions
identical to that computed for 0.25 ␮m sec⫺1. On the other hand, can give rise to interface shapes with more favorable curvature
the secondary vortex is driven by the temperature variation near 关2,78兴. Moreover, for GaSe the melting temperature is uncertain
the interface, which depends on interface curvature. Figure 3 by about 23°C 共cf. 关79–84兴 and compare references cited in Table
shows that the interface is somewhat more deformed than for 0.25 3兲. Thus, the dependence of the solidification process on whether
␮m sec⫺1, but maintains its concave-solid shape; the maximum the melting temperature lies midway between T h and T c is of
deflection is ⌬z/r i ⫽0.152. For U⫽1.0 ␮m sec⫺1, Fig. 2共b兲 shows interest.
that the maximum streamfunction value, which occurs in the sec- We first consider hot- and cold-zone temperatures of 1050 and
ondary vortex, is significantly higher than for 0.25 ␮m sec⫺1. 共The 900°C, respectively, as used by Singh et al. 关39,40兴, with the same
maximum ␺ values in the primary vortex are nearly unchanged overall temperature difference as for the standard case, to which
between U⫽0.25 and 1.0 ␮m sec⫺1.兲 Figure 3 shows that the we compare the results. We again take T M ⫽937°C.
interface deflection continues to increase, reaching ⌬z/r i Figure 6 shows that, at U⫽0.25 ␮m sec⫺1, the secondary vor-
⫽0.215. For U⫽3.0 ␮m sec⫺1, Fig. 2共c兲 shows that the flow in tex adjacent to the interface is highly distended, compared to the
the secondary vortex has intensified still more. Figure 3 shows standard 共‘‘centered’’兲 case. This is due to the fact that the maxi-
that the interface deflection is significantly higher, with ⌬z/r i mum ampoule-wall temperature gradient lies more than two radii
⫽0.437. The volume of the nearly isothermal region immediately above the interface, rather than near the interface 共as in the stan-
above the interface is considerably larger than at lower growth dard case兲. As a consequence, the primary vortex is shifted up-
rates. ward 共with little change in form兲, and the secondary vortex driven
When the maximum ampoule-wall temperature gradient is by it is free to occupy a larger volume. The separation streamline
doubled to 60°C cm⫺1, thermal variation in the liquid and solid is 共shown as a dotted curve兲 between the primary and secondary
considerably more localized. Figures 4共a–c兲 shows that the sec- vortices now lies about 1.5 diameters above the interface, com-
ondary vortex is confined to a volume near the interface somewhat pared to about 0.6 diameters in the centered case. The distended
smaller than for 30°C cm⫺1. This diminution is accompanied by secondary vortex persists to U⫽3.0 ␮m sec⫺1. As in the centered
an increase in the maximum magnitude of the streamfunction in case, the temperature distribution and the form of the flow it
the secondary vortex, which more than doubles 共compared to the drives 共although not the strength兲 are essentially independent of U
Fig. 7 Interface shapes for dT b „0…Õ dz Ä30°C cmÀ1, with T c
Ä900 °C, T h Ä1050°C, and T M Ä937°C. U Ä0.25 ␮m secÀ1,
⌬ z Õ r i Ä0.146; U Ä0.50 ␮m secÀ1, ⌬ z Õ r i Ä0.190; U Ä1.0
␮m secÀ1, ⌬ z Õ r i Ä0.274; U Ä3.0 ␮m secÀ1, ⌬ z Õ r i Ä0.563.
Fig. 6 Isotherms and streamlines for dT b „0…Õ dz Ä30°C cmÀ1,

U Ä0.25 ␮m secÀ1, with T c Ä900°C, T h Ä1050°C, and T M
Ä937°C: ␺ minÄÀ0.0292, ␺ maxÄ0.0538.
over 0.25–3.0 ␮m sec⫺1. Figure 7 shows that at each growth rate,

the interface deflection is 20–30 percent higher than for the stan-
dard case.
On the other hand, when we use the ‘‘alternate’’ melting tem-
perature of 960°C 关79–84兴 with the standard T h and T c , the flow
and interface shape 共not shown兲 are quite similar to the corre-
sponding ‘‘standard-case’’ results at U⫽0.25, 0.5, 1.0, and 3.0
␮m sec⫺1. Although this ‘‘alternate- T M ’’ case is also not cen-
tered, it differs from the previous uncentered case in that the
maximum vertical temperature gradient occurs more than one di-
ameter below the interface 共i.e., in the solid兲, as opposed to in the
liquid. Consequently, the primary vortex is somewhat closer to the
interface than in the standard case, rather than farther away as in
the other uncentered computations. The result is a slight axial
compression of the secondary vortex. At each growth rate, the
interface deflection differs by only a few percent from the corre-
sponding value computed for T M ⫽937°C.
3.3 Effect of Neglecting Anisotropy or Temperature-

Dependence of Properties. The importance of solid-phase ther-
mal conductivity anisotropy can be clearly identified by consider-
ing the fictitious case in which the conductivity is taken to be
isotropic. We take the conductivity to be one-third of the trace of Fig. 8 Isotherms and streamlines for a fictitious material with
the conductivity tensor (k iso⫽(2k t ⫹k c )/3), an approximation in isotropic solid-phase conductivity k isoÄ tr „ks …Õ3 and
excellent agreement with experiment and computations for many dT b „0…Õ dz Ä30°C cmÀ1. „a… U Ä0.25 ␮m secÀ1: ␺ minÄÀ0.0712,
microscopically anisotropic polycrystalline materials 关85兴. ␺ maxÄ0.893Ã10À4 ; „b… U Ä1.0 ␮m secÀ1: ␺ minÄÀ0.0507, ␺ max
The streamlines and isotherms shown in Figs. 8共a–c兲 and inter- Ä0.0; and „c… U Ä3.0 ␮m secÀ1: ␺ minÄÀ0.0572, ␺ maxÄ0.117.
perature exceeds T M above the liquid/solid/ampoule junction, it
warms and rises. For the real material 共and for the isotropic case
at 3.0 ␮m sec⫺1兲 the interface curvature is opposite, and so is the
flow on the interface. One consequence of this flow reversal is that
dopants or impurities rejected at the interface 共i.e., with segrega-
tion coefficients less than unity兲 will be transported radially out-
ward along the interface, and will tend to accumulate near the
ampoule wall, as for pyridine-doped benzene 关9兴 under growth
conditions that gave a convex-solid interface. 共Conversely, for
solutes with segregation coefficients greater than unity, which are
preferentially incorporated into the solid, the solid-phase mass
fraction will be higher near the centerline.兲
It is also useful to understand how the results depend on
whether one accounts for the temperature dependence of the vis-
cosity and liquid and solid-phase thermal conductivities, since for
many materials, detailed data on the temperature dependence of
these properties is not readily available. To make such an assess-
ment, we take all of the properties to be independent of T, except
the liquid density, which we allow to depend linearly on tempera-
Fig. 9 Interface shapes for dT b „0…Õ dz Ä30°C cmÀ1 for a ficti- ture, as before. The computed flows and interface shapes 共not
tious material with isotropic solid-phase thermal conductivity shown兲 are essentially unchanged, with the centerline interface
k isoÄ tr „ks …Õ3 and dT w „0…Õ dz Ä30°C cmÀ1 . U Ä0.25 ␮m secÀ1, deflections at each growth rate differing from the corresponding
⌬ z Õ r i ÄÀ0.177; U Ä0.50 ␮m secÀ1, ⌬ z Õ r i ÄÀ0.159; U Ä1.0 standard 共variable-property兲 case by less than 1 percent.
␮m secÀ1, ⌬ z Õ r i ÄÀ0.125; U Ä3.0 ␮m secÀ1, ⌬ z Õ r i Ä4.9Ã10À3 .
3.4 Effect of Localized Ampoule-Wall Heating. Here, we
explore the possibility of using the ampoule-wall temperature pro-
file to manipulate the flow and interface shape. The results of §3.1
face shapes shown in Fig. 9 reveal several key differences com- show that increasing the maximum gradient in the hyperbolic tan-
pared to the anisotropic case 共Figs. 2共a–c兲 and 3兲. First, for U gent ampoule-wall temperature profile 共1兲 can significantly reduce
⫽0.25 and 1.0 ␮m sec⫺1, the interface is now decidedly convex- interface curvature, especially at higher growth rates.
solid, rather than concave-solid as for the real material. For 3.0 Specifically, we consider ampoule-wall temperature gradients
␮m sec⫺1, the interface is nearly horizontal, but slightly concave- of the form
solid, with ⌬z/r i ⫽4.91⫻10⫺3 共about 4 percent of the deflection
for the real material兲. Second, for U⫽0.25, and 1.0 ␮m sec⫺1, the 2 /r 2
vortical flow has not split into counter-rotating vortices. Rather, T b 共 z 兲 ⫽ 21 关 T c ⫹T h ⫹ 共 T h ⫺T c 兲 tanh ␭z 兴共 1⫹ ␹ e ⫺ ␴ 共 z⫺z 0 兲 i 兲,
the flow in the small cell near the interface and the much larger (9)
co-rotating cell above are divided by a separatrix, rather than by a
separation streamline, as is the case for the anisotropic material. corresponding to localized heating near z⫽z 0 . The quantities ␹ ,
Third, at all growth rates considered, the isotherms in the solid z 0 /r i , and ␴ measure the amplitude, location, and degree of lo-
display noticeable curvature well below the interface. Away from calization of the additional heating. The distribution 共9兲 corre-
the immediate vicinity of the interface, the isotherms are essen- sponds to a ‘‘multiplicative’’ Gaussian ‘‘bump’’ superimposed on
tially independent of growth rate, as was the case for the real the underlying hyperbolic tangent temperature profile. It should be
material 共Fig. 2兲. This, and the thin thermal boundary layers in the possible to obtain distributions qualitatively similar to 共9兲 using a
liquid adjacent to the wall, again indicate that heat transfer in this ring heater centered about an axial position z 0 in the ‘‘adiabatic’’
system is conduction-dominated, and is unaffected by either flow zone of a three-zone Bridgman furnace.
or ampoule translation. For ␹ ⫽0.1, z 0 /r i ⫽⫺0.05, and ␴ ⫽2, Figs. 10共a–c兲 show how
These profound differences are due to the differences in the flow depends on growth rate. At U⫽0.25 ␮m sec⫺1, the flow
k zz,s /k rr,s 共unity for the isotropic case, as opposed to 0.12 for the differs from the standard case 共Fig. 2共a兲兲 in that a third toroidal
real material兲, and of k zz,s /k l 共2.83 for the isotropic case, com- vortex has formed between the interface and secondary vortex.
pared to 0.48 for the real material兲. The higher ratio of vertical to 共Note that the senses of rotation of the primary and secondary
radial conductivity in the solid allows for much more efficient vortices are unaffected by the local heating.兲 The effect of the
conduction down into the ‘‘isotropic’’ solid than in the anisotropic third vortex, whose sense of rotation is opposite to that of the
case, and reduces the outward radial flux. As discussed by Feigel- secondary vortex normally adjacent to the interface, will be to
son and Route 关2兴, this is largely responsible for the qualitative reverse the direction of flow along the interface. This is expected
differences in interface shape. Specifically, in the isotropic case, to alter dopant transport in the liquid, and consequently segrega-
axial conduction through the solid is more efficient than in the tion in the solid. As U increases to 0.50 ␮m sec⫺1 and subse-
anisotropic case, so that liquid penetrates less deeply on the cen- quently to 1.0 ␮m sec⫺1, a fourth vortex appears between the third
terline. As U increases, however, latent heat must be removed one and the interface. Its sense of rotation is the same as that of
more rapidly, and the depth of liquid penetration increases until the secondary vortex, but with a much lower strength. Its appear-
by 3.0 ␮m sec⫺1, the interface is slightly concave-solid. ance leads to reversion to radially inward interfacial flow.
The lower radial conductivity in the isotropic case 共two-thirds For U⫽0.25, 0.5, and 1.0 ␮m sec⫺1, Fig. 11 shows that local-
of the radial value in the anisotropic case兲 allows for development ized heating significantly reduces the curvature of concave-solid
of somewhat larger radial temperature gradients in the solid, giv- interfaces, especially at high growth rates. We have not yet, how-
ing rise to greater isotherm curvature away from the interface. ever, found conditions for which a convex-solid interface shape is
The nature of the vortical flow near the interface is directly achieved.
related to the interface shape. At small growth rates, when the
interface shape is convex-solid in the isotropic case, hot fluid from 3.5 Zero-Gravity Simulations. Simulations absent gravity
the upper part of the melt cools as it falls, and continues ‘‘down- are of interest for two reasons. First, initial experiments on micro-
hill’’ along the interface as it moves radially outward to the wall. gravity Bridgman growth of an anisotropic semiconductor have
As this downgoing fluid approaches the wall, on which the tem- recently been reported 关86兴. The material studied, tellurium, has
Fig. 11 Interface shapes „below… for localized ampoule-wall
heating with ␹ Ä0.1, ␴ Ä2, z 0 Õ r i ÄÀ0.05, and ␭Ä0.4 cmÀ1 „cor-
responding to dT b „0…Õ dz Ä30°C cmÀ1 when ␹ Ä0…. U Ä0.25
Ä1.0 ␮m secÀ1, ⌬ z Õ r i Ä0.156. Upper curves are for ␹ Ä0: U
Ä0.25 ␮m secÀ1, ⌬ z Õ r i Ä0.120; U Ä0.50 ␮m secÀ1, ⌬ z Õ r i
Ä0.152; U Ä1.0 ␮m secÀ1, ⌬ z Õ r i Ä0.215.
Fig. 10 Isohterms and streamlines for localized ampoule-wall

heating with ␹ ÄÀ0.1, z 0 Õ r i ÄÀ0.05, ␴ Ä2, and ␭Ä0.4 cmÀ1, cor-
responding to dT b „0…Õ dz Ä30°C cmÀ1 when ␹ Ä0. „a… U Ä0.25
␮m secÀ1: ␺ minÄÀ0.0515, ␺ maxÄ0.0387; „b… U Ä0.5 ␮m secÀ1:
␺ minÄÀ0.0481, ␺ maxÄ0.0389; and „c… U Ä1.0 ␮m secÀ1: ␺ min
ÄÀ0.0425, ␺ maxÄ0.0402.
hexagonal symmetry, like GaSe, so that its thermal conductivity is

isotropic about the c axis. Second, elimination of gravity in our
computations lays bare its role in determining the shape of the
interface and structure of the flow.
As indicated in §2, the difference between the liquid and solid
densities gives rise to solidification-shrinkage driven flow even
absent gravity. For U⫽1.0 ␮m sec⫺1, Fig. 12 shows that the
shrinkage-driven flow is normal to the interface, with no recircu-
lation. 共As indicated in §2, the streamlines shown are for a frame
moving with the ampoule; streamlines for a frame moving with
the interface would be exactly normal to the interface.兲 Results for
U⫽0.25 ␮m sec⫺1 共not shown兲 are similar, with less interface
deflection, and maximum and minimum streamfunction values ap-
proximately one-fourth those shown for 1.0 ␮m sec⫺1. The weak-
ness of the flow, as indicated by the small magnitudes of the
maximum and minimum values of the streamfunction, strongly
suggests that dopant transport will be largely by diffusion, as for
zero-g growth of pyridine-doped benzene 关9兴. We can use the
maximum magnitude of the streamfunction to compute a mass
transfer Péclet number based on the velocity u and a diffusion
coefficient of 2⫻10⫺5 cm2 sec⫺1. For U⫽1.0 ␮m sec⫺1, this
gives Pe D,u⫽( ␯ M /r i ) 兩 u兩 r i /D⫽390 max兩␺兩⫽0.03, compared to Fig. 12 Isotherms and streamlines for zero g at dT b „0…Õ dz
Pe D,u⫽27 at the same growth rate for normal gravity. We thus Ä30°C cmÀ1, U Ä1.0 ␮m secÀ1: ␺ minÄÀ7.79Ã10À5 , ␺ max
see that mass transfer will be dominated by diffusion at zero g. ÄÀ5.09Ã10À8 .
The liquid-phase temperature distributions differ from their driven flow. This is a consequence of the smaller Péclet number
normal-gravity counterparts in several respects. First, the zero- for GaSe (4.5⫻10⫺3 – 5.4⫻10⫺2 , depending on growth rate兲,
gravity isotherms in the primary vortex exhibit more curvature compared to 0.11–0.33 for growth of benzene.
near the centerline than in the normal-gravity case, due to less The ability to use localized heating to control interface shape
radial mixing associated with ‘‘countercurrent’’ normal-gravity and the direction and strength of the flow adjacent to the interface
flow 共see Fig. 2兲. Second, the zero-gravity isotherms show more provides a means to influence dopant distributions and reduce
curvature along the centerline in the secondary vortex. The lower twinning and other crystal defects over a range of growth rates.
curvature in the normal-gravity case is due to significant upward This approach points the way to tailoring the ampoule-wall tem-
flow along the centerline, which reduces isotherm curvature asso- perature profile so as to achieve not only desirable interface
ciated with the curved interface immediately below. shapes 关53,87兴, but also buoyancy-driven flows conducive to fa-
The interface shapes 共not shown兲 are very similar to those com- vorable solid-phase dopant distributions.
puted for normal gravity at the same growth rates, differing by Finally, the availability of a complete set of measured values of
less than 7 percent at the four growth rates considered. It is clear the thermal expansion coefficients and elastic coefficients 关16兴 for
that while thermally-driven buoyant convection will play a domi- GaSe provides the basis for calculating the stress distribution in
nant role in determining solute distributions, it has little effect on Bridgman-grown boules of this anisotropic material.
the interface shape and solid-phase temperature distribution.
5 Conclusions
4 Discussion This first analysis and computation of flow, heat transfer, and
The present results are qualitatively different from those for interface shape in vertical Bridgman growth of an anisotropic
benzene 关5兴, the only other material with anisotropic solid-phase semiconductor shows that when the c axis of GaSe is aligned with
thermal conductivity for which simulations of vertical Bridgman the growth direction, anisotropy can have important conse-
growth have been performed. Computations 共§3.3兲 for a fictitious quences. Interface shapes computed using experimental aniso-
material with all properties the same as those of GaSe, except for tropic thermal conductivity data are decidedly concave-solid, as
an isotropic thermal conductivity, show that the interface shape is opposed to the convex-solid shapes predicted using an isotropic
strongly influenced by anisotropy. Computations for benzene 关5兴 conductivity appropriate to a polycrystalline or amorphous mate-
also revealed that thermal anisotropy played an important role in rial at low growth rates, and slightly concave-solid shapes pre-
determining interface shape. One fundamental difference between dicted for isotropic conductivity at higher growth rates. For
benzene and GaSe is that for benzene the ratio of the conductivity concave-solid interfaces corresponding to the anisotropy of the
parallel to the growth direction to the conductivity perpendicular real material, the flow on the interface is radially inward toward
to the growth direction was 2.2, while for GaSe it is about 0.12. the centerline, while for the convex-solid interface shapes, the
An important consequence is that, unlike the benzene case, we flow is outward to the ampoule wall. The different flows and
have found no monotonic ampoule-wall temperature profile that interface shapes will lead to profoundly different solid-phase dop-
gives a convex-solid interface shape. As for the case of benzene, ant or impurity distributions, which we are currently investigating.
the isotherms are insensitive to the growth rate, except near the The results also show that the interface shape is not convex-
interface. solid for any of the growth conditions investigated. However, by
Our far-field computational boundary in the liquid is placed far use of a ring heater in the ‘‘adiabatic’’ zone between the hot and
enough from the interface that the vertical extent of the primary cold zones, one or two additional vortices 共beyond the two that
vortex is determined by the thermophysical properties and growth exist absent the additional heating兲 can be introduced in the region
conditions 共including thermal boundary conditions on the am- overlying the interface. Depending on growth rate and other con-
poule wall兲. As a consequence, the computed flow should faith- ditions, the flow direction on the interface can be reversed com-
fully model flows in long ampoules used to grow boules with low pared to the case without additional heating, leading to a change
degrees of axial segregation 关9兴. in the location of solute accumulation. Such local heating can also
The computed interface shapes are qualitatively similar to those reduce interfacial curvature.
earlier calculated for vertical Bridgman growth of Ge 关19兴, with
the deflection at the centerline increasing monotonically with
Acknowledgments
growth rate. At the higher growth rates considered, our computed
dimensionless interface deflections are in excess of twice the larg- The authors gratefully acknowledge the comments, suggestions,
est values computed earlier for Ge. and information provided by Drs. Nils C. Fernelius, Valeriy G.
The qualitatively different 共convex-solid, and at 3.0 ␮m sec⫺1, Voevodin, and two anonymous reviewers. We also very much
nearly planar兲 interface shapes obtained using an isotropic thermal appreciate the efforts of several people, including Donna Swis-
conductivity in the solid emphasize the importance of properly cher, Irina Bereznaya, Barbara Loomis, Tina Chrzastowski, and
accounting for anisotropy in simulations of vertical Bridgman Joseph Bentsman to obtain a copy of Reference 关62兴, without
growth of materials with significant thermal anisotropy. The re- which this work would not have been possible. The authors grate-
versal of flow will have the effect of changing the predicted loca- fully acknowledge support of the Microgravity Science and Ap-
tion of maximum solute enrichment from the centerline to the plications Division of the National Aeronautics and Space Admin-
ampoule wall for dopants rejected at the interface, and from the istration through Grant NAG3-1121 during the early stages of the
wall to the centerline for dopants preferentially incorporated into development of the code used herein. The computations were per-
the solid. formed using the facilities of the National Center for Supercom-
The fact that the flow and interface shape do not significantly puting Applications.
depend on whether account is taken of the temperature depen-
dence of the viscosity and conductivities bodes well for prediction References
of vertical Bridgman growth for materials for which less extensive
关1兴 Feigelson, R. S., and Route, R. K., 1990, ‘‘Improved Yield of Bridgman
thermophysical property data is available than for GaSe. Grown AgGaSe2 Crystals using Shaped Crucibles,’’ J. Cryst. Growth, 104, pp.
Unlike the benzene case 关5兴, zero-gravity simulations show that 789–792.
elimination of buoyancy-driven convection does not lead to a 关2兴 Feigelson, R. S., and Route, R. K., 1980, ‘‘Vertical Bridgman Growth of
convex-solid interface shape. This is again a consequence of the CdGeAs2 with Control of Interface Shape and Orientation,’’ J. Cryst. Growth,
49, pp. 261–273.
high lateral conductivity compared to the conductivity in the 关3兴 Prazak, M., and Holecek, S., 1995, ‘‘Directional Solidification of AlZn Eutec-
growth direction. Also, unlike the benzene case 共see Figs. 4 and 8 tic in Microgravity Conditions,’’ Cryst. Res. Technol., 30, pp. 927–932.
of Ref. 关5兴兲, the isotherms are affected very little by the buoyancy- 关4兴 McGhie, A. R., and Sloan, G. J., 1976, ‘‘Impurity Distribution in Organic
Crystals I. The System Naphthalene/2-Chloronaphthalene,’’ J. Cryst. Growth, and Chkunina, V. N., 1982, ‘‘Influence of Gamma and Neutron Irradiation on
32, pp. 60–67. the Photoluminescence Spectra of p-type GaSe Single Crystals,’’ Sov. Phys.
关5兴 Lee, H., and Pearlstein, A. J., 2000, ‘‘Simulation of Vertical Bridgman Growth Semicond., 16, pp. 729.
of Benzene, a Material with Anisotropic Solid-Phase Thermal Conductivity,’’ 关35兴 Shigetomi, S., Ikari, T., and Nakashima, H., 1993, ‘‘Optical and Electrical
J. Cryst. Growth, 209, pp. 934–952. Properties of Layer Semiconductor p-GaSe Doped with Zn,’’ J. Appl. Phys.,
关6兴 Gau, C., and Viskanta, R., 1985, ‘‘Effect of Crystal Anisotropy on Heat Trans- 74, pp. 4125–4129.
fer During Melting and Solidification of a Metal,’’ ASME J. Heat Transfer, 关36兴 Lee, W.-S., Kim, N.-O., and Kim, B.-I., 1996, ‘‘Optical Properties of
107, pp. 706–708. GaSe:Er3⫹ Single Crystals,’’ J. Mater. Sci. Lett., 15, pp. 1644–1645.
关7兴 Weaver, J. A., and Viskanta, R., 1989, ‘‘Effects of Anisotropic Heat Conduc- 关37兴 Singh, N. B., Suhre, D. R., Rosch, W., Meyer, R., Marable, M., Fernelius, N.
tion on Solidification,’’ Numer. Heat Transfer, 15A, pp. 181–195. C., Hopkins, F. K., Zelmon, D. E., and Narayanan, R., 1999, ‘‘Modified GaSe
关8兴 Huang, C. E., Ewell, D., and Feigelson, R. S., 1983, ‘‘Influence of Thermal Crystals for Mid-IR Applications,’’ J. Cryst. Growth, 198Õ199, pp. 588–592.
Conductivity on Interface Shape During Bridgman Growth,’’ J. Cryst. Growth, 关38兴 Micocci, G., Serra, A., and Tepore, A., 1997, ‘‘Electrical Properties of n-GaSe
64, pp. 441–447. Single Crystals Doped with Chlorine,’’ J. Appl. Phys., 82, pp. 2365–2369.
关9兴 Lee, H., and Pearlstein, A. J., 2000, ‘‘Simulation of Radial Dopant Segregation 关39兴 Singh, N. B., Narayanan, R., Zhao, A. X., Balakrishna, V., Hopkins, R. H.,
in Vertical Bridgman Growth of Pyridine-Doped Benzene, a Surrogate for Suhre, D. R., Fernelius, N. C., Hopkins, F. K., and Zelmon, D. E., 1997,
Binary Organic Nonlinear Optical Materials,’’ J. Cryst. Growth, 218, 354– ‘‘Bridgman Growth of GaSe Crystals for Nonlinear Optical Applications,’’
352. Mater. Sci. Eng., B, 49, pp. 243–246.
关10兴 Kasriel, R. H., 1971, Undergraduate Topology, W. B. Saunders, Philadelphia, 关40兴 Singh, N. B., Suhre, D. R., Balakrishna, V., Marable, M., Meyer, R., Fernelius,
PA, p. 116. N., Hopkins, F. K., and Zelmon, D., 1998, ‘‘Far-Infrared Conversion Materi-
关11兴 Fernelius, N. C., Singh, N. B., Suhre, D. R., and Balakrishna, V., 1999, als: Gallium Selenide for Far-Infrared Conversion Applications,’’ Prog. Cryst.
‘‘Modified Gallium Selenide Crystals for High Power Nonlinear Optical Ap- Growth Charact. Mater., 37, pp. 47–102.
plications,’’ U.S. Patent 5,980,789, issued Nov. 9, 1999. 关41兴 Anis, M. K., 1981, ‘‘The Growth of Single Crystals of GaSe,’’ J. Cryst.
关12兴 Fernelius, N. C., Hopkins, F. K., and Ohmer, M. C. 1999, ‘‘Nonlinear Optical Growth, 55, pp. 465–469.
Crystal Development for Laser Wavelength Shifting at AFRL Materials Di- 关42兴 Anis, M. K., 1988, ‘‘Thermopower Measurements in p-GaSe Single Crystals
rectorate,’’ in Operational Characteristics and Crystal Growth of Nonlinear Parallel and Perpendicular to the c-Axis,’’ Int. J. Electron., 65, pp. 215–221.
Optical Materials, Proc. SPIE, 3793, SPIE, Bellingham, Wash. pp. 2–8. 关43兴 Segura, A., Guesdon, J. P., Besson, J. M., and Chevy, A., 1979, ‘‘Photovoltaic
关13兴 Lieth, R. M. A., 1977, ‘‘III-VI Compounds,’’ in Preparation and Crystal Effect in InSe. Application to Solar Energy Conversion,’’ Rev. Phys. Appl.,
Growth of Materials with Layered Structures, R. M. A. Lieth, ed., Reidel, 14, pp. 253–257.
Dordrecht, pp. 225–254. 关44兴 Parfeniuk, C., Weinberg, F., Samarasekera, I. V., Schvezov, C., and Li, L.,
关14兴 Gouskov, A., Camassel, J., and Gouskov, L., 1982, ‘‘Growth and Character- 1992, ‘‘Measured Critical Resolved Shear Stress and Calculated Temperature
ization of III-VI Layered Crystals like GaSe, GaTe, InSe, GaSe1⫺x Tex and and Stress Fields during Growth of CdZnTe,’’ J. Cryst. Growth, 119, pp.
Gax In1⫺x Se,’’ Prog. Cryst. Growth Charact., 5, pp. 323–413. 261–270.
关15兴 Maschke, K., and Lévy, F., 1983, ‘‘III-VI Compounds,’’ in Numerical Data 关45兴 Brice, J. C., 1973, The Growth of Crystals from Liquids, North-Holland, Am-
and Functional Relationships in Science and Technology, Landolt-Börnstein, sterdam, p. 209.
New Series, Group III, 17f, O. Madelung, ed., Springer-Verlag, Berlin, pp. 关46兴 Cardetta, V. L., Mancini, A. M., and Rizzo, A., 1972, ‘‘Melt Growth of Single
9–102. Crystal Ingots of GaSe by Bridgman-Stockbarger’s Method,’’ J. Cryst.
关16兴 Fernelius, N. C., 1994, ‘‘Properties of Gallium Selenide Single Crystal,’’ Prog. Growth, 16, pp. 183–185.
Cryst. Growth Charact. Mater., 28, pp. 275–353. 关47兴 Sampaio, H., Gouskov, A., and Arguello, Z. P., 1977, ‘‘Orientations of the
关17兴 Holmes, D. E., and Gatos, H. C., 1981, ‘‘Convective Interference and ‘Effec- Basal Plane of Single Crystals of GaSe Grown by Vertical Bridgman Tech-
tive’ Diffusion-Controlled Segregation during Directional Solidification under nique,’’ J. Cryst. Growth, 41, pp. 275–277.
Stabilizing Vertical Temperature Gradients; Ge,’’ J. Electrochem. Soc., 128, 关48兴 Shtanov, V. I., Komov, A. A., Tamm, M. E., Atrashenko, D. V., and Zlo-
pp. 429–437. manov, V. P., 1998, ‘‘Phase Diagram of the Gallium-Selenium System and
关18兴 Chang, C. J., and Brown, R. A., 1983, ‘‘Radial Segregation Induced by Natu- Photoluminescence Spectra of GaSe Crystals,’’ Dokl. Chem., 361, pp. 140–
ral Convection and Melt/Solid Interface Shape in Vertical Bridgman Growth,’’ 143.
J. Cryst. Growth, 63, pp. 343–364. 关49兴 Gouskov, L., Gouskov, A., Lemos, V., May, W., and Sampaio, H., 1977,
关19兴 Adornato, P. M., and Brown, R. A., 1987, ‘‘Convection and Segregation in ‘‘Electrical Properties of GaTex Se1⫺x Crystals,’’ Phys. Status Solidi A, 39, pp.
Directional Solidification of Dilute and Non-dilute Binary Alloys: Effects of 65–71.
Ampoule and Furnace Design,’’ J. Cryst. Growth, 80, pp. 155–190. 关50兴 Sulewski, P. E., Bucher, E., Stücheli, N., Oglesby, C. S., Friemelt, K., Vögt,
关20兴 Adornato, P. M., and Brown, R. A., 1987, ‘‘Petrov-Galerkin Methods for M., Baumann, J. R., and Kloc, C., 1992, ‘‘Search for Giant Franz-Keldysh-
Natural Convection in Directional Solidification of Binary Alloys,’’ Int. J. Like Effects in GaSe and Other Layered Semiconductors,’’ Appl. Phys. A, 54,
Numer. Methods Fluids, 7, pp. 761–791. pp. 79–83.
关21兴 Kim, D. H., and Brown, R. A., 1989, ‘‘Models for Convection and Segregation 关51兴 Sakai, E., Nakatani, H., Tatsuyama, C., and Takeda, F., 1998, ‘‘Average En-
in the Growth of HgCdTe by the Vertical Bridgman Method,’’ J. Cryst. ergy Needed to Produce an Electron-Hole Pair in GaSe Nuclear Particle De-
Growth, 96, pp. 609–627. tectors,’’ IEEE Trans. Nucl. Sci., 35, pp. 85–88.
关22兴 Kim, D. H., and Brown, R. A., 1991, ‘‘Modelling of the Dynamics of HgCdTe 关52兴 Pfeiffer, M., and Mülhberg, M., 1992, ‘‘Interface Shape Observation and Cal-
Growth by the Vertical Bridgman Method,’’ J. Cryst. Growth, 114, pp. 411– culation in Crystal Growth of CdTe by the Vertical Bridgman Method,’’ J.
434. Cryst. Growth, 118, pp. 269–276.
关23兴 Kuppurao, S., Brandon, S., and Derby, J. J., 1995, ‘‘Modeling the Vertical 关53兴 Lan, C. W., and Ting, C. C., 1996, ‘‘A Study of the Interface Control of
Bridgman Growth of Cadmium Zinc Telluride: I. Quasi-Steady Analysis of Vertical Bridgman Crystal Growth using a Transparent Multizone Furnace,’’
Heat Transfer and Convection,’’ J. Cryst. Growth, 155, pp 93–102. Chem. Eng. Commun., 145, pp. 131–143.
关24兴 Liang, M. C., and Lan, C., 1996, ‘‘Three-Dimensional Convection and Solute 关54兴 Zhang, H., Zheng, L. L., Prasad, V., and Larson, D. J., 1998, ‘‘Local and
Segregation in Vertical Bridgman Crystal Growth,’’ J. Cryst. Growth, 167, pp. Global Simulations of Bridgman and Liquid-Encapsulated Czochralski Crystal
320–332. Growth,’’ ASME J. Heat Transfer, 120, pp. 865–873.
关25兴 Lan, C. W., and Chen, F. C., 1996, ‘‘A Finite Volume Method for Solute 关55兴 Anis, M. K., and Piercy, A. R., 1977, ‘‘Growth of Single Crystals of GaSe
Segregation in Directional Solidification and Comparison with a Finite Ele- with Natural Facets at Large Angles to the Layers,’’ Phys. Status Solidi A, 44,
ment Method,’’ Comput. Methods Appl. Mech. Eng., 131, pp. 191–207. pp. K5–K6.
关26兴 Lan, C. W., 1999, ‘‘Effects of Ampoule Rotation on Flows and Dopant Seg- 关56兴 Glazov, V. M., Makhmudov, S., and Mavlonov, S., 1972, ‘‘Thermal Expan-
regation in Vertical Bridgman Growth,’’ J. Cryst. Growth, 197, pp. 983–991. sion and Volume Changes During the Melting of Gallium and Indium Se-
关27兴 Ma, N., and Walker, J. S., 2000, ‘‘A Model of Dopant Transport During lenides’’ 共translated兲, Izv. Akad. Nauk Tadzh. SSR, Otd. Fiz.-Mat. Geol.-
Bridgman Crystal Growth with Magnetically Damped Buoyant Convection,’’ Khim. Nauk, No. 1, pp. 20–24.
ASME J. Heat Transfer, 122, pp. 159–164. 关57兴 Suzuki, H., and Mori, R., 1974, ‘‘Phase Study on Binary System Ga-Se,’’ Jpn.
关28兴 Jandl, S., Brebner, J. L., and Powell, B. M., 1976, ‘‘Lattice Dynamics of J. Appl. Phys., 13, pp. 417–423.
GaSe,’’ Phys. Rev. B, 13, pp. 686–693. 关58兴 Dieleman, J., Sanders, F. H. M., and van Dommelen, J. H. J., 1982, ‘‘The
关29兴 Bastow, T. J., Campbell, I. D., and Whitfield, H. J., 1981, ‘‘A 69Ga, 115In NQR Phase Diagram of the Ga-Se System,’’ Philips J. Res., 37, pp. 204–229.
Study of Polytopes of GaS, GaSe and InGe,’’ Solid State Commun., 39, pp. 关59兴 Dieleman, J., and Engelfriet, R. G., 1971, ‘‘The Phase Diagram of the System
307–311. Ga1⫺x Sex for 0.5⭐x⭐0.6 and 300 K⭐T⭐1500 K,’’ J. Less-Common Met.,
关30兴 Cox, E. G., Cruickshank, D. W. J., and Smith, J. A. S., 1958, ‘‘The Crystal 25, pp. 231–233.
Structure of Benzene at ⫺3°C,’’ Proc. R. Soc. London, Ser. A, 247, pp. 1–21. 关60兴 Zavrazhnov, A. Y., Turchen, D. N., Goncharov, E. G., and Prigorodova, T. A.,
关31兴 Guseinov, G. D., Abdullayeva, S. G., Ramazanzade, A. M., Ismailov, M. Z., 1999, ‘‘Scanning of T-X Projections of Phase Microdiagrams based on Data
Viscakas, J. K., Vaitkus, J. J., and Baltramiejūnas, R. A., 1975, ‘‘Anisotropy on Gas Solubility in Melts. Homogeneity Area of GaSe at Premelting Tem-
of Lattice Heat Conductivity of Complex Chalcogenides,’’ Phys. Lett., 54A, peratures,’’ Russ. J. Gen. Chem., 69, pp. 1692–1697.
pp. 378–380. 关61兴 Fedorov, V. I., and Machuev, V. I., 1972, ‘‘Thermal Conductivity of Selenium
关32兴 Tiller, W. A., 1963, ‘‘Principles of Solidification,’’ The Art and Science of and of Indium and Gallium Selenides in the Liquid and Solid States,’’ Sov.
Growing Crystals, J. J. Gilman, ed., Wiley, New York, pp. 276–342. Phys. Semicond., 6, pp. 142–144.
关33兴 Huang, C. E., Elwell, D., and Feigelson, R. S., 1984, ‘‘Computation of Stress 关62兴 Glazov, V. M., Makhmudov, S., and Mavlonov, S., 1978, ‘‘Temperature De-
in Bridgman Crystals,’’ J. Cryst. Growth, 69, pp. 275–280. pendence of the Viscosity of Gallium and Indium Selenides’’ 共translated兲,
关34兴 Abdullaev, G. B., Abasova, A. Z., Zaitov, F. A., Lepnev, L. S., Stafeev, V. I., Elektricheskie Svoistva Slozhnykh Poluprovodnikov i Kristallov, S. M. Mav-
lonov and R. A. Karieva, eds., Donish, Dushanbe, pp. 28–41. 关76兴 Rustamov, P. G., Melikova, Z. D., Nasirov, Y. N., and Alidzhanov, M. A.,
关63兴 Takeda, S., Okazaki, H., and Tamaki, S., 1982, ‘‘Specific Heat of Liquid In-Te 1969, ‘‘Preparation of Single Crystals of Solid Solutions of Alloys of the
Alloys,’’ J. Phys. C, 15, pp. 5203–5210. System GaS-GaSe and Study of Their Physical Properties,’’ Inorg. Mater., 5,
关64兴 Takeda, S., Tamaki, S., Takano, A., and Okazaki, H., 1983, ‘‘Specific Heat of pp. 750–752.
Liquid Ga-Te Alloys,’’ J. Phys. C, 16, pp. 467–471. 关77兴 Guseinov, G. D., and Rasulov, A. I., 1966, ‘‘Heat Conductivity Study of GaSe
关65兴 Mills, K. C., 1976, ‘‘Molar Heat Capacities and Enthalpies of Transition for Monocrystals,’’ Phys. Status Solidi, 18, pp. 911–922.
InSe共c兲, InSe1.2 and In2Se3,’’ High Temp.-High Press., 8, pp. 225–230. 关78兴 Brandon S., and Derby, J. J., 1992, ‘‘Heat Transfer in Vertical Bridgman
关66兴 Mamedov, K. K., Kerimov, I. G., Kostryukov, V. N., and Mekhtiev, M. I., Growth of Oxides: Effects of Conduction, Convection, and Internal Radia-
1967, ‘‘Specific Heat of Gallium Selenide and Thallium Selenide,’’ Sov. Phys. tion,’’ J. Cryst. Growth, 121, pp. 473–494.
Semicond., 1, pp. 363–364. 关79兴 Rustamov, P. G., Il’yasov, T. M., Safarov, M. G., and Sadykhova, S. A., 1979,
关67兴 Goldsmith, A., Waterman, T. E., and Hirschhorn, H. J., 1961, Handbook of ‘‘Projection of the Liquidus of the As, Ga储Se, Te System,’’ Russ. J. Inorg.
Thermophysical Properties of Solid Materials, Vol. III, Ceramics, Macmillan, Chem., 24, pp. 263–266.
New York, pp. 891–896.
关80兴 Il’yasov, T. M., and Rustamov, P. G., 1982, ‘‘Chemical Interaction and Glass
关68兴 Lewis, G. N., and Randall, M., 1961, Thermodynamics, 2nd ed., revised by K.
Formation in Chalcogenide Systems of the As2X3-GaX Type,’’ Russ. J. Inorg.
S. Pitzer and L. Brewer, McGraw-Hill, New York.
Chem., 27, pp. 1500–1503.
关69兴 Singh, N. B., Henningsen, T., Balakrishna, V., Suhre, D. R., Fernelius, N.,
关81兴 Rustamov, P. G., Melikova, Z. D., Safarov, M. G., and Alidzhanov, M. A.,
Hopkins, F. K., and Zelmon, D. E., 1996, ‘‘Growth and Characterization of
Gallium Selenide Crystals for Far-Infrared Conversion Applications,’’ J. 1965, ‘‘Solid Solutions in the System GaS-GaSe,’’ Inorg. Mater., 1, pp. 387–
Cryst. Growth, 163, pp. 398–402. 389.
关70兴 Manfredotti, C., Murri, R., Rizzo, A., Galassini, S., and Ruggiero, L., 1974, 关82兴 Rustamov, P. G., Babaeva, B. K., and Luzhnaya, N. P., 1965, ‘‘The Interaction
‘‘Deep Hole Traps in p-type GaSe Single Crystals,’’ Phys. Rev. B, 10, pp. of Gallium with Selenium,’’ Inorg. Mater., 1, pp. 775–776.
3387–3393. 关83兴 Muschinsky, W. P., and Pawelenko, N. M., 1969, ‘‘Untersuchung des Systems
关71兴 Nakatani, H., Sakai, E., Tatsuyama, C., and Takeda, F., 1989, ‘‘GaSe Nuclear GaSe-InSe,’’ Krist. Tech., 4, No. 2, pp. K5–K7.
Particle Detectors,’’ Nucl. Instrum. Methods Phys. Res. A, 283, pp. 303–309. 关84兴 Palatnik, L. S., and Belova, E. K., 1996, ‘‘Investigation of the Ga-Se Phase
关72兴 Rizzo, A., DeBlasi, C., Catalano, M., and Cavaliere, P., 1988, ‘‘Dislocations in Diagram,’’ Inorg. Mater., 2, pp. 657–659.
AIIIBVI Single Crystals,’’ Phys. Status Solidi A, 105, pp. 101–112. 关85兴 Kumar, S., and Singh, R. N., 1995, ‘‘Thermal Conductivity of Polycrystalline
关73兴 De Blasi, C., Manno, D., and Rizzo, A., 1989, ‘‘Convergent-Beam Electron Materials,’’ J. Am. Ceram. Soc., 78, pp. 728–736.
Diffraction Study of Melt- and Vapour-Grown Single Crystals of Gallium 关86兴 Parfeniev, R. V., Farbshtein, I. I., Shulpina, I. L., Yakimov, S. V., Shalimov,
Chalcogenides,’’ Il Nuovo Cimento, 11D, pp. 1145–1163. V. P., Turchaninov, A. M., Ivanov, A. I., and Savin, S. F., 2000, ‘‘Solidifica-
关74兴 Manfredotti, C., Mancini, A. M., Murri, R., Rizzo, A., and Vasanelli, L., 1977, tion of Anisotropic Semiconducting Material—Tellurium under Microgravity
‘‘Electrical Properties of p-Type GaSe,’’ Il Nuovo Cimento, 39B, pp. 257– Conditions,’’ Mater. Sci. Forum, 329Õ330, pp. 297–304.
268. 关87兴 Batur, C., Srinivasan, A., Duval, W. M. B., and Singh, N. B., 1995, ‘‘Control
关75兴 Lendvay, E., Kuhn, A., Chevy, A., and Ceva, T., 1971, ‘‘Dislocation Etching of Crystal Growth in Bridgman Furnace,’’ Prog. Cryst. Growth Charact.
of GaSe Single Crystals,’’ J. Mater. Sci., 6, pp. 305–308. Mater., 30, pp. 217–236.
J. H. Walther
Molecular Dynamics Simulation
Postdoctoral Fellow
e-mail: walther@inf.ethz.ch
of Nanodroplet Evaporation
P. Koumoutsakos1 Molecular dynamics simulations are used to study the sub-critical evaporation of a
Professor nanometer-size droplet at 300 K and 3 MPa. Classical molecular dynamics techniques
e-mail: petros@inf.ethz.ch are combined with an adaptive tree data structure for the construction of the neighbor
lists, allowing efficient simulations using hundreds of thousands of molecules. We present
Institute of Computational Sciences, a systematic convergence study of the method demonstrating its convergence for heat
ETH Zentrum, conduction problems in submicron scales. These high resolution simulations compute
Weinbergstrasse 43, values of the evaporation coefficient that are in excellent agreement with theoretical
CH-8092 Zürich, Switzerland predictions. 关DOI: 10.1115/1.1370517兴
Keywords: Nanofluidics, Nanodroplet, Molecular Dynamics Simulation
1 Introduction number of molecules to approach the macro-scale physics. To

validate this approach the present study involves subcritical
Phenomena associated with droplet dynamics are of fundamen-
evaporation of a droplet in a quiescent vapor allowing direct com-
tal importance to non-premixed combustion studies. Such phe-
nomena include the formation of sprays and droplets, droplet coa- parison with the D 2 evaporation law.
lescence and breakup, and droplet evaporation and combustion We conduct a series of high resolution simulations using sev-
关1–4兴. At subcritical conditions the droplet evaporation is well eral tens of thousands of computational molecules by efficiently
employing tree data structures. We present a systematic conver-
described by the classical D 2 evaporation law 关5兴
gence study to examine the influence of the different parameters
dD 2 imposed by the numerical method, including the size of the com-
⫽⫺ ␤ v , (1) putational domain, cutoff radius, droplet diameter, heating fre-
dt quency and size of heating region, initialisation period, and time
where ␤ v is the evaporation coefficient, and D is the droplet di- step size. It is shown that molecular dynamics simulations con-
ameter. At supercritical conditions the continuum analysis of the verge to the D 2 evaporation law to desired accuracy by using
problem is faced with suitable modeling of transport and thermo- large number of computational molecules. The paper is organized
dynamic properties, equations of state, and proper specification of as follows: Section 2 outlines the numerical method including a
the interface 关6,7兴. Using molecular dynamics simulations the description of the applied tree data structure and of the heating
modeling issues reduce to the specification of proper interatomic algorithm. Section 3 presents the convergence study and the pre-
potentials. dicted evaporation rates.
Molecular dynamics simulations of nanometer-size droplets in
equilibrium with a vapor have been applied for the study of the
droplet pressure and density profile, and to the calculation of the 2 Molecular Dynamics
surface tension as function of vapor temperature and droplet size
关8–11兴. Studies of non-equilibrium systems have included con- The evaporation of an argon droplet in its own vapor is studied
densation and evaporation of argon slabs 关12–14兴, evaporation of using molecular dynamics simulations. The atoms are assumed to
argon clusters 关15兴, and evaporation of argon droplets 关16,17兴, interact via the 12–6 Lennard-Jones pair potential
冋冉冊冉冊册
which recently have been extended to studies of supercritical
␴ 12
␴ 6
evaporation of liquid oxygen 关18兴. Related molecular dynamics V 共 r 兲 ⫽4 ⑀ ⫺ , (2)
simulations also include studies of droplet collision cf. Svanberg r r
et al. 关19兴, Sikdar and Chung 关20兴 and Murad and Law 关21兴 and
the cluster formation of a Lennard-Jones fluid during anisotropic where r is the distance between the atoms. For argon, the zero
expansion Ashurst and Holian 关22兴. energy distance ␴ is 3.4 Å, and the depth of the potential ⑀ is
The molecular dynamics simulations by Long, Micci, and 120k B , where k B is Boltzmann’s constant. The N atoms move
Wong 关16兴 involved sub-critical evaporation of an argon droplet according to Newton’s law
in an argon vapor at high Knudsen numbers. At a droplet size of d 2 rគ p
5 nm 共droplet consisting of 1000 atoms兲, they found good agree- m ⫽⫺ⵜគ V 共 rគ p 兲 , p⫽1, . . . ,N (3)
ment with the theoretical correlation by assuming an evaporation dt 2
coefficient of 0.3. and
Little 关17兴 simulated the sub and super-critical evaporation of
nanometer-size droplets using parameters similar to the present N
study. For droplet sizes of 8 nm 共using 5587 atoms兲, the evapo-
ration coefficient was found within 20 percent of the D 2 evapora-
ⵜ
គ V共 r p兲⫽ 兺 ⵜគ V 共兩 rគ ⫺rគ 兩兲 ,
q⫽1
p q p⫽1, . . . ,N, (4)
tion law.
where m is the mass of the atom. Numerical integration of Eq. 共3兲
In this paper we consider the possibility of studying non-
premixed combustion from the molecular level, by applying large nominally requires O(N 2 ) operations, where N is the number of
atoms, but is rendered an O(N) algorithm by the short range
1 nature of the potential. The computational problem is reduced
Also at CTR, NASA Ames 202A-A-1, Moffett Field, CA-94035, USA.
then to efficiently identifying the neighboring particles at each
HEAT TRANSFER. Manuscript received by the Heat Transfer Division September 10, computational element. In the present work the neighbors are con-
1999; revision received November 20, 2000. Associate Editor: T. Avedisian. sidered for cutoff radii of r c / ␴ ⫽2.5, 5, and 10.
Fig. 1 Sketch of computational domain and heating element
The system of equations is integrated in time using the leap-

frog scheme and a constant number of atoms and in a constant
volume. Periodic boundary conditions are imposed in all spatial
directions.
For a single droplet evaporating into a vapor, the far-field tem-
perature (T̃ *
v ) boundary condition is Fig. 2 Example of a two-dimensional tree created for 16 atoms
and allowing one particle per box
v 共 r 兲 →T̃ v as r→⬁,
T* * (5)
which is enforced by a momentum scaling of the atoms located
sufficiently far away from the droplet 关16,23,24兴. The velocity is
scaled according to build by traversing the tree bottom-up, starting at the next coarsest
level where all 共at most 8兲 boxes are colleagues. At higher levels
i ⫽vi
v new old
冑 3T v k B N h
2E kin
, (6)
of the tree the colleagues of a box are searched from the children
of colleagues of the box’s parent. This effectively limits the
search to 6 3 boxes for each box. The colleagues 共marked with an
where E kin is the total kinetic energy of the N h atoms found in the ⫻兲 of the shaded boxes are shown in Fig. 3. The neighbor list for
heating region (r i ⬎r h ), and r i is the radial co-ordinate of the the childless boxes is finally constructed by considering the col-
atom measured from the center of the computational box cf. Fig. leagues and their descendents of the childless box cf. Fig. 4. The
1. The size (r h ) of the heating element and the heating frequency construction of the tree and the neighbor list takes less than 15
are adjusted to minimize thermal noise and to obtain a good ap- percent of the total CPU time.
proximation of Eq. 共5兲.
Non-dimensional quantities are marked with an asterisk and are
based on ␴, ⑀, k B , and m. Specifically, time is non-
dimensionalised by ␶ * ⫽ ␶ 冑⑀ /(m ␴ 2 ). A non-dimensional time
step of ␦ ␶ * ⫽0.005 (⬇10 fs) is used unless otherwise specified.
2.1 Tree Algorithm. Traditionally, molecular dynamics
simulations use the Verlet neighbor list in combination with a
linked list 关25–27兴 to reduce the O(N 2 ) operations required by the
direct implementation of the force evaluation cf. Eq. 共4兲. How-
ever, for systems involving large density variations 共as in the
present case兲, the regular mesh imposed by these methods often
results in inefficient load balancing on parallel computers 关17,28兴.
The present code adopts a different strategy, by using an adaptive
tree data structure to sort the atoms and to build the neighbor list.
The tree is constructed at every time step by recursively dividing
the computational box in eight equal sized boxes. The division of
the boxes is terminated if the size of the box is less than 2r c , or
if the number of atoms in the box is less than some prescribed
value. The latter criterion allows for efficient computations of
problems with large density variations. A threshold value of 40
atoms per box was used in this work. Figure 2 shows an example
of a two-dimensional 共quad兲 tree created for 16 atoms 共shown as
circles兲 using a threshold value of one atom per box. The resulting
tree has four levels.
The list of neighboring boxes is constructed for each box that
has no further subdivisions 共childless boxes兲. Two boxes are said
to be neighbors if they share at least one corner. The neighbor list
is constructed in two steps. First, neighboring boxes at the same Fig. 3 Example of colleagues of boxes in the two-dimensional
level in the tree 共colleagues兲, are found for each box in the tree 共a tree created for the 16 atoms shown in Fig. 2. The colleagues of
box has a maximum of 26 colleagues兲. The list of colleagues is the filled boxes are marked with an Ã.
Table 1 List of spatial systems investigated. D * Ä D Õ ␴ is the
initial droplet diameter, L * Ä L Õ ␴ is size of the computational
domain, N drop and N vapor are the initial number of atoms in the
droplet and vapor phases respectively. ␣ Ä L Õ D is the relative
size of the computational domain, and r * h Ä r h Õ ␴ the position of
the heating element.
Fig. 4 Example of neighbors of boxes in the two-dimensional

tree created for the 16 atoms shown in Fig. 2. The neighbors of
the filled boxes are marked with an Ã. The interactions of the
particles in the two „filled… neighboring boxes are computed at
the level of the larger box „left… utilizing the symmetry of Eq. „4….
To allow efficient computations, the atoms are mapped onto the

boxes using the so called Morton ordering 关29,30兴, in which atoms
in neighboring boxes are stored in consecutive memory locations
共cf. Fig. 5兲. This ordering can furthermore be utilized in a parallel
domain decomposition of the atoms onto the processors cf. e.g.
关31兴. Another technique currently being investigated uses the
The system is subsequently relaxed for 25 or 50 non-dimensional
lower levels of the tree to obtain information of the spatial distri-
time units by heating the complete system to a temperature of 0.83
bution of the atoms and thus guiding the decomposition. After the
at every ␶ h / ␦ ␶ time step, where 1/␶ h is the heating frequency.
decomposition the remaining levels of the tree can be constructed
During this initial heating the change in droplet diameter is less
independently.
than 2 percent confirming that the droplet is in equilibrium with
Periodic boundary conditions are handled by explicitly copying
the surrounding vapor. After the relaxation, the temperature of the
the particles in the outer regions to a ‘‘ghost layer’’ outside the
‘‘far-field’’ vapor atoms is increased to the desired value of
computational box before the construction of the tree.
T v* ⫽2.5.
2.2 Initialization. The atoms corresponding to the liquid The size of the computational box (L⫻L⫻L) is chosen suffi-
and vapor phases are initially placed on a face-centered-cubic ciently large to avoid large density and pressure increases during
共f.c.c.兲 lattice with a liquid and vapor density of ␳ l* ⫽0.715 and the evaporation, and to accommodate the temperature far-field
␳*v ⫽0.0292, respectively, corresponding to a non-dimensional
boundary condition. Using a non-dimensional box size of ( ␣
* ⫽0.83 cf. 关32兴. The droplet is ‘‘cut’’
saturation temperature of T sat ⬅L/D) 6 to 8, where D is the droplet diameter results in pressure
from the cubic f.c.c. lattice into a spherical shape to reduce the rises of 0.012 to 0.0285 共15–40 percent pressure increase兲, which
required relaxation period. are both well below p * c .
The atoms are assigned a Maxwellian velocity distributed with Five spatial systems are investigated including three different
the desired initial temperature (T l* ⫽T * droplet diameters 共D⫽8, 13, and 17 nm兲 and three different sizes
v ⫽T sat) corresponding to a
*
of the computational domain cf. Table 1. The number of atoms
saturation pressure of p * ⫽0.0076 and well below the critical
range from 5769 and 13260 to 51105 and 105480 for the liquid
* ⫽0.1154 cf. 关33兴. At these conditions
pressure for argon of p crit and vapor phases, respectively. In all the cases, the heating was
the droplet will be in equilibrium with the surrounding vapor. The performed on atoms located well beyond the region of the droplet
initial mean drift velocity is computed and subtracted accordingly. 共r h / ␴ ⬎25, 30, and 40兲.
2.3 Diagnostics. The density and temperature profiles pre-
sented in the following are instantaneous profiles obtained by
sampling the enumeration of atoms and the associated kinetic en-
ergy in concentric shells of constant radial spacing around the
system’s center of mass. A total of fifty shells were found to give
a good resolution of the extracted fields.
The size of the droplet is given by the number of atoms found
in the liquid phase as 关17兴
D 2⬇ 6冉 N dropm
␲␳ l 冊 2/3
, (7)
where a spherical shape has been assumed. Using a simple Carte-

sian binning of the atoms, and assigning atoms in bins with a
sufficiently high density to the liquid phase was found to give
insufficient resolution. Instead, the local density is computed at
the position of each atom using a spherical binning. A radius of
interrogation of 1.5r c provides sufficient statistics while at the
same time retaining the spatial resolution. Atoms with a local
density exceeding a threshold density ( ␳ * c ) are assigned to the
Fig. 5 Atoms in neighboring boxes are stored in consecutive liquid phase. The threshold value was found to determine the size
memory locations to allow efficient computations using Morton of the drop to within a constant and hence not to influence the
ordering predicted evaporation rate. Thus, using a fixed value of ␳ * c
where p v is the vapor pressure, M the molar mass, and ⍀ ( 1,1) * the
collision integral.
Since 共␳D兲 varies slowly with temperature, ( ␳ v Dv ) can replace
( ␳ s Ds ), and Dv is the self-diffusion of the vapor. Using the pa-
rameters for argon 共c p ⫽548 J/kg K, and h f g ⫽88 kJ/kg兲关32兴, and
(⍀ ( 1,1) * ⫽0.996) 关34兴 the transfer number is close to unity and the
diffusion coefficient is 623⫻10⫺9 m2/s.
The self-diffusion coefficient of the bulk vapor has also been
computed in separate equilibrium molecular dynamics simulations
for system sizes of 1372 and 2916 atoms, respectively. The self-
diffusion is computed from the Green-Kubo relation 关26兴
冕
N
1 ⬁
D⫽
3N 0
兺 vគ 共 t 兲 • vគ 共 0 兲 dt,
j
j j (11)
where vគ j (t) is the velocity of the jth atom at time t. The predicted
values are 592⫻10⫺9 m2/s and 639⫻10⫺9 m2/s for the two sys-
tem sizes and within 5 percent of the theoretical value given by
Eq. 共10兲.
3 Results
A range of simulations were conducted to study the consistency
and convergence of the MD approach and to determine the influ-
ence of the cutoff radius, the temporal and spatial resolution, and
the heating frequency. The list of simulations is shown in Table 2,
and the computational cost for each case is listed in Table 2. The
present tree code was compared with a traditional cell index code
Fig. 6 Droplet density and temperature profiles during accom- for System I and V using cutoff values of 2.5 and 5.0, respec-
modation and evaporation of the 5768 atom droplet „Case E01…. tively. The cell index was found to be 20–30 percent faster for the
„—: ␶ * Ä25; – – –: ␶ * Ä30; - - -: ␶ * Ä100… smaller cutoff value and for the smaller system 共System I兲,
whereas the tree code is approximately 40 percent faster for the
larger system 共System V兲 using a cut-off value of 5.0. The com-
parison was performed on a SGI 共Octane R10k兲 workstation. Dur-
⫽0.298, and applying this procedure to the initial f.c.c. lattice ing the simulations the droplet remained close to the geometrical
resulted in deviations in number of liquid atoms of less than 3 center of the computational box, and hence no drift adjustment
percent. was required.
2.4 Evaporation. After the initial relaxation, the reduced 3.1 Convergence Study. The sensitivity of the different nu-
temperature of the vapor is raised to 2.5 corresponding to a pres- merical parameters is studied for the radial density and tempera-
sure of 0.071. The droplet is heated by the vapor until the satura- ture profiles during the evaporation of the 5766 atom droplet.
tion temperature is reached at T * ⫽1.156, when the droplet starts
3.1.1 Spatial Convergence. Simulations are conducted to
to evaporate. The radial density and temperature profiles during
study the influence of the cutoff radius using r c / ␴ ⫽2.5, 5.0, and
heating and evaporation at ␶ * ⫽25, 30, and 100 are shown in Fig. 10.0, respectively. The influence on the non-dimensional radial
6 for Case I. At t * ⫽30 the center of the droplet is seen to be density and temperature profiles at a non-dimensional time of 200
slightly sub-cooled at a temperature of 0.83. is shown in Fig. 7. The simulation using the smaller cutoff radius
At these subcritical conditions the droplet is expected to evapo- (r c / ␴ ⫽2.5) clearly exhibits increased evaporation and lower
rate according to the D 2 evaporation law 共1兲 droplet density due to the discontinuous cutoff of the potential and
d共 D2兲 the reduced surface tension 关35兴. The mean difference between the
⫽⫺ ␤ v , simulation using a cutoff radius of 2.5 and the simulations using
dt higher cutoff value is 12 percent, whereas the difference between
where the evaporation coefficient ␤ v is given by 关5兴 the simulations of 5.0 and 10.0 is less than 1.0 percent. Similar but
冉冊
less severe trends can be observed in the temperature profile
8 ␳ s Ds 共Fig. 7兲 exhibiting difference less than 1.0 percent between the
␤ v⫽ ln共 1⫹B 兲 , (8)
␳l simulations.
Three different sizes of the computational box, ␣ ⫽6, 7, and 8
where ␳ s and Ds are the density and self-diffusion at the surface are tested, and differences of 2.9 percent and 5.3 percent are ob-
of the droplet. The transfer number B is served in the temperature profiles for a non-dimensional time of
c p 共 T v ⫺T s 兲 100 and 200, respectively 共not shown兲. The corresponding devia-
B⫽ , (9) tions in the density profiles are less than 0.5 percent, consistent
hfg
with a prediction of the evaporation rate that is insensitive to ␣.
where c p and h f g are the specific heat and heat of vaporization,
3.1.2 Temporal Convergence. The temporal convergence is
and T v and T s are the temperature of the vapor and the droplet
studied using different non-dimensional time step sizes of 0.005
surface temperature, respectively. The coefficient of self-diffusion
共Case E01兲 and 0.010 共Case E08兲, respectively. The differences in
is found from 关34兴
both the density and temperature profiles are less than 1.7 percent
冑T 3v /M 共not shown兲. Similar deviations are observed for the length of the
D⫽2.6280⫻10⫺22 , (10) initialization period 共Cases E01 and E09 using 25 and 50 non-
p v ␴ 2 ⍀ 共 1,1兲 * 共 T v* 兲 dimensional time units, respectively兲. The effect of doubling the
Table 2 List of simulation cases. The ‘‘system’’ refers to the spatial systems listed in Table 1, r c Õ ␴
is the cutoff radius, ␣ Ä L Õ D the size of the computational box, ␦ ␶ * the non-dimensional time step,
␶ h Õ ␦ ␶ the heating frequency, ␶ i Õ ␦ ␶ the initialization period, CPU time per time step „in seconds on a
Sun Ultra 2 workstation…, and ␤ v is the predicted evaporation rate. The theoretical evaporation rate is
1.7Ã10À7 m2Õs.
initialization period results in a difference in the profiles of less

than 1.7 percent 共not shown兲. Thus, an initialization period of 25
is used throughout.
3.1.3 Heating Frequency. The effect of the heating fre-
quency (1/␶ h ) is studied using three different heating rates:
␶ h / ␦ ␶ ⫽100, 500, 1000 共Cases: E01, E10, and E11兲, and the dif-
ference in the density and temperature profiles are found to be less
than 0.5 percent and 1.3 percent, respectively. One simulation was
conducted to study the initial heat contents of the system. In this
Fig. 7 Convergence of the density and temperature profiles as

function of the cutoff radius for the 5768 droplet at ␶ * Ä200. —: Fig. 8 Evaporation curve for the 5768 droplet. ¿: E01 „ r c Õ ␴
r c Õ ␴ Ä2.5 „Case E01…; – – –: r c Õ ␴ Ä5.0 „Case E02…; - - -: r c Õ ␴ Ä2.5…; Ã: E02 „ r c Õ ␴ Ä5.0…; *: E03 „ r c Õ ␴ Ä10.0…; 䊐: E05 „ r c Õ ␴
Ä10.0 „Case E03… Ä2.5…; —: Theory.
Fig. 11 Evaporation curve for the 22360 atom droplet. ¿: E06;
—: Theory.
The ‘‘numerical’’ heating and evaporation caused by the dis-

continuous cutoff of the potential is seen for the simulations E01
and E05 as compared to the E02 and E03 cases.
Using Eq. 共8兲 and a mean liquid density of 0.6, see e.g., Fig. 7
gives a theoretical evaporation coefficient of 1.7⫻10⫺7 m2/s. The
simulated values extracted from Fig. 8 are 1.43⫻10⫺7 m2/s and
1.13⫻10⫺7 m2/s for the small and large cutoff simulation 共cf.
Table 2兲. Hence, the predicted evaporation coefficients are within
Fig. 9 Density and temperature profiles during evaporation for

the 5867 atom droplet „Case E02…. —: ␶ * Ä100; – – –: ␶ * Ä200;
- - -: ␶ * Ä300.
case the heating was stopped after the initialization and the droplet
reached an equilibrium with a radius close to the initial.
3.2 Evaporation Coefficient. The non-dimensional evapo-
ration rate (D 2 /D 20 ) for the 5867 atom droplet simulation is shown
in Fig. 8 for the different cutoff values. D 0 is the initial droplet
diameter. The initial relaxation during the first 25 non-
dimensional time units, and the subsequent heating of the droplet
including condensation of vapor on the droplet 关36兴, is clearly
visible until ␶ * ⬇100. As the droplet reaches saturation, the
evaporation follows the D 2 evaporation law.
Fig. 12 Time history of the 22360 atom droplet during evapo-

ration „vapor not shown…. From left to right: ␶ * Ä100, 300, and
500.
Fig. 10 Time history of the 5867 atom droplet during evapora-

tion „Case E02… „vapor not shown…. From left to right: ␶ * Fig. 13 Evaporation curve for the 51104 atom droplet. ¿: E07;
Ä100, 200, and 300. —: Theory.
where D 0 is the initial droplet diameter extrapolated from Figs. 8,
11, and 13. The evaporation rate consequently collapses as shown
in Fig. 15.
4 Conclusion
The sub-critical evaporation of a nanometer-size droplet has
been simulated using the method of molecular dynamics. The far-
field boundary condition for the temperature is implemented by a
regional heating technique, allowing correct prescription of far-
field conditions.
An adaptive tree data structure is employed to sort the particles
and to build the neighbor lists in order to reduce the computa-
tional cost. The efficiency is further enhanced by mapping the
particles consecutively in memory using Morton ordering.
Convergence of the method has been demonstrated by varying
the numerical parameters such as droplet size, domain size, cutoff
range, time step size, initialisation length, and heating frequency.
Three argon droplets with diameters of approximately 8, 13,
and 17 nm corresponding to 5768, 22360, and 51105 liquid atoms,
respectively, were evaporated into an argon vapor at a non-
Fig. 14 Time history of the 51104 droplet during evaporation. dimensional temperature of 2.5 and a pressure of 0.071. For the 8
From left to right: ␶ * Ä100, 400, and 700. nm droplet, the evaporation rate was found within 35 percent of
the theoretical predictions, whereas the evaporation rate of the two
larger droplets deviated less than 5 percent.
In the investigated parameter range, the largest differences were
observed for the simulations using different cutoff values resulting
in increasing the evaporation rate for the smaller cutoff values.
The results of the simulations demonstrate that molecular dy-
namics simulations can be a consistent numerical method for the
simulation of thermodynamics of nano-scale phenomena. New
computational techniques and algorithms allow us to conduct rou-
tine calculations using tens of thousands and hundred of thou-
sands of atoms, thus reaching more realistic behavior of the sys-
tem being simulated.
Nomenclature
Greek
Fig. 15 Evaporation rate. —: 5678 atom droplet; – – –: 22310 ␣ ⫽ non-dimensional box size
atom droplet; - - -: 51104 atom droplet. ␤v ⫽ evaporation coefficient
␦␶ ⫽ non-dimensional time step
⑀ ⫽ energy scale of Lennard-Jones potential
␳ ⫽ density
16 percent to 34 percent of the theoretical value. Deviations of the ␴ ⫽ zero energy distance of Lennard-Jones potential
same magnitude were also found by Little 关17兴, using similar ␶ ⫽ time
number of atoms. ␶l ⫽ droplet life time
The corresponding radial density and temperature profiles are ␶h ⫽ heating time interval
shown in Fig. 9. The instantaneous position of the liquid atoms ⍀ ( 1,1) ⫽ collision integral
during evaporation 共Case E02兲 as determined from the local den- Roman
sity method is shown in Fig. 10. Atoms originally in the vapor or
B ⫽ transfer number
liquid phases are coloured light gray and black, respectively. The
D ⫽ self-diffusion coefficient
atoms are plotted in scale as spheres with a radius of ␴.
Ds ⫽ self-diffusion coefficient at the surface of the droplet
Higher resolution simulations were run in order to establish the
Dv ⫽ self-diffusion coefficient of the vapor
convergence of the method with respect to theoretically predicted
D ⫽ droplet diameter
values. The evaporation curve, density and temperature profiles,
E kin ⫽ kinetic energy
and snapshots for the 22360 and 51104 atom droplets are shown
L ⫽ size of computational box
in Figs. 11–12 and 13–14, respectively. The simulated evapora-
N ⫽ total number of atoms
tion coefficients are 1.76⫻10⫺7 m2/s and 1.78⫻10⫺7 m2/s cf. Nh ⫽ number of atoms in heating region
Table 2, both in excellent agreement with the theoretical value. N drop ⫽ number of atoms in the droplet
Thus, the simulations converge to the theoretical values as the T ⫽ temperature
number of droplet atoms is increased beyond 104 . V ⫽ potential
To combine the evaporation curves for the different droplet cp ⫽ specific heat
sizes, time is non-dimensionalized using the macro-scale droplet hfg ⫽ heat of vaporisation
life time, kB ⫽ Boltzmann’s constant
m ⫽ atomic mass
␤v r ⫽ distance
␶ l⫽ . (12) rc ⫽ cutoff radius
D 20 rh ⫽ distance from center of box to heating region
ri ⫽ distance from center of box to i-th atom 关13兴 Matsumoto, M., 1996, ‘‘Molecular Dynamics of Liquid Surfaces,’’ Mol.
Simul., 16, pp. 209–217.
p ⫽ pressure 关14兴 Zhakhovskiı̌, V. V., and Anisimov, S. I., 1997, ‘‘Molecular-Dynamics Simu-
v ⫽ velocity lation of Evaporation of a Liquid,’’ JETP, 84, No. 4, pp. 734–745.
T ⫽ temperature 关15兴 Rytkönen, A., Valkealahti, S., and Manninen, M., 1997, ‘‘Melting and Evapo-
T sat ⫽ saturation temperature ration of Argon Clusters,’’ J. Chem. Phys., 106, No. 5, pp. 1888–1892.
t ⫽ time 关16兴 Long, L. N., Micci, M. M., and Wong, B. C., 1996, ‘‘Molecular Dynamics
Simulations of Droplet Evaporation,’’ Comput. Phys. Commun., 96, pp. 167–
Superscripts 172.
关17兴 Little, J. K., 1996, ‘‘Simulation of Droplet Evaporation in Supercritical Envi-
* ⫽ non-dimensional quantity ronments Using Parallel Molecular Dynamics,’’ Ph.d. thesis, The Pennsylvania
State University.
Subscripts 关18兴 Kaltz, T. L., Long, L. N., Micci, M. M., and Little, J. K., 1998, ‘‘Supercritical
0 ⫽ initial value Vaporization of Liquid Oxygen Droplets Using Molecular Dynamics,’’ Com-
bust. Sci. Technol., 136, pp. 279–301.
crit ⫽ critical 关19兴 Svanberg, M., Marković, N., and Petterson, J. B. C., 1998, ‘‘Collision Dynam-
v ⫽ vapor ics of Large Water Clusters,’’ J. Chem. Phys., 108, No. 14, pp. 5888–5897.
l ⫽ liquid 关20兴 Sikdar, S., and Chung, J. N., 1997, ‘‘A Quasimolecular Approach for Discrete
c ⫽ cutoff Study of Droplet Collision,’’ Int. J. Comput. Fluid Dyn., 8, pp. 189–200.
关21兴 Murad, S., and Law, C. K., 1999, ‘‘Molecular Simulation of Droplet Collision
h ⫽ belonging to heating region in the Presence of Ambient Gas,’’ Mol. Phys., 96, No. 1, pp. 81–85.
i ⫽ initialisation 关22兴 Ashurst, W. T., and Holian, B. L., 1999, ‘‘Droplet Size Dependence Upon
sat ⫽ saturation Volume Expansion Rate,’’ J. Chem. Phys., 111, No. 6, pp. 2842–2843. 共Let-
ters to the Editor兲.
关23兴 Grönbeck, H., Tománek, D., Kim, S. G., and Rosén, A., 1997, ‘‘Does Hydro-
References gen Pre-Melt Palladium Clusters?,’’ Chem. Phys. Lett., 264, pp. 39–43.
关1兴 Glassman, I., 1994, ‘‘A Look Forward: The Next 25 Years,’’ Combust. Sci. 关24兴 Westergren, J., Grönbeck, H., Kim, S.-G., and Tománek, D., 1997, ‘‘Noble
Technol., 98, pp. 217–222. Gas Temperature Control of Metal Clusters: A Molecular Dynamics Study,’’ J.
关2兴 Brezinsky, K., 1994, ‘‘The Next Twenty Five Years of Combustion Research: Chem. Phys., 107, No. 8, pp. 3071–3079.
One Researcher’s Perspective,’’ Combust. Sci. Technol., 98, pp. 237–243. 关25兴 Verlet, L., 1967, ‘‘Computer Experiments on Classical Fluids,’’ Phys. Rev.,
关3兴 Sirignano, W. A., 1993, ‘‘Fluid Dynamics of Sprays—1992 Freeman Scholar 159, pp. 98–103.
Lecture,’’ J. Fluids Eng., 115, pp. 345–378. 关26兴 Allen, M. P., and Tildesley, D. J., 1987, Computer Simulation of Liquids,
关4兴 Aggarwal, S. K., and Peng, F., 1995, ‘‘A Review of Droplet Dynamics and Clarendon Press, Oxford.
Vaporization Modeling for Engineering Calculations,’’ ASME J. Eng. Gas 关27兴 Grest, G. S., and Dünweg, B., 1989, ‘‘Vectorized Link Cell FORTRAN Code
Turbines Power, 117, pp. 453–461. for Molecular Dynamics Simulations for a Large Number of Particles,’’ Com-
关5兴 Kuo, K. K.-Y., 1986, Principles of Combustion, John Wiley & Sons, New put. Phys. Commun., 55, pp. 269–285.
York. 关28兴 Plimpton, S., 1995, ‘‘Fast Parallel Algorithms for Short-Range Molecular Dy-
关6兴 Shuen, J. S., Yang, V., and Hsiao, C. C., 1992, ‘‘Combustion of Liquid-Fuel namics,’’ J. Comput. Phys., 117, pp. 1–19.
Droplets in Supercritical Conditions,’’ Combust. Flame, 89, pp. 299–319. 关29兴 Samet, H., 1989, The Design and Analysis of Spatial Data Structures,
关7兴 Yang, V., Lin, N. N., and Shuen, J.-S., 1994, ‘‘Vaporization of Liquid Oxygen Addison-Wesley Publishing Company.
共LOX兲 Droplets in Supercritical Hydrogen Environments,’’ Combust. Sci. 关30兴 Cottet, G.-H., and Koumoutsakos, P., 2000, Vortex Methods: Theory and
Technol., 97, pp. 247–270. Practice, Cambridge University Press, New York.
关8兴 Protsenko, S. P., and Skripov, V. P., 1977, ‘‘Molecular-Dynamics Calculation 关31兴 Salmon, J. K., and Warren, M. S., 1993, ‘‘Skeletons From the Treecode
of Thermodynamic Properties and Structure System of Liquid Argon Nuclei,’’ Closet,’’ J. Chem. Phys., 111, pp. 136–155.
Sov. J. Low Temp. Phys., 3, No. 1, pp. 1–4. 关32兴 Rabinovich, V. A., Vasserman, A. A., Nedostup, V. I., and Veksler, L. S.,
关9兴 Rusanov, A. I., and Brodskaya, E. N., 1977, ‘‘The Molecular Dynamics Simu- 1988, Thermophysical Properties of Neon, Argon, Krypton, and Xenon, Hemi-
lation of a Small Drop,’’ J. Colloid Interface Sci., 62, No. 3, pp. 542–555. sphere.
关10兴 Powles, J. G., Fowler, R. F., and Evans, W. A. B., 1983, ‘‘A New Method for 关33兴 Reynolds, W. C., 1979, Thermodynamic Properties in SI, Stanford University.
Computing Surface Tension Using a Drop of Liquid,’’ Chem. Phys. Lett., 96, 关34兴 Hirschfelder, J. O., Curtiss, C. F., and Bird, R. B., 1967, Molecular Theory of
No. 3, pp. 289–292. Cases and Liquids, 4th ed., John Wiley & Sons, New York.
关11兴 Thompson, S. M., Gubbins, K. E., Walton, J. P. R. B., Chantry, R. A. R., and 关35兴 Mecke, M., Winkelmann, J., and Fischer, J., 1997, ‘‘Molecular Dynamics
Rowlinson, J. S., 1984, ‘‘A Molecular Dynamics Study of Liquid Drops,’’ J. Simulation of the Liquid-Vapor Interface: The Lennard-Jones Fluid,’’ J.
Chem. Phys., 81, No. 1, pp. 530–542. Chem. Phys., 107, No. 21, pp. 9264–9270.
关12兴 Yasuoka, K., Matsumoto, M., and Kataoka, Y., 1994, ‘‘Evaporation and Con- 关36兴 Margerit, J., and Sero-Guillaume, O., 1996, ‘‘Study of the Evaporation of a
densation at a Liquid Surface of Argon,’’ J. Chem. Phys., 101, No. 9, pp. Droplet in Its Stagnant Vapor by Asymptotic Matching,’’ Int. J. Heat Mass
7904–7911. Transf., 39, No. 18, pp. 3887–3898.
Monte Carlo Study of Phonon
Transport in Solid Thin Films
Including Dispersion and
Sandip Mazumder
e-mail: sm@cfdrc.com Polarization
Assoc. Mem. ASME
CFD Research Corporation, The Boltzmann Transport Equation (BTE) for phonons best describes the heat flow in
215 Wynn Drive, solid nonmetallic thin films. The BTE, in its most general form, however, is difficult to
Huntsville, AL 35805 solve analytically or even numerically using deterministic approaches. Past research has
enabled its solution by neglecting important effects such as dispersion and interactions
Arunava Majumdar between the longitudinal and transverse polarizations of phonon propagation. In this
6185 Etcheverry Hall, article, a comprehensive Monte Carlo solution technique of the BTE is presented. The
Department of Mechanical Engineering, method accounts for dual polarizations of phonon propagation, and non-linear dispersion
University of California, relationships. Scattering by various mechanisms is treated individually. Transition be-
Berkeley, CA 94720-1740 tween the two polarization branches, and creation and destruction of phonons due to
e-mail: majumdar@me.berkeley.edu scattering is taken into account. The code has been verified and evaluated by close
examination of its ability or failure to capture various regimes of phonon transport
ranging from diffusive to the ballistic limit. Validation results show close agreement with
experimental data for silicon thin films with and without doping. Simulation results show
that above 100 K, transverse acoustic phonons are the primary carriers of energy in
silicon. 关DOI: 10.1115/1.1377018兴
Keywords: Conduction, Heat Transfer, Microscale, Monte Carlo, Non-Equilibrium, Thin

Films
1 Introduction these conditions, a temperature gradient is established within the

medium, and the overall transport process is diffusion-like. This
During the past decade, aggressive scale-down in the feature
regime 共diffusion limit兲 can be described accurately using the
sizes of electronic devices has renewed interest in the physics of
Fourier Law of heat conduction. If the characteristic dimension of
energy carriers in solid crystalline materials. The central focus of the medium is significantly smaller than the mean free path of the
the device physics community, however, has been on electron phonons 共ballistic limit兲, scattering events are rare, and thermody-
transport, which is primarily responsible for the electronic perfor- namic equilibrium may not exist in the material. Here, the Fourier
mance of a material. The miniaturization of electronic devices, Law of heat conduction is invalid. The effective mean free path of
coupled with faster processing speeds, results in a very large phonons in undoped silicon is approximately 300 nm 关4兴. This is
quantity of heat being generated per unit volume in these devices. already much larger than the characteristic dimension of some of
The efficient removal of heat from such devices is a daunting task, the devices used today. This implies that the transport phenomena
and overheating is one of the most common causes of device in such devices cannot be described using diffusion
failure. Recent trends show that thermal issues are rapidly becom- approximations.
ing the central focus. In order to address this problem, it is first The Boltzmann Transport Equation 共BTE兲 is the most suitable
necessary to understand the fundamental physics of heat conduc- candidate for describing phonon transport in crystalline solids.
tion in microscale solid thin films. This is attributed to its ability to correctly describe both equilib-
Heat conduction in crystalline materials occurs by lattice vibra- rium and non-equilibrium phenomena. In the 1950s and 1960s,
tions, which produce quanta of energy called phonons. A detailed when computers were in their infancy, solid state physicists at-
discussion on the wave versus particle nature of energy transport tempted to obtain closed-form analytical solutions of the BTE for
in solids may be obtained from classical texts by Ziman 关1兴, Kittel phonon transport 关5–11兴. This, however, is possible only after a
关2兴, or from more recent ones by Tien et al. 关3兴. The phonons number of simplifying assumptions, some of which are not justi-
travel through space and engage in anharmonic interactions with fiable from a physical standpoint, and therefore, result in signifi-
one another 共phonon-phonon scattering兲, and with electrons, im- cant discrepancies with experimentally observed behavior. For ex-
purities, and geometric boundaries. The average distance a pho- ample, Ju and Goodson 关4兴 pointed out that if phonon dispersion
non travels before it interacts with another phonon is its mean free in silicon is neglected, the average mean free path of phonons is
path. The drift 共or ballistic movement兲 of phonons result in devia- only about 43 nm, as opposed to 300 nm when dispersion is
tion from thermodynamic equilibrium. Phonon-phonon scattering accounted for. With the advent of modern high-speed computers,
helps restore thermodynamic equilibrium. Other forms of scatter- it is now possible to remove these assumptions, and solve the BTE
ing do not necessarily restore thermodynamic equilibrium. If the in its most general form. Over the last decade or so, there has been
characteristic size of a solid material is much larger than the mean tremendous advancement in the development of solution tech-
free path of the resident phonons, the number of scattering events niques for the BTE for charge carriers 关12–16兴. However, limited
is large, and local thermodynamic equilibrium is restored. Under progress has been made in the area of solution of the BTE for
phonon transport. Recent works by Majumdar et al. 关17兴, Chen
and Tien 关18兴, Goodson 关19兴, and Chen 关20兴 have presented solu-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division April 24, tion strategies for both the diffusion and the ballistic limit. How-
2000; revision received January 20, 2001. Associate Editor: D. Poulikakos. ever, in such calculations dispersion was neglected, and the dual
polarizations of phonon propagation were replaced by a single summation in Eq. 共2兲 is performed, is the first Brillouin zone.
‘‘average’’ 共or ‘‘effective’’兲 polarization branch. Furthermore, Optical phonons do not contribute significantly to thermal trans-
such calculation strategies can only be adopted for considerably port 关2,5兴. They are eliminated from the summation over the vari-
simplified geometry. If all these assumptions are removed, it is ous modes for our current needs. For a large crystal, the wave
extremely difficult to solve the BTE for phonons by a determin- vector space is very dense, and can be assumed to be continuous.
istic approach in an arbitrary geometry, simply because the num- This enables replacement of the summation in Eq. 共2兲 by an inte-
ber of independent variables is too large, and would render any gral, yielding
冕冉冊
kind of discretization scheme too complex to be practical. Further-
1 dK
more, the nonlinear scattering events are difficult to incorporate
without an overall relaxation time approximation 关3兴, in which
E•V⫽ 兺 p K
具n典⫹
2
ប␻
共 2 ␲ /L 兲 3
. (3)
case, individual scattering events cannot be treated in isolation.
The alternative is to solve the BTE using stochastic or Monte The integration over wave-vector space can be transformed to an
Carlo techniques. Monte Carlo techniques have been used with integration over frequency space by invoking the dispersion rela-
success for electron transport simulations, and they have also been tion, K⫽K( ␻ ,p), and assuming that the Brillouin zone is isotro-
used for phonon transport by Klistner 关21兴 and by Peterson 关22兴. pic, yielding
Peterson’s work assumes the linear Debye theory and may be
considered useful as a starting point. Some of the ideas in this
article are borrowed from Peterson’s work 关22兴. However, Peter-
E⫽ 兺p
冕冉␻
具n典⫹
1
2冊ប␻D共 ␻ 兲d␻, (4)
son’s work is far from complete, and makes the same simplifying where D( ␻ ) is the density of state, and is expressed as 关2兴
assumptions, discussed earlier. Klistner’s work focuses on the bal-
listic limit only. It is now necessary to develop comprehensive K 2 dK
D共 ␻ 兲⫽ . (5)
solution strategies for the BTE phonon transport to study thermal 2␲2 d␻
behavior of solid state devices, and to bring phonon transport
computations at par with its electronic counterpart. This is the first The quantity D( ␻ )d ␻ represents the number of vibrational states
attempt at a comprehensive Monte Carlo solution technique for between ␻ and ␻ ⫹d ␻ . The second term within parenthesis in Eq.
the solution of the BTE for phonons. The objective is to present a 共4兲 is independent of temperature, and can be summed up to yield
solution technique and algorithm, which removes most assump- the so-called zero-point energy. Finally, the group velocity of
tions that have been made in the past, and is sufficiently general to phonons is defined as 关2兴:
be adopted for future modeling work in this area. It is demon- Vg ⫽ⵜ k␻ . (6)
strated here that modeling of this nature can help extract critical
information pertaining to the physics of phonon transport. The above six equations completely describe the state of a pho-
non, and are sufficient for the realization of the numerical scheme
to be discussed in Section 3.
2 Theory
2.2 BTE for Phonons. The Boltzmann Transport Equation
Although, the following material may be found in any funda- is a transport equation describing the transport of an ensemble of
mental text on solid state physics, it is necessary to briefly discuss particles 共or stochastic samples兲. In the case of phonons, it as-
the underlying theory of phonons and the BTE of phonons, prior sumes the following form 关3兴:
冋册
to discussion of numerical techniques pertaining to the solution of
⳵f ⳵f
兺⬘ 关 ⌽ 共 K,K⬘ 兲 f 共 K⬘ 兲 ⫺⌽ 共 K⬘ ,K兲 f 共 K兲兴 ,
the BTE for phonons.
⫹Vg •ⵜ f ⫽ ⫽
2.1 Lattice Vibrations and Phonons. A crystal lattice can ⳵t ⳵t scat K
be visualized as a complex three-dimensional array of masses (7)
共representing individual atoms兲 and springs 共representing atomic where f (t,r,K) is the distribution function of an ensemble of
bonds兲. The deviation from the equilibrium position of an indi- phonons, and r is the position vector describing the position of an
vidual atom within the lattice result in vibrational waves, which individual stochastic particle. The left-hand side of Eq. 共7兲 repre-
propagate through the crystal and carry energy. The vibrational sents drift, and causes departure from equilibrium, while the scat-
energy of the atoms is quantized, and each quantum is called a tering term on the right-hand side of Eq. 共7兲 restores equilibrium.
phonon. A phonon is an ensemble of travelling waves having a The distribution function is, in general, a function of seven inde-
certain frequency, ␯, a certain group velocity, Vg , and carrying pendent variables, namely, time, three space variables, and three
with it energy equal to h ␯ , where h is the Planck constant. wave-vector variables. ⌽共K,K⬘兲 is the scattering rate from state
Under thermodynamic equilibrium, the temperature of a me- K⬘ to K, and is usually a nonlinear function of K. These com-
dium is well defined. At a temperature, T, the equilibrium phonon plexities render the solution of the BTE extremely difficult by
occupation number, 具 n 典 , is given by the Bose-Einstein distribu- deterministic means.
tion 关2兴
3 Monte Carlo Solution Technique
1
冋册
具n典⫽ , (1) In a Monte Carlo solution technique, samples are drawn from
ប␻ the six individual stochastic spaces. These include the three wave-
exp ⫺1
k BT vector components and the three position vector components. The
samples 共phonons, in this case兲 are then allowed to drift 共or unre-
where ␻ is the angular frequency 共radians/s兲 of the phonon strained motion兲 and scatter in time, and their statistics is col-
共⫽2␲␯兲, ប⫽h/2␲ , and k B is the Boltzmann constant. The total lected at various points in time and space, and processed to extract
vibrational energy in a material of volume V⫽L 3 is written as 关2兴 the necessary information.
兺兺冉具n典⫹ 2 冊 ប␻,
1 3.1 Initial Conditions.
E•V⫽ (2)
p K
Number of Phonons. The first step is to initialize the states of
where E is the energy per unit volume of the material. The sum- the ensemble of phonons that are to be used for the Monte Carlo
mations in Eq. 共2兲 have been performed over all polarizations of simulation. Prior to initialization, it is first necessary to determine
the various modes of phonon propagation 共longitudinal optical, how many phonons need to be initialized within each control vol-
transverse optical, longitudinal acoustic and transverse acoustic兲, ume. The number of phonons per unit volume, and polarization p,
p, and wave-vector, K. The wave-vector space over which the excluding the zero-point number, is given by 关from Eq. 共4兲兴
Nb
N共 p 兲⫽ 兺具n共 ␻
i⫽1
0,i ,p 兲典 D 共 ␻ 0,i ,p 兲 ⌬ ␻ i , (8)
where the integral over frequency space has been replaced by a

discrete summation. N b is the number of spectral intervals or
bands into which the frequency space is broken into, and ⌬␻ is
the spectral band width. The central frequency of the ith spectral
band is ␻ 0,i . The number of phonons, resulting from Eq. 共8兲, is
usually an extremely large number 共for example, at 80 K, silicon
has approximately 4.5⫻1022 phonons/m3 兲. In practice, it is not
feasible to simulate that many phonons, and therefore, a scaling
factor is required. First a decision is made as to how many
phonons will be traced 共an input to the code兲. Let this number be Fig. 1 Pictorial representation of phonon initial position, P , in
N prescribed . The code then calculates a scaling factor from the „a… two-dimensional triangular element, and „b… three-
expression dimensional tetrahedral element
N actual
W⫽ , (9)
N prescribed
pose. At low temperatures, where the frequency distribution is
where N actual is computed using Eq. 共8兲 after summing over all highly nonlinear, it may be worthwhile to have more than 1000
polarizations. This scaling factor, W, is then stored permanently, spectral intervals. However, this will increase computational cost,
and used for the remainder of the computational process. Each and will result in poor statistical sampling within some spectral
stochastic sample used during the simulation actually represents intervals. Furthermore, as our computations will show, 1000 in-
an ensemble of W phonons. During the simulation, of course, the tervals, even at low temperatures, result in less than 2 percent
total number of samples within the computational domain may errors in solutions.
change depending on the boundary conditions. The number of phonons per unit volume in the ith spectral
Position. The numerical scheme to initialize the position of a interval is given by
phonon depends on the type of control volume cells used for N i ⫽ 具 n 共 ␻ 0,i ,LA 兲典 D 共 ␻ 0,i ,LA 兲 ⌬ ␻ i
spatial sampling of the phonons. For example, if all control vol-
umes are rectangular orthogonal hexahedra, the initialization con- ⫹2 具 n 共 ␻ 0,i ,TA 兲典 D 共 ␻ 0,i ,TA 兲 ⌬ ␻ i , (11)
sists of drawing three random numbers to assign a random set of where the equilibrium distribution is now evaluated at the central
coordinates (x,y,z) within the hexahedron. This may well be suf- frequency of the spectral interval in question. In order to deter-
ficient for the simulation of simple geometries. However, since mine the frequency of a phonon, the normalized cumulative num-
our goal is to develop a solution technique, which is to be used ber density function is first constructed
eventually for the thermal analysis of entire semiconductor de-
vices, it is necessary to adopt a more general strategy. In general, 兺 ik⫽1 N k
a spatial bin 共or cell兲 may be an arbitrary polyhedron in three- F i⫽ N . (12)
兺 k⫽1
b
Nk
dimensional and arbitrary polygon in two-dimensional. It is al-
ways possible to dissect this cell into its smallest element, which A random number, R, is then drawn. If F i⫺1 ⬍R⬍F i , then the
is a tetrahedron in three-dimensional and triangle in two- phonon belongs to the ith spectral interval. A blind search to
dimensional. Details of such procedure are beyond the scope of locate i would be far too expensive. In the current scheme, the
this article, and may be obtained from references on unstructured search is conducted using a bisection algorithm, which requires
grid generation 关23兴. The position vector, rP , of any point P, the execution of at most 11 such checks. Having determined the
within a triangle 共in two-dimensional兲 or tetrahedron 共in 3D兲 is spectral interval, the actual phonon frequency is then given by
expressed as
⌬␻i
rP ⫽ 再 rA ⫹R 1 a⫹R 2 b
rA ⫹R 1 a⫹R 2 b⫹R 3 c
2D
3D
, (10)
␻ ⫽ ␻ 0,i ⫹ 共 2R⫺1 兲
2
where R is the same random number used to select the spectral
, (13)
where R 1 , R 2 , and R 3 are random numbers between zero and interval from Eq. 共12兲. A random number relation has been em-
unity, and rA is the position vector of point A. In addition, R 1 and ployed to distribute the frequencies linearly between ␻ 0,i ⫺⌬ ␻ /2
R 2 must obey the inequality: R 1 ⫹R 2 ⬍1. The vectors a, b, and c and ␻ 0,i ⫹⌬ ␻ /2. The assumption here is that the phonon fre-
are illustrated in Fig. 1. quency distribution is linear between ␻ 0,i ⫺⌬ ␻ /2 and ␻ 0,i
⫹⌬ ␻ /2. The advantage of this approach is that instead of having
Frequency. As a first step, the dispersion relationship of the discrete frequencies of ␻ 0,i , we now have an almost continuous
material is provided to the code in the form of a table. From this frequency space.
data, the maximum cut-off frequencies of the longitudinal acous-
tic 共LA兲 and transverse acoustic 共TA兲 branches are determined. Polarization. Any phonon, initialized within the computa-
They are denoted by ␻ max,LA and ␻ max,TA , respectively. The fre- tional domain has to be assigned to either a LA branch or a TA
quency space between zero and ␻ max,LA is then discretized into branch. To determine which branch the phonon will belong to,
spectral intervals, ⌬ ␻ i ⫽ ␻ max,LA /Nb , where N b is the number of first, the Bose–Einstein distribution is utilized to calculate the
spectral intervals. It was determined by trial-and-error studies, that probability of LA to TA phonons at the ith spectral interval 共or
for most cases, 1000 equal-sized spectral intervals serves the pur- frequency兲
N i 共 LA 兲具 n 共 ␻ 0,i ,LA 兲典 D 共 ␻ 0,i ,LA 兲

P i 共 LA/TA 兲 ⫽ ⫽ , (14)
N i 共 TA 兲 ⫹N i 共 LA 兲 2 具 n 共 ␻ 0,i ,TA 兲典 D 共 ␻ 0,i ,TA 兲 ⫹ 具 n 共 ␻ 0,i ,LA 兲典 D 共 ␻ 0,i ,LA 兲
where the equilibrium number of phonons is computed using the frequency distribution being governed by Bose-Einstein statistics.
thermodynamic temperature of the cell 共or spatial bin兲 in which The state of emitted phonons is completely independent of the
the phonon is to be initialized, and i is the spectral interval to state of incident phonons. In the numerical scheme, any phonon
which the phonon belongs 共determined by the procedure outlined incident on an isothermal wall is tallied, and then deleted from the
in the previous sub-section兲. Next, a random number between zero simulation. This provides the net incident energy during a time
and unity is drawn. If the random number is less than interval ⌬t. The number of phonons emitted from a boundary face
P i (LA/TA), then the phonon belongs to the LA branch. Other- of area, A, and temperature, T, is given by
冉兺兺
wise, it belongs to a TA branch.
Nb
Wave Vector. Once the phonon frequency has been sampled, N face⫽A•⌬t• 具 n 共 ␻ 0,i ,p 兲典具 Vg 共 ␻ 0,i 兲 •n̂ 典
the wave-vector can be determined directly using the dispersion p i⫽1
冊
relationship. In general, given a frequency, ␻, an interpolation can
be performed within the three-dimensional dispersion table to find
K x , K y , and K z . In most cases, however, dispersion relationships ⫻D 共 ␻ 0,i ,p 兲 ⌬ ␻ i , (17)
are available only in the 共100兲 and 共111兲 directions, and it is,
perhaps, best to assume isotropy, as a first step. If we assume
where Vg ( ␻ 0,i )•n̂ represents the phonon velocity normal to the
isotropy, then K x is first determined 共and this is equal to K y and
face 共Fig. 2兲. In general, phonons will be emitted from a face in all
K z 兲. It then follows that the magnitude of the wave-vector 兩 K兩
directions 共over a solid angle 2␲兲. The quantity Vg ( ␻ 0,i )•n̂, there-
⫽ 冑3K x in three-dimensional, and 兩 K兩 ⫽ 冑2K x in two-dimensional
fore, has to be directionally averaged 共as indicated by the angled
cases. The direction of the wave-vector is expressed as
冉冊
brackets in Eq. 共17兲兲 in order to obtain the average normal veloc-
sin ␪ •cos ␺ ity to the boundary face. In two-dimensional, for example, Eq.
ŝ⫽ sin ␪ •sin ␺ , (15) 共17兲, after appropriate directional averaging, reduces to
冉兺兺冊
cos ␪ Nb
A
where ␺ ⫽2 ␲ R 2 , cos ␪⫽2R1⫺1, and K⫽兩K兩ŝ, R 1 and R 2 being N face⫽ •⌬t• 具 n 共 ␻ 0,i ,p 兲典 Vg 共 ␻ 0,i 兲 D 共 ␻ 0,i ,p 兲 ⌬ ␻ i ,
random numbers between zero and unity. ␪ and ␺ are polar and ␲ p i⫽1
azimuthal angles, respectively, and are shown in Fig. 2. In two- (18)
dimensional cases, only one random number is necessary, and the
direction vector is given by where N face is the actual number of phonons that should be emit-
冉冊
ted into the computational domain from the boundary face. The
cos ␺ representative number emitted in the simulation would be
ŝ⫽ . (16)
sin ␺ N face /W.
The emission location of a phonon is determined by drawing
Once the initialization is complete, the next task is to provide random numbers, and distributing the phonons uniformly over the
boundary conditions for the simulation scheme. entire face. In three-dimensional, a basic surface element is a tri-
3.2 Boundary Conditions. Since the primary focus of this angle, and the procedure described earlier for volumetric two-
article is heat transport, various kinds of thermal boundary condi- dimensional elements can be used for three-dimensional surface
tions will be discussed in this section. The two most common elements. The frequency of the emitted phonon, the magnitude of
types are isothermal and prescribed heat flux boundary conditions, the wave-vector, and the polarization of the phonon can be deter-
of which a prescribed heat flux of zero results in a special kind of mined by following a procedure similar to the one discussed in
boundary, namely an adiabatic boundary. In this article, only the Section 3.1. The difference is that the volumetric number density
treatment of isothermal and adiabatic boundaries is discussed. The has to be replaced everywhere by face number flux. The direction
treatment of conjugate walls 共material interfaces兲 is required for of the wave-vector depends on the surface normal of the surface
the study of superlattices, and may be found elsewhere 关20,24,25兴. in question, and is expressed as follows:
An isothermal wall is analogous to a black wall for photon
radiation. All phonons that strike it are thermalized 共analogous to ŝ⫽sin ␪ cos ␺ t̂1⫹sin ␪ sin ␺ t̂2⫹cos ␪ n̂, (19)
absorption by a blackbody兲. Depending on its temperature, the
boundary also emits phonons into the computational domain, the where n̂ is the unit surface normal, and t̂1 and t̂2 are unit surface
tangents 共Fig. 2兲, which must be perpendicular to each other. The
two surface tangents and the surface normal must follow the right-
handed system. First, the surface normal is computed using
Gauss-divergence theorem. Next, one of the surface tangents is
computed using the coordinates of the vertices of the triangular
face in question. The second tangent is computed from the rela-
tionship: t̂1⫻t̂2⫽n̂. This ensures that both the above criteria are
satisfied.
An adiabatic wall acts as a perfect reflector for the phonons.
Most surfaces, irrespective of their polish, act as partially specular
reflectors. An excellent discussion on surface reflection character-
istics and their impact on boundary scattering of phonons may be
found in Klistner 关21兴. In the current numerical scheme, the de-
gree of specularity, d, can be prescribed. The value of d may vary
between zero 共completely diffuse reflection兲 and unity 共com-
pletely specular兲. The former scenario results in intense boundary
scattering, while the latter poses no resistance to the flow of
phonons. The preceding discussion clarifies the point that in the
current numerical procedure, boundary scattering is not treated
indirectly using so-called geometric factors 关5,7,8兴 and boundary
Fig. 2 Coordinate system showing the direction of phonon scattering time-scales, but rather, by actual interaction of a tra-
emission and the associated angles versing phonon with the true geometric boundary. When a phonon
strikes an adiabatic boundary, a random number is first drawn. If where G is the reciprocal lattice vector. During any three-phonon
this random number is less than the prescribed value of d, the process, energy and momentum must both be satisfied simulta-
phonon is specularly reflected using the relation neously. This disallows certain types of transitions between the
two polarization branches. It can be shown mathematically 关5,7兴
ŝr ⫽ŝi ⫹2 兩 ŝi •n̂兩 n̂, (20) that the only allowable ones are LA⇔LA⫹TA and LA⇔TA
⫹TA. In principle, it is possible to incorporate three-phonon scat-
where ŝi and ŝr are the direction vectors of the incident and re-
tering by treating each phonon as a potential candidate for scat-
flected phonon, respectively. If this number is larger than d, the
tering and allowing it to interact with every other phonon in its
phonon is reflected diffusely at the surface using Eq. 共19兲.
vicinity. The frequency and wave-vector combinations of the in-
3.3 Scattering. Phonons move inside a crystalline material teracting phonons would then automatically decide, subject to sat-
with velocity Vg and engage in scattering as they move. Scattering isfaction of Eq. 共23兲, if an interaction between them is at all pos-
is caused by lattice boundaries 共intrinsic scattering兲, defects or sible. This is the most fundamental approach to computing a
dislocations, charge carriers such as electrons, and by external phonon’s lifetime. However, in a Monte Carlo scheme, since only
boundaries or material interfaces. Electron-phonon scattering may limited number of samples are traced, the ␻ and K spaces are
alter the thermal conductivity of a material under certain situa- generally far from being continuous to the degree that all possible
tions, but this effect is not well understood. For the current study, triadic interactions can be accounted for. Secondly, such compu-
such electron-phonon or phonon-free-carrier scattering is ne- tations would be immensely expensive, and not acceptable for
glected. All other types of scattering can potentially affect the practical problems. The problem is circumvented using perturba-
thermal conductivity, and, therefore, heat transport in a crystalline tion analysis. The equilibrium distribution is first perturbed lin-
material. In the current numerical scheme, the events occurring early, and all triadic interactions are considered at a given tem-
within a time-step ⌬t is split into two distinct sequential phases, perature, allowing the system to come back to equilibrium. By this
namely drift1 and scattering. During the drift phase, phonons way, phonon lifetimes can be computed. Discussion on such tech-
move linearly from one location to another. Their positions are niques may be obtained from early work by Klemens 关5兴. Subse-
tracked using an explicit first-order time integration quent analyses, including polarization and normal scattering ef-
fects, were performed by Holland 关7,8兴 for a variety of important
r共 t⫹⌬t 兲 ⫽r共 t 兲 ⫹Vg •⌬t. (21) semiconductor materials. The final outcome of such analysis is
expressions for phonon lifetimes, for the two polarization
Following the drift phase, phonons are tallied within each spatial
branches, as a function of frequency and temperature.
bin, and the energy of each spatial bin is computed and stored.
Once the scattering time-scales for the two polarization
This will, henceforth, be referred to as the target energy of the
branches are known as a function of frequency and temperature,
cell, and will be denoted by E T . At this juncture, the pseudo-
the three-phonon scattering processes can be implemented in the
temperature, T̃, of each spatial bin is computed by numerical in- following manner. First, based on the time-scale of scattering and
version of Eq. 共4兲: the time-step used, a probability of scattering is computed: P NU
Nb ⫽1⫺exp(⫺⌬t/␶NU) 关22兴. The time-scale, ␶ NU , is an effective
E TW ប ␻ 0,i D 共 ␻ 0,i ,p 兲 ⌬ ␻ i
V
⫽ 兺兺p i⫽1 exp关 ប ␻ 0,i /k B T̃ 兴 ⫺1
. (22) time-scale which accounts for both normal and Umklapp pro-
cesses, i.e., ␶ ⫺1 ⫺1 ⫺1
NU ⫽ ␶ U ⫹ ␶ N , where ␶ U and ␶ N are time-scales for
Umklapp and normal processes, respectively. Next, a random
The pseudo-temperature is an artificial temperature that may be number is drawn and compared with P NU . If the random number
computed for every spatial bin assuming that thermodynamic is less than P NU , the phonon is scattered. Scattering is imple-
equilibrium exists in that bin. It is necessary to compute the prob- mented by resetting the phonon frequency, wave-vector, and po-
ability of LA to TA phonons within each spectral bin. For this, Eq. larization. The frequency is sampled from the equilibrium distri-
共14兲 is used, in which the thermodynamic temperature is replaced
bution computed at T̃. Once the frequency is known, the wave-
by the pseudo-temperature. Although it is not clear at this point
vector can be determined using the procedure outlined in Section
why these quantities should be computed, it will become clear
3.1. The polarization is determined using the LA/TA probability
once the scattering algorithm is discussed. During the second
phase of the scheme, phonons within a spatial bin are allowed to based on T̃, which was computed and stored after the drift phase.
engage in three-phonon scattering processes, and are also scat- Resetting the state of the phonon essentially thermalizes the pho-
tered by impurities when they are present. non and helps restore equilibrium, which is the sole function of
three-phonon scattering processes. When the frequency of a cer-
Three-Phonon Scattering (Normal and Umklapp Processes). tain phonon is changed, energy conservation is violated. After all
Phonons engage in three-phonon elastic interactions 关2兴 of two scattering processes have been completed, the actual energy of
types: Normal scattering and Umklapp scattering. The latter di- each spatial bin is again tallied. This actual energy, E A , is likely
rectly poses resistance to energy transport, while the former indi- to be different from the cell’s target energy E T , due to violation
rectly affects the transport process by altering the frequency dis- of energy conservation. A creation/destruction scheme is next
tribution of the resident phonons. An immense volume of implemented. In this scheme, phonons are continuously added
literature is available, which studies the details of three-phonon and/or deleted from a cell until the actual energy of the cell
interactions of all types in a variety of different materials. At the matches the target energy of the cell within a certain tolerance.
heart of three-phonon interactions lie the conservation laws of Investigations revealed that a tolerance level equal to ប ␻ max,TA is
energy and momentum: adequate. The added or deleted phonons are drawn from the equi-
librium distribution, so that the overall effect of the phonon
␻ 1⫹ ␻ 2↔ ␻ 3 共 Energy Conservation; All兲 creation/destruction process is also to restore thermodynamic
equilibrium. Depending on the phonon lifetimes, during a given
K1 ⫹K2 ↔K3 共 Momentum Conservation; Normal兲 (23) time period, all phonons within the computational domain may
not scatter. At low temperature, for example, very few of them
K1 ⫹K2 ↔K3 ⫹G will scatter. Thus, although the scattering algorithm may restore
共 Momentum Conservation; Umklapp兲 , local thermodynamic equilibrium, this may only be attained par-
tially, thereby retaining the non-equilibrium nature of transport.
1
‘‘Drift’’, as described here, is not to be confused with motion under the influence
This scattering scheme captures all of the essential physics of
of an external force, such as an electric field. Here, ‘‘drift’’ implies unrestrained three-phonon scattering processes. It allows transition between
ballistic motion from point A to point B. polarization branches, conserves energy, and helps restore ther-
modynamic equilibrium. Momentum conservation is violated, as profile is one that may be obtained in the case of radiative equi-
would be by Umklapp scattering. The shortcoming of the scheme librium 关17兴. The second case would result in a linear temperature
lies in that it cannot treat normal and Umklapp scattering in iso- profile, as would be predicted by the Fourier law of heat
lation. The authors are currently in the process of investigating conduction 关3兴.
procedures where this may be possible. Simulations were performed in a simple one-dimensional gal-
lium arsenide film of thickness L⫽400 nm. The left end of the
Scattering by Impurities. Scattering by impurities, defects and film was maintained at T L ⫽20 K, while the right end was main-
dislocations are treated in the Monte Carlo scheme in isolation tained at T R ⫽10 K, and symmetry boundary conditions at the top
from normal and Umklapp scattering. This is difficult in determin- and bottom surfaces. The Debye temperature of GaAs is approxi-
istic approaches. The time-scale for scattering due to impurities, mately 345 K 关8兴. It is well known, that at temperatures well
␶ i , is used to compute a probability of scattering, and a selection below the Debye temperature, the heat conduction is governed by
based on a new random number is performed, as discussed earlier. the Stefan-Boltzmann law 关3兴. Under this condition, the tempera-
In this case, however, the frequency of the scattered phonon is left
ture of the entire film would assume a constant value of 关 (T L4
unaltered 共implying that the magnitude of the wave-vector is also
left unaltered兲. The direction of the wave-vector is changed, the ⫹T 4R )/2兴 1/4 关17兴, with discontinuities at the two boundaries. A
new direction vector being described by Eq. 共15兲 or 共16兲. The comparison of the numerical results against analytical results is
time-scale for scattering by impurities is expressed using a simple shown in Fig. 3. The maximum error between the two solutions is
model proposed by Vincenti and Kruger 关26兴 approximately 2 percent. At low temperature, the equilibrium dis-
tribution of phonons 共Eq. 共1兲兲 is strongly nonlinear 关2兴. The small
␶ ⫺1
i ⫽ ␣ ␴␳ 兩 Vg 兩 , (24) error in solution is actually caused by the fact that in the numeri-
cal scheme, this non-linear distribution is approximated by a
where ␣ is a constant of the order of unity, ␳ is the defect density
piecewise linear function of N b ⫽1000 segments. The accuracy of
per unit volume, and ␴ is the scattering cross-section expressed as
temperature profile in the ballistic limit proves that the numerical
关17兴
treatment of isothermal boundaries, described earlier, is accurate.
␴⫽␲r2 冉冊 ␹4
␹ 4 ⫹1
, (25)
Verification of the code in the diffusion limit was performed
both for one-dimensional and two-dimensional films, since ana-
lytical solutions are obtainable in both cases. For the one-
where r is the atomic radius of the impurity, and ␹ ⫽r 兩 K兩 . In the dimensional case, a simulation were performed in a gallium ars-
current study, it is assumed that the main effect of an impurity enide film with T R ⫽300 K and T L ⫽500 K. At these temperatures,
atom is point scattering. Impurities may alter hole concentrations intrinsic scattering is very strong, and transport is in the diffusion
in the material, resulting in additional scattering, but such effects limit. It was estimated that the average mean free path of the
are neglected, primarily due to the lack of understanding of the resident phonons is approximately 0.05.L. The dispersion rela-
physics underlying such scattering. tionship for GaAs was used from Waugh and Dolling 关27兴. The
The scattering procedure, outlined above, employs a relaxation scattering time-scale expressions were adopted from Bhandari and
time approximation for the various scattering events. Although Verma 关9兴. Impurity and boundary scattering effects were ne-
this is not as accurate as using the selection rules directly to com- glected. The simulation was performed with just 20,000 samples.
pute the nonlinear scattering rates, the Monte Carlo procedure The steady-state temperature profiles are depicted in Fig. 4, which
does allow treatment of each scattering event in isolation, and this exhibits a good match. The noisy nature of the numerical profile is
is very important when accounting for the various regimes of due to statistical errors, to be discussed in detail in later sections.
transport. The physics of various types of scattering are quite The two-dimensional simulations were performed in a 400nm ⫻
different. For example, intrinsic scattering results in frequency 400nm square geometry with four different temperatures on the
and polarization change, while defect scattering changes the di- four different walls. The material was GaAs, and 100,000 samples
rection of the propagating phonon only. Therefore, it would be were used for initialization, and 100 spatial bin were used. The
grossly incorrect to treat all scattering processes in a similar man- results are illustrated in Figs. 5共a兲–共c兲, where the quantity, ‘‘Er-
ner by employing an overall relaxation time, computed using the ror_T’’, in 5共c兲 is the difference in temperature between analytical
Mathiessen rule. This is a clear advantage of using the Monte
Carlo approach over deterministic approaches used in almost all
past studies.
This completes the description of the models and the numerical
scheme underlying solution of the BTE for phonons. The com-
mercial engineering analysis software, CFD-ACE⫹, provided the
framework on which the numerical scheme was developed. In
choosing a time-step for simulation of phonon transport, it must
be ascertained that the chosen time-step is smaller than the small-
est scattering time-scale. Furthermore, the time-step should be
small enough so that the fastest traveling phonon does not travel a
distance, which exceeds the characteristic spatial bin size.
4 Results and Discussion

The code was first tested to ensure that it reproduces the correct
limiting behavior and trends. Following such verification, it was
used to compute the thermal conductivity of pure and doped sili-
con films between the temperature of 10–300 K.
4.1 Verification of Numerical Scheme. As discussed ear-
lier, there exist two well-defined limiting scenarios: 共1兲 lⰇL 共bal-
listic limit兲; and 共2兲 lⰆL 共diffusion limit兲, where l is the average
mean free path of the phonons, and L is the characteristic film Fig. 3 Comparison of temperature profile obtained by Monte
thickness. The first case results in ballistic transfer of phonons Carlo solution of the BTE for phonons with analytical results in
between the boundaries of the film, and the resulting temperature the ballistic limit
Fig. 4 Comparison of temperature profile obtained by Monte
Carlo solution of the BTE for phonons with analytical results in
the diffusion limit Fig. 6 Temperature profiles in a one-dimensional GaAs film
for various degrees of specularity „d… of the film boundaries
and Monte Carlo solutions. It is seen that the maximum error is

about 20 K, which corresponds to about 6.7 percent. The maxi- purities were deactivated. The degree of specularity at the top and
mum error regions, as expected, are clustered around the junction bottom boundaries was set to unity, so that the boundaries do not
of the 600 K and 300 K walls, where the thermal gradients are pose any resistance to the phonons. The results are shown in Fig.
largest. These results lend validity to the numerical implementa- 7. It is seen that as the defect density is increased, the temperature
tion of intrinsic scattering, described in Section 3.3. drop across the film increases, as would be expected.
Following the verification of the two limits, the boundary and The above exercises lend credibility and help gain confidence
impurity scattering mechanisms were tested individually to ascer- in the various models and their numerical implementation. Fol-
tain that they reproduce correct behavior, at least qualitatively. lowing these exercises, the code was used to perform thermal
First, simulations were performed in GaAs films with cross-plane conductivity studies in silicon thin films.
共left to right兲 thickness of 1 ␮m and in-plane 共bottom to top兲
thickness of 100 nm. The left boundary was set to 20 K and the 4.2 Validation Study for Silicon Thin Films. Silicon is
right boundary was set to 10 K. The top and bottom boundaries probably the most important electronic material to date. Several
represent adiabatic scattering walls. The degree of specularity of researchers have performed studies on silicon in the past, and as a
these surfaces was varied from zero to unity. Simulation results result its dispersion characteristics are well known. Furthermore,
are shown in Fig. 6. All other forms of scattering were deacti- Holland 关7兴 has provided comprehensive guidelines pertaining to
vated. As the degree of specularity is decreased, boundary scat- phonon lifetimes in silicon. Although, a large number of thermal
tering becomes increasingly strong, and a temperature gradient is conductivity measurements of silicon has been made in the past,
established across the film. In the absence of intrinsic scattering, most of these data are for thick samples, where transport is in the
these temperature profiles are not likely to be linear 关21兴, as seen diffusion limit, and important effects such as the effect of tem-
in Fig. 6. perature and boundary scattering become mute. For study of such
Another set of simulations was performed for the same film, important effects, it is important to use thin-film data. The data
where all other forms of scattering, other than scattering by im- used here for validation was collected recently by Asheghi 关28兴,
Fig. 5 Comparison of temperature profile obtained by Monte Carlo solution of the BTE for phonons with analytical
results in the diffusion limit for a two-dimensional geometry: „a… analytical solution, „b… Monte Carlo solution of BTE, and
„c… difference in „a… and „b…. The left wall was set to 600 K, the top wall to 500 K, the right wall to 400 K, and the bottom
wall to 300 K.
Fig. 9 Thermal conductivity across a 3 ␮m pure silicon film:
Fig. 7 Temperature profiles in a GaAs film for various levels of numerical predictions with and without boundary scattering
impurity „r… compared to experimental data †28‡
and to our knowledge, is the most comprehensive and reliable example, if a simulation is performed between 10 K and 30 K, the
data available. The dispersion relationship for silicon in the 共001兲 average film temperature is about 25.4 K,2 and consequently, it is
direction is available from early studies by Brockhouse 关29兴, and assumed that the thermal conductivity computed by this simula-
the acoustic branches have been reproduced and shown in Fig. 8, tion is the thermal conductivity at 25.4 K. The simulations were
for the sake of completeness. allowed to proceed until steady state; following this, a time aver-
The relaxation times for the various three-phonon scattering aging over 10 time-steps was performed to extract the final results
processes were adopted from Holland 关7兴 of the computation. Soon after the dispersion relationship of the
␶ ⫺1
NU ⫽B L ␻ T
2 3
共 LA,Normal⫹Umklapp兲 material is read, the code computes the highest and lowest veloc-
ity that any resident phonon may have. Theoretically, the lowest
␶ ⫺1
N ⫽B TN ␻ T
4
共 TA,Normal兲 (26) velocity is zero 共at the edge of the Brillouin zone兲. However, due
再
to discretization of the K-space, the lowest velocity is non-zero in
0 共 TA,Umklapp for ␻ ⬍ ␻ 1/2兲 the numerical treatment. From this lowest velocity, and the char-
␶ ⫺1
U ⫽
B TU ␻ 2 /sinh 冉冊
ប␻
k BT
共 TA,Umklapp for ␻ ⬎ ␻ 1/2兲
, acteristic film dimension, the time required to reach steady state
can be estimated. For the current study, all simulations were per-
formed up to twice this estimated steady-state time to ensure that
where ␻ 1/2 is the frequency corresponding to K/K max⫽0.5. For the solutions are indeed steady state solutions. Another indication
silicon, ␻ 1/2⫽2.417⫻1013 radians/s 共Fig. 8兲. The constants B L , of steady state is that the total number of phonons 共i.e., the total
B TN , and B TU can be obtained directly from Holland 关7兴. A series energy兲 in the computational domain becomes constant, barring
of simulations were performed for a 3 ␮m thick silicon film at statistical fluctuations.
different temperatures all the way from 10 K to 300 K. The dif- Figure 9 shows the results of the simulations. At low tempera-
ference in temperature between the two boundaries of the film was tures, the mean free path of the phonons due to intrinsic lattice
always maintained at 20 K and it is assumed that the computed scattering is much larger than the film dimension, and the trans-
thermal conductivity from the simulation is the representative port is ballistic. The thermal conductivity, in such cases, is arbi-
thermal conductivity at the average temperature of the film. For trarily large. Heat conduction, in the low temperature regime, is
purely dominated by boundary scattering. The phonons impinge
on the film boundaries directly and scatter. When boundary scat-
tering was introduced, it was found the thermal conductivity drops
drastically. Of course, the magnitude is determined by the com-
bined effect of the degree of specularity of the scattering surfaces
and the distance by which they are apart. For the current study,
neither the distance between the scattering planes, nor the degree
of specularity of the scattering planes, was known. First, the dis-
tance between the planes was fixed at 10 ␮m. Next, the degree of
specularity was calibrated to match the experimental thermal con-
ductivity value at 25 K. The calibrated value for d was found to be
0.6. This value was finally used for the remaining simulations at
all other temperatures. At higher temperatures, Umklapp scatter-
ing is dominant, and the effect of boundary scattering is negli-
gible. It is clear from Fig. 9 that the match between experimental
and numerical result is excellent. It must be emphasized that the
results have been plotted on a linear scale 共as opposed to all pre-
vious research, where differences between model predictions and
Fig. 8 Dispersion relation „only acoustic branches shown… of 2

At this temperature, transport is still in the ballistic limit, and the temperature of
silicon in the „001… direction „adopted from Brockhouse †29‡… the medium is the average of the fourth power of the boundary temperatures.
Table 2 Statistical errors associated with thermal conductivity
calculations
framework of a Monte Carlo simulation, the role of the LA and

TA phonons can be easily isolated by sampling their contributions
Fig. 10 Thermal conductivity across a 3 ␮m boron-doped
to the overall heat flux at each boundary, separately. The results
„1025 atomsÕm3 … silicon film: numerical predictions compared
to experimental data †28‡. The same plot is plotted on a loga- are shown in Table 1, for the same silicon film described earlier.
rithmic scale and shown in the inset. From Table 1, it is clear that at low temperature, almost all of the
LA phonons that are emitted from the hot boundary of the film
reach the cold boundary. As the temperature of the film is in-
creased, although the number of LA phonons increase, their life-
experimental data are hardly noticeable because they are always times are much shorter, and a large fraction of them transition to
plotted on a logarithmic scale兲 to highlight the fact that the match TA phonons, indicating that LA⇒TA⫹TA is probably the most
is indeed excellent. dominant three-phonon process. Above 100 K, the TA phonons
The task of predicting thermal conductivity of doped silicon are the primary energy carriers. The results shown here corrobo-
films was undertaken next. Asheghi 关28兴 has measured thermal rate remarks by Bhandari and Verma 关9兴 who claimed that at high
conductivity of boron-doped silicon films at three different doping temperature transport in GaAs is dominated by TA phonons.
levels of 1023, 1024, and 1025 atoms/m3 . The constant, ␣, in Eq.
共24兲 was first calibrated using the data for 1024 atoms/m3 . The 4.3 Numerical Error Estimates. The results, presented in
calibration was done by performing a least square fit of the data at the preceding sections, have statistical errors associated with
all temperatures, and yielded a value of ␣⫽7.7. This value was them. The errors depend primarily on the number of stochastic
then used to compute the thermal conductivity for a doping level samples used in the simulation 共i.e., N prescribed兲, and the number of
of 1025 atoms/m3 . The results are shown in Fig. 10. Some discrep- scattering events that occur. To estimate the magnitude of the
ancy is observed between predicted results and experimental data. statistical errors, simulations were performed at each temperature
This may be attributed to the simplistic model used for the relax- with ten different random number seeds. Computations were per-
ation time for impurity scattering. formed only at seven selected temperatures to keep the total com-
One critical issue, which has remained unresolved over the past putational time within reasonable limits. Depending on the num-
several decades, is the relative role of LA and TA phonons in ber of scattering events, each simulation takes between two to
energy transport at various temperatures. Past studies 关9,30兴 have three hours on a DEC Alpha 500 MHz workstation. Following
tried to isolate the role of the two branches, but no firm conclusion simulations at each temperature, the raw data was analyzed to
has been reached. Part of the reason is that these analytical studies compute the mean thermal conductivity and its standard deviation.
have neglected transition between the two branches, and accord- The results are shown in Table 2. In order to compare the errors at
ing to Klemens 关30兴, this could result in misleading deductions, different temperatures, it is necessary to run the simulations at all
especially for materials in which dispersion is strong. In the temperatures with the same number of samples at steady state.
This, however, is difficult to control because the number of
samples changes due to scattering. For the current study, an effort
was made to attain a steady state count of 100,000 for each simu-
Table 1 Role of LA phonons in Silicon at various tempera-
tures
lation. This was done by trial-and-error adjustment of N prescribed .
However, as evident in Table 2, this goal was only achieved par-
tially. With an increase in temperature, the statistical errors in-
crease. This is because the number of scattering events increases
with temperature, and larger number of samples is necessary to
account for all possible interactions accurately. Beyond about 100
K, the number of scattering events are so large that further change
in that number does not affect the statistical accuracy
significantly.
5 Summary and Conclusions

The BTE for phonons best describe the non-equilibrium trans-
fer of heat in thin films, where the mean free path of the energy-
carrying phonons may be larger than the film thickness. In this
article, a detailed solution technique for the BTE was presented. A
novel three-phonon scattering algorithm has been presented,
which accounts for transition of phonons between the polarization
branches, and frequency dependent phonon lifetimes. The model ␴ ⫽ scattering cross section 关m2兴
was used to compute the thermal conductivity of silicon thin films ␴ k ⫽ standard deviation in thermal conductivity calcula-
between 10–300 K. Numerical predictions exhibit very close tions 关W/m•K兴
match with experimental data for pure silicon. For doped silicon ␶ NU ⫽ combined relaxation time for intrinsic scattering 关s兴
films, it was found that a model, which accounts for the mass ␶ i ⫽ relaxation time for impurity scattering 关s兴
difference between the base and the dopant atoms, is necessary to ␶ N ⫽ relaxation time for normal scattering 关s兴
produce a better match. It was also found that TA phonons play a ␶ U ⫽ relaxation time for Umklapp scattering 关s兴
dominant role in heat transport in silicon above 100 K. Future ␻ ⫽ angular frequency 关radians/s兴
research will focus on enhancing the scattering model to allow ␻ max,LA ⫽ maximum cut-off frequency for LA branch 关radians/
treatment of normal and Umklapp scattering as isolated events, s兴
and the implementation of truly anisotropic Brillouin zones. It is
␻ max,TA ⫽ maximum cut-off frequency for TA branch 关radians/
our hope that this preliminary effort will provide a sound platform
s兴
for future research pertaining to numerical solution of the BTE for
phonons. ␻ 0,i ⫽ central frequency of i-th band 关radians/s兴
⌬␻ ⫽ band width 关radians/s兴
␺ ⫽ azimuthal angle
Acknowledgments
This work was funded by the National Science Foundation
through a SBIR Phase I award 共Award #9960172; Program Of-
ficer: Dr. Cynthia Ekstein兲. The authors gratefully acknowledge
this support. The authors would also like to thank Drs. Mehdi
References
Asheghi and Kenneth Goodson of the Department of Mechanical 关1兴 Ziman, J. M., 1960, Electrons and Phonons, Oxford University Press, London.
关2兴 Kittel, C., 1986, Introduction to Solid State Physics, John Wiley & Sons Inc.,
Engineering, Stanford University for providing the original files
Sixth Edition.
containing the experimental data. 关3兴 Tien, C. L., Majumdar, A., and Gerner, F. M., eds., 1998, Microscale Energy
Transport, Taylor and Francis.
Nomenclature 关4兴 Ju, Y. S., and Goodson, K. E., 1999, ‘‘Phonon Scattering in Silicon Films with
Thickness of Order 100 nm,’’ Appl. Phys. Lett., 74, No. 20, pp. 3005–3007.
A ⫽ area of boundary face 关m2兴关5兴 Klemens, P. G., 1958, ‘‘Thermal Conductivity and Lattice Vibrational
B L , B TN , B TU ⫽ constant in relaxation time expression Modes,’’ in Solid State Physics, 7, F. Seitz and D. Turnball, eds., Academic
d ⫽ degree of specularity Press, NY.
D( ␻ ) ⫽ density of state 关m⫺3•s兴关6兴 Callaway, J., 1959, ‘‘Model for Lattice Thermal Conductivity at Low Tem-
peratures,’’ Phys. Rev., 113, No. 4, pp. 1046–1051.
E ⫽ energy per unit volume 关J/m3兴关7兴 Holland, M. G., 1963, ‘‘Analysis of Lattice Thermal Conductivity,’’ Phys.
f ⫽ distribution function Rev., 132, No. 6, pp. 2461–2471.
F ⫽ cumulative number density function 关8兴 Holland, M. G., 1964, ‘‘Phonon Scattering in Semiconductors from Thermal
G ⫽ reciprocal lattice vector 关m⫺1兴 Conductivity Studies,’’ Phys. Rev., 134, No. 2A, pp. A471–A480.
h ⫽ Planck constant (⫽6.6262⫻10⫺34 J•s) 关9兴 Bhandari, C. M., and Verma, G. S., 1965, ‘‘Role of Longitudinal and Trans-
verse Phonons in Lattice Thermal Conductivity of GaAs and InSb,’’ Phys.
ប ⫽ h/2␲ (⫽1.05459⫻10⫺34 Js) Rev., 140, No. 6A, pp. A210–A214.
kB ⫽ Boltzmann constant 关10兴 Guyer, R. A., and Krumhansl, J. A., 1966, ‘‘Solution of the Linearized Phonon
(⫽1.38062⫻10⫺23 J/K) Boltzmann Equation,’’ Phys. Rev., 148, No. 2, pp. 766–778.
关11兴 Hardy, R. J., 1970, ‘‘Phonon Boltzmann Equation and Second Sound in Sol-
k̄ ⫽ ensemble averaged thermal conductivity
ids,’’ Phys. Rev. B, 2, No. 4, pp. 1193–1206.
关W/m•K兴关12兴 Moglestue, C., 1982, ‘‘Monte Carlo Particle Modeling of Small Semiconduc-
K ⫽ component of wave vector 关m⫺1兴 tor Devices,’’ Comput. Methods Appl. Mech. Eng., 30, pp. 173–208.
K ⫽ wave vector 关m⫺1兴关13兴 Jacoboni, C., and Reggiani, L., 1983, ‘‘The Monte Carlo Method for the So-
L ⫽ characteristic length 关m兴 lution of Charge Transport in Semiconductors with Applications to Covalent
N ⫽ number density 共number/volume兲关m⫺3兴 Materials,’’ Rev. Mod. Phys., 55, No. 3, pp. 642–705.
关14兴 Lugli, P., Bordone, P., Reggiani, L., Reiger, M., Kocevar, P., and Goodnick, S.
具n典 ⫽ phonon occupation number M., 1989, ‘‘Monte Carlo Studies of Nonequilibrium Phonon Effects in Polar
n̂ ⫽ unit surface normal Semiconductors and Quantum Wells. I. Laser Photoexcitation,’’ Phys. Rev. B,
Nb ⫽ number of spectral bands 39, No. 11, pp. 7852–7865.
N face ⫽ number of phonons emitted from a face 关15兴 Fischetti, M. V., and Laux, S. E., 1988, ‘‘Monte Carlo Analysis of Electron
p ⫽ polarization Transport in Small Semiconductor Devices Including Band-Structure and
P NU ⫽ probability of intrinsic scattering Space-Charge Effects,’’ Phys. Rev. B, 38, No. 14, pp. 9721–9745.
关16兴 Fischetti, M. V., and Laux, S. E., 1993, ‘‘Monte Carlo Study of Electron
P(LA/TA) ⫽ probability of LA versus TA Transport in Silicon Inversion Layers,’’ Phys. Rev. B, 48, No. 4, pp. 2244–
r ⫽ position vector 2274.
R, R 1 , R 2 , R 3 ⫽ random numbers between 0 and 1 关17兴 Majumdar, A., 1993, ‘‘Microscale Heat Conduction in Dielectric Thin Films,’’
ŝ ⫽ direction of wave-vector ASME J. Heat Transfer, 115, No. 7, pp. 7–16.
t ⫽ time 关s兴关18兴 Chen, G., and Tien, C. L., 1993, ‘‘Thermal Conductivities of Quantum Well
Structures,’’ J. Thermophys. Heat Transfer, 7, No. 2, pp. 311–318.
⌬t ⫽ time-step size 关s兴
关19兴 Goodson, K. E., 1996, ‘‘Thermal Conduction in Nonhomogeneous CVD Dia-
T ⫽ thermodynamic temperature 关K兴 mond Layers in Electronic Microstructures,’’ ASME J. Heat Transfer, 118, pp.
T̃ ⫽ pseudo temperature 关K兴 279–286.
t̂1, t̂2 ⫽ unit surface tangents 关20兴 Chen, G., 1998, ‘‘Thermal Conductivity and Ballistic Phonon Transport in the
Cross-Plane Direction of Superlattices,’’ Phys. Rev. B, 57, No. 23, pp. 14958–
V ⫽ volume 关m3兴 14973.
Vg ⫽ group velocity of phonon 关m/s兴关21兴 Klistner, T., VanCleve, J. E., Henry, E. F., and Pohl, R. O., 1988, ‘‘Phonon
W ⫽ scaling factor Radiative Heat Transfer and Surface Scattering,’’ Phys. Rev. B, 38, No. 11,
pp. 7576–7594.
Greek 关22兴 Peterson, R. B., 1994, ‘‘Direct Simulation of Phonon-Mediated Heat Transfer
␣ ⫽ calibrated constant 关Eq. 24兴 in a Debye Crystal,’’ ASME J. Heat Transfer, 116, pp. 815–822.
关23兴 Mavriplis, D. J., 1995, ‘‘Unstructured Mesh Generation and Adaptivity,’’ In-
␯ ⫽ frequency 关Hz兴 stitute for Computer Applications in Science and Engineering, Technical Re-
⌽ ⫽ scattering rate 关s⫺1兴 port #TR-95-26.
␪ ⫽ polar angle 关24兴 Cahill, D. G., 1997, ‘‘Heat Transport in Dielectric Thin Films and at Solid-
␳ ⫽ defect density 关atoms/m3兴 Solid Interfaces,’’ Microscale Thermophys. Eng., 1, pp. 85–109.
关25兴 Sverdrup, P. G., Ju, Y. S., and Goodson, K. E., 1999, ‘‘Sub-Continuum Simu- 关28兴 Asheghi, M., 2000, ‘‘Thermal Transport Properties of Silicon Films,’’ Thesis
lations of Heat Conduction in Silicon-on-Insulator Transistors,’’ IMECE 1999, in Mechanical Engineering, Stanford University.
Nashville, TN, HTD 363–3. 关29兴 Brockhouse, B. N., 1959, ‘‘Lattice Vibrations in Silicon and Germanium,’’
关26兴 Vincenti, W. G., and Kruger, C. H., 1977, Introduction to Physical Gas Dy- Phys. Rev. Lett., 2, pp. 256.
namics, Robert Kreiger Publ., New York, NY.
关30兴 Klemens, P. G., 1981, ‘‘Theory of Lattice Thermal Conductivity: Role of
关27兴 Waugh, J. L., and Dolling, G., 1963, ‘‘Crystal Dynamics of Gallium Ars-
enide,’’ Phys. Rev., 132, pp. 2410–2412. Low-Frequency Phonons,’’ Int. J. Thermophys., 2, No. 1, pp. 55–62.
Integrative Thermodynamic
Optimization of the Crossflow
Heat Exchanger for an Aircraft
Jose V. C. Vargas
Adrian Bejan
Environmental Control System
Department of Mechanical Engineering This paper documents the process of determining the internal geometric configuration of
and Materials Science,
a component by optimizing the global performance of the installation that uses the com-
Duke University, Box 90300,
ponent. The example chosen is the crossflow heat exchanger used in the environmental
Durham, NC 27708-0300
control system of a modern aircraft. The optimization of global performance is achieved
by minimizing the total entropy generation rate of the installation. There are three de-
grees of freedom in the heat exchanger configuration (the length-to-width and height-to-
David L. Siems width aspect ratios, and the separator plate spacing ratio), which is subjected to two
The Boeing Company,
global constraints: total component volume, and total wall material volume (or weight/
MC S106-7075, PO Box 516,
density) of wall material. Numerical results show how the optimal configuration responds
Saint Louis, MO 63166-0519
to changes in specified external parameters such as volume, weight, Mach number, dif-
fuser inlet cross-sectional area, and the pressure at which the cabin air is initially bled
from the engine compressor. It is shown that the optimal configuration is robust and that
major features such as the ratios of channel spacings and flow lengths are relatively
insensitive to changes in some of the external parameters. It is also shown that the
optimal heat exchanger geometry is insensitive to the thermodynamic irreversibility
caused by discharging the used ram air into the ambient. 关DOI: 10.1115/1.1375811兴
Keywords: Geometry, Heat Exchangers, Optimization, Second Law, Thermodynamics
1 Introduction architecture of the component emerges as a consequence of the

global optimization of the entire system, subject to the constraints
The thermodynamic optimization of heat exchangers has at- faced by the entire system.
tracted considerable attention during the past two decades, and is Objectives other than thermodynamic optimization can be pur-
now an established subfield in heat transfer and engineering ther- sued as an integrative design, for example, minimum total cost, or
modynamics. It is now recognized as the mechanism for the gen- minimum total weight. The integrative design methodology can
eration of architecture in flow systems everywhere, in engineering be summarized as follows: an entire system can be conceived
and nature 关1兴. Thermodynamic imperfection is due to currents from the beginning as a system designed to achieve certain global
共fluid, heat兲 that must overcome resistances. Better performance at objectives optimally, not as an ensemble of already existing parts
the global level is achieved when the resistances are minimized 关15兴.
together, i.e., when the imperfection is distributed in space opti- When should the global optimization of the system be based on
entropy generation minimization? In brief, the methods of exergy
mally 关1兴. Optimal distribution of imperfection represents flow
analysis 共EA兲, entropy generation minimization 共EGM兲 and ther-
architecture, or constructal design. moeconomics 共TE兲 are the most established changes that have
Heat exchangers have often been subjected to thermodynamic taken place in modern engineering thermodynamics during the
optimization 共or entropy generation minimization兲 in isolation, past three decades 关16–25兴. The emphasis is now on identifying
i.e., removed from the larger installation that uses them. Examples the mechanisms and system components that are responsible for
are the counterflow, crossflow, parallel flow and phase-change thermodynamic losses 共EA兲, the sizes of these losses 共EA兲, mini-
heat exchanger optimizations documented in 关2–13兴. mizing the losses subject to the global constraints of the system
The existing work is based on the important assumption that the 共EGM兲, and minimizing the total costs associated with building
component 共heat exchanger兲 can be optimized in isolation. The and operating the energy system 共TE兲. Thermodynamic optimiza-
successful execution of this step is known as the principle of tion 共EGM兲 may be used by itself 共without cost minimization兲 in
the preliminary stages of design 关26兴, in order to identify trends
thermodynamic isolation, and depends on a careful understanding
and the existence of optimization opportunities. The optima and
of how the component interacts with the rest of the installation. structural characteristics identified based on thermodynamic opti-
Paradoxical conclusions result when this principle is not respected mization can be made more realistic through subsequent refine-
共for an illustration, see 关14兴, pp. 606–609兲. ments based on global cost minimization.
The much surer alternative is to pursue the thermodynamic op- Thermodynamic optimization can be useful by itself in cases
timization of the heat exchanger as a component installed in the where the total cost of the installation is dominated by the cost
larger system. This consists of optimizing the larger system by due to thermodynamic irreversibility. The classical example of
varying the geometric features of the component. This is the inte- this kind is cryogenics, or refrigeration at very low temperatures,
grative approach used in this paper. The result is that the entire where the power requirement is substantial and proportional to the
entropy generated in the cold space 关16,27兴. Another example is
the subject of this paper: environmental control systems for air-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division June 9, 2000; craft.
revision received January 20, 2001. Associate Editor: D. Poulikakos. To see why aircraft energy systems can be conceptualized on
Fig. 1 Environmental control system with compression and
expansion on the engine-air side „ ṁ e …, and heat exchanger for
cooling with ram air „ ṁ a …
Fig. 2 The T - s diagrams of the processes undergone by the
engine-air and ram-air streams
the basis of thermodynamic optimization, consider their objectives
and physical constraints. Power and refrigeration systems are as-
semblies of streams and hardware 共components兲. The size of the Ẇ t ⫽Ẇ c . The purpose of the expansion through the turbine is to
hardware is always constrained 共e.g., weight, volume兲. Each decrease the pressure and temperature to the specified cabin con-
stream carries exergy 共useful work content兲, which is the life ditions ( P 4 ,T 4 ). The air flow rate ṁ e is also specified by the
blood of the power system, i.e., another form of the fuel brought environmental design of the cabin.
on board and burned in order to drive the system. Exergy is de- The lower stream ṁ a shown in Fig. 1 is the ram air used on the
stroyed 共or entropy is generated兲 whenever streams interact with cold side of the crossflow heat exchanger. The ram air enters the
each other and with components. The design objectives are: 共i兲 to system at ambient conditions ( P a ,T a ), and initially it has the
optimize streams and components so that they destroy the least same speed as the aircraft (V a ). It is decelerated to the state
exergy subject to constraints, and 共ii兲 to make sure that the opti- ( P a,1 ,T a,1 ,V a,1) in a diffuser, and continues through the heat ex-
mized entities ‘‘match,’’ or can be ‘‘fitted’’ together 共wrapped changer, where it is heated to the state ( P a,2 ,T a,2 ,V a,2). Before
around each other兲 into a new integrative design of the larger being discharged into the ambient, the ṁ a stream is accelerated to
system. What emerges is a design, i.e., the visible structure that the state ( P a,out ,T a,out ,V a,out ). The ram air mass flow rate ṁ a is
reflects the optimization principle and the various constraints. not specified.
The new feature of the optimization approach presented herein The processes undergone by the engine-air and ram-air streams
is that it determines the entire geometry of the subject component are shown on their respective T-s diagrams in Fig. 2. Each pro-
共in this case, a heat exchanger兲. The optimal geometry is deduced cess is irreversible. The upper diagram shows that the compres-
from the minimization of the thermodynamic imperfection of the sion process 共1兲–共2兲 is not isentropic: the isentropic efficiency is
installation in which the component is used. This is in contrast to ␩ c2 ⬍1. Similarly, the expansion 共3兲–共4兲 is characterized by the
the optimization methodology that is currently used in the aircraft turbine isentropic efficiency ␩ t ⬍1. The flow 共2兲–共3兲 on the hot
industry 共i.e., the trade study兲, which attempts to select analyti- side of the heat exchanger is accompanied by the frictional pres-
cally the optimal geometry from some arbitrarily defined and lim- sure drop ⌬ P e . The preliminary compression 共a兲–共1兲 executed in
ited field of candidates. the jet-engine compressor is not shown; its isentropic efficiency is
␩ c1 ⬍1.
The lower diagram of Fig. 2 shows the deceleration process
2 Model (a) – (a,1) through the diffuser, where the isentropic efficiency is
Consider the model with two streams shown in Fig. 1. The ␩ d . The pressure drop experienced by the ram air while flowing
upper stream ṁ e is originally drawn from the ambient ( P a ,T a ), on the cold side of the heat-exchanger surface is ⌬ P a . The isen-
and is compressed to the state ( P 1 ,T 1 ) in the low-pressure stages tropic efficiency of the nozzle acceleration process (a,2) – (a,out)
of the jet engine. This ‘‘engine’’ air stream (ṁ e ) is a fraction of is ␩ n .
the air flow rate processed by the jet-engine compressor. The To be true to the integrative design philosophy, the model
stream ṁ e is compressed further in a separate compressor, which should have included ‘‘everything’’ that is affected by, or is
raises its pressure and temperature ( P 2 ,T 2 ). Next, the stream coupled to the environmental control system. One step toward
temperature is lowered to T 3 in a crossflow heat exchanger, and more realism and greater complexity would have been to include
then the stream is expanded through a turbine. The shaft power the jet-engine compressor 共Fig. 1兲 in the greater system that is
produced by the turbine is used to drive the separate compressor, subjected to entropy generation analysis and minimization. We
stopped short of doing this, because of time and computational
cost limitations. We felt that the environmental control system of
Fig. 1 is large enough 共certainly larger than the heat exchanger兲 to 冉
N S2 ⫽ ln
␶1
␶a
⫺b ln
P̃ 1
P̃ a
冊冉 ⫹ ln
␶2
␶1
⫺b ln
P̃ 2
P̃ 1
冊冉
⫹ ln
␶3
␶2
⫺b ln
P̃ 3
P̃ 2
冊
冉冊冉冊
illustrate the integrative approach, i.e., why this approach differs
from optimizing the heat exchanger alone. 1 1 ␶ a,out
⫹ ln ⫺b ln ⫹␮ ⫺1 . (7)
␶3 P̃ 3 ␶a
3 Entropy Generation Analysis
The difference between the N S1 and N S2 expressions consists only
The objective is to select and arrange the geometric features of in the last terms, ␮ ln(␶a,out /␶a) versus ␮ 关 ( ␶ a,out / ␶ a )⫺1 兴 . Both
the system so that its penalty 共entropy generation, exergy destruc- terms increase monotonically with ␶ a,out . The value of ␶ a,out is
tion兲 to the aircraft power output is minimized. The analysis de- the same, regardless of how we evaluate the total entropy genera-
veloped in this section establishes the relation between the en- tion rate: there is one ␶ a,out for each geometric configuration of
tropy generation rate and the physical features of the system. To the crossflow heat exchanger. We will show numerically that be-
start with, the entropy generation rate due to the complete system cause of this feature the minimization of N S1 leads to the same
defined in Fig. 1 is optimal geometry as the minimization of N S2 . The minimized
Ṡ gen,1⫽ṁ e 冋冉 c p ln
T1
Ta
⫺R ln
P1
Pa 冊冉
T2
⫹ c p ln ⫺R ln
T1
P2
P1 冊 N S1 and N S2 values, however, are different as soon as ␶ a,out dif-
fers from ␶ a .
To minimize the entropy generation rate is equivalent to mini-
冉
⫹ c p ln
T3
T2
⫺R ln
P3
P2
T4
冊冉
⫹ c p ln ⫺R ln
T3
P4
P3 冊册 e
mizing the pressure P 1 , or the compressor power that produces
the ṁ e stream. This equivalence is proclaimed by the Guoy-
Stodola theorem 关16,17兴.
⫹ṁ a c p ln冉 T out

T in
⫺R ln
P out
P in 冊 a
. (1)
In the EGM method the entropy generation calculation is
complemented by the analysis of the heat flow and fluid flow
through each component. This analysis accounts for the physical
The terms collected in this expression correspond to the sequence parameters and size constraints of the component. For the first
of components recognized in Figs. 1 and 2. The two fluids are component of our system—the engine compressor (a) – (1) in
modeled as ideal gases with constant specific heat, (R,c p ) e and Fig. 1—these effects are embodied in the specified compressor
(R,c p ) a . It is convenient to introduce the nondimensional vari- efficiency
ables T 1,re v ⫺T a
␩ c1 ⫽ . (8)
P T T 1 ⫺T a
P̃⫽ ␶⫽ (2)
P4 T4
Noting that T 1,re v̇ ⫽T a ( P 1 / P a ) b , Eq. 共5兲 becomes
冋冉冊册
Ṡ gen ṁ a c pa b
N S⫽ ␮⫽ , (3) ␶a P̃ 1
ṁ e c pe ṁ e c pe ␶ 1⫽ ␶ a⫹ ⫺1 . (9)
where ( P 4 ,T 4 ,ṁ e ) are the fixed cabin conditions, N S is the en- ␩ c1 P̃ a
tropy generation number, and ␮ is the ratio of the two capacity
Next, for the separate compressor 共1兲–共2兲 the definition of the
rates. Equation 共1兲 becomes
isentropy efficiency ␩ c2 and the first law analysis yield
冉
N S1 ⫽ ln
␶1
␶a
⫺b ln
P̃ 1
P̃ a
冊冉 ⫹ ln
␶2
␶1
⫺b ln
P̃ 2
P̃ 1
冊冉 ⫹ ln
␶3
␶2
⫺b ln
P̃ 3
P̃ 2
冊 ␶ 2⫽ ␶ 1⫹
␶1
␩ c2
冋冉冊册 P̃ 2
b
⫺1 (10)
冉冊
P̃ 1
冋冉冊册
1 1 ␶ a,out
⫹ ln ⫺b ln ⫹ ␮ ln , (4) ␶1 P̃ 2
b
␶3 P̃ 3 ␶a w c⫽ ⫺1 . (11)
␩ c2 P̃ 1
where b⫽R/c p . It is assumed that the two streams have the same
b value, namely b⫽0.286, which corresponds to ␥ ⫽c p /c v ⫽1.4. In Eq. 共11兲 the compressor power input w c is dimensionless, w c
An alternative to Eq. 共4兲 is to include in the entropy generation ⫽Ẇ c /(ṁ e c pe T 4 ).
rate calculation the contribution made by the discharging of the The heat transfer between the two streams in the crossflow heat
spent stream (T a,out ) into the ambient (T a ). Equation 共1兲 is re- exchanger is described by the effectiveness-NTU relation 关28兴
placed by
Ṡ gen,2⫽ṁ e 冋冉 c p ln
T1
Ta
⫺R ln
P1
Pa
T2
冊冉
⫹ c p ln ⫺R ln
T1
P2
P1 冊 ␧⫽1⫺exp兵 ␮ N 0.22关 exp共 ⫺ ␮ ⫺1 N 0.78兲 ⫺1 兴其 ,
which is based on the assumption that ṁ e c pe is the smaller of the
(12)
冉冊冉冊册
two capacity rates 共i.e., ␮ ⬎1兲, and where N is the number of heat
T3 P3 T4 P4 transfer units, N⫽UA/(ṁ e c pe ). The definition of the effective-
⫹ c p ln ⫺R ln ⫹ c p ln ⫺R ln
T2 P2 T3 P3 e
ness ␧ produces two relations between the inlet and outlet tem-
peratures of the two streams,
⫹ṁ a c p ln 冉 T out
T in
⫺R ln
P out
P in 冊 a
⫹ṁ a c p ln
Ta
T a,out
⫹
Q̇ e
Ta
, ␶ 2 ⫺ ␶ 3 ⫽␧ 共 ␶ 2 ⫺ ␶ a,1兲 (13)
␧
(5) ␶ a,2⫺ ␶ a,1⫽ 共 ␶ 2 ⫺ ␶ a,1兲 . (14)
␮
where the cooling experienced by the discharged air is
Equations 共12兲–共14兲 hold for ␮ ⬎1; the corresponding set for ␮
Q̇ e ⫽ṁ a c p 共 T a,out ⫺T a 兲 . (6) ⬍1 is not listed, for the sake of brevity. The pressure drop along
Using the dimensionless variables defined in Eqs. 共2, 3兲, we re- the hot side of the heat transfer surface, ⌬ P̃ e ⫽ P̃ 2 ⫺ P̃ 3 , is ana-
write Eq. 共5兲 as a second entropy generation number lyzed later in this section.
Downstream of the crossflow heat exchanger is the turbine 共3兲–
共4兲: the definition of the efficiency ␩ t , and the first law analysis of
the turbine yield
1
冋
⌬ P i ⫽ G 2i v i,in 共 K c,i ⫹1⫺ ␴ 2i 兲 ⫹2
2 冉
v i,out
v i,in
⫺1 ⫹ f i 冊
Ai vi
A c,i v i,in
␶ 3 ⫽1⫹ ␩ t ␶ 3 共 1⫺ P̃ ⫺b
3 兲 (15) ⫺ 共 1⫺ ␴ 2i ⫺K e,i 兲
v i,out
v i,in
册, 共 i⫽e,a 兲 , (27)
w t ⫽ ␩ t ␶ 3 共 1⫺ P̃ ⫺b
3 兲. (16) where
The dimensionless turbine power output is defined as w t 1 A c,i
⫽Ẇ t /(ṁ e c pe T 4 ). Finally, since the turbine drives the separate v i⫽ 共v ⫹ v i,out 兲 ␴ in ⫽ . (28)
2 i,in A f ,i
compressor, w t ⫽w c , the first-law analysis of the turbine and
compressor 共together兲 requires In these relations G i , v i , A i , f i , ␴ i , K c,i , K e,i , A c,i , and A f ,i ,
represent in order, the mass velocity (G i ⫽ṁ i /A c,i ), the specific
␶ 2 ⫺ ␶ 1 ⫽ ␶ 3 ⫺1. (17) volume ( v i ⫽R i T i / P i ), the total heat transfer area, the friction
On the cold side of the system, we analyze the course followed factor, the cross-section contraction ratio, the contraction loss co-
by the ram air. This begins with the diffuser (a) – (a,1), for which efficient, the enlargement loss coefficient, the flow cross-section
the ␩ d definition and the first law of thermodynamics require and the stream cross-section before entering the duct. The friction
factor is available as a function of the Reynolds number based on
T a,1rev⫺Ta hydraulic diameter, f i ⫽ f i (Rei), either as empirical correlations
␩ d⫽ (18) for parallel-plate channels 关30兴 or as tabulated data for finned
Ta,1⫺T a
surfaces 关29兴.
1 2 The same heat exchanger sources provide the heat transfer co-
c p,a 共 T a,1⫺T a 兲 ⫽ 共 V ⫺V 2a,1兲 , (19) efficient information needed for evaluating the number of heat
2 a
transfer units N. The flow of heat across the A i surface overcomes
where T a,1re v ⫽T a ( P a,1 / P a ) b . The dimensionless version of Eqs. three thermal resistances in series,
共18, 19兲 is 1 ṁ e c pe
冉 1 tw 1
冊
冋冉冊册
⫽ ⫽ṁ e c pe ⫹ ⫹ . (29)
␶ a,1 1 P̃ a,1
b
N UA ␩ 0,e h e A e k w A w ␩ 0,a h a A a
⫺1⫽ ⫺1 (20) In this expression ␩ 0,i , t w , k w , and A w are the fin efficiencies of
␶a ␩d P̃ a the finned surfaces, the thickness of the wall penetrated by heat
␶ a,1 1 transfer, the wall thermal conductivity, and the average heat trans-
⫺1⫽ 共 Ṽ 2a ⫺Ṽ 2a,1兲 . (21) fer surface, A w ⫽(A e ⫹A a )/2. The heat transfer coefficients
␶a 2 (h e ,h a ) are available in dimensionless form, as the respective
Stanton number versus the channel Reynolds number,
The dimensionless bulk velocity Ṽ⫽V/(c pa T a ) 1/2 is proportional
h i /(c pi G i )⫽St i (Rei).
to the Mach number M ⫽V/( ␥ a R a T a ) 1/2, or Ṽ⫽M ( ␥ a ⫺1) 1/2. In-
voking the conservation of the mass flow rate ṁ a in every cross-
section A c and the ideal gas model, namely ṁ a ⫽ ␳ VA c and ␳ 4 Parallel-Plate Heat Exchanger
⫽ P/(RT), we write the Mach number as
The preceding analysis completes the model, and allows us to
M⫽
ṁ a R a
PA c ␥ a
T 冉冊 1/2
. (22)
relate N S1 to the irreversibilities and physical sizes of all the com-
ponents. The numerical minimization of the entropy generation
rate can proceed, but first we must specify the type of the heat
In sum, when the geometry and operating conditions are specified, transfer surface built into the crossflow heat exchanger. This as-
Eqs. 共20兲, 共21兲, and 共22兲 pinpoint state (a,1), or P̃ a,1 and ␶ a,1 . sumption is necessary in order to access the f i and St i information
The heating process (a,1) – (a,2) experienced by the ṁ a stream that is available in the heat exchanger literature.
is governed by the effectiveness-NTU relations 共12兲–共14兲. The In this optimization study we assumed the simplest and best
pressure drop ⌬ P̃ a ⫽ P̃ a,1⫺ P̃ a,2 is given by Eq. 共27兲 later in this documented surface type: parallel plates of spacings B e and B a ,
section. as shown in Fig. 3. For the laminar regime (ReDh⬍2300) we used
Finally, the irreversible flow through the nozzle is described by the friction factor and heat transfer coefficient correlations 共e.g.,
the efficiency and first-law relations 关30兴兲,
T a,2⫺T a,out 24 h i D h,i
␩ n⫽ (23) f i⫽ ⫽8.235, (30)
T a,2⫺T a,out re v ReD h,i ki
1 2 where D h,i ⫽2B i , ReDh,i⫽GiDh,i /␮i , and i⫽e,a. The correlations

c pa 共 T a,2⫺T a,out 兲 ⫽ 共V ⫺V 2a,2兲 . (24) used for the turbulent regime were 关30兴
2 a,out
⫺1/4
f i ⫽0.079 ReD 共 2300⬍ReD h,i ⬍2⫻104 兲 (31)
Recognizing T a,out re v ⫽T a,2( P a / P a,2) b and the nondimensional h,i
variables employed earlier, we can nondimensionalize Eqs. 共23,
h i D h,i 共 f i /2兲共 ReD h,i ⫺103 兲 Pri
24兲 as ⫽ 共 2300⬍ReD h,i ⬍5⫻106 兲 .
冋冉冊册
b
ki i ⫺1 兲
1⫹12.7共 f i /2兲 1/2共 Pr2/3
␶ a,out P̃ a (32)
1⫺ ⫽ ␩ n 1⫺ (25)
␶ a,2 P̃ a,2 The height H ea of the elemental channel unit is defined as the
height of the assembly composed of one engine-air passage and
␶ a,2 ␶ a,out 1 2 one ram-air passage,
⫺ ⫽ 共 Ṽ a,out ⫺Ṽ 2a,2兲 . (26)
␶a ␶a 2 H ea⫽B e ⫹B a ⫹2t w . (33)
For the pressure drops along the counterflow (⌬ P e ,⌬ P a ) we If n is the number of elemental units in the stack, the height of the
use the classical formulation 关29兴 heat exchanger core is
Table 1 Physical values used as reference case in the numeri-
cal optimization of the overall system
Fig. 3 The geometry of a crossflow heat exchanger core with

parallel-plate channels
L z ⫽nH ea (34) metric features of the core. The first variable identified in this way
The total volume of the core is constrained, V t ⫽L x L y L z , con- was the plate spacing ratio B e /B a . Figure 4 shows that an optimal
stant. The length scale V 1/3
t is used for the purpose of nondimen-
spacing ratio exists over a wide domain covered by other specified
sionalizing all the lengths that characterize this geometry, parameters of the larger installation, for example, the bleeding
pressure P̃ 1 共Fig. 4共a兲兲, the cross-sectional area of the diffuser
共 B̃ e ,B̃ a , t̃ w ,H̃ ea ,L̃ x ,L̃ y ,L̃ z 兲 ⫽ 共 B e ,B a ,t w ,H ea ,L x ,L y ,L z 兲 /V 1/3
t inlet Ã cd ⫽A cd /1m 2 共Fig. 4共b兲兲, and the Mach number M 共Fig.
(35) 4共c兲兲.
such that the volume constraint reads In the designs optimized in Fig. 4 the shape of the core and the
number of elemental channels are fixed: L̃ x ⫽5, L̃ y ⫽1, L̃ z ⫽0.2,
L̃ x L̃ y L̃ z ⫽1 (36) and n⫽10. The other parameters that had to be specified are the
Another constraint that we attach to the geometric optimization reference values collected in Table 1, for example, t̃ w ⫽t w /V 1/3 t
procedure is the total volume 共or weight兲 of the wall material, ⫽0.6 mm/1 m⫽0.0006, hence ␾ ⫽0.06.
2nL x L y t w . This second constraint is best represented by the vol- Figure 4 reveals a potentially cost-saving feature of the thermo-
ume fraction ␾ (Ⰶ1) occupied by the wall material in the entire dynamic optimization: the optimized ratio B e /B a is relatively in-
volume, ␾ ⫽2nL x L y t w /V t , constant, or sensitive to changes in some of the specified system parameters
共e.g., P̃ 1 and M, Figs. 4共a兲 and 4共c兲兲. The order of magnitude of
t̃ w B e /B a is 1. This lack of sensitivity lends robustness to the opti-
␾ ⫽2n ⫽2nL̃ x L̃ y t̃ w 共 constant兲 . (37) mized design: we pursued the robustness feature in subsequent
L̃ z stages of this study. On the other hand, the optimal ratio B e /B a
According to Fig. 3, the architecture of the heat exchanger core decreases as the diffuser inlet area Ã cd increases 共Fig. 4共b兲兲. This
is determined completely if we specify six dimensions: trend is expected, because a larger Ã cd captures a larger ram-air
L x , L y , L z , B e , B a , and t w . There are two constraints, Eqs. flow rate, which must be accommodated by a larger B a spacing.
共36兲 and 共37兲, and the reasonable assumption that the wall thick- The optimization of the ratio B e /B a was built as the first 共in-
ness is a specified parameter ( t̃ w Ⰶ1) dictated by the availability nermost兲 loop in the more complex numerical optimization rou-
of standard sizes of sheet metal. In conclusion, the geometric de- tine that was developed ultimately for the entire installation. The
sign of the core has three degrees of freedom. next step in the search for additional geometric degrees of free-
dom that would permit the minimization of N S1 was the focus on
5 Entropy Generation Minimization the geometric aspect ratios of the core, L z /L x and L y /L x . We
assigned fixed values to these two ratios, and repeated the optimi-
The model described in Sections 3 and 4 allows the computa- zation with respect to B e /B a . This search is summarized in Figs.
tion of the two entropy generation numbers, N S1 and N S2 , defined 5共a,b兲, which show that the N S1 minimum with respect to B e /B a
by Eqs. 共4兲 and 共7兲, respectively. This is possible once physical persists.
values 共Table 1兲, a set of external parameters 共M, Ã cd , ␾, and P̃ 1 兲 From this point on we pursued the geometric optimization of
and the set of geometric parameters 共L y /L x , L z /L x , and B e /B a 兲, the core by varying three dimensionless numbers, the ratios
are chosen for the overall system. Equations 共8兲–共37兲 are then B e /B a , L y /L x , and L z /L x . In Fig. 6 we verified that there exists
combined to form a nonlinear system of eleven nondimensional a thermodynamic optimum with respect to the aspect ratio L y /L x .
equations and eleven unknowns 共␶ a,1 , P̃ a,1 , ␶ a,2 , P̃ a,2 ␶ 2 , P̃ 2 , In this test we fixed B e /B a and L z /L x along with the other speci-
␶ 3 , P̃ 3 , ␧, Ã c,out , and ␶ a,out 兲. This system was solved by func- fied parameters (M ,Ã cd , P̃ 1 , ␾兲. The required number of elemen-
tional iteration. A maximum tolerance of 10⫺6 was imposed on tal units in the core, n, decreases as L y /L x increases in Fig. 6.
储 R 储 , which is the residual norm of the resulting system of nonlin- Discrete points are plotted on this figure, because we made calcu-
ear equations. The numerical optimization was performed by lations only for integer values of n.
varying the set of geometric parameters 共L y /L x , L z /L x , and A similar conclusion follows from Fig. 7, where L z /L x varies
B e /B a 兲, for each set of external parameters 共M, Ã cd , ␾ and P̃ 1 兲, while B e /B a and L y /L x are held fixed. The necessary number of
and identifying the optimal set of geometric parameters such that elemental units increases as L z /L x increases. An optimal ratio
N S1 and N S2 were minimum. L z /L x exists such that the global entropy generation number N S1
The critical part of this phase of the work was the identification is minimum.
of the degrees of freedom 共the variables兲 that would permit the The bulk of the numerical work was centered on the minimiza-
minimization of N S1 , i.e., the optimal selection of important geo- tion of N S1 with respect to all three geometric parameters, B e /B a ,
Fig. 5 The minimization of entropy generation rate with re-
spect to the ratio of channel spacings „ B e Õ B a … when the aspect
ratios of the heat exchanger core are fixed „ L y Õ L x , L z Õ L x …
at the end of the two-way optimization (L y /L x ,B e /B a ,n), and

how they vary with the third geometric parameter (L z /L x ).
The results of minimizing the global irreversibility N S1 with
respect to the three geometric degrees of freedom are presented in
Fig. 9. On the abscissa we varied one of the externally specified
parameters, P̃ 1 . The optimized architecture of the heat exchanger
core is represented by the ratios (B e ,B a ,L y /L x ,L z /L x ) opt3way .
The figure also reports the required number of elemental units
n opt3way , and the minimized entropy generation rate N S1,mmm .
Fig. 4 The minimization of entropy generation rate with re-

spect to the ratio of channel spacings „ B e Õ B a …, and the effect
of varying P̃ 1 , Ã cd , and M
L y /L x , and L z /L x . The numerical procedure was organized into

three nested loops, however, it was not executed blindly. Each
intermediate optimum 共e.g., with respect to B e /B a , L y /L x , or
L z /L x 兲 was verified to make sure that it represents a realistic
geometry. Each step in the code was monitored. Examples of this
intermediate testing are presented in Figs. 8共a,b兲, which summa-
rize the results at the end of the two-way optimization, namely,
the optimization with respect to B e /B a and L y /L x . The subscript
‘‘mm’’ indicates that N S1,mm was already minimized twice.
Figure 8共a兲 shows the effect of varying the third aspect ratio,
L z /L x , and that N S1,mm can indeed be minimized for a third time. Fig. 6 The minimization of entropy generation rate with re-
Figures 8共a,b兲 also report the corresponding geometric parameters spect to the heat exchanger aspect ratio L y Õ L x
Fig. 9 The effect of P̃ 1 on the minimization of N S 1 with respect
Fig. 7 The minimization of entropy generation rate with re- to all three degrees of freedom, B e Õ B a , L y Õ L x , and L z Õ L x
spect to the heat exchanger aspect ratio L z Õ L x
in the calculation (N S2 ), the minimized irreversibility (N S2,mmm )

This bank of geometric results can be enlarged by repeating the is greater than when the mixing with the ambient is not accounted
procedure for finite ranges of other external parameters, M, Ã cd for 共N S1,mmm , Fig. 9兲. Similar conclusions are provided by Figs.
and ␾. 10–12, which show the effect of remaining external parameters
An important test was conducted by repeating the entire three- 共Ã cd ,M,␾兲.
way optimization by using the entropy generation number N S2 of
Eq. 共7兲. The results are nearly the same as in Fig. 9. The use of 6 Discussion
N S2 instead of N S1 makes no difference with respect to the result- Taken together, Figs. 8–12 document the three-way optimized
ing optimized architecture (B e /B a ,L y /L x ,L z /L x ) opt3way . For the architecture of the crossflow heat exchanger over a wide range of
purpose of deriving geometric form from thermodynamic optimi- external parameters 共P̃ 1 ,Ã cd ,M,␾兲. These results hold for the ref-
zation, it is sufficient to consider the global system upstream of erence properties listed in Table 1, where the total volume was set
the point of discharge into the ambient, T a,out , Fig. 1. On the at V t ⫽1 m3. In a subsequent phase of this study we relaxed this
other hand, when the thermal mixing with the ambient is included
Fig. 10 The effect of Ã cd on the minimization of N S 1 with re-

spect to all three degrees of freedom, B e Õ B a , L y Õ L x and L z Õ L x
Fig. 8 Summary of results at the end of the minimization of

N S 1 with respect to two of the three degrees of freedom, B e Õ B a Fig. 11 The effect of M on the minimization of N S 1 with respect
and L y Õ L x to all three degrees of freedom, B e Õ B a , L y Õ L x , and L z Õ L x
channels 共B e and B a 兲, the fin geometry for the heat transfer sur-
faces, the separator plate material, and the separator plate thick-
ness (t w ). To be consistent with the optimization approach devel-
oped herein, smooth plates were selected as the heat transfer
surfaces. Steel was selected as the separator plate material based
on the required operating temperatures. Values for separator plate
spacings 共B a and B e 兲 and separator plate thickness were set
共fixed兲 the same as those found in typical cross-flow heat ex-
changers that are used for applications similar to the design re-
quirements. The resulting ‘‘typical industry’’ heat exchanger ge-
ometry is shown in Fig. 14.
The industry approach produces the core dimensions listed on
the left side of Table 2, where the total volume and solid-fraction
volume are Ṽ t ⫽0.043 and ␾ ⫽0.06. The remaining external pa-
rameters are M ⫽0.85, Ã cd ⫽0.01, P̃ 1 ⫽2 and the values listed in
Fig. 12 The effect of ␾ on the minimization of N S 1 with respect Table 1. The right side of Table 2 shows the corresponding di-
to all three degrees of freedom, B e Õ B a , L y Õ L x , and L z Õ L x mensions recommended by the optimization method described in
this paper. Noteworthy are the bottom two lines, which show that
the traditional design recommends a geometry that generates more
volume constraint, and varied V t over the range 0.5–2 m3. This than three times as much entropy as the configuration optimized in
variation is indicated by the dimensionless volume this paper. Again, this conclusion does not depend on whether the
total entropy generation rate is calculated as N S1 or as N S2 . The
Vt two geometries are compared further in Fig. 14.
Ṽ t ⫽ (38)
1m 3 It is worth noting that, from the aircraft industry’s point of
view, the results presented in this paper must be regarded as being
indicated on the abscissas of Fig. 13. The optimized geometric
features and the three-way minimized entropy generation rate de- of a preliminary nature. This is due to 共1兲 the specific heat ex-
pend on the total volume. The three-way minimized entropy gen- changer configuration that is optimized, and 共2兲 the weight and
volume constraints that are imposed in the optimization approach.
eration rate of the entire system increases as Ṽ t increases. Two
The selected heat exchanger configuration has smooth plates for
dimensionless parameters of the optimized geometry (L z /L x ,n)
heat transfer surfaces, while cross-flow heat exchangers designed
are insensitive to changes in Ṽ t when Ṽ t ⬎1. for aircraft applications typically have fins on these surfaces.
To illustrate the improvement in heat exchanger performance These fins greatly increase the wetted area of the heat transfer
provided by this optimization approach, an optimum heat ex- surfaces, thereby improving the heat exchange effectiveness 共i.e.,
changer geometry that was defined by it has been compared to a heat transfer performance兲. The fins also structurally support and
corresponding ‘‘typical industry’’ geometry 共i.e., an alternate ge- stabilize the plates that separate the flow channels 共the walls that
ometry that was defined by a design method used currently in the
are bounded by the heat transfer surfaces兲. This allows the plates
aerospace industry兲. Design requirements for this comparison
to be very thin, further improving the heat exchange effectiveness.
were selected arbitrarily for a typical operating condition for the
The work presented in this paper will have to be extended to
subject heat exchanger type and application. These requirements
address finned heat transfer surfaces for it to be directly useful in
were specifically the mass flow rate, inlet temperature and pres-
industry applications. Similarly, the weight and volume con-
sure, and exit temperature and pressure on both sides of heat
straints that are imposed arbitrarily on the heat exchanger in this
exchanger. A ‘‘typical industry’’ heat exchanger geometry that
optimization approach will have to be relaxed, in favor of deter-
satisfies these requirements was then determined by applying the
mining the heat exchanger size that minimizes the aircraft’s take-
heat exchanger design algorithm 共implemented in a computer pro-
off gross weight.
gram兲 that was developed in the Integrated Environmental Control
When completed, the results of this work will provide industry
System 共ICES兲 project 关31兴, a contracted effort sponsored by the
Air Force. In addition to the design requirements listed above, five with a structured process for synthesizing the thermodynamic op-
other conditions had to be specified to execute this algorithm: the timum geometry for a cross-flow heat exchanger directly from the
separator plate spacings for both the engine and ram-air flow physical laws. Eventually, with still further efforts in this direc-
tion, it may be possible to extend the EGM methodology to the
point where it can provide this same capability for entire, complex
thermodynamic systems. Such a capability would be of great ben-
efit to industry, as described in 关32兴.
We close with a comment on engineering thermodynamics, the
status of which was reviewed in Section 1. In this paper we relied
on entropy generation minimization 共EGM兲 because it is the sim-
plest, divorced from costs. We recognized up front 共Section 1兲 the
limitations and orientative value of the method. Thermoeconomics
authors may illustrate the integrative approach by using thermo-
economics tools. This work should probably be done, because
industry needs integrative and constructal design关1兴.
Although valuable for what it is and what it has to offer, ther-
moeconomics tends to obscure the thermodynamics portion of its
foundation. Economics 共costs兲 considerations creep without notice
into thermodynamics, where they do not belong. Because of costs,
we are tempted to think that some components are more important
than others, or that the entropy generation in the heat exchanger
Fig. 13 The effect of the total volume on the minimization of does not have the same thermodynamics value as the entropy
N S 1 with respect to all three degrees of freedom, B e Õ B a , generation in the remaining components. This is improper. En-
L y Õ L x , and L z Õ L x tropy generation is entropy generation, and it is pure thermody-
Fig. 14 The two geometries compared in Table 2: the traditional design versus the thermodynamically
optimal geometry described in this paper.
namics. It is the only—the universal—measure of imperfection in

a thermodynamic sense, no matter where it occurs in the system.
It is due to currents that overcome resistances. The road to maxi-
mum system performance at the global level is by balancing all
Table 2 Comparison between a design produced by tradi- the resistances together, i.e., by distributing the imperfection op-
tional methods in industry, and the geometric configuration optimally through the space occupied by the system. This is how the
timized thermodynamically in this paper
flow system acquires its architecture 关1兴.
Acknowledgment
This material is based upon work supported by the Air Force
Office of Scientific Research under Contract No. F49620-98-C-
0007. Any opinions, findings and conclusions or recommenda-
tions are those of the authors and do not necessarily reflect the
views of the Air Force Office of Scientific Research. The authors
thank Prof. G. Tsatsaronis for his insightful and constructive com-
ments on the original version of the manuscript.
Nomenclature
A ⫽ heat transfer area
Ac ⫽ cross-sectional area
Af ⫽ stream cross-section before entering the duct
b ⫽ exponent, R/c p
B ⫽ channel spacing
cp ⫽ specific heat at constant pressure
Dh ⫽ hydraulic diameter
f ⫽ friction factor
G ⫽ mass velocity
h ⫽ heat transfer coefficient
H ea ⫽ thickness of elemental unit 关3兴 London, A. L., and Shah, R. K., 1983, ‘‘Costs and Irreversibilities in Heat
Exchanger Design,’’ Heat Transfer Eng., 4, pp. 59–73.
k ⫽ thermal conductivity 关4兴 Sekulic, D. P., and Herman, C. V., 1986, ‘‘One Approach to Irreversibility
K c,e ⫽ contraction and enlargement coefficients Minimization in Compact Crossflow Heat Exchanger Design,’’ Int. Commun.
L x,y,z ⫽ lengths Heat Mass Transfer, 13, pp. 23–32.
ṁ ⫽ mass flow rate 关5兴 Sekulic, D. P., 1986, ‘‘Entropy Generation in a Heat Exchanger,’’ Heat Trans-
M ⫽ Mach number, Eq. 共22兲 fer Eng., 7, pp. 83–88.
关6兴 Poulikakos, D. 1980, ‘‘Fin Geometry for Minimum Entropy Generation
n ⫽ number of elemental units Rate,’’ M.S., thesis, University of Colorado, Boulder, Co.
N ⫽ number of heat transfer units 关7兴 Poulikakos, D., and Johnson, J. M., 1989, ‘‘Second Law Analysis of Com-
NS ⫽ entropy generation number, Eqs. 共3兲, 共4兲 and 共7兲 bined Heat and Mass Transfer Phenomena in External Flow,’’ Energy, 14, pp.
P ⫽ pressure 67–73.
关8兴 Krane, R. J., 1987, ‘‘A Second Law Analysis of the Optimum Design and
Pr ⫽ Prandtl number
Operation of Thermal Energy Storage Systems,’’ Int. J. Heat Mass Transf., 30,
Q̇ e ⫽ heat transfer rate pp. 43–57.
R ⫽ ideal gas constant 关9兴 Witte, L. C., and Shamsundar, N. A., 1983, ‘‘A Thermodynamic Efficiency
Re ⫽ Reynolds number Concept for Heat Exchanger Devices,’’ J. Eng. Power, 105, pp. 199–203.
关10兴 Witte, L. C., 1988, ‘‘The Influence of Availability Costs on Optimal Heat
Ṡ gen ⫽ entropy generation rate Exchanger Design,’’ J. Heat Transfer, 110, pp. 830–835.
St ⫽ Stanton number 关11兴 Paoletti, S., Rispoli, F., and Sciubba, E., 1989, ‘‘Calculation of Exergetic
tw ⫽ wall thickness Losses in Compact Heat Exchanger Passages,’’ ASME AES, 10-2, pp. 21–29.
T ⫽ temperature 关12兴 Benedetti, P., and Sciubba, E., 1993, ‘‘Numerical Calculation of the Local
U ⫽ overall heat transfer coefficient Rate of Entropy Generation in the Flow Around a Heated Finned Tube,’’
ASME HTD, 266, pp. 81–91.
v ⫽ specific volume 关13兴 Sekulic, D. P., and Shah, R. K., 1995, ‘‘Thermal Design Theory of Three-
V ⫽ velocity Fluid Heat Exchangers,’’ Adv. Heat Transfer, 26, pp. 219–328.
Vt ⫽ total volume 关14兴 Bejan A., 1997, Advanced Engineering Thermodynamics, Second Edition,
w ⫽ dimensionless power Wiley, New York.
关15兴 Gambill, J. M., Wiese, D. E., Claeys, H. M., Matulich, D. S., and Weiss, C. F.,
Ẇ ⫽ power 1993, ‘‘Integrated Aircraft Thermal Management and Power Generation,’’
Greek Symbols SAE Paper No. 932055, presented at the 23rd International Conference on
Environmental Systems, Colorado Springs, CO, July 12–15.
␥ ⫽ ratio of specific heats, c p /c v 关16兴 Bejan, A., 1982, Entropy Generation through Heat and Fluid Flow, Wiley,
⌬ P ⫽ pressure difference New York.
关17兴 Bejan, A., 1996, Entropy Generation Minimization, CRC Press, Boca Raton,
␧ ⫽ effectiveness FL.
␩ ⫽ efficiency 关18兴 Feidt, M., 1987, Thermodynamique et Optimization Énergetique des Systémes
␩ 0 ⫽ fin efficiency et Procedés, Technique et Documentation, Lavoisier, Paris.
␮ ⫽ ratio of capacity rates, Eq. 共3兲关19兴 Krane, R. J., ed., 1994, Thermodynamics and the Design, Analysis, and Im-
␴ ⫽ contraction ratio provement of Energy Systems 1994, AES-Vol. 33, ASME, New York.
关20兴 Krane, R. J., ed., 1995, Thermodynamics and the Design, Analysis, and Im-
␶ ⫽ dimensionless temperature, Eq. 共2兲 provement of Energy Systems 1995, AES-Vol. 35, ASME, New York.
␾ ⫽ solid volume fraction 关21兴 Lazzaretto, A., and Tsatsaronis, G., 1997, ‘‘On the Quest for Objective Equa-
tions in Exergy Costing,’’ Proceedings of the ASME Advanced Energy Systems
Subscripts Division, M. L. Ramalingam, J. L. Lage, V. C. Mei, and J. N. Chapman, eds.,
a ⫽ ram air AES-Vol. 37, pp. 197–210.
关22兴 Moran, M. J., 1982, Availability Analysis: A Guide to Efficient Energy Use,
c1 ⫽ engine compressor Prentice-Hall, Englewood Cliffs, NJ.
c2 ⫽ compressor, Fig. 1 关23兴 Moran, M. J., and Sciubba, E., 1994, Exergetic Analysis: Principles and Prac-
d ⫽ diffuser tice, J. Eng. Gas Turbines Power, 116, pp. 285–290.
e ⫽ engine air 关24兴 Richter, H. J., ed., 1993, Thermodynamics and the Design, Analysis and Im-
in ⫽ inlet provement of Energy Systems 1993, HTD-Vol. 266, ASME, New York.
关25兴 Stecco, S. S., and Moran, M. J., 1992, Energy for the Transition Age, Nova
n ⫽ nozzle Science, New York.
out ⫽ outlet 关26兴 Bejan, A., Tsatsaronis, G., and Moran, M., 1996, Thermal Design and Opti-
rev ⫽ reversible mization, Wiley, New York.
t ⫽ turbine 关27兴 Ahern, J. E., 1980, The Exergy Method of Energy Systems Analysis, Wiley,
New York.
w ⫽ wall 关28兴 Incropera, F. P., and DeWitt, D. P., 1990, Fundamentals of Heat and Mass
Superscripts Transfer, 3rd ed., Wiley, New York.
关29兴 Kays, W. M., and London, A. L., 1984, Compact Heat Exchangers, 3rd ed.,
( ⬃ ) ⫽ dimensionless, Eqs. 共2兲, 共35兲 and 共36兲 McGraw-Hill, New York.
关30兴 Bejan, A., 1993, Heat Transfer, Wiley, New York.
关31兴 Dieckmann, R. R., Watson, A. C., and Glover, S. F., 1972, ‘‘Development of
References Integrated Control System Designs for Aircraft,’’ Vol. I-ECS Design, Techni-
关1兴 Bejan, A., 2000, Shape and Structure, from Engineering to Nature, Cambridge cal Report AFFDL-TR-72-9, McDonnel Douglas Corporation, St. Louis, MO,
University Press, Cambridge, UK. pp. 58–84.
关2兴 Bejan, A., 1977, ‘‘The Concept of Irreversibility in Heat Exchanger Design: 关32兴 Bejan, A., and Siems, D. L., 2001, ‘‘The Need for Exergy Analysis and Ther-
Counterflow Heat Exchangers for Gas-to-Gas Applications,’’ J. Heat Transfer, modynamic Optimization in Aircraft Development,’’ Exergy—An Interna-
99, pp. 374–380. tional Journal, 1, in press.
An Improved Design and Rating
Analyses of Counter Flow Wet
Cooling Towers
Jameel-ur-Rehman Khan Cooling towers are one of the largest heat and mass transfer devices that are in common
use. In this paper, we present a detail model of counter flow wet cooling towers. The
Syed M. Zubair1 authenticity of the model is checked by experimental data reported in the literature. The
smzubair@kfupm.edu.sa values of number of transfer units (NTU) and tower effectiveness (␧) obtained from the
model were compared with the commonly described models. Appreciable difference in
Mechanical Engineering Department, NTU and ␧ values is found if the resistance to heat transfer in the water film and non-
King Fahd University of Petroleum & Minerals, unity of Lewis number is considered in the calculations. The results demonstrate that the
Dhahran 31261, Saudi Arabia errors in calculating the tower effectiveness could be as much as 15 percent when con-
sidering the effect of air-water interface temperature. A procedure for the use of the
model in designing and rating analyses of cooling towers is demonstrated through ex-
ample problems. The limiting performance of the cooling towers; that is effectiveness
equal to one, is explained in terms of air-approach temperature. The model is also used
for obtaining the maximum possible mass-flow rate ratio of water-to-air, for different
operating conditions. 关DOI: 10.1115/1.1376395兴
Keywords: Cooling Towers, Direct Contact, Heat Transfer, Heat Exchangers, Modeling
Introduction the equations for heat and water vapor transfer. He showed the
utility of total heat or enthalpy difference as a driving force to
Cooling towers are widely used in most industrial power gen-
allow for both sensible and latent heats. The basic postulations
eration units, refrigeration and air conditioning plants, chemical,
and approximations that are inherent in Merkel’s theory are 共1兲
petrochemical, and petroleum industries to reject waste heat to the the resistance for heat transfer in the liquid film is negligible, 共2兲
environment. In particular, steam power plants reject heat at about the mass flow rate of water per unit of cross sectional area of the
twice the rate at which electric power is generated. For a given tower is constant, i.e., there is no loss of water due to evaporation,
water-cooling load, a wet cooling tower needs only about one- 共3兲 the specific heat of air-stream mixture at constant pressure is
fourth the amount of contact surface when compared with that of same as that of the dry air, and 共4兲 the Lewis number for humid
a dry tower. It should be noted that dry towers 共or air-cooled heat air is unity. It is important to note that the formulation and imple-
exchangers兲 are economically capable of cooling the water to
within about 10°C of the ambient dry-bulb temperature 关1兴. Such
temperature levels are often too high for cooling water require-
ments of most industrial processes. The water consumption rate of
a wet cooling tower system is only about 5 percent that of a
once-through system, making it the least expensive system to op-
erate with purchased water supplies. Additionally, the amount of
heated water discharged 共blowdown兲 is very small, so that the
ecological effect is reduced. Lastly, cooling towers can economi-
cally cool water to within 3 to 6°C of the ambient wet-bulb tem-
perature 关1–2兴.
The cooling towers, as shown in Fig. 1, generally consist of
large chambers loosely filled with trays or decks of wooden
boards as slats or of PVC material. The water to be cooled is
pumped to the top of the tower, where it is distributed over the top
deck by sprays or distributor troughs made of wood or PVC ma-
terial. It then falls and splashes from deck-to-deck down through
the tower. Air is permitted to pass through the tower horizontally
due to wind currents or is drawn vertically upward 共countercur-
rent兲 to the falling water. In the case of countercurrent towers the
air motion may be due to the natural chimney effect of the warm
moist air in the tower or may be caused by fans at the bottom
共forced draft兲 or at the top 共induced draft兲 of the tower.
A basic theory of cooling tower operation was first proposed by
Walker et al. 关3兴. The practical use of basic differential equations,
however, was first presented by Merkel 关4兴, in which he combined
1
Corresponding author.
HEAT TRANSFER. Manuscript received by the Heat Transfer Division January 11,
2000; revision received January 20, 2001. Associate Editor: V. P. Carey. Fig. 1 Schematic of a wet counter flow cooling tower
mentation of Merkel’s theory in cooling tower design and rating is • negligible heat transfer from the tower fans to air or water
presented and discussed in most unit operations and process heat streams
transfer textbooks. • constant water and dry air specific heats
Webb 关5兴 performed a unified theoretical treatment for thermal • constant heat and mass transfer coefficients throughout the
analysis of cooling towers, evaporative condensers and evapora- tower
tive fluid coolers. In this paper specific calculation procedures are • constant value of Lewis number throughout the tower
outlined for sizing and rating each type of evaporative exchanger. • water lost by drift is negligible
In another paper Webb and Villacres 关6兴 described three computer • uniform temperature throughout the water stream at any cross
algorithms that have been developed to perform rating calcula- section
tions of three evaporatively cooled heat exchangers. The algo- • uniform cross-sectional area of the tower
rithms are particularly useful for rating commercially available
heat exchangers at off-design conditions. The heat and mass trans- From steady-state energy and mass balances on an incremental
fer characteristics of a particular heat exchanger is derived from volume 共refer to Fig. 2兲, the following differential equation may
the manufacturer’s rating data at the design point. be written
Jaber and Webb 关7兴 presented an analysis that shows how the ṁ a dh⫽ṁ w dh f ,w ⫹ṁ a dWh f ,w . (1)
theory of heat exchanger design may be applied to cooling towers.
They demonstrated that the effectiveness 共␧兲 and number of trans- We may also write the water energy balance in terms of the heat-
fer units 共NTU兲 definitions are in very good agreement with those and mass-transfer coefficients, h c and h D , respectively, as
used for the heat exchanger design, and are applicable to all cool- ṁ w dh f ,w ⫽h c A V dV 共 t w ⫺t 兲 ⫹h D A V dV 共 W s,w ⫺W 兲 h f g,w (2)
ing tower operating conditions. It should be noted that Jaber and
Webb 关7兴 did not consider heat-transfer resistance in the air-water and the air side water-vapor mass balance as
interface and the effect of water evaporation on the air-process ṁ a dW⫽h D A V dV 共 W s,w ⫺W 兲 . (3)
states, along the vertical length of the tower. The results are only
applicable to Lewis number equal to one. Furthermore they em- By substitution Lewis number as Le⫽h c /h D c pa in Eq. 共2兲, we
ployed Merkel’s approximation of replacing the sum of, single- obtain
phase heat transfer from the water-air interface to the air and the ṁ w dh f ,w ⫽h D A V dV 关 Lec pa 共 t w ⫺t 兲 ⫹ 共 W s,w ⫺W 兲 h f g,w 兴 . (4)
mass transfer 共evaporation of water兲 at the interface with the en-
thalpy as a driving potential. It is important to note that the accu- It should be noted that we have defined Lewis number in Eq. 共4兲,
racy of results obtained from such an assumption depends upon similar to the definition that is used by Braun et al. 关8兴 and
the method of calculating the moist air properties. Threlkeld 关13–14兴; however, Jaber and Webb 关7兴 and El-
Braun et al. 关8兴 presented effectiveness models for cooling tow- Dessouky et al. 关9兴 have used Le⫽Sc/Pr, commonly used in heat
ers and cooling coils. The models utilize existing effectiveness and mass-transfer literature. In this regard, we prefer to stick to
relationship developed for sensible heat exchangers with modified the notation of Threlkeld 关13–14兴, that is considered as one of the
definitions for number of transfer units and the fluid capacitance standard references in the cooling tower literature. Combining
rate ratio. Results of the models were compared with those of Eqs. 共1兲, 共3兲, and 共4兲, we get
more detailed numerical solutions to the basic heat and mass
dh 共 t w ⫺t 兲
transfer equations and experimental data. They also did not con- ⫽Le c pa ⫹h g,w . (5)
sider the effect of air-water interface temperature; however, they dW 共 W s,w ⫺W 兲
did consider the effect of water evaporation on the air process Using the approximation of constant c pa , we have
states along the vertical length of the tower. The results are only
presented for Lewis number equal to unity.
Dessouky et al. 关9兴 presented a solution for the steady-state
counter-flow wet cooling tower with new definitions of tower ef-
fectiveness and number of transfer units. Their model is essen-
tially a modified version of Jaber and Webb’s model with the
inclusion of Lewis number, which appears as a multiplication fac-
tor to the enthalpy driving potential. They did consider the effect
of interface temperature and Lewis number, however, the effect of
water evaporation on the air process states, along the vertical
length is not considered. Furthermore they used an approximate
equation for calculating the moist air enthalpy, which was ob-
tained by curve fitting the tabulated thermodynamic properties of
saturated air-water vapor mixture. It is important to note that the
calculation of moist air properties should be accurate to obtain
reliable results. The objective of this paper is to consider together
the effects of Lewis number, the heat-transfer resistance in the
air-water interface, and the effect of water evaporation on the air
process states, along the vertical length of the tower. In this regard
we have used reliable air-water thermodynamic property equa-
tions that are developed by Hyland and Wexler 关10–11兴.
Mathematical Formulation
A schematic of a counterflow-cooling tower showing the im-
portant states is given in Fig. 2. The major assumptions that are
used to derive the basic modeling equations may be summarized
as 关12–14兴:
• heat and mass-transfer in a direction normal to the flows only
• negligible heat and mass transfer through the tower walls to Fig. 2 Mass and energy balance of a wet counter flow cooling
the environment tower
The above equation can be used for obtaining the interface tem-
perature. However for large values of E the interface and bulk
water temperature are equal.
Detail Model
A computer program is written for solving Eqs. 共1兲–共6兲 numeri-
cally. In this program properties of air-water vapor mixture and
moist air are needed at each step of numerical calculation. These
properties are obtained from the property equations given in Hy-
land and Wexler 关10–11兴, which are also used by ASHRAE 关2兴 in
computing air-water vapor thermodynamic properties. The pro-
gram gives the dry-bulb temperature, wet-bulb temperature of air,
water temperature and humidity ratio of air at each step of nu-
merical calculation starting from air-inlet to air-outlet values. If
the value of (h D A V ) is known, the required tower volume may be
obtained by using 关13–14兴
Fig. 3 Water operating line on enthalpy-temperature diagram

V⫽
ṁ a
h DA V 冕 Wi
Wo dW
W s,w ⫺W
. (10)
indicating the effect of tie line „ E ÄÀ h c , w Õ h D … on saturated The integral in the above equation is solved numerically. The
moist air enthalpy number of transfer units of the tower are calculated by
NTU⫽h D A V V/ṁ a ⫽ 冕Wi

Wo dW
W s,w ⫺W
. (11)
h s,w ⫺h⫽c pa 共 t w ⫺t 兲 ⫹h 0g 共 W s,w ⫺W 兲 . (6)
The cooling tower effectiveness 共␧兲 is defined as the ratio of ac-
Equation 共5兲 may then be written as tual energy to the maximum possible energy transfer.
dh 共 h s,w ⫺h 兲 h o ⫺h i
⫽Le ⫹ 共 h g,w ⫺h 0g Le兲 . (7) ␧⫽ (12)
dW 共 W s,w ⫺W 兲 h s,w,i ⫺h i
It should be noted that Eq. 共7兲 describes the condition line on General correlations for heat and mass transfer of cooling towers
the psychometric chart for the changes in state for moist air pass- in terms of physical characteristics do not exist. It is usually nec-
ing through the tower. For given water temperatures (t w,i , t w,o ), essary to correlate the tower performance data for specific tower
Lewis number 共Le兲, inlet condition of air and mass flow rates, designs. Mass transfer data are typically correlated in the form 关1兴
冉冊
Eqs. 共1兲 and 共6兲 may be solved numerically for exit conditions of n
h DA VV ṁ w
both the air and water stream. The solution is iterative with re- ⫽c , (13)
spect to the air humidity ratio and temperatures 共W, t, and t w 兲. At ṁ w ṁ a
each iteration, Eqs. 共1兲–共6兲 can be integrated numerically over the where c and n are empirical constants specific to a particular
entire tower volume from the air inlet to outlet by a procedure tower design. Multiplying both sides of the above equation by
similar to that described in references 关12–14兴. Initially inlet con- (ṁ w /ṁ a ) and considering the definition for NTU 共refer to Eq.
dition of air and the condition of saturated air in equilibrium with 共11兲兲 gives the empirical value of NTU as
冉冊
leaving water temperature are known. The quantity dh/dW is cal-
n⫹1
culated by using Eq. 共7兲. Then in the second step enthalpy of inlet ṁ w
air is increased by a small amount (h i ⫹⌬h). The resulting incre- NTUem ⫽c . (14)
ṁ a
mental change in water temperature is calculated by the equation:
⌬t w ⫽⫺(ṁ a /ṁ w c w )(⌬h⫺Wh f ,w ). Next, the state of saturated air The coefficients c and n of the above equation were fit to the
in equilibrium with water temperature is calculated at this in- measurements of Simpson and Sherwood 关15兴 for four different
creased water temperature. The procedure is repeated until we get tower designs over a range of performance conditions by Braun
the water temperature equal to the water inlet temperature. et al. 关8兴. The experimental and empirical values were compared
In deriving Eqs. 共1兲–共6兲, it was assumed that there is no resis- with the values obtained by our detail model, and the results
tance to heat flow in the interface between air and water. In other are shown in Table 1. As can be seen from the Table, the calcu-
words, the interface temperature was assumed to be equal to the lated and empirical values of NTU are well within the acceptable
bulk water temperature. However, for heat transfer to take place limits. Also the wet-bulb temperature of outlet air (t wb,o ), calcu-
between air and water, the temperature of the interface film should lated from the present model is compared with the experimental
be less than the bulk water temperature, as shown in Fig. 3. In that values reported in Simpson and Sherwood 关15兴, the two values are
case all the terms in Eqs. 共1兲–共6兲 with the subscripts (s,w) will be very close to each other 共within ⫾0.6 percent兲. Therefore, the
replaced by (s,int). Webb 关5兴 assumed that t w is nearly equal to
(t int⫹0.5).
Figure 3 shows both enthalpies of the saturated air-water vapor Table 1 Experimental and calculated values of cooling tower
parameters
mixture and tower operating line as a function of water tempera-
ture. Considering the short distance between h s,w and h s,int on the
saturation curve as a straight line, following simple relationship
can be easily deduced 关9兴
h s,w ⫺h⫽h s,w ⫺h s,int⫹E 共 t w ⫺t int兲 , (8)
where E is slope of the tie line and is constant for a given cooling
tower. This slope is given by
E⫽⫺h c,w /h D . (9)
model developed in this paper can be used to either design or rate
counter-current wet cooling towers. A classical problem of de-
signing a cooling tower is to calculate the required tower volume
共or NTU兲, given the type of packing, the wet bulb temperature of
the entering air, water inlet temperature, water outlet temperature
and mass-flow rates of water and air. On the other hand, in a
typical rating problem the exit cooled water temperature is ob-
tained given volume of the tower 共or NTU兲, water inlet tempera-
ture, inlet air dry-bulb temperature, inlet air wet bulb temperature
and the mass flow rates of water and air. The heat and mass
transfer coefficients h c , h c,w , and h D , required for calculating Le
and NTU can be determined experimentally. Webb 关5兴 have re-
ported experimental values of the above coefficients, as functions
of Reynolds number and packing parameters. The designer of
cooling towers should determine experimentally, the above coef-
ficients based on the given packing geometry.
Design of Cooling Towers

Effect of Le and E. In the literature, Merkel’s model is
mostly used for design and rating calculations of cooling towers.
However, the difference in values of NTU and ␧ predicted by
Merkel’s model and detail thermodynamic model is significant for
different values of E and Le. Figure 4 is a plot between NTU and
E for different values of Le. The plot is drawn for the following
set of input data which is also used by Dessouky et al. 关9兴: t w,i
⫽35.0°C, t w,o ⫽30.0°C, h i ⫽76.6 kJ/kg, and ṁ w /ṁ o ⫽1.00. The
figure shows that for Le⫽1.0 both the detail model and the Des-
souky’s model are predicting about the same value of NTU. How-
ever for Le equal to 0.87 and 1.2, there is an appreciable differ-
ence between the two models. Figures 5共a兲 and 5共b兲 are plots
between the effectiveness 共␧兲 and the non-dimensional tempera-
ture ratio 共R兲 defined as
Fig. 5 Comparisons of effectiveness ␧ with temperature ratio
t w,i ⫺t w,o R by varying the water-outlet temperature t w , o for different val-
R⫽ (15) ues of mass-flow rate ratios. Slope of the tie line E Ä11.1 „Fig.
t w,i ⫺t wb,i 5„a……, and E Äⴥ „Fig. 5„b…….
for different values of the mass flow rate ratio (ṁ w /ṁ a ). The plot
is generated for the following set of input conditions which
are also considered in the experimental work of Simpson and the bulk water temperature. Clearly for lower values of E, Des-
Sherwood 关15兴: t db,i ⫽29.0°C, t wb,i ⫽21.11°C, t w,i ⫽28.72°C, souky’s model and detail model predict different values of ␧,
Le⫽0.9 and h D A V ⫽3.025 kg/s•m3. In this figure, R is varied by however, for E⫽⬁, as expected, both the models predict same
varying the value of t w,o from t w,i to t wb,i . It should be noted that values of ␧. Table 2 summarizes the values of NTU and ␧ for
as the value of t w,o decreases from t w,i to t wb,i the value of R different values of Lewis number Le and E, obtained from de-
tailed model, Braun’s model, Dessouky’s model and Jaber 共or
increases from 0.0 to 1.0. The dashed lines in both the figures Merkel’s兲 model. It should be noted that Jaber’s model with Le
represent Dessouky’s model. Figure 5共a兲 is drawn for E⫽11.1, ⫽1.0 is equivalent to Merkel’s model. The values given in the
which is used by Baker and Shryock 关16兴, while Fig. 5共b兲 is drawn table is obtained by using the input data also shown on the table.
with E⫽⬁ that corresponds to the interface temperature equal to As can be seen from the table that some results are fairly close in
comparison and others are quite different.
Effect of Water Outlet Temperature. Figure 6共a兲 is a plot
between the effectiveness and non-dimensional temperature ratio
R, for different values of the mass flow rate ratio (ṁ w /ṁ a ). The
input data used in this figure is same as that of Fig. 5. As ex-
pected, the figure shows that as the value of t w,o decreases, the
effectiveness of the tower ␧ increases. This increase of ␧ with R is
higher for larger mass flow rate ratios. Figure 6共b兲 is the corre-
sponding plot between R and air-approach temperature that is de-
fined as the temperature difference between the water inlet tem-
perature and wet bulb temperature of outlet air (t w,i ⫺t wb,o ). It is
important to note that this difference in temperature is a measure
of the closeness to saturation condition of the cooling tower. For
example, lower the temperature difference, the higher will be ef-
fectiveness and vice versa. When this temperature difference is
equal to zero, the effectiveness of the tower is equal to 1.0. This
figure clearly demonstrates the reason why for higher mass flow
rate ratios, the effectiveness of the tower is approaching unity for
values of R less than 1.0, as is demonstrated in the case of
ṁ w /ṁ a ⫽2.0. However for ṁ w /ṁ a ⫽0.5, the tower never reaches
Fig. 4 Comparisons of NTU with slope of tie line E Ä the saturation point. Figure 7 shows the plot between NTU and R
À h c , w Õ h D for different values of Lewis number Le for the above set of input data. The figure shows that for
Table 2 Values of NTU and ⑀ obtained from different models
for various values of Le and E , –design calculations*
Fig. 7 Comparisons of NTU with temperature ratio R by vary-

ing water-outlet temperature t w , o
ṁ w /ṁ a ⫽2.0, the increase in NTU is less for R⭐0.4; however,

beyond this point the value of NTU increases rapidly. This implies
that to obtain lower values of t w,o , a large volume of cooling
tower is required. It should be noted that NTU represents the size
of cooling towers. Also Fig. 6共a兲 shows that for every value of R,
there is one value of ṁ w /ṁ a at which the effectiveness of the
tower is equal to 1.0. This value of (ṁ w /ṁ a ) max is defined as the
maximum possible value of mass flow rate ratio, any other value
greater than this value will not increase the tower effectiveness.
Figure 8 is the plot between (ṁ w /ṁ a ) max and R for the above set
of input data. This figure shows that for lower values of R, the
maximum mass flow rate ratio required to obtain ␧⫽1.0 is very
high. However, for R⭓0.1, these values of mass flow rate ratio
decrease rapidly. This can be explained from the fact that for a
particular mass flow rate ratio, values of ␧ increase linearly with
R, 共refer Fig. 6共a兲兲. It should also be noted that the values of
(ṁ w /ṁ a ) max are independent of Le for all values of R. This can be
explained due to the fact that ␧ of the tower depends on the inlet
dry and wet bulb temperatures of air and the water inlet and outlet
Fig. 6 Comparisons of effectiveness ␧ with temperature ratio

R by varying water-outlet temperature t w , o „Fig. 6„a……, and air Fig. 8 Variations of maximum mass-flow rate ratio with tem-
approach temperature „ t w , i À t wb , o … with R „Fig. 6„b……. perature ratio R by varying the water-outlet temperature t w , o
Fig. 11 Variations of maximum mass-flow rate ratio with tem-
perature ratio R by varying the air inlet wet-bulb temperature
t wb , i
comparing Fig. 6共a兲 with Fig. 9共a兲, we note that for same mass
flow rate ratios, the tower effectiveness increases rapidly with
t w,o . Figure 9共b兲 is the plot between (t w,i ⫺t wb,o ) and R. As ex-
plained earlier in Fig. 6, Fig. 9共b兲 demonstrates the reason why for
higher mass flow rate ratio ṁ w /ṁ a ⫽2.0, the effectiveness of the
tower is equal to unity for R⭓0.45. This explains the behavior of
Fig. 9 Comparisons of effectiveness ␧ with temperature ratio Fig. 9共a兲. Figure 10 shows the plot between NTU and R, while
R by varying air inlet wet-bulb temperature t wb , i „Fig. 9„a……, and Fig. 11 is the plot between (ṁ w /ṁ a ) max and R for the above set of
air approach temperature „ t w , i À t wb , o … with R „Fig. 9„b…… input data. These figures show a behavior similar to those of Figs.
7–8, with only different numbers, indicating the influence of inlet
air wet bulb temperature. Similar to Fig. 8, the values of
temperatures. Because these temperatures are input for designing (ṁ w /ṁ a ) max are independent of Le for all values of R due to the
cooling towers, the effectiveness will be constant and will not reasons discussed earlier.
vary with Le.
Effect of Inlet Air Wet Bulb Temperature. Figure 9共a兲 is a Rating of Cooling Towers
plot between effectiveness and temperature ratio R for different
Table 3 summarizes the values of t w,o and ␧ for different values
values of (ṁ w /ṁ a ). These plots are generated for the following
of Lewis number Le and E, obtained from the detail model,
set of input data: t db,i ⫽29.0°C, t w,o ⫽24.22°C, t w, j ⫽28.72°C,
Braun’s model, Dessouky’s model and Jaber 共or Merkel’s兲 model.
and h D A V ⫽3.025 kg/s•m3. We emphasize that inlet air wet bulb As can be seen from this table that there is no appreciable differ-
temperature (t wb,i ) is varied from 10.56°C to t w,o to produce these ence in the water outlet temperatures predicted by various models.
plots. In this case, as expected, the tower effectiveness increases
with R and this increase is high for higher mass flow rates. On Effect of Inlet Air Wet Bulb Temperature. Figure 12共a兲
is the plot between the tower effectiveness 共␧兲 and inlet air wet
bulb temperature (t wb,i ) for different values of mass flow rate
ratios. This figure is generated for the following set of input
data: t db,i ⫽29.0°C, t w,i ⫽28.72°C, ṁ a ⫽1.187 kg/s, h D A V
⫽3.025 kg/s•m3, and V⫽0.697 m3. Figure 12共b兲 is the corre-
sponding plot between (t w,i ⫺t wb,o ) versus t wb,i . In these plots,
the values of t wb,i was varied from 11 to 27°C. It is interesting to
note that for the values of t wb,i considered, the tower effectiveness
remain almost constant and there is no appreciable change in
(t w,i ⫺t wb,o ) with ṁ w /ṁ a , particularly at high values of t wb,i .
Figure 13 is the plot between water outlet temperature (t w,o ) and
inlet air wet bulb temperature (t wb,i ) for the above set of input
data. This figure shows that t w,o changes by about 12.0°C for the
range of t wb,i considered in the figure, i.e., for the case of
ṁ w /ṁ a ⫽0.5. However as the mass flow rate ratio increases, the
increase in t w,o decreases, as can be seen for ṁ w /ṁ a ⫽0.60 and
⫽0.75 for which the increase in t w,o are 11 and 9°C, respectively.
Effect of Water Inlet Temperature. Figure 14共a兲 is the plot
between the tower effectiveness 共␧兲 and the water inlet tem-
perature (t w,i ) for different values of mass flow rate ratios. This
Fig. 10 Comparisons of NTU with temperature ratio R by vary- figure is generated for the following set of input data: t db,i
ing air inlet wet-bulb temperature t w , b , i ⫽29.0°C, t wb,i ⫽21.1°C ṁ a ⫽1.187 kg/s, h D A V ⫽3.025 kg/s•m3,
Table 3 Values of t w , o and ⑀ obtained from different models
for various values of Le and E , –rating calculations*
Fig. 13 Comparisons of water outlet temperature t w , o with in-

let air wet bulb temperature t wb , i
and V⫽0.697 m3. Figure 14共b兲 is the corresponding plot between

water outlet temperature (t w,o ) and water inlet temperature (t w,i ).
It should be noted that there is an appreciable change in effective-
ness of the cooling tower for the range of t w,i considered, unlike
Fig. 12共a兲 where there is no significant change in ␧ for the range
of t wb,i considered. Thus, indicating that changes in water-inlet
temperature has relatively more effect on ␧ of the cooling tower,
compared to changes in t wb,i for the same tower configuration.
Figure 14共b兲 shows that the change in t w,o for the range of t w,i
considered, decreases with the decrease in mass-flow rate ratios.
Also, for a particular mass flow rate ratio this change in t w,o is less
Fig. 12 Comparisons of effectiveness ␧ with inlet air wet bulb Fig. 14 Comparisons of effectiveness ␧ with water inlet tem-
temperature t wb , i „Fig. 12„a……, and air approach temperature perature t w , i „Fig. 14„a……, and water outlet temperature t w , o with
„ t w , i À t wb , o … with R „Fig. 12„b…… t w , i „Fig. 14„b…….
perature. It is also demonstrated that the inlet air dry-bulb tem-
perature has a negligible effect on the tower performance, which
is typical of wet cooling tower.
Acknowledgments
The authors acknowledge the support provided by King Fahd
University of Petroleum and Minerals for this research project.
Nomenclature
A V ⫽ surface area of water droplets per unit volume of the
tower, m2/m3
c pa ⫽ specific heat at constant pressure of moist air,
kJ/kga •K
c w ⫽ specific heat of water, kJ/kgw •K
E ⫽ slope of the ‘‘tie’’ line, kJ/kgw •K
h ⫽ enthalpy of moist air, kJ/kga
h c ⫽ convective heat-transfer coefficient of air, kW/m2•K
h c,w ⫽ convective heat-transfer coefficient of water, kW/m2•K
h D ⫽ convective mass-transfer coefficient, kgw /m2 •s
h f ⫽ specific enthalpy of saturated liquid water, kJ/kgw
h f ,w ⫽ specific enthalpy of water evaluated at t w , kJ/kgw
h g ⫽ specific enthalpy of saturated water vapor, kJ/kgw
h 0g ⫽ specific enthalpy of saturated water vapor evaluated at
0°C, kJ/kgw
h f g,w ⫽ change-of-phase enthalpy (h f g,w ⫽h g,w ⫺h f ,w ), kJ/kgw
h s,w ⫽ enthalpy of saturated moist air evaluated at t w , kJ/kga
h s,int ⫽ enthalpy of saturated moist air evaluated at t int , kJ/kga
Le ⫽ Lewis number (Le⫽h c /h D c p,a )
ṁ a ⫽ mass flow rate of dry air, kga /s
Fig. 15 Comparisons of effectiveness ␧ with inlet air dry bulb ṁ w ⫽ mass flow rate of water, kgw /s
temperature t db , i „Fig. 15„a……, and water outlet temperature t w , o NTU ⫽ number of transfer units
with t db , i „Fig. 15„b…… Pr ⫽ Prandtl number
R ⫽ dimensionless temperature range 共refer to Eq. 共15兲兲
Sc ⫽ Schmidt number
compared to the corresponding change shown in Fig. 13, implying t ⫽ dry-bulb temperature of moist air, °C
that changes in t wb,i has relatively more effect on t w,o compared to t int ⫽ air-water interface temperature, °C
changes in t w,i for the same tower. t w ⫽ water temperature, °C
V ⫽ volume of tower, m3
W ⫽ humidity ratio of moist air, kgw /kga
Effect of Inlet Air Dry Bulb Temperature W s,w ⫽ humidity ratio of saturated moist air evaluated at t w ,
Figure 15共a兲 is the plot between the tower effectiveness 共␧兲 and kgw /kga
inlet air dry-bulb temperature (t db,i ) for different values of mass ␧ ⫽ effectiveness
flow rate ratios, while Fig. 15共b兲 is the plot between water outlet Subscripts
temperature (t w,o ) and inlet air dry-bulb temperature (t db,i ).
These plots are drawn for the following set of input data: a ⫽ moist air
t w,i ⫽28.72°C, t wb,i ⫽21.1°C ṁ a ⫽1.187 kg/s, h D A V ⫽3.025 db ⫽ dry-bulb
kg/s•m3, and V⫽0.697 m3. These curves, as expected, demon- em ⫽ empirical
strate that there is a negligible effect of changes in inlet air dry g,w ⫽ vapor at water temperature
bulb temperature on the performance of cooling towers, which is i ⫽ inlet
a typical characteristic of wet-cooling towers 关1兴. int ⫽ air-water interface
max ⫽ maximum
o ⫽ outlet
Conclusions s,w ⫽ saturated moist air at water temperature
A more realistic detail model for the steady-state operation of a w ⫽ water
counter flow wet cooling tower is described with respect to ex- w,i ⫽ water inlet
ample problems. In this model, we have considered the effect of w,o ⫽ water outlet
water evaporation on the air process states, the resistance of heat wb,i ⫽ wet-bulb inlet
transfer in the water film and the non-unity of the Lewis number. wb,o ⫽ wet-bulb outlet
The data obtained from the application of the model showed that Superscripts
a substantial error can be made when the resistance to heat trans- cal ⫽ calculated
fer in the water film is neglected and the Lewis number is consid- exp ⫽ experimental
ered to be unity. However, the magnitude of errors in calculating
tower thermal parameters is a strong function of the ratio between
the heat and mass transfer coefficients that are used in the calcu- References
lations. The design and rating analyses of the wet cooling tower, 关1兴 ASHRAE, 1983, ASHRAE Equipment Guide, chap. 3, American Society of
Heating, Refrigerating, and Air-Conditioning Engineers, Inc., Atlanta, GA.
which is considered in the experimental work of Simpson and 关2兴 ASHRAE, 1989, ASHRAE Handbook of Fundamentals, American Society of
Sherwood 关15兴 are explained in more detail. The variation of Heating, Refrigerating, and Air-Conditioning Engineers, Inc., Atlanta, GA.
tower effectiveness is investigated in terms of air approach tem- 关3兴 Walker, W. H., Lewis, W. K., McAdams, W. H., and Gilliland, E. R., 1923,
Principles of Chemical Engineering, 3rd ed., McGraw-Hill Inc., New York. Properties of the Saturated Phases of H2O from 173.15 to 473.15 K,’’
关4兴 Merkel, F., 1925, ‘‘Verdunstungshuhlung,’’ Zeitschrift des Vereines Deutscher ASHRAE Trans., 89, No. 2, pp. 500–519.
Ingenieure (V.D.I.), 70, pp. 123–128. 关11兴 Hyland, R. W., and Wexler, A., 1983, ‘‘Formulations for the Thermodynamic
关5兴 Webb, R. L., 1984, ‘‘A Unified Theoretical Treatment for Thermal Analysis of Properties of Dry Air from 173.15 to 473.15 K, and of Saturated Moist Air
Cooling Towers, Evaporative Condensers, and Fluid Coolers,’’ ASHRAE from 173.15 to 372.15 K, at Pressure to 5 MPa,’’ ASHRAE Trans., 89, No. 2,
Trans., 90, No. 2, pp. 398–415. pp. 520–535.
关6兴 Webb, R. L., and Villacres, A., 1984, ‘‘Algorithms for Performance Simula-
关12兴 Myers, R. J., 1967, ‘‘The Effect of Dehumidification on the Air Side Heat
tion of Cooling Towers, Evaporative Condensers, and Fluid Coolers,’’
Transfer Coefficient for a Finned-Tube Coil,’’ M.Sc. thesis, University of Min-
ASHRAE Trans., 90, No. 2, pp. 416–458.
nesota, Minneapolis, MN.
关7兴 Jaber, H., and Webb, R. L., 1989, ‘‘Design of Cooling Towers by the
Effectiveness-NTU Method,’’ ASME J. Heat Transfer, 111, No. 4, pp. 837– 关13兴 Threlkeld, J. L., 1972, Thermal Environmental Engineering, 2nd ed. Prentice-
843. Hall Inc., New Jersey.
关8兴 Braun, J. E., Klein, S. A., and Mitchell, J. W., 1989, ‘‘Effectiveness Models 关14兴 Kuehn, T. H., Ramsey, J. W., and Threlkeld, J. L. 1998, Thermal Environmen-
for Cooling Towers and Cooling Coils,’’ ASHRAE Trans., 95, No. 2, pp. tal Engineering, 3rd ed. Prentice-Hall Inc., New Jersey.
164–174. 关15兴 Simpson, W. M., and Sherwood, T. K., 1946, ‘‘Performance of Small Me-
关9兴 El-Dessouky, H. T. A., Al-Haddad, A., and Al-Juwayhel, F., 1997, ‘‘A Modi- chanical Draft Cooling Towers,’’ Refrigerating Engineering, 52共6兲, pp. 525–
fied Analysis of Counter flow Cooling Towers,’’ ASME J. Heat Transfer, 119, 543 and 574–576.
No. 3, pp. 617–626. 关16兴 Baker, D. R., and Shryock, H. A., 1961, ‘‘A Comprehensive Approach to the
关10兴 Hyland, R. W., and Wexler, A., 1983, ‘‘Formulations for the Thermodynamic Analysis of Cooling Tower Performance,’’ Trans. ASME, 83, pp. 339–349.
Heat Transfer Enhancement to the
Peiwen Li
Industrial Technology Researcher
Mechanical Engineering Laboratory AIST, MITI,
Drag-Reducing Flow of Surfactant
Japan
The Energy Conservation Center, Japan Solution in Two-Dimensional
Yasuo Kawaguchi
Senior Researcher
Channel With Mesh-Screen
Mechanical Engineering Laboratory AIST, MITI,
Japan
Namiki 1-2, Tsukuba, Ibaraki, Japan
Inserts at the Inlet
e-mail: m4050@mel.go.jp
The heat transfer enhancement of drag-reducing flow of high Reynolds number in a
two-dimensional channel by utilizing the characteristic of fluid was studied. As the net-
Hisashi Daisaka works of rod-like micelles in surfactant solution are responsible for suppressing the
Graduate Student turbulence in drag-reducing flow, destruction of the structure of networks was considered
Keio University, Japan to eliminate the drag reduction and prevent heat transfer deterioration. By inserting wire
mesh in the channel against the flow, the drag-reducing function of the micellar structure
Akira Yabe in surfactant aqueous solution was successfully switched off. With the Reynolds number
Director of Research Planning Office close to the first critical Reynolds number, the heat transfer coefficient in the region
Mechanical Engineering Laboratory AIST, MITI, downstream of the mesh can be improved significantly, reaching the same level as that of
Japan water. The region with turbulent heat transfer downstream of the mesh becomes smaller
as the concentration of surfactant in the solution increases. Three types of mesh of
Koichi Hishida different wire diameter and opening space were evaluated for their effect in promoting
Professor heat transfer and the corresponding pressure loss due to blockage of the mesh. The
Keio University, Japan turbulent intensities were measured downstream from the mesh by using a Laser Doppler
Velocimetry (LDV) system. The results indicated that the success of heat transfer en-
Masanobu Maeda hancement is due to the strong turbulence promoted by the mesh which destroys the
Professor network of rod-like micelles by applying high shear stress and thus relaxing the shear
Keio University, Japan induced state (SIS). 关DOI: 10.1115/1.1370518兴
Keywords: Channel Flow, Enhancement, Fluids, Heat Transfer, Non-Newtonian
1 Introduction transfer enhancement mechanism of surfactant solution, and there

is no generally accepted efficient way to enhance the heat transfer
Toms 关1兴 and Mysels 关2兴 independently identified that adding
of drag-reducing surfactant solution. Therefore, the present ex-
small amounts of polymers or surfactants can result in a signifi-
perimental work attempted to find an effective way of enhancing
cant reduction of the friction drag of turbulent flow. Due to such
the heat transfer for drag-reducing flow of surfactant solution.
reduction of frictional drag of turbulent flow, the pumping power
Of the methods described in the literature to enhance the heat
consumed for transporting fluid can be reduced significantly. The
behavior of friction factor for various aqueous solutions of poly- transfer of surfactant solution, the conventional way tried by Sato
mers and surfactants has been investigated extensively in the last et al. 关14兴 to modify the heat transfer surface works reasonably
fifty years 关3–7兴. Among various applications, the addition of well, but with a large penalty of pressure loss. In industrial appli-
drag-reducing additives to the circulating water of district heating cations, minimal modification of the existing heat exchangers is
and cooling 共DHC兲 systems is a promising one. This is because important from economic considerations if surfactant additives are
the solution is enclosed in the circulation system and there is no to be used in DHC systems. In this study, devices were set in front
loss of the additives, thus protecting the environment from con- of the heat transfer section against the flow to destroy the micellar
tamination by the additives. In such cases, polymers are less use- structures and so eliminate their drag-reducing effect and enhance
ful due to serious and irrecoverable degradation of the drag reduc- heat transfer, since micellar structures in surfactant solution are
tion effect caused by mechanical stress at the circulation pump. responsible for the suppression of turbulence 关15,16兴. We selected
Surfactants however, are less affected by mechanical forces to a mesh as the device because of its simple geometry and low cost.
their drag reducing ability and have been widely accepted as drag- In the work of Gasljevic and Matthys 关17兴, a plug of laminated
reducing additives in DHC systems 关8–10兴. mesh was expected to flatten the velocity profiles in a pipe flow to
Because of the suppression of turbulence in the drag-reducing achieve a higher heat transfer coefficient. Because of enormous
flow, the heat transfer is also reduced significantly and even more pressure loss and limited heat transfer enhancement, they con-
seriously than that of friction factor, which thus adversely affects cluded that intensive local energy dissipation might not be a vi-
the performance of heat exchangers in DHC systems 关11,12兴. This able means of heat transfer augmentation in heat exchangers.
is unacceptable because the serious reduction of heat transfer may However, the turbulence characteristic downstream of the plug
lower the exergy efficiency of the district heating and cooling was not discussed in their work. In view of the mechanism of
system even if there is conservation of energy from drag reduction turbulence suppression in drag-reducing flow, the production of
关13兴. However, there have not been sufficient studies on the heat turbulence should be an important issue in the study of heat trans-
fer enhancement. Therefore, it is worth studying how the mesh
size, surfactant concentration and Reynolds number affect the heat
HEAT TRANSFER. Manuscript received by the Heat Transfer Division October 26, transfer enhancing effect as this has not been sufficiently investi-
1999; revision received November 20, 2000. Associate Editor: S. S. Sadhal. gated elsewhere. It is also important to know the turbulence quan-
tities in the downstream region of the mesh plug to understand the reducing region because we are interested in transportation of
mechanism of promoting turbulence in drag-reducing flow of sur- fluid under the condition of drag reduction and elimination of this
factant solution. condition in heat exchangers.
It has been revealed that the observed low value of friction of
drag-reducing flow is due to the suppression of turbulence and
reduction of Reynolds’ shear stress 关5兴 in the flow by adding 2 Experimental Apparatus and Procedure
surfactant. Therefore, it is crucial to restore turbulence in the flow A schematic view of the circulation system used in the present
to enhance the heat transfer of drag-reducing flow. It has been study is shown in Fig. 1. A tank with a volume of 2 m3 served as
often stated in the literature and is generally accepted that the the reservoir for the surfactant aqueous solution. Because the tem-
rod-like 共or worm-like兲 micelles’ networks make the surfactant perature of the surfactant solution influences the critical Reynolds
solution a viscoelastic fluid and cause the turbulence in the flow to number sensitively 关24兴, a thermostat was installed to control the
be suppressed at high Reynolds number 关18兴. The rod-like mi- temperature of the solution in the tank through heating by an
celles were reported 关19–22兴 to exist at conditions above a critical electric heater and cooling by tap water in a coil. To ensure uni-
micelle concentration 共CMC兲 in surfactant solutions and to aggre- formity of temperature in the tank, an agitator with rotating im-
gate as networks or super structures under the shear induced state peller was installed. The temperature fluctuation of surfactant so-
共SIS兲关15,23兴. Conversely, relaxation of the shear induced state lution in the tank was thus controlled to within 0.1°C around the
共SIS兲 and destruction of the rod-like micelles’ networks may be an set value.
efficient way to promote turbulence in the drag-reducing flow at The two-dimensional water channel was 6 meters in length,
high Reynolds number. The degradation of drag reduction at very which was connected properly by three sections of 2 meters each.
high Reynolds number beyond a critical Reynolds number is a The flow cross-section was 500 mm in spanwise width and 40 mm
good example, in which the large wall shear stress over the critical in height 共H兲. One section served as a heat transfer section. The
wall shear stress is crucial, which is the threshold at which the channel walls were of 20 mm thickness and made of transparent
shear induced state starts to relax. Nevertheless, because the sur- acrylic resin except the heating surface, to allow the Laser Dop-
factant micellar solution has a non-Newtonian feature, the special pler Velocimetry 共LDV兲 system to be used to measure the two-
mechanism of suppressing turbulence by using surfactant addi- dimensional velocity components. A settling chamber equipped
tives in high Reynolds number flow should be considered when with a distributor and a quarter-circular contraction nozzle was
studying how to enhance the heat transfer of the drag-reducing used to provide a uniform velocity profile. A honeycomb rectifier
flow. of 150 mm length was installed at the entrance to remove large
When surfactant solution flows against the wire mesh, the fluid eddies. A collection chamber with a diffuser connected to the
element passing through the opening may experience a short path outlet of the channel was installed to avoid flow instability caused
with high shear stress, which can result in the destruction of the by flow separation at the channel outlet. The flow from the col-
rod-like micelles’ networks. Then, as the fluid temporarily be- lection chamber flowed through an electro-magnetic flow meter
comes a Newtonian fluid, high Reynolds number flow naturally and returned to the reservoir through plastic pipes. The resolution
becomes turbulent flow and we can expect high heat transfer com- of the flow meter was 0.01 m3/min. Pressure taps were set on the
parable to that of water without drag reducers. Three types of fine mid-line of the bottom wall.
wire mesh were used in the present experimental investigation to The heating surface was 1.6 meters in length located 4.3 meters
destroy the micellar structure. The mesh plug is expected to pro- downstream from the entrance of the channel. A stainless steel
mote turbulence downstream within a sufficiently long region, and film heater was attached to the inner side of the bottom wall. The
therefore the heat transfer surface of existing heat exchangers electric current and voltage through the film heater were measured
need not be modified. In other words, one may treat the flow to calculate the uniform wall heat flux. Under the film heater,
downstream of the mesh as normal turbulent flow within a certain thermocouples were set on the mid-line of the channel span to
region. measure the streamwise distribution of local temperature of the
As the factors affecting the formation and destruction of the heating wall. Because the film thickness was 20 ␮m, the tempera-
rod-like micelles’ networks, the effects of surfactant concentra- ture difference on the film’s thickness was ignored. All the experi-
tion, mesh dimensions, the Reynolds number or velocity of the ments were done at a room conditioning temperature of 30°C,
flow, the critical Reynolds number or critical wall shear stress,
etc. on heat transfer are investigated. The turbulence quantities
downstream of the mesh plug were also measured to provide im-
portant information about the variation of turbulent structures in
order to clarify the heat transfer enhancing mechanism of drag-
reducing surfactant solution of high Reynolds number. Even if the
rod-like micelles may temporarily be broken up by mesh inser-
tions, they still tend to aggregate as networks under SIS down-
stream of the mesh. This process may have a significant effect on
the turbulence variation downstream of the mesh, therefore, mea-
surement of turbulence quantities in the region downstream of the
mesh is of significance for studying the formation of networks of
the rod-like micelles.
Using three types of mesh, we set out to select the optimum one
with the greatest heat transfer enhancement at minimum pressure
loss caused by the mesh plug. Therefore, the pressure loss caused
by the mesh plug was also investigated in the study.
Note that the critical Reynolds number for drag-reducing flow
is an onset value such that when the flow Reynolds number ex-
Fig. 1 Schematic of experiment facility „Notations:
ceeds it, the drag-reducing ability starts to degrade gradually till it
1-2Dchannel; 2-mesh plug; 3-Optical window for LDV laser
completely disappears at a larger Reynolds number. A flow with shoot; 4-Pressure transducer; 5-Filter; 6-Contractor;
Reynolds number not larger than the critical Reynolds number is 7-Deffuser; 8-Flow meter; 9-Tank; 10-Agitator; 11-Cooling coil;
called drag-reducing flow 共DR flow兲. The Reynolds numbers in- 12-Heater; 13-Thermometer; 14-Thermostat; 15-Pump; 16-
vestigated in this study are limited to those within the drag- Valve…
Table 1 The dimensions of wire mesh qw
h⫽ (3)
共 T w ⫺T f 兲
When the channel was asymmetrically heated, the Nusselt num-

ber may be slightly lower than that of the symmetrically heated
channel. The effect of asymmetric heating decreases with an in-
creasing Prandtl number. For water of asymmetric heating, Nus-
selt number was reported to be 10 percent lower than the symmet-
ric heating case in the literature 关25–27兴.
The properties of the surfactant solution are based on the mean
temperature of the flow. It is a common practice to use the
which was close to the inlet temperature of the test section. The thermo-physical properties of solvent 共here it is water兲 in data
maximum heat loss from channel wall to outside was estimated to reduction for surfactant solutions. In the most application-oriented
be less than 0.5 percent. studies 关11,17,28兴 such a treatment is convenient for comparing
The three types of wire mesh were commercially available, the friction and heat transfer results under the same situations of
made of stainless steel and had dimensions as listed in Table 1. As flow and heat flux between the solvent and the surfactant solution.
schematically shown in Fig. 2, the mesh sheet was attached to a On the other hand, the surfactant solution has viscoelastic charac-
wire frame and arranged with its cells against the main flow. In teristics and much remains to be done to clarify its rheological
case of including several sheets of mesh in one plug, the clearance properties 关20兴. The surfactant solutions may show higher viscos-
between two sheets was 5 mm. The coordinate shown in Fig. 2 is ity and large errors may arise if this is not taken into account in
adopted in the later discussions. the pure theoretical analysis. The height of the channel, H, was
An LDV measurement system was installed on a mobile plat- adopted as the characteristic length for Reynolds number and
form, which could be positioned streamwise arbitrarily to measure Nusselt number.
the two-dimensional velocity at different stations downstream The surfactant tested in our experiment was cetyltrimethyl am-
from the mesh plug. The LDV system consisted of an argon-ion monium chloride 共CTAC兲, which belongs to the cationic group of
laser, standard DANTEC 55X optics system working in a two- surfactants and is less affected by calcium or sodium ions natu-
color three-beam mode, front scattering system with two photo- rally existing in tap water. Its effective temperature in drag reduc-
multipliers, and two counter-processors. The probe volume had a tion ranges from 10°C to 45°C. Although we did not obtain data
length of 1.3 mm and a beam waist diameter of 0.08 mm. A very of the critical micelles concentration 共CMC兲 of CTAC, the signifi-
small amount of polyethylene beads, 0.05 ppm with diameter of 5 cant drag reduction shown at 25 ppm and 30°C in our experiments
␮m and density of 950 kg/m3, was used as scattering particles for described later is the evidence that the CMC of CTAC at around
the laser. Typically, 8000 data were acquired at one spatial point 30°C is at least lower than 25 ppm. This is because the presence
in the measurement of velocity. of rod-like micelles can cause a drag-reducing effect. The chemi-
The pressure drop was measured with a high precision pressure cal formula of CTAC is C16H33N共CH3兲3Cl with molecular weight
transducer at a resolution of 0.1 Pa. With the measured pressure of 320.00 g/mole. Local tap water was used as solvent in the test.
drop over a certain streamwise distance, L, the wall shear stress The same weight concentration of sodium salicylate
and Fanning friction factor are given by (NaSal;HOC6H4COONa,160.10 g/mole) was added to the solu-
⌬ PH tion to provide organic ions. The organic counter-ions from the
␶ w⫽ (1) sodium salicylate greatly affect the CMC of that particular surfac-
2L tant system material 关20,21兴. For simplicity, the composition of
␶w solution was indicated only by the surfactant concentration in this
f⫽ . (2) work, but the same mass concentration of NaSal was always in-
共 1/2兲 ␳ U 2b cluded in the solution.
One thermo-couple was set in the settling chamber to measure Uncertainties of the calculated results were estimated by using
the inlet mean temperature of the flow. A data acquisition unit, IM the root-sum-square method of Kline and McClintock 关29兴. By
DA 100-01E 共Yokogawa Electric Co.兲, was connected to a per- calibration of measuring instruments, the uncertainties of mea-
sonal computer that automatically converted the thermo-couple’s sured values for temperature, pressure drop, flow rate of water of
output voltage into temperature. Temperatures were monitored to surfactant solution, the voltage and resistance of the electric film
ensure that data in the steady state was being acquired. All the heater were determined to be ⫾0.1°C, ⫾1.0 Pa, ⫾7.5
temperature readings were obtained and averaged over several ⫻10⫺3 m3 /min, ⫾1.4003 V, and ⫾0.01 ⍀, respectively. The dis-
minutes. The outlet fluid temperature was estimated from the heat tance of the pressure taps, the length of the heating plate, channel
balance between the electric power and temperature increment of height, and spanwise width have uncertainties of ⫾0.7 mm, ⫾0.7
the flow. The local heat transfer coefficient is defined as follows: mm, ⫾0.1 mm, and ⫾0.14 mm, respectively. The standard
DANTEC LDV system was well calibrated and the uncertainties
for mean velocities and velocity fluctuations were estimated to be
⫾0.2 percent and ⫾1.0 percent, respectively. We also checked the
results of many measurements at one point with running condi-
tions unchanged: the values of mean velocity, root mean square of
velocity fluctuations and Reynolds stress all showed very reliable
repeatability with deviations from average value of no larger than
⫾0.2 percent, ⫾1.0 percent, and ⫾1.4 percent, respectively. Then,
the uncertainty analysis of the reduced results was conducted.
For convenience, the uncertainties of the experimental results
that were calculated from measured values were expressed as per-
centages compared to the values of the results. The pressure drop
measured for the mesh plug has uncertainty of less than ⫾0.5
percent. The bulk velocity and Reynolds number both have uncer-
tainties of ⫾1.5 percent. Because of the small value of pressure
Fig. 2 Two-dimensional channel and mesh plug „unit:mm… drop of drag-reducing flow, the wall shear stress, ␶ w , and fric-
tional factor, f, have uncertainties of ⫾3.5 and ⫾4.7 percent, re-
spectively. All the above uncertainties of velocity, Reynolds num-
ber, wall shear stress and frictional factor are calculated under the
Reynolds numbers of 2.0⫻104 . In most cases in the experiments,
the Reynolds numbers were larger than 2.0⫻104 and therefore the
above uncertainties will be smaller, however, at the minimum
Reynolds number of 1.0⫻104 they may be as large as ⫾3.0 per-
cent, ⫾3.0 percent, ⫾8.1 percent, and ⫾10.0 percent, respec-
tively. The uncertainties of heat transfer coefficients are affected
significantly by uncertainties of temperature and the heat flux.
With the heat loss taken into consideration, the uncertainties of
heat transfer coefficients are estimated to be less than ⫾6 percent
for drag-reducing flow. Because of small temperature differences
between heating wall and bulk flow, the uncertainties of heat
transfer coefficients for water can be ⫾10.0 percent although we
used larger heat flux than that for surfactant solution. When the
heat transfer was effectively enhanced in drag-reducing flow, the Fig. 4 Critical wall shear stress versus concentration of sur-
local heat transfer coefficients close to that of water had uncer- factant in smooth channel „ T bÄ30°C… „region I „Re
tainties as high as ⫾13.0 percent. The uncertainties of Nusselt ËRec1…—DR flow; region II „Rec1ËReËRec2…—post DR flow; re-
numbers are essentially almost equal to the uncertainties of heat gion III „ReÌRec2…—turbulent flow…
transfer coefficients. The surfactant concentrations have uncer-
tainties of less than ⫾0.5 percent.
3 Results and Discussion rod-like micelles in the solutions with high concentration must
occur at a high threshold of Reynolds number or, more correctly,
3.1 Friction Behavior Without Plug. As a preliminary wall shear stress.
work, tests on the friction factors of the flow of solutions at dif- Related to the critical Reynolds number is the critical wall shear
ferent surfactant concentration were conducted at a constant bulk stress. It was proved that the unique threshold for the disintegra-
temperature of 30°C in a smooth channel to obtain fundamental tion of the aggregation of rod-like micelles as a network is wall
data. The friction factors are shown in Fig. 3. For comparison, the shear stress, as evidenced by the dependency of the critical Rey-
friction factors given by Dean’s equation 关30兴 for two- nolds number on the hydraulic diameter of flow duct reported by
dimensional turbulent flow of water with Reynolds number rang- Ohlendorf et al. 关20兴, Gasljevic and Matthys 关32兴, Usui et al. 关33兴,
ing from 6.0⫻103 to 6.0⫻105 are plotted in the figure by the solid and our previous work 关34兴. The critical wall shear stress is shown
line. Dean’s equation is as follows: in Fig. 4 for surfactant solution at several concentrations and tem-
perature of 30°C.
f ⫽0.073 Re⫺0.25 共 Dean’s equation 关 30兴兲 . (4) In order to judge the wall shear stress necessary to turn drag-
The dashed line in Fig. 3 shows the minimum friction factor for reducing flow to completely turbulent flow for surfactant solution
drag-reducing flow given by Virk’s equation 关31兴 indicating the of different concentrations, it is necessary to define another criti-
maximum drag reduction asymptote as follows: cal wall shear stress at which the friction factor of surfactant so-
lution reaches that of water. For distinction as shown in Fig. 4, we
1/冑 f ⫽19 log共 Re 冑 f 兲 ⫺32.4 共 Virk’s equation 关 31兴兲 . name this critical wall shear stress as the ‘‘second critical wall
(5) shear stress,’’ and the wall shear stress at which the drag-reducing
With increasing surfactant concentration, the maximum drag flow starts to degrade as the ‘‘first critical wall shear stress.’’ The
reduction is shifted to higher Reynolds number. This indicates that second critical wall shear stress and the corresponding critical
corresponding to the large concentration, there is a large critical Reynolds number represents the point at which the flow com-
Reynolds number. It was pointed out by Ohlendorf et al. 关20兴 that pletely loses its drag-reduction ability and becomes turbulent
the length of rod-like micelles increases with increasing surfactant flow. Hereafter, we call the flow at Reynolds number smaller than
concentration. The critical Reynolds number is actually a thresh- the first critical Reynolds number as drag-reducing flow shown in
old at which the network of rod-like micelles starts to disintegrate. Fig. 4 as region 1, and that larger than the second critical Rey-
In drag-reducing flow, rod-like micelles in the network align in nolds number as turbulent flow 共region III兲. The flow at Reynolds
the flow direction, therefore, the disassociation of the network of number between the first critical Reynolds number and the second
critical Reynolds number will be called post-drag-reducing flow
共region II兲.
Both of the two critical wall shear stresses increase with the
increment of surfactant concentration. However, the second criti-
cal wall shear stress increases at a larger gradient. Thus, the dif-
ference between the second critical wall shear stress and the first
critical wall shear stress increases with the increment of surfactant
concentration. It can be deduced that promoting turbulence in the
flow with higher concentration of surfactant will be more difficult
because both of its two critical shear stresses are higher than that
of the solution with lower concentration of surfactant.
3.2 Heat Transfer

Channel Without Plug. In order to provide a basis for com-
parison of heat transfer coefficient in the later discussions of the
heat transfer enhancement effect of the mesh, tests of heat transfer
in a smooth channel were conducted. Figure 5 shows the results of
mean Nusselt number versus Reynolds number for water and sur-
Fig. 3 Friction factors versus Reynolds number in smooth factant solutions with inlet temperature of 30°C in the channel
channel „ T bÄ30°C… without mesh. The heat transfer data from water flow agrees very
Fig. 6 Local Nusselt number of water and surfactant solution
in smooth channel „ T inÄ30°C…
Fig. 5 Mean Nusselt number versus Reynolds number in

smooth channel „ T inÄ30°C… Channel With Micellar Structure Destruction Device—Mesh
Plug. Three types of mesh plug were used in the experiment to
destroy the micellar structure in order to promote turbulence in the
channel. Their effects in promoting heat transfer will be discussed
well with the Gnielinski’ equation 关35兴, which is suitable for the hereafter. Because the turbulent water flow already contains
heat transfer in the hydraulic fully developed and thermally devel- strong turbulence, the mesh plug has a minor enhancing effect on
oping region and the hydraulic diameter was used as the charac- the heat transfer of water flow in the channel only at a very short
teristic length for Nusselt number and Reynolds number in the distance from the entrance, as shown in Fig. 7. However, for
equation. surfactant solution at the concentration of 30 ppm and certain
Reynolds number, there is a significant improvement of the heat
NumD ⫽0.012共 Re0.87
D ⫺280 兲 Pr b 1⫹ 共 D/L 兲 c
0.4 2/3
transfer as seen in Fig. 8. Downstream of the mesh plug, the local
heat transfer coefficient reaches almost the same value as that of
共 Gnielinski’s equation 关 35兴兲 . (6)
water. When considering this different heat transfer enhancing
The heat transfer coefficients of drag-reducing surfactant solu- effect to water and surfactant solution by the mesh, it is necessary
tion are rather low compared to that of water flow at the same to discuss the mechanism of heat transfer enhancement by mesh
Reynolds numbers. Also, the Nusselt number of drag-reducing insertion for water and surfactant solution. The rod-like micelle
flow shows relatively weak increment with increasing Reynolds has a diameter in the order of several nanometers, and its length
number, which reflects the suppression of turbulence and thus the may be several tens of times the diameter, which increases with
nature of the flow is between laminar flow and turbulent flow 关5兴. increasing concentration 关20兴. In the work of Lu et al. 关18兴, en-
Because of the same level of turbulence suppression under the tangled rod-like micelles, or thread-like as they called it, were
three concentrations, i.e., 30 ppm, 60 ppm, and 90 ppm, their heat observed in the quiescent state and the shear thickening was visu-
transfer reduction and Nusselt number are almost the same if the alized by Hu et al. 关15兴 as evidence of the shear-induced struc-
Reynolds numbers do not exceed their corresponding first critical tures of micelles in surfactant solution, which is often called net-
Reynolds number. Due to the difference of the critical Reynolds work structures of rod-like micelles. The large suppression of
number, the Nusselt number for case of 30 ppm shows a signifi- turbulence in a turbulent flow at high Reynolds number is gener-
cant increase when the Reynolds number is above 3.0⫻104 , while ally believed due to the suppression of small-scale turbulence ed-
the cases of 60 ppm and 90 ppm still keep a low Nusselt number. dies by the network structures of rod-like micelles. In the bursting
Putting those heat transfer data in the drag-reducing region to- and growth of these eddies, extensional motions dominate
gether for concentrations from 30 ppm to 90 ppm, we obtained a 关15,16,18兴. The result that heat transfer was enhanced downstream
correlation 共see Eq. 共7兲兲 similar to that of Gnielinski’s equation of the mesh strongly suggested that turbulence was recovered in
关35兴 for predicting the Nusselt number for the present results. 95
percent of the experimental values have deviations of less than 10
percent from the empirical equation proposed in this study for the
tested solutions. The Reynolds number in Eq. 共7兲 ranges from
1.5⫻104 to 4.0⫻104 . However, when Eq. 共7兲 is used for solutions
of surfactant concentration ranging from 30 ppm to 90 ppm, the
Reynolds number must be limited to not larger than the first criti-
cal Reynolds number of the solutions respective to its concentra-
tion of surfactant. L in Eq. 共7兲 is the length of the two-dimensional
heating surface.
Num ⫽0.28共 Re0.42⫺6 兲 Pr0.4b 1⫹ 共 D/L 兲 2/3c (7)

The local Nusselt numbers for the solutions at three different
concentrations are shown in Fig. 6. There is a minor increment of
local Nusselt number for the 30 ppm case in comparison with
other concentrations at the entrance region. Because the Reynolds
number in Fig. 6 belongs to the drag-reducing region for any of
the three concentrations, most of the local heat transfer coeffi- Fig. 7 Local Nusselt number of water flow downstream of
cients show no significant difference from one another. mesh „ T inÄ30°C…
Fig. 9 Enhanced local Nusselt number with number of mesh
Fig. 8 Enhanced local Nusselt number by different types of
sheet included in a plug „T inÄ30°C; C mÄ30 ppm…
mesh „T inÄ30°C; C mÄ30 ppm…
this region; the results of turbulence measurements will be dis- celles’ structure by those mesh sheets in the plug. It is clear that
cussed later in this work. One may also deduce that the recovery the plug with more mesh sheets gives a better heat transfer en-
of turbulence is due to the destruction of the micellar structure of hancement. However, the Reynolds number of the surfactant flow
the rod-like micelles. Mesh enforces high shear stress and has is of primary importance; at a small Reynolds number the heat
both possibilities to generate turbulence and destroy the network transfer enhancement by the mesh plug is limited even when 5
structures of micelles, but the later is more important for flow of sheets are included in the plug. At a larger Reynolds number the
surfactant solution in this case. It is a high shear stress rather than plugs with 3 or 5 mesh sheets all gave satisfactory heat transfer
high shear rate which destroys the shear alignment of the micelles, enhancement.
i.e., destroys the network structure of the micelles 关15,16兴. In order to evaluate the overall performance of the mesh plug in
Due to the strong tendency of the rod-like micelles to aggregate enhancing the heat transfer of surfactant solution, Fig. 10 depicts
to form a network once the high shear stress is relieved, the flow the mean Nusselt number versus the Reynolds number at a sur-
far from the mesh plug tends to recover as drag-reducing flow factant concentration of 30 ppm. There is a clear trend that when
because the high shear stress does not last for a long distance and the Reynolds number of the flow is closer to the critical Reynolds
thus there is a decrease of the heat transfer coefficient far down- number, for instance larger than 2.2⫻104 , every type of mesh
stream. Because the three types of mesh have different wire di- plug demonstrates a tremendous effect in enhancing the heat
ameter and opening rate as listed in Table 1, they exert different transfer. With Reynolds number between 2.4⫻104 to 3⫻104 ,
magnitudes of destruction on the rod-like micelles’ network. Thus even one sheet of A-type mesh can promote the heat transfer
the heat transfer recovery from the stage of turbulent to drag- coefficient to three times that of the smooth channel. This sug-
reducing flow covers a different length of region. The plug of the gests that after temporary destruction, the formation of a new
A-type mesh has the minimum length of recovery region, while network of rod-like micelles is strongly affected by the velocity or
the C-type mesh has the maximum. At a larger Reynolds number, Reynolds number of the flow. In other words, the promotion of
the heat transfer augmented region is enlarged and also demon- turbulence by the mesh can only take effect above a certain level
strates less difference among the three types of mesh in promoting of kinetic energy of the flow itself. Therefore, an adequate Rey-
turbulence in drag-reducing flow. The mesh sheets are supported nolds number or velocity has to be considered for improving the
by a mesh holder made of 2.0 mm wire. To examine the effects of heat transfer performance by using a mesh plug in surfactant so-
the holder, heat transfer using the holder without mesh was also lution. In the works of Gasljevic and Matthys 关17兴 on the heat
tested. Since there was no heat transfer improvement in this test, it transfer of pipe flow, the solution tested had a large concentration
is believed that the heat transfer was enhanced by the mesh of surfactant. It can be deduced that the critical Reynolds number
through modification of the fluid or destruction of the micelles’ would be very high, but the Reynolds numbers they tested were
structure. relatively low and therefore the heat transfer enhancement effect
Because the A-type mesh has a relatively larger opening rate by laminated mesh plug was very limited in their work.
which impedes the flow less and it effectively enhances heat trans- Heat transfer tests for solutions at higher concentrations of sur-
fer if the Reynolds number is large enough, plugs using several factant were conducted in this study. Figures 11 and 12 show
sheets of A-type mesh were devised. The heat transfer perfor- some of the results at several Reynolds numbers. Because the
mance of surfactant solution with different number of mesh sheets solution having a high concentration of surfactant have longer
in the plug is shown in Fig. 9. The different level of heat transfer rod-like micelles which correspond to high critical Reynolds num-
enhancement reflects the different times of destruction to the mi- ber or critical wall shear stress, destruction of the structure of the
Fig. 10 Mean Nusselt number versus Reynolds number using
different types of plug „T inÄ30°C; C mÄ30 ppm…
rod-like micelles’ network becomes more difficult, or in other

words, the formation of the network structure of the rod-like mi-
celles’ downstream is much easier in such a case. This can be seen
from the significant differences in the effectiveness of heat trans-
fer enhancement by using a plug including 5 sheets of A-type Fig. 12 Enhanced local Nusselt number with 5 sheets of A
mesh for the solution with surfactant concentration of 40 ppm and type mesh included in a plug „T inÄ30°C; C mÄ60 ppm…
60 ppm. The mesh plug actually has a very limited effect here in
enhancing the heat transfer at the Reynolds numbers that other-
wise work well for the 30 ppm case. Nevertheless, it is possible to
improve the effectiveness of heat transfer enhancement by the
mesh plug by increasing the Reynolds number.
Based on the above results, the mechanism of mesh insertion to
destroy the structures of rod-like micelles’ network is discussed
hereafter. The mesh wires apply impact and large shear stress to
the flow of surfactant solution, and, thus, destroy the micelles’
network. If partial similarity with solid destruction holds, there
might be a threshold velocity for effective destruction of the mi-
celles’ network. For solution at high concentrations of surfactant,
the rod-like micelles are longer, and the same effective destruction
as that of low concentrations of surfactant can only be achieved
by high flow speed or using more mesh wires. On the one hand,
less destruction of the super structures of the rod-like micelles’
network, like the case of one sheet of A-type mesh for the solution
of 30 ppm surfactant, cannot produce enough turbulence. How-
ever, excessive destruction or impact on the super structure of the
rod-like micelles’ network, like the case of five sheets of A-type
mesh in the 30 ppm surfactant solution, may not promote the
turbulence proportional to that by one sheet of A-type mesh. This
is because once the micellar structure is destroyed and turbulence
is promoted, more mesh sheets are not necessary, and like that in
water flow the mesh cannot be used to enhance heat transfer.
The turbulence may be promoted by one-time destruction of the
rod-like micelles’ network by a mesh with fine holes, however,
several times of destruction by a mesh with large holes may have
the same effect. Taking the mesh wire as an edge in cutting the
flowing network of rod-like micelles, the ratio of the mesh’s total
wire length to flow area 共including wire兲 is 1.33, 3.77, and 11.9
mm⫺1, respectively for the three types of mesh, A, B, and C. The
A-type and B-type meshes have the same wire diameter and three
sheets of A-type mesh may have roughly the same edge-to-area
ratio as that of one sheet of B-type mesh. This may help to explain
Fig. 11 Enhanced local Nusselt number with 5 sheets of A the same level of heat transfer enhancement by one sheet of
type mesh included in a plug „T inÄ30°C; C mÄ40 ppm… B-type and three sheets of A-type mesh as shown in Fig. 10. A
Fig. 13 Turbulent quantities of drag-reducing isothermal flow downstream of plug with 5 sheets of A-type mesh included „T b
Ä30°C; ReÄ2.2Ã104 …
more detailed investigation of the proper mesh dimensions for From the heat transfer results, it is known that the mesh plug
better destruction of the micelles’ networks to promote turbulence cannot enhance the heat transfer of the solution with surfactant
is a subject for further study. concentration of 60 ppm at Reynolds number of 2.2⫻104 . The
very low turbulence intensities shown in Fig. 13 for the 60 ppm
3.3 Variation of Turbulence Intensities Downstream from solution exhibit turbulence suppression and loss of turbulent dif-
Mesh Plug. The turbulence intensities of the flow downstream fusivity. This observation is the direct reason for the low heat
of the mesh plug were measured to obtain direct proof that strong transfer coefficients shown in the previous section. However, the
turbulence was promoted when the heat transfer coefficient was solution at a surfactant concentration of 30 ppm shows almost the
enhanced. All the measurements of two-dimensional velocity for same intensity of turbulence as that of water downstream of the
isothermal flow by LDV were conducted in the channel with a mesh. In the 30 ppm case, the shift of turbulence intensity from
plug of 5 sheets of A-type mesh. Figure 13 shows the variation of the stage of water-like to that of the case of the 60 ppm solution
three components of turbulence stress downstream from the plug. demonstrated that the flow recovers from turbulent flow to drag-
For comparison, the normalized turbulence properties of water, reducing flow downstream. It is thus believed that the improve-
and the solutions with surfactant concentration of 30 ppm and 60 ment of heat transfer coefficient resulted from the promoted tur-
ppm are plotted together at four streamwise positions. The Rey- bulence in the flow. Because the rod-like micelles’ networks are
nolds number was 2.2⫻104 . responsible for the suppression of turbulence in the surfactant so-
Fig. 15 Heat transfer enhancement versus pressure drop us-
ing mesh plug „T inÄ30°C; C mÄ30 ppm…
shown in Fig. 15. The numbers attached to symbols in the figure

represent different Reynolds number as seen from the legend col-
umn. The vertical axis represents the percentage of heat transfer
coefficient of surfactant solution to that of water flow. The hori-
zontal axis represents the pressure loss caused by the plug, which
is the cost for the heat transfer enhancement.
At low Reynolds numbers the heat transfer enhancement effect
Fig. 14 Equivalent length of pressure loss resulted by mesh is very limited as seen in the figure, but the effect increases with
plug „T inÄ30°C; C mÄ30 ppm… increasing Reynolds number. As an example, in order to improve
the mean heat transfer coefficient to 85 percent of that of water
flow at Reynolds number of 2.8⫻104 , one sheet of A-type mesh
lution, we deduce that after the destruction of the rod-like mi- is enough, while to achieve 95 percent three sheets of A-type
celles’ networks there will be no suppression and thus turbulence mesh may be appropriate. Overall, the A-type mesh is recom-
re-appears. However, if the high shear stress does not remain in mended in the present study for its good performance of heat
the whole flow region, the rod-like micelles’ networks can be transfer enhancement and low level of pressure loss.
reorganized under the shear induced state which allows the recov-
ery of the drag reducing effect.
3.4 Pressure Drop by Mesh Plug. As well as the heat Conclusions
transfer enhancement, the mesh plug also results in pressure loss The heat transfer and the mechanism of effective heat transfer
in the flow. It is desired to enhance the heat transfer at a low cost enhancement for drag-reducing surfactant solution was studied,
of pressure loss, otherwise the reduction of pressure drop obtained and the following conclusions were drawn.
by the drag-reducing effect will be offset. In order to derive the
value of pressure drop that resulted only from the plug, the pres- 1 As the result of preliminary experiments, the second critical
sure drop of surfactant solution over 4 meters of channel length Reynolds number or wall shear stress was adopted, which repre-
including the mesh plug was tested, and the same measurement sents the upper limit of transition at which the drag-reducing flow
was conducted for a smooth channel without plug. The difference completely changes into turbulent flow. The second critical wall
of the two pressure measurements was calculated and used to shear stress is proportional to the concentration of surfactant in the
obtain the pressure drop that resulted from the mesh plug. Figure solution. The difference between the second critical wall shear
14 shows the dimensionless pressure drop that resulted from the stress and the first critical wall shear stress increases with increas-
mesh plug versus Reynolds number for 30 ppm surfactant solu- ing concentration of surfactant in the solution, which indicates
tion. It is seen that the A-type single mesh gives the minimum that to destroy the network of rod-like micelles for promoting
pressure drop. The plug including one sheet of B-type mesh re- turbulence high shear stress is needed for solutions with high con-
sulted in the same pressure drop as that produced by 5 sheets of centration of surfactant. Within the range of concentrations tested,
A-type mesh. the heat transfer coefficients of drag-reducing flow are rather low
In order to estimate the pressure loss caused by the plug, an and have no significant difference if the Reynolds number is lower
equivalent length 共Le兲 of straight smooth channel with fully de- than the corresponding critical Reynolds number. A correlation
veloped water flow was calculated for the convenience of estimat- for Nusselt number against Reynolds number was obtained for the
ing the pressure loss when using a mesh plug to enhance heat present two-dimensional channel, which is suitable for surfactant
transfer in heat exchangers. With increasing Reynolds number, concentrations from 30 ppm to 90 ppm and Reynolds number
turbulence arises that results in the increase of Le as seen in lower than their respective critical Reynolds number.
Fig. 14. 2 The effectiveness of three types of wire mesh forming a plug
for promoting turbulence in enhancing heat transfer was studied.
3.5 Overall Evaluation of Heat Transfer Enhancement In general, the Reynolds number and concentration of surfactant
and Pressure Loss by Mesh Plug. To estimate the overall ef- significantly affect the effectiveness of heat transfer enhancement
fectiveness of heat transfer enhancement using a mesh plug in by using the mesh. When the Reynolds number is close to the first
surfactant solution, the mean heat transfer coefficient of surfactant critical Reynolds number, the mesh plug can promote the heat
solution by using a mesh was normalized by that of water flow as transfer coefficient of drag-reducing flow effectively. Therefore,
the mesh that is effective for enhancing heat transfer for a solution T f ⫽ mean temperature of flow in heat
of low concentration may not be effective for a solution of high transfer section 共°C兲
concentration under the same Reynolds number. T in ⫽ inlet temperature of flow in heat trans-
3 It is deduced that due to the destruction of the rod-like fer section 共°C兲
micelles’ networks by the mesh plug there will be no suppression T w ⫽ wall temperature 共°C兲
of turbulence in a certain region downstream of the mesh. u ⬘ ⫽ root mean square of turbulent velocity
Through measurement using LDV, the same level of turbulence fluctuation in streamwise direction
intensities as that of water flow in surfactant solution was clarified 共m/s兲
in the case that heat transfer was enhanced significantly. If Rey- v ⬘ ⫽ root mean square of turbulent velocity
nolds number does not exceed the first critical Reynolds number, fluctuation in normal direction 共m/s兲
the turbulence gradually decreases further downstream of the ⫺u v ⫽ kinetic Reynolds shear stress 共m2/s2兲
mesh plug, which demonstrates that the flow recovers from turbu- U b ⫽ bulk velocity 共m/s兲
lent to drag-reducing flow. The recovery region, however, has x ⫽ streamwise distance from starting point
different length depending on the Reynolds number and concen- of heating plate as shown in Fig. 2 共m兲
tration of surfactant. y ⫽ distance in normal direction apart from
4 An overall evaluation was conducted on the principle of ef- heating wall as shown in Fig. 2 共m兲
fective heat transfer enhancement with low cost of pressure loss
caused by the mesh plug. The A-type mesh was shown to be Greek Symbols
preferable. At a Reynolds number close to the critical Reynolds ␳ ⫽ density of solvent 共Kg/m3兲
number for a surfactant concentration of 30 ppm, the A-type mesh ␭ ⫽ thermal conductivity of solvent 共W/m•K兲
plug promotes the mean heat transfer coefficient effectively on the ␯ ⫽ kinematic viscosity of solvent 共m2/s兲
present 1.6 meter long heating plate, while the pressure loss ␶ w ⫽ wall shear stress 共with subscription of c1 is the first
caused by the plug is relatively small. critical wall shear stress, c2 the second critical wall
shear stress兲共Pa兲
Acknowledgment
The first author is grateful for the support of the industrial tech- References
nology research program of the New Energy and Industrial Tech- 关1兴 Toms, B. A., 1948, ‘‘Some Observations on the Flow of Linear Polymer
nology Development Organization 共NEDO兲 of Japan. Solutions Through Straight Tubes at Large Reynolds Numbers,’’ Proceedings
of 1st Int. Congr. on Rheology, 2, pp. 135–141.
关2兴 Mysels, K. J., 1949, US Patent 2492173, December 27, 1949.
Nomenclature 关3兴 Gyr, A., and Bewersdorff, H. W., 1995, Drag Reduction of Turbulent Flows by
Additives, Kluwer Academic Publishers, The Netherlands.
C m ⫽ mass concentration of surfactant and 关4兴 Berman, N. S., and Cooper, E. E., 1972, ‘‘Stability Studies in Pipe Flows
counter-ion material in solutions 共ppm兲 Using Water and Dilute Polymer Solutions,’’ AIChE J., 18, pp. 312–320.
C p ⫽ specific heat 共kJ/kg K兲关5兴 Kawaguchi, Y., Tawaraya, Y., Yabe, A., Hishida, K., and Maeda, M., 1996,
D ⫽ hydraulic diameter of the two- ‘‘Turbulent Transport Mechanism in a Drag-reducing Flow with Surfactant
Additive Investigated by Two Component LDV,’’ Proceedings of 8th Interna-
dimensional channel 共m兲 tional Symposium on Application of Laser Techniques to Fluid Mechanics, pp.
f ⫽ fanning friction factor 共with subscrip- 29.4.1–29.4.7.
tion of c1 corresponds to the first criti- 关6兴 Toh, K. H., and Ghajar, A. J., 1988, ‘‘Heat Transfer in the Thermal Entrance
cal Reynolds number and c2 the sec- Region for Viscoelastic Fluid in Turbulent Pipe Flows,’’ Int. J. Heat Mass
Transf., 31, No. 6, pp. 1261–1267.
ond critical Reynolds number兲关7兴 Inaba, H., and Haruki, N., 1996, ‘‘Drag Reduction and Heat Transfer Charac-
␶ w /(0.5␳ U 2b ) teristics of Water Solution with Surfactant in a Straight Pipe,’’ Proceedings of
h ⫽ heat transfer coefficient 共W/m2K兲 the 3rd KSME-JSME Thermal Engineering Conference, pp. 215–220.
关8兴 Bewersdorff, H. W., and Ohlendorf, D., 1988, ‘‘The Behavior of Drag-
H ⫽ channel height 共m兲 Reducing Cationic Surfactant Solutions,’’ Colloid and Polymer Science, 266,
L ⫽ length of heating section; distance be- No. 10, pp. 941–953.
tween pressure taps 共m兲关9兴 Bewersdorff, H. W., 1989, ‘‘Drag Reduction in Surfactant Solutions,’’ Pro-
Le ⫽ equivalent length of smooth channel ceedings of IUTAM Symposium, Structure of Turbulence and Drag Reduction,
edited by A. Gyr, Zurich/Switzerland, pp. 293–312.
with fully developed water flow over 关10兴 Myska, J., and Zakin, J. L., 1996, ‘‘Comparison of Flow Behavior of Poly-
which the pressure loss equals to that meric and Cationic Surfactant Drag-reducing Additives,’’ Fluid Engineering
resulted by mesh plug, Division, Summer Meeting, FED-Vol. 237, ASME, New York, pp. 165–168.
Le⫽⌬p plug /⌬p w⫺1m (m) 关11兴 Pollert, J., Zakin, J. L., Myska, J., and Kratochvil, P., 1994, ‘‘Use of Friction
Reducing Additives in District Heating System Field Test at Kladono-
Nu;Num ;NumD ;Numw ⫽ local Nusselt number; mean Nusselt Krocehlavy, Czech Republic,’’ Proceedings of Int. District Heating and Cool-
number; mean Nusselt number using ing 1994 Conference, pp. 141–156.
hydraulic diameter as characteristic 关12兴 Steiff, A., and Klopper, K., 1996, ‘‘Application of Drag-Reducing Additives in
length; mean Nusselt number of water District Heating Systems,’’ Fluid Engineering Division, Summer Meeting,
FED-Vol. 237, ASME, New York, pp. 235–242.
without mesh plug 关13兴 Li, P. W., Kawaguchi, Y., and Yabe, A., 2000, ‘‘Feasibility Study of New
ppm ⫽ part per million Heat Transportation System with Drag-Reducing Surfactant Additives,’’ Sym-
⌬ P ⫽ pressure drop 共Pa兲 posium on Energy Engineering in the 21st Century, Hong Kong.
⌬ P plug ⫽ pressure drop resulted by mesh-plug 关14兴 Sato, K., Mimatsu, J., and Kumada, M., 1998, ‘‘Drag Reduction and Heat
Transfer Augmentation of Surfactant Additives in a Two-Dimensional Chan-
共Pa兲 nel Flow,’’ Proceedings of the 35th Japan National Heat Transfer Symposium,
⌬ P w⫺1m ⫽ pressure drop of fully developed water pp. 693–694.
flow over one meter of smooth channel 关15兴 Hu, Y., Boltenhagen, P., and Pine, D. J., 1998, ‘‘Shear Thickening in Low-
共Pa兲 concentration Solutions of Worm-like Micelles—共I兲 Direct Visualization of
Transient Behavior and Phase Transitions,’’ J. Rheol., 42„5…, pp. 1185–1207.
q w ⫽ wall heat flux 共W/m2兲关16兴 Hu, Y. T., Boltenhagen, P., Matthys, E., and Pine, D. J., 1998, ‘‘Shear Thick-
Pr ⫽ Prandtl number v ␳ Cp/␭ ening in Low-Concentration Solutions of Wormlike Micelles. II. Slip, Frac-
Re;ReD ⫽ Rynolds number U b H/ v ;U b D/ v 共with ture, and Stability of the Shear-Induced Phase,’’ J. Rheol., 42„5…, pp. 1209–
subscription of c1 corresponds to the 1226.
关17兴 Gasljevic, K., and Matthys, E. F., 1997, ‘‘Experimental Investigation of Ther-
first critical wall shear stress and c2 mal and Hydrodynamic Development Regions for Drag-Reducing Surfactant
the second critical wall shear stress兲 Solutions,’’ ASME Journal of Heat Transfer, 119, pp. 80–88.
T b ⫽ temperature of bulk flow 共°C兲关18兴 Lu, B., Li, X., Zakin, J. L., and Talmon, Y., 1997, ‘‘A Non-Viscoelastic Drag
Reducing Cationic Surfactant System,’’ J. Non-Newtonian Fluid Mech., 71, 关27兴 Kostic, M., and Hartnett, J. P., 1986, ‘‘Heat Transfer to Water Flowing Tur-
pp. 59–72. bulently through a Rectangular Duct With Asymmetric Heating,’’ Int. J. Heat
关19兴 Chaffey, C. E., and Porter, G. S., 1984, ‘‘Steady Shear Flow of Solutions of Mass Transfer, 29, p. 1283.
Rod-like Macromolecules,’’ J. Rheol., 28, pp. 249–272. 关28兴 Q, Y. Y., Kawaguchi, Y., Lin, Z. Q., Erwing, M., Christensen, R. N., and
关20兴 Ohlendorf, D., Interthal, W., and Hoffmann, H., 1986, ‘‘Surfactant System for Zakin, J. L., 1999, ‘‘Enhance Heat Transfer in Drag Reducing Surfactant So-
Drag Reduction: Physico-Chemical Properties and Rheological Behavior,’’ lutions,’’ Proceedings of the 11th European Drag Reduction Working Meet-
Rheol. Acta, 25, pp. 468–486. ing, Prague, Czech Republic, p. 42.
关21兴 Lindner, P., Bewersdorff, H. W., Hee, R., Sittart, P., Thiel, H., Langowski, J., 关29兴 Kline, S. J., and McClintok, F. A., 1953, ‘‘Describing Uncertainties in Single-
and Oberthur, R., 1990, ‘‘Drag-Reducing Surfactant Solutions in Laminar and Sample Experiments,’’ Mech. Eng. 共Am. Soc. Mech. Eng.兲, 75, pp. 3–8.
Turbulent Flow Investigated by Small-angle Neutron Scattering and Light 关30兴 Dean, R. B., 1978, ‘‘Reynolds Number Dependence of Skin Friction and Other
Scattering,’’ Prog. Colloid Polym. Sci., 81, pp. 107–112. Bulk Flow Variables in Two-Dimensional Rectangular Duct Flow,’’ J. Fluids
关22兴 Usui, H., 1997, ‘‘Turbulent Control by Functional Fluids,’’ J. Jpn. Soc. Fluid Eng., 100, pp. 215–223.
Mech., 16, pp. 105–109.
关31兴 Virk, P. S., 1975, ‘‘Drag Reduction Fundamentals,’’ AIChE J., 21, No. 4, pp.
关23兴 Hu, Y., and Matthys, E. F., 1995, ‘‘Characterization of Micellar Structure
625–656.
Dynamics for a Drag-reducing Surfactant Solution under Shear, Normal Stress
关32兴 Gasljevic, K., and Matthys, E. F., 1995, ‘‘On the Diameter Effect for Turbu-
Studies and Flow Geometry Effects,’’ Rheol. Acta, 34, pp. 450–4600.
关24兴 Kawaguchi, Y., Daisaka, H., Yabe, A., Hishida, K., and Maeda, M., 1997, lent Flow of Drag-reducing Surfactant Solutions,’’ Developments and Appli-
‘‘Existence of Double Diffusivity Fluid Layers and Heat Transfer Character- cations of Non-Newtonian Flows, FED-Vol. 231/MD-Vol. 66, ASME, New
istics in Drag-Reducing Channel Flow,’’ Proceedings of 2nd International York, pp. 237–243.
Symposium on Turbulence Heat and Mass Transfer, edited by K. Hanjalic and 关33兴 Usui, H., Itoh, T., and Saeki, T., 1998, ‘‘On Pipe Diameter Effects in Surfac-
T. W. J. Peeters, Delft, pp. 157–166. tant Drag-reducing Pipe Flows,’’ Rheol. Acta, 37, pp. 122–128.
关25兴 Barrow, H., 1961, ‘‘Convection Heat Transfer Coefficients for Turbulent Flow 关34兴 Li, P. W., Daisaka, H., Kawaguchi, Y., Yabe, A., Hishida, K., and Maeda, M.,
between Parallel Plates With Unequal Heat Fluxes,’’ Int. J. Heat Mass Transf., 1998, ‘‘Study on Heat Transfer of Surfactant Solution as Drag-reducing
1, p. 306. Flow,’’ Proc., of the 35th Japan National Heat Transfer Symposium, 3, pp.
关26兴 Hatton, A. P., and Quarmby, A., 1963, ‘‘The effect of Axially Varying and 691–692.
Unsymmetrical Boundary Conditions on Heat Transfer With Turbulent Flow 关35兴 Gnielinski, V., 1976, ‘‘New Equations for Heat and Mass Transfer in Turbu-
Between Parallel Plates,’’ Int. J. Heat Mass Transf., 6, p. 903. lent Pipe and Channel Flow,’’ Int. Chem. Eng., 16-2, pp. 359–368.
Using Porous Fins for Heat
S. Kiwan
Transfer Enhancement
Assistant Professor
e-mail: kiwan@just.edu.jo
This work introduces a novel method that enhances the heat transfer from a given surface
by using porous fins. The thermal performance of porous fins is estimated and compared
with that of the conventional solid fins. It is found that using porous fin of porosity ⑀ may
M. A. Al-Nimr enhance the performance of an equal size conventional solid fin and, as a result, save 100
Professor
e-mail: malnimr@just.edu.jo
⑀ percent of the fin material. The effect of different design and operating parameters on
the porous fin thermal performance is investigated. Examples of these parameters are Ra
number, Da number, and thermal conductivity ratio. It is found that more enhancement in
Mechanical Engineering Department,
the porous fin performance may be achieved as Ra increases especially at large Da
Jordan University of Science and Technology,
numbers. Also, it is found that there is an optimum limit for the thermal conductivity ratio
Irbid-Jordan
beyond which there is no further improvement in the fin performance.
关DOI: 10.1115/1.1371922兴
Keywords: Conduction, Finned Surfaces, Heat Transfer, Natural Convection, Porous

Media
Introduction less solar collector 关11,12兴 and to improve the performance of

heat exchangers 关10兴. Porous fins can be manufactured using high
The fins industry has been seeking ways to reduce the size and
thermal conductivity materials such as Aluminum, Copper, and
cost of fins. This demand is often justified by the high cost of the
Silver.
high-thermal-conductivity metals that are employed in the manu-
In the present work, the thermal performance of porous fins is
facture of finned surfaces and by the cost associated with the
investigated numerically and a comparison between their perfor-
weight of the fin especially in airplanes and motorcycles applica- mance and that of the solid fins is conducted. In the analysis, the
tions. Brinkman-Forchheimer-extended Darcy model is adopted to
The reduction in the size and cost of fins is achieved by the simulate the fluid behavior inside the porous fins. This model is
enhancement of the heat transfer carried out by the fins. The en- used to study the effect of several operating and design parameters
hancement of heat transfer from fins has become an important on the thermal performance of the fin.
factor that has captured the interest of many researchers. Enhance-
ment of heat transfer from fin can be accomplished through the
following techniques: 共1兲 increasing the surface area to volume Analysis
ratio, 共2兲 increasing the thermal conductivity of the fin, and 共3兲
increasing the convective heat transfer coefficient between the Referring to Fig. 1, an array of porous fins are attached on a hot
surface of the solid fin and the surrounding fluid. surface in order to enhance the heat transfer from the hot surface
Regarding the first technique, large number of works have been to the cold one. In order to facilitate the solution of the governing
conducted to find the optimum shape of fins 关1–9兴. This approach equations, several assumptions are adopted. These assumptions
is based on splitting a certain dimension of the fins in an optimal include: 共1兲 the fins have infinite depth and the array ends effects
way providing that the total volume of the fin material is fixed. are neglected; 共2兲 laminar flow with no internal heat generation;
For example, Duffin 关1兴 has used variational calculus to find the 共3兲 no viscous dissipation; 共4兲 the porous medium is homoge-
optimum fin shape. The optimum profile shape of fins with neous, isotropic, and saturated with a single-phase fluid; 共5兲 both
temperature-dependent conductivity has been determined by Jany the fluid and the solid matrix have constant physical properties;
and Bejan 关2兴. The optimum dimensions for a plate fin with fixed 共6兲 the surface radiant exchanges are neglected; and 共7兲 the solid
volume and transversal laminar boundary layers have been deter- matrix and the fluid are assumed to be at local thermal equilibrium
mined by Bejan 关3兴. An overview of the fin optimum shaping with each other. The interactions between the porous medium and
issue has been presented by Snider and Kraus 关4兴. Poulikakos and the clear fluid is simulated by the Darcy-Brinkman-Forchheimer
Bejan 关5兴 have shown that optimum fin shapes and dimensions formulation and the continuity of velocity and stresses at the in-
can be determined also based on purely thermodynamic grounds. terface 关14兴. Using the dimensionless parameters given in the no-
Under the unidirectional heat transfer assumption, fin optimization menclature, the equations of continuity, motion and energy, for
has been carried out by 关6–7兴. Also, the optimization process has both porous and fluid domains, are reduced to the following non-
been carried out by 关8–9兴 under the effect of variable convection dimensional equations, respectively
heat transfer coefficient. ⳵U1 ⳵V1
In the present work we intend to use porous fins to enhance the ⫹ ⫽0 (1)
fins thermal performance. Due to their larger effective surface ⳵X ⳵Y
area, porous fins have better thermal performance as compared to
the performance of equal weight conventional solid ones.
In the literature, porous substrates of high thermal conductivity
U1
⳵U1
⳵X
⫹V 1
⳵U1
⳵Y
⫽⫺ ⑀ 21
⳵ P1
⳵X
⫹ ⑀ 1 Pr2 冋
⳵ 2U 1 ⳵ 2U 1
⳵X2
⫹
⳵Y 2
⫺ 册
⑀ 21 Pr2
Da 1
U
⫺ ⑀ 21 AU 1 冑U 21 ⫹V 21 ⫹ ⑀ 21 Ra2Pr2␪ 1
have been used to improve the thermal performance of different
(2)
thermal systems 关10–13兴. As an example, porous substrates have
been used to improve the performance of conventional and tube-

U1
⳵V1
⳵X
⫹V 1
⳵V1
⳵Y
⫽⫺ ⑀ 21
⳵ P1
⳵Y
⫹ ⑀ 1 Pr2 冋
⳵ 2V 1 ⳵ 2V 1
⳵X 2 ⫹
⳵Y 2 ⫺ 册
⑀ 21 Pr2
Da 1
V
⫺ ⑀ 21 AV 1 冑U 21 ⫹V 21
HEAT TRANSFER. Manuscript received by the Heat Transfer Division January 20,
2000; revision recieved July 14, 2000. Associate Editor: A. Bejan. (3)
U 1 ⫽U 2 , V 1 ⫽V 2 , ␪ 1⫽ ␪ 2 ,
⳵U1 ⳵U2 ⳵V1 ⳵V2 ⳵␪1 ⳵␪2
⫽ , ⫽ , kR ⫽ .
⳵Y ⳵Y ⳵Y ⳵Y ⳵Y ⳵Y
For solid conventional fins, all boundary conditions are the same
except: at X⫽0 and any Y
⳵U2 ⳵V2 ⳵␪1 ⳵␪2
⫽ ⫽ ⫽ ⫽0.
⳵X ⳵X ⳵X ⳵X
At X⫽W and 0⬍Y ⬍L
⳵␪1 ⳵␪2
U 2 ⫽V 2 ⫽0, kR ⫽
Fig. 1 Schematic diagram for the fins configuration ⳵X ⳵X
At Y ⫽L and 0⬍X⬍W
U1
⳵␪1
⳵X
⫹V 1
⳵␪1
⳵Y
⫽␣R
⳵X2
⫹ 冋
⳵ 2␪ 1 ⳵ 2␪ 1
⳵Y 2 册 (4) U 2 ⫽V 2 ⫽0, kR
⳵␪1 ⳵␪2
⳵Y
⫽
⳵Y
At Y ⫽0 and 0⬍X⬍W
⳵U2 ⳵V2
⫹ ⫽0 (5) ␪ 1 ⫽1,
⳵X ⳵Y
冋册
⳵U2 ⳵U2 ⳵ P2 ⳵ 2U 2 ⳵ 2U 2 where
U2 ⫹V 2 ⫽⫺ ⫹Pr2 ⫹ ⫹Ra2Pr2␪ 2
⳵X ⳵Y ⳵X ⳵X2 ⳵Y 2 k1 w h l m
(6) k R⫽ , W⫽ , H⫽ ⫽1, L⫽ , M⫽ .
k2 h h h h
U2
⳵V2
⳵X
⫹V 2
⳵V2
⳵Y
⫽⫺
⳵ P2
⳵Y
⫹Pr2
⳵ 2V 2 ⳵ 2V 2
⳵X2
⫹
⳵Y 2 冋册 (7) Solution Methodology
The equations and boundary conditions described above are
U2
⳵␪2
⳵X
⫹V 2
⳵␪2
⳵Y
⫽
⳵X2
⫹ 冋
⳵ 2␪ 2 ⳵ 2␪ 2
⳵Y 2 册 (8)
solved by using the finite element code FIDAP 7.06a 关15兴. The
code solved for the velocity, pressure and temperature fields. A
mixed pressure formulation is used 共see 关15兴 for more details兲.
and the governing equations of the conventional solid fin are re- The meshes used in the final runs are tested for mesh indepen-
duced to the following conduction equation in the solid domain: dent solution. Two criteria are considered in getting a mesh-
⳵ 2␪ 1 ⳵ 2␪ 1 independent solution; no significant change 共less than 1 percent兲
⫹ ⫽0, (9) in the solution fields and in the heat flux calculations. All meshes
⳵X2 ⳵Y 2 are graded to provide smaller elements near the boundaries, the
where interfaces between porous and clear domains and the surfaces of
symmetry boundary conditions as shown in Fig. 2. This is done
K1 F 1h g ␤ 2 ⌬Th 3 for accurate temperature predictions in order to resolve the steep
Da⫽ , A⫽ , Ra2⫽ ,
h2 冑K 1 ␣ 2␯ 2 velocity and temperature gradients there. All boundary elements
are two-node linear, while all non-boundary elements are four-
␯2 ␣1 node, quadrilateral elements. Several mesh sizes ranging from
Pr2⫽ , ⌬T⫽T h ⫺T c , ␣ R⫽ . 1000 up to 4500 elements were used to study the problem under
␣2 ␣2
consideration. It is found for all cases studied here that increasing
In Eqs. 共1兲–共9兲, subscripts 1 and 2 refer to the porous and clear the mesh size beyond 4000 elements resulted in less than flux
domains, respectively. calculations. Thus the solutions obtained using 4500-element
In the case under consideration, the momentum and energy mesh can be considered mesh independent solutions.
equations assume the following boundary conditions: To validate the numerical scheme used in the present study, a
At Y ⫽0 and W⬍X⬍M comparison with relevant result is made. This comparison is car-
ried out for the problem of natural convection heat transfer in a
U 2 ⫽0, V 2 ⫽0, ␪ 2 ⫽1. square cavity 关16兴. This case is close to the problem under con-
At Y ⫽0 and 0⬍X⬍W sideration with no porous fin. The result of this comparison
showed that the numerical model predictions for the velocity and
U 1 ⫽0, V 1 ⫽0, ␪ 1 ⫽1. temperature fields are accurate within 2 percent.
At Y ⫽H and any X
U 2 ⫽0, V 2 ⫽0, ␪ 2 ⫽0. Thermal Performance of the Porous Fins
At X⫽0 and any Y and at X⫽M and any Y The improvement in the fin thermal performance can be evalu-
ated in terms of the ratio q p /q s which is the ratio of the heat
⳵U1 ⳵U2 ⳵V1 ⳵V2 ⳵␪1 ⳵␪2 transfer carried out from the hot surface using porous fins to that
⫽ ⫽ ⫽ ⫽ ⫽ ⫽0.
⳵X ⳵X ⳵X ⳵X ⳵X ⳵X carried out using solid conventional fin. It is clear from the dimen-
sionless governing equations that the performance of the porous
At X⫽W and 0⬍Y ⬍L fin depends on the following dimensionless parameters:
U 1 ⫽U 2 , V 1 ⫽V 2 , ␪ 1⫽ ␪ 2 , ⑀1 , Pr2 , Da, A, Ra2 , ␣R , kR , W, M, L.
⳵U1 ⳵U2 ⳵V1 ⳵V2 ⳵␪1 ⳵␪2 Wherever we have to fix their values, the above parameters are
⫽ , ⫽ , kR ⫽ .
⳵X ⳵X ⳵X ⳵X ⳵X ⳵X assumed as
At Y ⫽L and 0⬍X⬍W Da⫽6⫻10⫺6 , L⫽0.125, Ra2 ⫽1⫻107 ,
Fig. 3 Simple simulation for the porous fin
convect the same amount of heat carried out by the original fin.
This implies that we may save 70 percent of the original fin total
material.
The above simulation overestimates the benefit obtained from
using porous fins since it assumes that the convective heat transfer
coefficient within the pores of the porous fin has the same value as
that at the outer surface of the solid fin. In reality, the convective
heat transfer coefficient within the porous material is much lower
than that at the outer surface of the solid fin.
Figure 5 shows a sample plot of the streamlines and isotherma
in the computational domain. It is clear from Fig. 共5a兲 that the
fluid penetrates inside the porous fin. Figure 6 shows the effect of
the thermal conductivity ratio k 1 /k 2 on the heat transfer ratio
q p /q s for different fin lengths. As predicted, increasing the ther-
mal conductivity ratio has the effect of increasing the heat transfer
ratio due to the increase in the effective thermal conductivity of
the porous fin. However, there is an optimum value for k 1 /k 2
beyond which there is no significant improvement in q p /q s . In-
creasing the fin effective thermal conductivity raises the fin tem-
perature until all parts of the fin attain a maximum temperature
which is the fin base temperature. At this limit, the fin convects
the maximum possible heat and no further improvement in its
performance can be achieved by increasing its effective thermal
conductivity. Also, it is clear from this figure that using porous
fins of porosity ⑀ improves the fin performance by a certain per-
centage depends on the fin length and on other operating and
design conditions. Improving the fin performance is not the main
achievement obtained by using porous fins. The main achieve-
ments is the saving of 100 ⑀ percent in the weight of the fin. As an
example, if a porous fin of porosity 0.1 has improved the conven-
Fig. 2 Finite element mesh for the flow field tional solid fin performance by 10 percent, this implies that we
may improve the conventional fin performance by about 22 per-
cent (⫽100((q p /q s )/(1⫺ ⑀ )⫺1)percent) if the same weight of
the conventional solid fin material is utilized in manufacturing
W⫽0.0125, M ⫽0.25, k R ⫽4400, Pr2 ⫽0.71 porous fins each has the same volume as the conventional solid
one.
␣ R ⫽2.5, A⫽1.0, ⑀ ⫽0.5. Using porous fin increases the effective surface area of the fin
through which the fin convects heat to the working fluid. Al-

The basic philosophy behind using porous fins is to increase the
effective surface area through which heat is convected to the am-
bient fluid. As shown in Fig. 3, porous fins may be simulated by
imagining that one single solid fin of fixed weight is divided to N
slides of equal sizes and which have the same total weight as the
original fin. Figure 4 shows the ratio of heat transfer carried out
by the N slides to that carried out by the original fin as a function
of number of slides N. The total heat transfer carried out by the N
slides is estimated by multiplying the heat transfer carried out
through one slide by N. It is clear from this figure that dividing the
fin to slides increases the heat transfer dramatically. The same
figure shows the material weight which can be saved if the re-
maining slides have to convect the same amount of heat carried
out by the original single fin. As an example, if the original fin is Fig. 4 Effect of slides number on the heat transfer ratio and on
divided to 5 slides, then 30 percent of these slides are sufficient to the fraction of saved material
Fig. 7 Effect of solid to fluid thermal conductivity ratio on fin
porosity
Another important observation may be drawn from Fig. 6

which is the fact that the optimum k 1 /k 2 ratio (⫽(k 1 /k 2 ) opt , at
which the porous fin attains its maximum performance, increases
as the fin length increases. As mentioned previously, increasing
the fin length causes less improvement in the porous fin perfor-
mance compared to the improvement in the conventional fin per-
formance.
A relation between the fin porosity, or the saving in the fin
weight, and the solid to fluid thermal conductivity ratio, can be
derived for each fin length. The optimum thermal conductivity
ratio (k 1 /k 2 ) opt , which is found directly from Fig. 6, is given as:
冉冊k1
k2 opt
⫽ ⑀ ⫹ 共 1⫺ ⑀ 兲
ks
k2
⬇ 共 1⫺ ⑀ 兲
ks
k2
Fig. 5 Sample plot for the „a… streamlines and „b… isotherms in
the flow field from which
though the effective thermal conductivity of the porous fin de- 冉冊

k1
k2 opt
creases, due to the removal of the solid material, the increase in ⑀ ⬇1⫺ .
the fin effective surface area overcomes this reduction. ks
Also, it is clear from Fig. 6 that increasing the fin length re- k2
duces the heat transfer ratio. Increasing the fin length causes more
improvement in the solid fin performance as compared to the im- Figure 7 shows the variation in ⑀ as a function of k s /k 2 . As clear
provement in the porous fin performance. As the fin length in- from this figure, and under a fixed fin performance, more saving
creases, the temperature of the part very near the tip approaches in the fin weight can be achieved by using fins having large k s /k 2
the working fluid temperature. This implies that the driving force ratio. However, this saving reaches to an asymptotic level beyond
for natural convection decreases, and as a result, the fluid can not which no further saving in the fin material is achieved. As the
penetrate through the pores of the porous domain. Both porous thermal conductivity of the solid material k s increases, the tem-
and conventional fins can not benefit so much from the increase in perature of the porous fin increases, and in the limit as k s /k f
the fin length. However, the reduction in the porous fin effective →⬁, the fin attains temperatures very close to the fin base tem-
thermal conductivity, due to the removal of the solid material, perature. At this limit, the convected heat through the fin reaches
makes the benefit obtained from increasing the porous fin length to a maximum value and no further increase in this heat may be
lower than that obtained from increasing the conventional solid fin achieved regardless of the increase in the k s /k 2 .
length. Figure 8 shows the effect of Ra number on the heat transfer
ratio at different Da numbers. As predicted, increasing Ra number
increases the heat transfer ratio q p /q s . Increasing Ra number
improves the convective heat transfer coefficient between the fin
and the working fluid. This improves the performance of both
porous and solid fins. However, in porous fins the convective heat
transfer coefficient interacts with the fin through a volumetric sur-
face area which is much larger than the conventional fin surface
area. As a result, any improvement in the convective heat transfer
coefficient causes much more improvement in the porous fin per-
formance as compared to the improvement in the conventional fin
performance. Also, it is clear from the same figure that the effect
of Ra number is more significant on fins having large Da number.
As the permeability of the porous fin increases, i.e., increasing Da
number, the working fluid ability to penetrate through the fin
pores and to convect heat increases. Porous fins having small Da
Fig. 6 Effect of thermal conductivity ratio on the heat transfer number behave as a solid fins due to their small permeability. As
ratio for different fin lengths a result, the ratio q p /q s is very close to 1 for very small Da
porous fin performance as compared to the improvement in the
conventional fin performance. The effect of Ra number is more
significant on fins having large Da number.
As predicted, Nu number increases as the thermal conductivity
ratio increases. However, there is an optimum limit for the ther-
mal conductivity ratio beyond which there is no further improve-
ment in Nu number. Also, it is found that Nu number increases as
the fin length increases.
Nomenclature
A ⫽ dimensionless coefficient of the microscopic inertia term,
F 1 h/ 冑K 1
Fig. 8 Effect of Rayleigh number on the heat transfer ratio for c ⫽ specific heat
different Darcy numbers Da ⫽ Darcy number, K 1 /h 2
F 1 ⫽ Forchheimer coefficient, 1.8/(180⑀ 5 ) 0.5
g ⫽ gravity constant
h ⫽ distance between the hot and the cold surfaces
H ⫽ dimensionless Distance between the hot and the cold
surfaces, h/h⫽1
k ⫽ thermal conductivity
k R ⫽ thermal conductivity ratio, k 1 /k 2
K 1 ⫽ permeability of the porous fin
l ⫽ fin length
L ⫽ dimensionless fin length, l/h
m ⫽ half the spacing between the center lines of two adjacent
fins
M ⫽ half the dimensionless spacing between the center lines
of two adjacent fins, m/h
N ⫽ number of slides
Nu ⫽ Nusselt number, 2q ⬙p m/⌬Tk 2
Fig. 9 Effect of thermal conductivity ratio on Nusselt number p ⫽ pressure
at different fin lengths P ⫽ dimensionless pressure, ph 2 / ␳ 2 ␣ 22
Pr ⫽ Prandtl number, ␯/␣
q ⫽ heat transfer
q 1 ⫽ heat transfer from the original fin
number and the effect of Ra number on this ratio is insignificant. q N ⫽ total heat transfer from the N slides
Figure 9 shows the effect of the thermal conductivity ratio on Nu q ⬙ ⫽ heat flux per unit area
number which is defined as q R ⫽ heat transfer ratio, q p /q s
2q ⬙p m r ⫽ radial coordinate
Nu⫽ , Ra2 ⫽ Rayleigh number of the clear domain, g ␤ 2 ⌬Th 3 / ␣ 2 ␯ 2
⌬Tk 2
T ⫽ temperature at any point
and which represents the dimensionless convective heat transfer T c ⫽ temperature of the cold surface
coefficient between the hot surface, including the porous fins, and T h ⫽ temperature of the hot surface
the cold surface. As predicted, Nu number increases as the ther- u ⫽ axial velocity
mal conductivity ratio increases. However, there is an optimum U ⫽ dimensionless volume averaged axial velocity, uh/ ␣ 2
limit for k 1 /k 2 beyond which there is no further improvement in v ⫽ transverse velocity
Nu number. As mentioned previously, significant increase in the V ⫽ dimensionless transverse velocity, v h/ ␣ 2
porous fin length causes insignificant improvement in the fin per- w ⫽ half width of the fin
formance. As a result, it is necessary to increase the optimum ratio W ⫽ dimensionless half width of the fin, w/h
(k 1 /k 2 ) opt to enable the fin to reach its maximum performance as x ⫽ axial coordinate
the fin length increases. X ⫽ dimensionless axial coordinate, x/h
Also, it is clear from the same figure that Nu number increases y ⫽ transverse coordinate
as the fin length increases. Increasing the fin length increases the Y ⫽ dimensionless transverse coordinate, y/h
heat transfer carried out from the fin and as a result, increases Nu Greek Symbols
number.
␣ ⫽ thermal diffusivity
␤ ⫽ coefficient of volumetric thermal expansion
Conclusion ⌬T ⫽ temperature difference, T h ⫺T c
⑀ ⫽ porosity or void ratio
This work introduces a novel method that enhances the heat ␪ ⫽ dimensionless temperature, (T⫺T c )/(T h ⫺T c )
transfer from a given surface by using porous fins. The thermal ␮ ⫽ dynamic viscosity
performance of porous fins is estimated and compared with that of ␯ ⫽ kinematic viscosity
the conventional solid fins. ␳ ⫽ density
It is found that using porous fin of porosity ⑀ may enhance the
performance of an equal size conventional solid fin and, as a Subscripts
result, save 100 ⑀ percent of the fin material. Also, it is found that 1 ⫽ porous domain properties
more enhancement in the fin performance may be achieved as Ra 2 ⫽ clear or fluid domain properties
number increases. As a result, any improvement in the convective c ⫽ cold
heat transfer coefficient causes much more improvement in the h ⫽ hot
p ⫽ porous fin properties 关8兴 Stachiewicz, J. W., 1969, ‘‘Effect of Variation of Local Film Coefficient on
Fin Performance,’’ ASME J. Heat Transfer, 91, pp. 21–26.
R ⫽ ratio 关9兴 Look, D. C., Jr., 1988, ‘‘Two-Dimensional Fin Performance: Bi 共Top Surface兲
s ⫽ solid fin properties ⭓ Bi 共Bottom Surface兲,’’ ASME J. Heat Transfer, 110, pp. 780–782.
关10兴 Alkam, M., and Al-Nimr, M. A., 1999, ‘‘Improving the Performance of
Double-Pipe Heat Exchangers by Using Porous Substrates,’’ Int. J. Heat Mass
References Transf., 42, pp. 3609–3618.
关1兴 Duffin, R., 1959, ‘‘A Variational Problem Relating to Cooling Fins,’’ J. Math. 关11兴 Al-Nimr, M. A., and Alkam, A., 1998, ‘‘A Modified Tubeless Solar Collector
Mech., 8, pp. 47–56. Partially Filled with Porous Substrate,’’ Renewable Energy, 13„2…, pp. 165–
关2兴 Jany, P., and Bejan, A., 1988, ‘‘Ernst Schmidt’s Approach to Fin Optimiza- 173.
tion: An Extension to Fins with Variable Conductivity and the Design of Ducts 关12兴 Alkam, M. K., and Al-Nimr, M. A., 1999, ‘‘Solar Collectors with Tubes Par-
for Fluid Flow,’’ Int. J. Heat Mass Transf., 31, pp. 1635–1644. tially Filled with Porous Substrates,’’ ASME J. Sol. Energy Eng., 121, pp.
关3兴 Bejan, A., 1984, Convection Heat Transfer, John Wiley and Sons, New York. 20–24.
关4兴 Snider, A. D., and Kraus, A. D., 1986, ‘‘The Quest for the Optimum Longi- 关13兴 Al-Nimr, M. A., and Alkam, M., 1997, ‘‘Unsteady Non-Darcian Forced Con-
tudinal Fin Profile,’’ ASME HTD, 64, pp. 43–48. vection Analysis in an Annulus Partially Filled with a Porous Material,’’
关5兴 Poulikakos, D., and Bejan, A., 1982, ‘‘Fin Geometry for Minimum Entropy ASME J. Heat Transfer, 119, pp. 799–804.
Generation in Forced Convection,’’ ASME J. Heat Transfer, 104, pp. 616– 关14兴 Vafai, K., and Kim, S. J., 1995, ‘‘On the Limitations of the Brinkman-
623. Forchheimer-Extended Darcy Equation,’’ Int. J. Heat Fluid Flow, 16, pp. 11–
关6兴 Irey, R. K., 1968, ‘‘Errors in the One-Dimensional Fin Solution,’’ ASME J. 15.
Heat Transfer, 90, pp. 175–176. 关15兴 Engelman, M. S., 1993, Fidap Theoretical Manual-Revision 7, 1st ed., Fluid
关7兴 Lau, W., and Tan, C. W., 1973, ‘‘Errors in One-Dimensional Heat Transfer Dynamics International, Inc., Evanston, IL.
Analyses in Straight and Annular Fins,’’ ASME J. Heat Transfer, 95, pp. 关16兴 del Vahl Davis, G., 1983, ‘‘Natural Convection in a Square Cavity–A Bench-
549–551. mark Solution,’’ Int. J. Numer. Methods Fluids, 3, pp. 249–264.
Precision Temperature Control
of High-Throughput Fluid Flows:
Theoretical and Experimental
Kevin M. Lawton
Analysis
Steven R. Patterson A precision method for attenuating temperature variations in a high-throughput control
e-mail: spatters@uncc.edu
fluid stream is described and analyzed. In contrast to earlier investigations, the present
study emphasizes heat transfer analysis of the constituent control device and derives
Russell G. Keanini theoretical descriptions of system responses to time-varying fluid temperatures. Experi-
ments demonstrate that the technique provides: (1) frequency-dependent attenuation
Center for Precision Metrology,
which is several orders of magnitude greater than that obtained via a perfect mixing
Department of Mechanical Engineering and
volume; (2) attenuation, over two decades of disturbance frequency, that reduces in-flow
Engineering Science, temperature variations by factors ranging from 10 to ⬇104 ; (3) asymptotic attenuation
University of North Carolina at Charlotte, greater than three orders of magnitude for spectral components having periods less than
Charlotte, NC 28223 the device thermal equilibrium time; and (4) attenuation which is fully consistent with
theoretical predictions. The model developed provides design criteria for tailoring system
performance. In particular, it is shown that for a given control stream flow rate, the
magnitude of maximal attenuation can be adjusted by varying the thermal resistance
between the flow and attenuating medium, while the range of frequencies maximally
attenuated can be adjusted by varying the product of thermal resistance and attenuating
medium heat capacity. The analysis and design are general and should prove useful in the
design and analysis of other high-throughput precision temperature control systems.
关DOI: 10.1115/1.1375810兴
Keywords: Control, Heat Transfer, Heat Exchangers
1 Introduction vacuum chamber providing ⫾15 ␮K control over tens of hours

Due to the inherent thermal sensitivity of materials and sys- 关12兴, precision temperature-controlled water baths providing ⫾25
tems, precision experiments and instrumentation generally require ␮K stability over tens of hours 关13,14兴, and a stirred water bath
precision temperature control. In some experiments, thermal ef- providing ⫾3.5 ␮K stability over tens of hours 关15兴.
fects can be the largest source of dimensional error and non- In contrast to resistance-based precision thermal control and
repeatability 关1兴. Likewise, thermal variability can limit the opera- precision thermal control in closed, non-circulating fluid systems,
tion of various sensors and devices, e.g., laser interferometers, relatively little work has been reported on precision thermal con-
autocollimators, and capacitance probes 关2兴. Temperature stabil- trol of flowing or recirculating control fluid streams. Lopez and
ity, i.e., the maintenance of a given temperature to a given degree Barron 关16兴 used temperature-controlled forced air to achieve
of precision, often constitutes a particularly challenging aspect of ⫾0.1 K control over liquid crystal samples. Ogasawara 关14兴 used
precision system design 关3兴. recirculating water to maintain a thermostatic water bath to within
Precision temperature control near room temperature has re- ⫾25 ␮K of a set point. This approach had also been used to
ceived significant attention. One class of controllers uses resistive control plasma tube temperatures in lasers 关17,18兴. Sydenham and
heaters to maintain a small enclosure near ambient. Due to the
Collins 关19兴, using recirculating water in a deep underground test
small size of these systems, high stability, on the order of ⫾10 ␮K
to ⫾100 ␮K over hours, days or months of time can be achieved. facility 共where the latter minimized background thermal varia-
Examples include ⫾20 ␮K stability in standard cells over several tions兲, achieved ⫾40 ␮K stability in a 10 m long steel standard
months 关4兴, ⫾60 ␮K stability in laser diodes over approximately over hundreds of hours.
one hour 关5兴, ⫾10 ␮K stability in geophysical accelerometers This paper describes the theoretical models, design, construc-
over several days 关6兴, and ⫾100 ␮K stability in semiconductor tion, and performance of a ‘‘thermal gradient attenuator’’ low-
lasers over tens of hours 关7兴. Resistance-based controllers provid- pass filter. This device is a critical component for producing a
ing milliKelvin stability have also been developed or proposed for high stability chilled water supply that can be used for precision
use in thermometry 关8兴, physical property measurement 关9,10兴, temperature control. By attenuating higher frequency temperature
strain measurement 关9兴, and optical microscopy 关11兴. oscillations in the control fluid, active heater controllers need only
Another class of controllers uses fluids to heat or cool enclo- reject lower frequency disturbances to achieve precise tempera-
sures near room temperature. Since these controllers have both ture control.
heating and cooling capabilities, they are particularly suited for This work extends previous investigations by developing a rela-
use with heat-generating devices. Examples include an oil-cooled tively simple approach capable of providing centiKelvin stability
over hours and milliKelvin stability over minutes, within recircu-
HEAT TRANSFER. Manuscript received by the Heat Transfer Division June 27, 2000; lating flows. Equally important, considerable attention is devoted
revision received January 20, 2001. Associate Editor: D. A. Zumbrunnen. to modeling system heat transfer and system dynamics.
2 Thermal Gradient Attenuator—Justification for
Need
A multi-purpose temperature-controlled chamber has been de-
signed for use in precision metrology experiments. The chamber
walls are channeled to allow flow of temperature-controlled
chilled water, which in turn provides thermal control of the cham-
ber interior. Although a heater controller is available to modulate
the chilled water temperature, because of limited bandwidth it
cannot reject the higher frequency temperature oscillations of the
available chilled water supply 共see Figs. 1 and 2兲. Since the de-
sired level of temperature control is order 0.1 mK, the large short-
term temperature deviations of the chilled water must be signifi-
cantly attenuated before entering the heater control system.
In order to attenuate these higher frequency oscillations, a ther-
mal gradient attenuator, shown schematically in Fig. 3, must be
placed in the chilled water stream. As shown in Figs. 1 and 2, the
spectrum for the entering chilled water stream contains significant
spectral components over the range from 0.04 to 10 mHz, with
peak rates of change of order 1 K/min. With dominant temporal
variations occurring at relatively low frequencies, effective attenu- Fig. 3 Thermal gradient attenuator configuration
ation requires that the attenuator exhibit large attenuation at the
relatively low frequency of 0.04 mHz.
For the purpose of analysis, the thermal gradient attenuator is
modeled as three subsystems, each characterized by a dominant 3 Lumped-Differential Model of the Tube Bank-
heat transfer mode and an associated dynamic. The core of the Attenuator Subsystem
device is identified as the tube bank-attenuator subsystem since it
The model of the tube bank-attenuator sub-system is based on
consists of an array of uniformly distributed identical tubes that
two radially-lumped, coupled differential equations which model
direct the flow of the control fluid past the attenuator medium. The
the time and axially-dependent temperature variations within the
entrance and exit volumes below and above the tube bank-
flowing control fluid and the surrounding attenuating medium.
attenuator are identified as the lower accumulator and upper ac-
The derivation of these equations proceeds in three steps: 共1兲 as-
cumulator subsystems. Subsystem and total device performance
sumptions relevant to the tube bank-attenuator are stated; 共2兲 or-
are characterized using transfer functions, which model the ratio
der of magnitude arguments are used to simplify the energy equa-
of the output and input temperatures as functions of frequency.
tion within the flow, the tube wall, and the surrounding
attenuating medium; and 共3兲 each simplified equation is radially
integrated, leading to the lumped working equations. Neto and
Cotta 关20,21兴 have developed a lumped-differential analysis of
double-pipe heat exchangers. The present formulation for a re-
stricted class of heat exchanger leads to two relatively simple,
integrable differential equations, while the more general approach
in Neto and Cotta, which incorporates a generalized integral trans-
form method 关22兴, leads to a large system of coupled differential
equations.
The class of heat exchangers modeled here is restricted by a set
of assumptions relating to the geometry, thermal properties of the
components, and the frequency range of the input disturbances.
These assumptions can be related to a single tube and the sur-
rounding attenuator medium. The length of the attenuator medium
is taken to be much larger than the radial dimension, and the ratio
of axial fluctuation length scale, w̄/ f , to radius, r, is sufficiently
small that radial conduction dominates axial conduction. Thus,
Fig. 1 Chilled water temperature and temperature derivative
versus time, 1.9 lÕmin flow rate w̄/ f ⳵T ⳵T
Ⰷ1 so that Ⰷ (1)
r ⳵r ⳵z
and
r/LⰆ1. (2)
It is also assumed that the attenuator medium is either a solid or a
stagnant fluid where advection is negligible. In the latter case, it is
assumed that the ratio of conductive to time-dependent tempera-
ture variations
uជ •ⵜT us
⬇ (3)
⳵ T/ ⳵ t f •L
are small, where the velocity scale, u s ⫽(g• ␤ •L•⌬T) 1/2, is deter-
mined by buoyancy.
Fig. 2 Chilled water spectrum and average FFT gradients Based on these assumptions, heat transfer, to first order, within
versus frequency; 1.9 lÕmin flow rate the attenuator medium is governed by the equation
⳵T
⳵t
1 ⳵
⫽␣a •
r ⳵r
r
⳵T
⳵r
. 冉冊 (4)
where R f is the fluid thermal resistance, and
b T 兩 r t ⫺T̄ a c
Similarly, conduction within the tube wall is governed by q 共 z,t 兲 ⫽ , (16)
Ra
⳵
r
⳵r ⳵r
⳵T
冉冊
⫽0 (5) where R a is the attenuator thermal resistance. Given the following
definition and equation
while within the fluid flow b T̄ f ⫺T 兩 r f c b T 兩 r f ⫺T 兩 r t c b T 兩 r t ⫺T̄ a c
⳵T
⳵t
⫹w 共 r 兲
⳵T
⳵z
1 ⳵
⫽␣a •
r ⳵r
r
⳵T
⳵r
, 冉冊 (6)
R⬅R f ⫹R t ⫹R a ⫽
q
⫹
q
⫹
q
关 T̄ f ⫺T̄ a 兴
where w(r) is the fully-developed laminar flow velocity profile ⫽ (17)
关23兴. For tubes with sufficiently high aspect ratio and small Rey- q
nolds number, the entry length comprises a small fraction of the and substituting Eqs. 共13兲 and 共15兲 into Eq. 共9兲 yields
total length, and the above equation is a good approximation for
the entire tube. ⳵ T̄ f ⳵ T̄ f 1
Equations 共4兲 through 共6兲 may be integrated radially to obtain ⫹w̄ ⫽ 共 T̄ a ⫺T̄ f 兲 , (18)
respective lumped equations for the mean temperature in the at- ⳵t ⳵z ␶f
tenuator, tube wall, and fluid flow regions where ␶ f ⫽RC f is the fluid time constant. Similarly, Eq. 共7兲
冉冏冏冊
simplifies to
⳵ T̄ a ␣a ⳵T ⳵T
⫽ ra ⫺r t (7) ⳵ T̄ a 1
⳵t 共 r a ⫺r t 兲 2 ⳵r ⳵r
ra rt ⫽ 共 T̄ f ⫺T̄ a 兲 , (19)
⳵t ␶a
rf
⳵T
⳵r
冏 rf
⫽r t
⳵T
⳵r
冏 rt
(8) where ␶ a ⫽RC a , is the attenuator time constant.
⳵ T̄ f
⳵t
⫹
1
r 2f
冕 0
rf
rw 共 r 兲
⳵T
⳵z
dr⫽
␣ ⳵T
r f ⳵r
冏 rf
, (9) 4 Tube Bank-Attenuator Transfer Function
Neglecting heat transfer through the outer wall of the attenuator
assembly, the transfer function of the entire tube bank-attenuator
where the mean temperatures are defined as is the same as that of a single tube, given proper flow rate scaling.
冕 ra
Thus, the transfer function of the entire tube bank follows from
1 Eqs. 共18兲 and 共19兲. A more compact result follows by introducing
T̄ a 共 z,t 兲 ⫽ r•T 共 r,z,t 兲 •dr (10)
r 2a ⫺r 2t rt the dimensionless space and time variables
冕
1 rf ␩ ⫽z/ 共 w̄ ␶ a 兲 and ␶ ⫽t/ ␶ a (20)
T̄ f 共 z,t 兲 ⫽ r•T 共 r,z,t 兲 •dr. (11) and the dimensionless heat capacity ratio
r 2f 0
␨ ⫽ ␶ a / ␶ f ⫽C a /C f . (21)
When the radial conduction time scale is much shorter than the
disturbance and axial convection time scales, radial temperature Substitution of Eqs. 共20兲 and 共21兲 into Eqs. 共18兲 and 共19兲 yields
gradients are quickly smoothed, and the integral in Eq. 共9兲 may be
approximated as ⳵ T̄ f ⳵ T̄ f
⫹ ⫽ ␨ 共 T̄ a ⫺T̄ f 兲 (22)
冕
⳵␶ ⳵␩
rf ⳵T ⳵ T̄ f
rw 共 r 兲 dr⬇r 2f w̄ , (12)
0 ⳵z ⳵z ⳵ T̄ a
⫽ 共 T̄ f ⫺T̄ a 兲 . (23)
where the radially averaged velocity is given by ⳵␶
冕 rf
Taking the Laplace transforms of Eqs. 共22兲 and 共23兲, with initial
2 conditions equal to zero, yields
w̄⫽ rw 共 r 兲 dr. (13)
r 2f 0
⳵ T̂ f
The integrated energy equation, Eq. 共8兲, implies that at any in- sT̂ f ⫹ ⫽ ␨ 共 T̂ a ⫺T̂ f 兲 (24)
stant, the local radial heat transfer rate, q⫽q(z,t), is constant ⳵␩
within the tube wall. By solving Eq. 共5兲, it is readily shown that sT̂ a ⫽ 共 T̂ f ⫺T̂ a 兲 , (25)
关 T 共 r f ,z,t 兲 ⫺T 共 r t ,z,t 兲兴 where s⫽2i ␲ f ␶ a , and f is the frequency in Hz. Eliminating T̂ a
q 共 z,t 兲 ⫽2 ␲ k t L
ln共 r t /r f 兲 yields
关 T 共 r f ,z,t 兲 ⫺T 共 r t ,z,t 兲兴 b T 兩 r f ⫺T 兩 r t c ⳵ T̂ f s
⫽ ⬅ , (14) ⫹ 共 s⫹ ␨ ⫹1 兲 T̂ f ⫽0. (26)
Rt Rt ⳵␩ s⫹1
where R t is the total tube thermal resistance along its length, L. This equation is readily solved to provide an expression for the
Similarly, heat transfer at the inner and outer tube boundaries are fluid temperature as a function of dimensionless length and di-
given as mensionless frequency
q 共 z,t 兲 ⫽
b T̄ f ⫺T 兩 r f c
⫽⫺r f k f L
⳵T
冏 , (15)
T̂ f 共 ␩ ,s 兲 ⫽T̂ f 共 ␩ ⫽0,s 兲 •exp 冋 ⫺s
s⫹1
共 s⫹ ␨ ⫹1 兲 ␩ . 册 (27)
Rf ⳵r rf Thus, the tube bank transfer function 关24兴 is given by
共4.2 MJ K⫺1 m⫺3兲 which is higher than that of aluminum, steel,
copper, and numerous other, more expensive materials.
The device configuration consists of an externally insulated 208
liter 共55-gallon兲 drum, 947 nylon tubes 共each 0.8 m long and 4.8
mm in diameter兲, and 4 acrylic spacer plates 共Fig. 3兲. Chilled
water enters the attenuator from the top, flows to the bottom
where it is distributed to the tube bank through a lower accumu-
lator, flows upward through the tubes into an upper accumulator,
and then exits. Nylon was chosen as the tube material because of
its low cost and suitable thermal resistance. The tubes are ar-
ranged in a hexagonal pattern with tube center spacing of approxi-
mately 16 mm. Thus, each tube is surrounded by an annulus of
attenuator medium having an effective radius of approximately 8
mm. The tube aspect ratio condition of Eq. 共2兲 is satisfied for this
Fig. 4 Theoretical model transfer functions versus frequency configuration since r/L⫽O(10⫺2 ). Also, the radially lumped
model is valid for frequencies below an upper frequency limit
determined by the inverse of the radial diffusion time scale
Tr a 共 s 兲 ⫽
T̂ f 共 ␩ ⫽ ␩ L ,s 兲
T̂ f 共 ␩ ⫽0,s 兲
⫽exp 冋 ⫺s
s⫹1
共 s⫹ ␨ ⫹1 兲 ␩ L 册 f 1⬇
␣a
共 rm兲2
⫽O 共 10⫺2 Hz兲 , (32)
冋册
where r m is the radius encompassing half of the area of the
⫺s• ␨ • ␩ L attenuator medium.
⫽exp •exp关 ⫺s• ␩ L 兴 , (28)
s⫹1 Because the in-tube temperature variations are small relative to
the dominant input temperature variations, the tube wall is ap-
where the latter form separates the effect of transport delay, proximately isothermal. Thus, the convection coefficient can be
exp(⫺s• ␩ L ), from other effects in the attenuator. Since only the computed from the laminar, thermally fully developed flow
real part of the exponent affects the magnitude of the transfer condition 关25兴
function, the transfer function magnitude can be expressed in the
form kf kf
h f ⫽Nu ⫽3.66 .
冋冉冊册
(33)
s Df Df
Trm a 共 s 兲 ⫽exp ⫺ ␨ • ␩ L •Re . (29)
s⫹1 Given h f , the fluid thermal resistance is
Figure 4 depicts the tube bank-attenuator transfer function for 1
several values of ␩ L , ␨, and ␶ a . For a given dimensionless length, Rf⫽ , (34)
␩ L , the heat capacity ratio, ␨, determines the magnitude of the h fAs
asymptotic attenuation. In addition, for a given ␩ L , the range of where A s is the convection surface area. The validity of the as-
frequencies optimally attenuated increases as ␶ a increases, so that sumption of thermally fully developed flow is supported by the
attenuation extends to lower disturbance frequencies. In particu- fact that the thermal development length 关23兴 for these parameters
lar, the asymptotic attenuation is proportional to is approximately 2 percent of the tube length.
The attenuator thermal resistance, R a , is based on the approxi-
Ca L 1 mation of steady state conduction through a stationary fluid. From
␨ • ␩ L⫽ ⫽ , (30)
C f 2w̄ ␶ a ṁc f R Eqs. 共3兲 and 共32兲, the ratio of the time derivative to radial diffu-
sion terms is small, so that the temperature field is approximately
which means that given a fluid with mass flow rate, ṁ, and spe- steady within the attenuator medium. Thus, the effective attenua-
cific heat, c f , the total thermal resistance between the fluid flow tor thermal resistance is
and the attenuator medium determines the asymptotic attenuation.
In terms of dimensional variables, the transfer function, Eq. 共29兲, ln共 r m /r t 兲
is R a⫽ . (35)
2 ␲ k aL
Trm a 共 f 兲 ⫽exp ⫺ 冋 1
ṁc f R
•Re 冉 2i ␲ f ␶ a
2i ␲ f ␶ a ⫹1
冊册 . (31) 6 Accumulator and Total Transfer Functions
In order to complete the model of the entire device, the transfer
This form more clearly shows the effect of ␶ a in determining the functions of the upper and lower accumulators must be developed.
frequency at which significant attenuation occurs. Given a thermal Chilled water is distributed to the tube bank by the lower accu-
resistance to achieve an asymptotic attenuation requirement, the mulator and collects prior to exiting in the upper accumulator. The
heat capacity of the attenuator, C a , determines the attenuator time lower accumulator feeds the tube bank through a perforated-tube
constant, which in turn determines the frequency range over distribution manifold containing numerous small holes 共Fig. 3兲.
which asymptotic attenuation is closely approached. The flow through each hole produces a small jet within the lower
accumulator, which in turn, induces strong mixing. It is thus as-
sumed that the lower accumulator functions as a perfectly mixed
5 Tube Bank-Attenuator Configuration and volume, with the corresponding lumped energy equation given by
Properties
In order to efficiently package a large heat capacity 共to attenu- dT̄ la
␳f V la ⫽Ṁ 共 T in 共 t 兲 ⫺T̄ la 兲 , (36)
ate a wide range of frequencies兲, the attenuator medium should dt
have large specific heat capacity 共heat capacity per volume兲. Other
considerations for the medium include cost and ease of distribut- where Ṁ is the total mass flow rate for all of the tubes, T in (t) is
ing the material about the tubes in order to provide the desired the time varying input temperature, V la is the lower accumulator
thermal resistance for optimal attenuation. Water fulfills these re- volume, and T̄ la is the average lower accumulator temperature.
quirements well since it is inexpensive and has a specific capacity Taking the Laplace transform yields the transfer function
Fig. 6 Drum attenuator transfer functions versus frequency;
measured, theoretical, perfect mixing, and diffusion for 2.5
lÕmin flow rate
In the region near 1 mHz, where time derivatives of temperature

are largest 共and are thus least conducive to precision thermal con-
trol兲, the attenuator provides in excess of 20 dB more attenuation
Fig. 5 Attenuator test setup than a perfect mixer. Likewise, in the lower portion of the fre-
quency range, f ⬍3 mHz, the attenuator provides significantly
more attenuation than does axial diffusion. Although axial diffu-
1 sion provides more attenuation for f ⬎3 mHz, this occurs beyond
Tr la 共 s 兲 ⫽ , (37) the point where the attenuator has already provided 60 dB of
1⫹ ␶ m s attenuation.
where ␶ m ⫽ ␳ f V la /Ṁ is the characteristic residence time for the Attenuator performance compares favorably with that obtained
flow in the lower accumulator. by Ogasawara 关14兴. Ogasawara’s device consisted of a fluid-filled
In contrast to the lower accumulator, mixing in the upper accu- enclosure in which spiral baffles promoted mixing of water. He
mulator is relatively weak. In addition, since the length, L ua , of reported that the device reduced temperature variations by a factor
the upper accumulator is much smaller than the radius, r ua , axial of 3. By contrast, and as shown in Fig. 6, the thermal gradient
conduction dominates the heat transfer. Considering the scales of attenuator provides attenuation on the order of 60 dB, correspond-
each term in the associated energy equation, and assuming that the ing to factor of reduction on the order of 103 . Interestingly, it
flow is essentially vertical and uniform across the upper accumu- appears that Ogasawara’s device can be modeled as a perfectly
lator, the governing equation is mixed volume; indeed, the order of attenuation obtained by
Ogasawara is consistent with the degree of attenuation associated
dT ua dT ua d 2 T ua with a perfect mixer.
⫹w̄ ua ⫽␣ f , (38) Four additional sets of attenuation data are shown in Fig. 7,
dt dz dz 2
with standard measurement 关26兴 and model uncertainties. Each
where w̄ ua is the uniform vertical velocity across the accumulator. data set, corresponding to a different chilled water flow rate, was
Taking the Laplace transform of Eq. 共38兲, the associated transfer obtained following a minor device modification 共where a seal was
function is installed to eliminate small leaks between the tube bank and the
lower accumulator兲. Consistent with the results shown in Fig. 6,
Tr ua 共 s 兲 ⫽exp共 ␭L ua 兲 , (39) the measurements in Fig. 7 match theoretical predictions to within
where ␭⫽(w̄ ua ⫺ 冑 w̄ 2ua ⫹4 ␣ 2 )/2␣ f
. Given Eqs. 共28兲, 共37兲, and 5 dB RMS. The largest discrepancies appear at very low frequen-
共39兲, the total transfer function of the entire attenuator device is cies where chilled water input temperature variations are small
and large attenuation is not needed. The theoretical envelopes
Tr T 共 s 兲 ⫽Tr la 共 s 兲 •Tr a 共 s 兲 •Tr ua 共 s 兲 . (40) representing model uncertainty are calculated via Eq. 共40兲 using
measured mean flow rates and mean attenuator water levels along
7 Experimental Setup
An experimental setup for testing attenuator performance is
shown in Fig. 5. A constant-displacement pump maintains a con-
stant flow rate of water through the system. A cartridge heater,
controlled through a computer-based data acquisition board and
connected to a variable power supply, is used to impose single-
frequency sinusoidal temperature variations on the flow. The com-
puter also captures input and output temperatures from the ther-
mal gradient attenuator using thermistors and a high-precision
digital multimeter 共via GPIB interface兲.
8 Transfer Function Results

Measured and theoretical transfer functions for the thermal gra-
dient attenuator are shown in Fig. 6, where attenuation is shown
for 18 imposed frequencies. The root mean squared 共RMS兲 differ-
ence between measured and theoretical values is approximately 3
dB. Also shown for reference is a transfer function corresponding Fig. 7 Measured and theoretical transfer functions versus fre-
to a perfectly mixed attenuator volume and a transfer function quency, with uncertainty bars on measured points and uncer-
corresponding to an attenuator dominated by axial conduction, tainty envelopes for theoretical curves, where envelopes reflect
where the total volume is equal to that of the experimental device. uncertainty in flow rates and accumulator water levels
with associated estimated uncertainties. Measurement uncertainty
is estimated by first expressing the system transfer function as
the ratio of the amplitudes of the output and input temperature
variations
T out 共 f 兲
Tr 共 f 兲 ⫽ . (41)
T in 共 f 兲
The uncertainty in the transfer function is then estimated by as-
suming that individual uncertainties are independent and normally
distributed, so that 关27兴
␴ Tr ⫽ 冋冉冊⳵ Tr
⳵ T out
2
• ␴ Tout
2
⫹ 冉冊册
⳵ Tr
⳵ T in
2
• ␴ Tin
2
1/2
, (42)
Fig. 9 Attenuator output temperature and temperature deriva-
tive versus time for the unfiltered chilled water stream with a
where superimposed 8.6 mHz sinusoid
␴ Tout
2
⫽ ␴ Tout
2
共 f 兲 ⫽ 共 S t • ␴ T 兲 2 ⫹ 共 T out 共 f 兲 • ␴ s 兲 2 (43)
␴ Tin
2
⫽ ␴ Tin
2
共 f 兲 ⫽ 共 S t • ␴ T 兲 2 ⫹ 共 T in 共 f 兲 • ␴ s 兲 2 (44)
ders of magnitude to order 1 mK/min, which is a suitably well
and ␴ T 共⫽0.25 ohm兲 is the thermistor measurement uncertainty, behaved source for precision temperature control.
S t (⫽1.77•10⫺3 K/ohm兲 is the nominal thermistor sensitivity, and
␴ s (⫽5•10⫺4 K/K兲 is the uncertainty of S t .
10 Summary and Conclusions
9 Attenuation of Broadband Thermal Variations The analysis, design, and performance of a precision thermal
The device performance was also tested with the inlet stream gradient attenuator has been presented. For purposes of analysis,
subject to broadband thermal disturbances. For this experiment, the system was decomposed into three subsystems: a lower accu-
in-house chilled water serves as the control fluid stream, where mulator where the control fluid collects and is passively mixed, a
the associated power spectrum in Fig. 2 shows that temperature tube bank-attenuator where the control fluid exchanges heat with a
variations are spectrally broadband. In addition, a sinusoidal 共8.6 surrounding, stagnant attenuation medium 共here, water兲, and an
mHz兲 thermal variation was superimposed on the control fluid upper accumulator where the control fluid again collects prior to
stream to provide additional higher frequency disturbance to fully exiting the device. Analytic expressions have been developed to
test the attenuator performance 共at 2.2 l/min flow rate兲. As shown describe the performance of each subsystem and the total system.
in Fig. 8, the resultant in-stream temperature exhibits large ampli- A set of experiments in which sinusoidal temperature distur-
tude 共3.5 K peak to peak兲 variations and equally large 共4 K/min兲 bances are imposed on the control fluid stream, have been per-
time rates of change. formed in order to evaluate device performance. Performance
Under these conditions, the device functions at a low-pass filter, measured over a disturbance frequency range from 0.1 mHz to 10
where high frequency disturbances are suppressed to magnitudes mHz, and at four representative control stream flow rates, is well
below measurement sensitivity 共for the 8.6 mHz variation兲, and described by the analytic model. At all but the lowest frequencies,
transmitted low frequency disturbances are reduced in magnitude; the system provides significantly greater attenuation than that pro-
see Fig. 9. The dominant variation in the output stream reflects the duced by a perfectly mixed volume. In addition, a comparison
slow periodic oscillation of the input at approximately 3 cycles with earlier work 关14兴 shows that the present device provides
every 2 hours 共⬇0.4 mHz兲. The attenuation of approximately 2 frequency-dependent attenuation which is two, and, at higher fre-
orders of magnitude for this low frequency 共Fig. 7兲 is consistent quencies, almost three orders of magnitude greater than the earlier
with the data in Figs. 8 and 9. device.
The most important result of Fig. 9 is the dramatic reduction in Finally, an experiment was performed to investigate the system
the short-term variations, or temperature derivative. Achieving performance under conditions where the control stream is subject
temperature control better than milliKelvin levels without filtering to broadband thermal disturbances. In this case, the system func-
the input stream in Fig. 8 would be very difficult because of the tions at a low-pass filter/attenuator, effectively eliminating spec-
large and rapidly changing disturbances. After attenuation by the tral components having periods shorter than the device thermal
attenuator device, the short-term deviations are reduced by 3 or- equilibration time, ␶ a , and attenuating lower frequency compo-
nents. Consistent with attenuation observed for spectrally pure
disturbances, it is found that unfiltered temperature oscillations
exhibit increasing attenuation with increasing spectral frequency.
The analytic model allows prediction of system response and
provides simple criteria for designing system performance. The
model shows that for fixed control stream flow rate, the magnitude
of maximal attenuation can be tuned by adjusting the total thermal
resistance between the stream and the attenuator medium. Given a
thermal resistance that produces a desired maximal attenuation,
the frequency range subject to maximal attenuation can be tuned
by adjusting the attenuator heat capacity. The analysis and design
are general and should prove useful in the design and analysis of
other high-throughput precision temperature control systems.
Acknowledgment
Fig. 8 Attenuator input temperature and temperature deriva-
tive versus time for an unfiltered chilled water stream with a This work was supported by the National Science Foundation,
superimposed 8.6 mHz sinusoid Grant No. DMI-9500390.
Nomenclature w̄ ⫽ average axial in-tube fluid velocity
w̄ ua ⫽ average axial velocity in the upper
␣a ,␣ f ⫽ attenuator and fluid diffusivities
accumulator
As ⫽ convection surface area
␨ ⫽ heat capacity ratio: attenuator to fluid
␤ ⫽ thermal expansion coefficient
C a ,C f ⫽ single tube attenuator and fluid total heat
capacities References
Df ⫽ fluid diameter 关1兴 Bryan, J., 1990, ‘‘International Status of Thermal Error Research,’’ Annals of
f ⫽ frequency, Hz the CIRP, 39, pp. 645–656.
fl ⫽ upper frequency limit determined by the 关2兴 Slocum, A. H., 1992, Precision Machine Design, Prentice Hall, NJ, p. 96.
inverse radial diffusion time scale 关3兴 Smith, S. T., and Chetwynd, D. G., 1997, Foundations of Ultraprecision
Mechanism Design, Gordon and Breach, Amsterdam, p. 84.
hf ⫽ fluid convection coefficient 关4兴 Cutkosky, R. D., and Field, B. F., 1974, ‘‘Standard Cell Enclosure with 20 ␮K
k a ,k f ,k t ⫽ attenuator, fluid, and tube conductivities Stability,’’ IEEE Trans. Instrum. Meas., 23, pp. 295–298.
l/min ⫽ liters per minute 关5兴 Lee, H. S., Yang, S. H., and Chung, N. S., 1990, ‘‘Temperature Controller
L ⫽ attenuator tube length Using an Error Signal Modulation,’’ Rev. Sci. Instrum., 61, pp. 1329–1331.
关6兴 Dratler, J., 1974, ‘‘A Proportional Thermostat with 10 Microdegree Stability,’’
L ua ⫽ upper accumulator length Rev. Sci. Instrum., 45, pp. 1435–1444.
ṁ ⫽ mass flow rate of water through each 关7兴 Esman, R. D., and Rode, D. L., 1983, ‘‘100 ␮K Temperature Controller,’’
tube Rev. Sci. Instrum., 54, pp. 1368–1370.
关8兴 Cutkowsky, R. D., and Davis, R. S., 1981, ‘‘Simple Control Circuit for Tem-
Ṁ ⫽ total mass flow rate of water through the perature Control and Other Bridge Applications,’’ Rev. Sci. Instrum., 52, pp.
attenuator device 1403–1405.
␩ ⫽ dimensionless axial coordinate 关9兴 Strem, R. B., Das, B. K., and Greer, S. C., 1981, ‘‘Digital Temperature Control
␩L ⫽ dimensionless tube length and Measurement System,’’ Rev. Sci. Instrum., 52, pp. 1705–1708.
关10兴 Williams, G. I., and House, W. A., 1981, ‘‘Thermistor Controlled Water/Oil
O() ⫽ order of magnitude of the expression Bath for Precision Measurements in the Range 0-30 C,’’ J. Phys. E, 14, pp.
␳f ⫽ fluid density 755–760.
r a ,r f ,r t ⫽ attenuator, fluid and tube outer radii 关11兴 Miller, R. J., and Gleeson, H. F., 1994, ‘‘A Computer Interfaced High-Stability
rm ⫽ radius encompassing half of the attenua- Temperature Controller and Heating Stage for Optical Microscopy,’’ Measure-
ment Science and Technology, 14, pp. 904–911.
tor area 关12兴 Sarid, D., and Cannell, D. S., 1974, ‘‘A 15 Microdegree Temperature Control-
r ua ⫽ upper accumulator outer radius ler,’’ Rev. Sci. Instrum., 45, pp. 1082–1089.
R ⫽ thermal resistance between the fluid flow 关13兴 Harvey, M. E., 1968, ‘‘Precision Temperature-Controlled Water Bath,’’ Rev.
and the lumped attenuator medium Sci. Instrum., 39, pp. 13–18.
关14兴 Ogasawara, H., 1986, ‘‘Method of Precision Temperature Control Using Flow-
R a ,R f ,R t ⫽ attenuator, fluid, and tube thermal resis- ing Water,’’ Rev. Sci. Instrum., 57, pp. 3048–3052.
tances 关15兴 Priel, Z., 1978, ‘‘Thermostat with a Stability of 3.5 ␮K,’’ J. Phys. E, 11, pp.
s ⫽ dimensionless Laplace transform vari- 27–30.
able 关16兴 Iglesias, M. C. L., and Baron, M., 1990, ‘‘Temperature Control in Small Air-
␶ ⫽ dimensionless time Heated Measuring Cells,’’ Rev. Sci. Instrum., 61, pp. 2245–2246.
关17兴 Ogasawara, H., and Nishimura, J., 1982, ‘‘Frequency Stabilization of Internal-
␶a ,␶ f ⫽ attenuator and fluid time constants Mirror He-Ne Lasers by a Flowing Water Method,’’ Appl. Opt., 21, pp. 1156–
␶m ⫽ characteristic residence time for the flow 1157.
in the lower accumulator 关18兴 Ogasawara, H., and Nishimura, J., 1983, ‘‘Frequency Stabilization of Internal-
Mirror He-Ne Lasers,’’ Appl. Opt., 22, pp. 655–657.
T ⫽ temperature 关19兴 Sydenham, P. H., and Collins, G. C., 1975, ‘‘Thermistor Controller with Mi-
T̄ a ,T̄ f ⫽ radially averaged mean attenuator and crokelvin Stability,’’ J. Phys. E, 8, pp. 311–315.
fluid temperatures 关20兴 Neto, F. S., and Cotta, R. M., 1992, ‘‘Counterflow Double-Pipe Heat Ex-
changer Analysis Using a Mixed Lumped-Differential Formulation,’’ Int. J.
T̂ a ,T̂ f ⫽ Laplace transformed mean attenuator Heat Mass Transf., 35, pp. 1723–1731.
and fluid temperatures 关21兴 Neto, F. S., and Cotta, R. M., 1993, ‘‘Dynamic Analysis of Double-Pipe Heat
T in ,T out ⫽ time varying input and output tempera- Exchangers Subjected to Periodic Inlet Temperature Disturbances,’’ Warme
and Stoffubertragung, 18, pp. 497–503.
tures 关22兴 Ozisik, M. N., and Murray, R. L., 1974, ‘‘On the Solution of Linear Diffusion
T̄ la ⫽ lower accumulator average temperature Problems With Variable Boundary Conditions,’’ ASME J. Heat Transfer, 96,
Tr ⫽ transfer function pp. 48–51.
关23兴 Incropera, F., and DeWitt, D., 1990, Introduction to Heat Transfer, Wiley,
Tr a ,Tr la ,Tr ua ,Tr T ⫽ attenuator, lower accumulator, upper New York, p. 432.
accumulator, and total transfer functions 关24兴 Dorf, R. C., and Bishop, R. H., 1995, Modern Control Systems, Addison-
Trm a ⫽ attenuator transform function magnitude Wesley, New York, p. 394.
us ⫽ buoyancy velocity scale 关25兴 Bejan, A., 1984, Convection Heat Transfer, Wiley, New York, p. 298.
关26兴 International Organization for Standardization, 1993, Guide to the Expression
uជ ⫽ vector velocity field of Uncertainty in Measurements, Geneva.
V la ⫽ volume of lower accumulator 关27兴 Beckwith, T. G., Marangoni, R. D., and Lienhard, J. H., 1993, Mechanical
w ⫽ axial velocity in the z-direction Measurements, Addison-Wesley, New York, p. 82.
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.
Detailed Heat Transfer Coefficient cool, and subjected to potential damage. Therefore, blade tip is
recognized as a critical area in turbine blade design. Blade tip
Distributions on a Large- failure is primarily caused by the hot leakage flow through the tip
gap. The leakage flow accelerates due to pressure difference be-
Scale Gas Turbine Blade Tip tween the pressure and the suction sides of a blade. This tip leak-
age flow is undesirable because it might lead to cracking and
Shuye Teng oxidation along the pressure side tip corner from mid-chord to
trailing edge. Blade tip clearance also has a significant influence
on turbine efficiency, thus, the designers improve efficiency by
Je-Chin Han either reducing the tip clearance gap if possible or implementing a
e-mail: jchan@mengr.tamu.edu more effective tip leakage sealing mechanisms. Whatever design
choice is selected for a particular turbine blade tip, accurate mea-
G. M. S. Azad surement of detailed flow field and heat transfer in the blade tip
region is extremely important. Reliable experimental data are also
Turbine Heat Transfer Laboratory, important for computational model development and efficient
cooling system design.
Department of Mechanical Engineering, Some studies have focused on the blade tip leakage flow for
Texas A&M University, aerodynamic purposes and studied the effects of blade tip leakage
College Station, TX 77843-3123 flow on blade performance. Studies by Bindon and Morphus 关1兴
and Bindon 关2兴 have contributed to the general understanding of
tip-gap flow patterns. Bindon made pressure and flow field mea-
surements on an axial turbine blade tip in a linear cascade under
Measurements of detailed heat transfer coefficient distributions on
low-speed conditions. He suggested that the pressure side near tip
a turbine blade tip were performed in a large-scale, low-speed
flow rolls over the tip due to strong active pressure gradient across
wind tunnel facility. Tests were made on a five-blade linear cas-
the tip. The leakage flow has a sink-like character on the pressure
cade. The low-speed wind tunnel is designed to accommodate the
side and a source-like character on the suction side near the tip.
107.49 deg turn of the blade cascade. The mainstream Reynolds
The tip leakage flow has a significant effect on the blade tip heat
number based on cascade exit velocity was 5.3⫻ 105 . Upstream transfer, the pressure side near-tip region heat transfer, and the
unsteady wakes were simulated using a spoke-wheel type wake suction side near-tip heat transfer. Moore et al. 关3兴 studied the
generator. The wake Strouhal number was kept at 0 or 0.1. The flow and heat transfer in turbine tip gaps. Yamamoto 关4兴 made
central blade had a variable tip gap clearance. Measurements flow measurements between adjacent blades in the passage using
were made at three different tip gap clearances of about 1.1 per- a five-hole pitot probe. Yaras et al. 关5兴 studied the effect of simu-
cent, 2.1 percent, and 3 percent of the blade span. Static pressure lated rotation on tip leakage and found a significant reduction in
distributions were measured in the blade mid-span and on the the gap mass flow rate due to rotation. Sjolander and Cao 关6兴
shroud surface. Detailed heat transfer coefficient distributions studied the flow field in an idealized turbine tip gap. Kaiser and
were measured on the blade tip surface using a transient liquid Bindon 关7兴 investigated a quantitative analysis of the effects of tip
crystal technique. Results show that reduced tip clearance leads clearance, tip geometry, and multiple stages on turbine stage effi-
to reduced heat transfer coefficient over the blade tip surface. ciency in a rotating turbine rig environment.
Results also show that reduced tip clearance tends to weaken the The earliest study on simulated two-dimensional rectangular tip
unsteady wake effect on blade tip heat transfer. heat transfer is done by Mayle and Metzger 关8兴. They studied heat
关DOI: 10.1115/1.1373655兴 transfer to a simulated tip with and without a rotating shroud.
Results showed that the effect of moving wall is negligible over
Keywords: Forced Convection, Heat Transfer, Measurement
Techniques, Turbines, Unsteady, Wakes the entire range of parameters considered in the study. The
pressure-driven flow through the tip gap mainly influences the
blade tip heat transfer. The experiments that neglected the effect
of blade rotation may be used to assess the blade tip heat transfer
Introduction as long as the flow conditions are similar. Metzger and Rued 关9兴
Turbine blades are exposed to high thermal loads associated presented heat transfer and flow results in the near tip region of
with high temperature gases from the combustor. Blade tips are the pressure surface using a two-dimensional rectangular tip simu-
one of the most susceptible regions to burnout due to this high lation and a sink flow. This model simulated a sink flow similar to
thermal load. They are exposed to hot gases on all sides 共including the sink like characteristics of the blade tip gap on the pressure
pressure side, suction side, and blade tip兲, typically difficult to side. Rued and Metzger 关10兴 presented heat transfer and flow
results in the near tip region of the suction surface using a two-
dimensional rectangular tip simulation and a source flow. This
HEAT TRANSFER. Manuscript received by the Heat Transfer Division June 27, 2000; model simulated a blade suction surface with a gap feeding a
revision received December 5, 2000. Associate Editor: M. Hunt. source flow similar to the flow exiting the tip gap on the suction
surface of an airfoil. There are some other studies on squealer or
grooved blade tip heat transfer because turbine blade tips are typi-
cally grooved chordwise to reduce tip flow and heat transfer. The
groove acts as a labyrinth seal to increase flow resistance and thus
reduce leakage flow that in turn reduces heat transfer. Chyu et al.
关11兴 and Metzger et al. 关12兴 studied heat transfer on grooved two-
dimensional rectangular tip model.
With the development of efficient numerical codes, some re-
searchers carried out numerical investigations to predict blade tip
heat transfer and flow field. Metzger et al. 关13兴 provided a numeri-
cal model to estimate tip and shroud heat transfer. Ameri and
Steinthorsson 关14,15兴 predicted rotor blade tip and shroud heat
transfer for a SSME 共Space Shuttle Main Engine兲 turbine. Ameri
et al. 关16兴 predicted flow and heat transfer on the tip of a GE-E3
first stage rotor blade for both a smooth tip and recessed tip.
Ameri et al. 关17兴 also predicted the effects of tip clearance and
casing recess on heat transfer and stage efficiency for several
squealer blade tip geometries. Most recently, Ameri and Bunker
关18兴 performed a computational study to investigate detailed heat
transfer distributions on blade tip surfaces for a large power gen- Fig. 1 Five blade cascade with the tip of the central blade
eration turbine. They compared and validated their model with the coated with liquid crystal and viewed by a RGB camera
experimental data of Bunker et al. 关19兴 obtained for the same
geometry.
There are comparatively limited data in the open literature on
blade tip heat transfer in a cascade environment. Metzger et al. similar to the one used by Ou et al. 关23兴, simulated the upstream
关13兴 used several heat flux sensors and measured local tip heat unsteady wake. The wake generator had 32 rods, each 0.63 cm in
fluxes on the flat tips at two different tip gaps in a rotating turbine diameter, to simulate the trailing edge of an upstream vane. The
rig. Yang and Diller 关20兴 reported local heat transfer coefficient at wake Strouhal number was adjusted by controlling the rod rota-
a single location on a turbine blade model with a recessed cavity tion speed (N). The blade configuration, scaled up five times,
in a stationary linear cascade. Bunker et al. 关19兴 studied the heat produced a velocity distribution typical of an advanced high-
transfer and flow on the first stage blade tip of a power generation pressure aircraft turbine blade row. The unsteady wakes are actu-
gas turbine. The blade tip geometry is that of a flat tip with a ally velocity deficiencies caused by the blockage of the main-
recessed shroud casing and a nominal tip clearance of approxi- stream flow by the rotating rods. The ensemble-averaged
mately 1 percent of the blade span. Tip surface heat transfer is turbulence intensity profiles at the cascade inlet 共Du et al. 关24兴兲
presented for sharp-edge and radius-edge tips, three clearances, show that intensity could be as high as 20 percent inside the wake,
and two freestream approach turbulence levels. This study pro- but the time mean averaged turbulence intensity was about 10.4
vided nearly full surface information on heat transfer coefficients percent for a wake Strouhal number (S) of 0.1. For cases without
in a non-rotating cascade environment with appropriate pressure the unsteady wake effect, the time mean averaged turbulence is
distribution on the blade tip and shroud model. Azad et al. 关21,22兴 about 0.7 percent.
studied the flow and heat transfer on the first stage blade tip of an Figure 2 presents a schematic view of the blade tip heat transfer
aircraft gas turbine 共GE E3 blade兲. The effects of tip-gap clearance setup. The cascade had five blades, each with an axial chord
and free-stream turbulence level on the detailed heat transfer co- length of 17 cm and a radial span of 25.4 cm. The blade spacing
efficient distributions under engine representative flow conditions was 17.01 cm at the cascade inlet and the throat-to-blade span
were obtained for both the plane and squealer tips using a tran- ratio was 0.2. The central blade was the test blade with variable
sient liquid crystal technique. They found that the overall heat tip gap clearances. The neighborhood blades did not have tip gap
transfer coefficients on the squealer tip are lower than that on the clearance. The test blade was made of non-conductive material
plane tip under the same conditions. 共RENSHAPE兲 to satisfy the semi-infinite solid assumption used in
The present study used a transient liquid crystal technique to the transient liquid crystal technique. It is a non-conductive insu-
measure the detailed heat transfer coefficient distributions on a lating material, similar to wood. The maximum blade thickness at
gas turbine blade tip. Tests were performed on a five-blade linear 50 percent span is 4.80 cm. The tip of the central test blade was
cascade in a large-scale, low-speed wind tunnel facility. The blade coated with a thin layer of liquid crystal for detailed heat transfer
configuration produced a velocity distribution typical of an air- measurement.
craft gas turbine blade. Pressure measurements were made in the Figure 3 shows the pressure tap locations on the test blade and
blade mid-span area as well as on the shroud surface, which is the shroud opposite the test blade. The tap holes had an outer and
opposite the blade tip surface. Heat transfer measurements were inner diameter of 1.65 mm and 0.83 mm, respectively. One set of
made on the blade tip surface for a tip gap clearance of 1.1 per- tap holes was located in the mid-span 共50 percent of the blade
cent, 2.1 percent, and 3 percent of the blade span 共25.4 cm兲, span兲 of the test blade. The pressure taps were placed on both the
respectively, under both steady and unsteady flow conditions. pressure and suction surfaces. Pressure taps were also placed on
the shroud surface, as shown in Fig. 3, to measure the pressure
distribution on the shroud surface, which is a good indicator of the
Experimental Apparatus tip surface pressure. A total of 48 pressure taps were used to
Figure 1 shows the schematic of the test section and the camera measure the shroud surface pressure. One subset of pressure taps
location. The test apparatus consisted of a low speed wind tunnel was located around the tip perimeter and along the central camber
with an inlet nozzle, a linear turbine blade cascade with the test line, and a second subset of taps was located 10 mm outside the
blade in the center, and a suction type blower. The wind tunnel tip edge. All these pressure taps were located on the shroud sur-
was designed to accommodate the 107.49 deg turn of the blade face opposite the tip surface. No pressure taps were placed on the
cascade. The cascade inlet mean velocity was about 20 m/s. The tip surface. These pressure distributions were useful in estimating
mean velocity increased 2.5 times from the inlet of the cascade to the tip leakage flow. Pressures were measured by an inclined
the exit. The inlet Mach number is 0.072. The peak Mach number manometer.
is 0.144 at X/SL⫽0.45. A spoked wheel type wake generator, Figure 4 presents the heater instrumentation for the blade tip
Fig. 3 Pressure tap locations on blade and shroud „unit: cm…
Fig. 2 Schematic view of blade tip heat transfer setup „unit: create tip gaps of the desired heights. The 1.1 percent, 2.1 percent,
cm… and 3.1 percent tip gaps are defined according to the span of the
guide blades.
Before the heat transfer experiment, the initial temperatures,
T i , were obtained by placing 36 thermocouples on the tip surface.
heat transfer experiment. The inside of the top 1/4 portion of the
blade was hollow for heater instrumentation. The blade tip portion
had a thickness of 0.64 cm, which was thick enough to meet the
one-dimensional semi-infinite model assumption used in the tran-
sient liquid crystal method. The blade tip was heated underneath
by a copper plate with a heating wire. Heating wire was uniformly
distributed over one side of a 0.1 cm thick copper plate. The other
side of the copper plate was tightly glued to the underside of the
blade tip portion. This ensured the uniform heating of the blade tip
region. However, there was no heating wire in the trailing edge
portion because the blade was too thin there. The blade shell had
a thickness of 0.48 cm. The inside surface of the blade shell was
instrumented with a thin foil heater. The foil had a thickness of
0.0038 cm. During the test, the blade shell was heated to a tem-
perature as high as the blade tip to prevent heat conduction from
the edge of the blade tip to the rest of the blade. After both the
heating wire and the foil heater had been instrumented, the top
hollow portion was filled with fiber glass insulation material. The
top portion was then glued together with the rest of the blade
model. The span of this test blade was 24.64 cm. This test blade
will provide a up to 3 percent tip clearance from the shroud. The
four blades for flow guidance 共as shown in Fig. 2兲 have a span of
25.4 cm and there is no tip clearance between the guide blades
and the shroud. The test blade was fastened to the bottom endwall
with screws. Different hard rubber gaskets of desired thickness
were placed between the test blade bottom and the endwall to Fig. 4 Blade tip heater instrumentation „unit: cm…
One set of thermocouples was located along the tip perimeter and
another along the central camber line. These temperatures were
then interpolated to get an initial temperature map over the whole
tip surface. The uniformity of the surface temperature with heat-
ing was within ⫾2°C. An interpolation scheme was used to fur-
ther reduce the temperature non-uniformity in the initial surface
temperature to within ⫾0.2°C. In the heat transfer experiment, all
thermocouples were taken away. The whole blade tip surface was
coated with a thin layer of liquid crystal 共Hallcrest, Type BM/
R32C5W/C17-10兲. Color to temperature calibration was done on
a copper bar coated with liquid crystal as suggested by Vedula
and Metzger 关25兴. The primary color display of liquid crystal is
from an initial colorless stage to red, green, blue, and then color-
less again. The color change temperatures to red, green, and blue
are 31.6°C, 32.7°C, and 37.2°C, respectively. The coating was a
light spray of several ␮m thickness. Ireland and Jones 关26兴
showed the response time for the liquid crystals to be only a few
milliseconds. The duration of the transient test is much longer
than the response time of the liquid crystal. The test blade tip
surface was heated uniformly by the copper plate with uniformly
distributed heating wire. The blade surface temperature was moni-
tored during heating using two embedded thermocouples placed
near the top portion of both the suction and pressure sides. In the
present study, the blade surface was preheated to a temperature Fig. 5 Static pressure distribution „total pressureÕlocal static
above the liquid crystal blue color 共37.2°C兲. The liquid crystal pressure… at the blade mid-span portion
color change time was measured using a high precision image
processing system. The system consisted of a RGB 共Red, Green,
and Blue兲 camera connected to a color frame grabber board in the Figure 6 presents the static pressure distributions in terms of
PC and a color video monitor. The composite signal from the P o / P on the shroud surface opposite the blade tip surface for the
camera was split into red, green, and blue components before three different tip gaps: 1.1 percent, 2.1 percent, and 3 percent.
passing on to the color frame grabber board inside the PC. An Pressure measurements were not made on the blade tip surface.
image processing software was used to measure the time of color
change of liquid crystals. Details on the image processing system
were presented by Teng et al. 关27兴. The suction blower was
started before the transient test. It took about 5 s for the flow to
reach a full cascade inlet speed of 25 m/s. The velocity quickly
increases from 15 m/s to 25 m/s in the last second of that 5 s. The
image system began to take data 5 s after the start of the blower.
Several surface temperature points were taken once more after the
suction blower had started for 5⫾1 seconds to calibrate the initial
temperature distribution. The blade surface was heated to a very
high temperature so that it would take long time to be cooled
down to the temperature where the liquid crystal changes color.
The transient test took an average of 60 to 90 s. So the uncertainty
caused by the accelerating of the mainstream flow at the very
beginning of the transient tests was controlled to be within 8.3
percent.
Flow Condition in the Cascade and Pressure Measure-

ments
Figure 5 presents the surface static pressure distribution mea-
sured at the midspan portion of a pressure tap instrumented blade.
Pressure distribution was measured using a Dwyer manometer
共Model No. 424兲. The uncertainty is respectively ⫾0.01 and ⫾0.1
inches of water when ⌬ P is less or larger than 2 inches of water.
The static pressure is presented in terms of P o / P, which is the
ratio of the total pressure ( P o ) and the local static pressure 共P兲.
Due to the suction wind tunnel, the local static pressure 共P兲 is
lower than the atmospheric pressure ( P o ). A higher value of
P o / P corresponds to a lower static pressure, while a lower value
corresponds to a higher static pressure. The static pressure on the
suction side decreases to about X/SL⫽0.45 and then increases
slightly to the exit. The static pressure on the pressure side is only
slightly less than the total pressure in the region from the leading
edge to X/ PL⫽0.4. After X/ PL⫽0.4, the static pressure of the
pressure side begins to decrease all the way to the exit. From Fig.
5 we see that, at 50 percent of the blade span, the maximum static
pressure difference between the pressure and suction sides occurs Fig. 6 Static pressure distributions on the shroud surface op-
at about X/SL⫽X/ PL⫽0.4. posite to the blade tip surface for three different tip gaps
Here, again, a higher value of P o / P corresponds to a lower static
pressure, while a lower value corresponds to a higher static pres-
sure. The contour plot in Fig. 6 helps explain the heat transfer
results on the tip surface. The blade shape black curve indicates
the blade location under the shroud. The P o / P distributions
clearly demonstrate the tip leakage flow path. The lower P o / P
value 共the higher static pressure兲 near the pressure side leading
edge portion indicates where the leakage flow enters the tip gap,
while the higher P o / P value 共the lower static pressure兲 at the
suction side near the trailing edge portion indicates that the leak-
age flow exits the tip gap through this location. The lowest static
pressure spot is located at about half chord length and in the
center of the blade tip surface. In the leakage flow path, a separa-
tion vortex is generated as the incoming flow separates off the
inner edge of the pressure side rim and exits near the suction side
trailing edge. Sjolander and Cao 关6兴 observed this separation vor-
tex and Ameri et al. 关16兴 also predicted this phenomenon. Due to
this separation vortex, the heat transfer rate increases in this re-
gion as will be shown in the heat transfer results. Figure 6 shows
that, for the smallest tip gap 共1.1 percent兲, the pressure gradient
was slightly increased in the front tip portion and greatly reduced
in the central portion, which caused the leakage flow to shift
slightly forward and be weakened on the central tip surface.
Heat Transfer Data Analysis

A transient liquid crystal technique is used to measure the de-
tailed heat transfer coefficients on the blade tip surface. The tech-
nique is similar to the one described by Teng et al. 关27兴. A one-
dimensional transient conduction into a semi-infinite solid with
convective boundary condition was assumed. The solution for sur-
face temperature was obtained as Fig. 7 Detailed Nusselt number distributions without unsteady
冋冉冊冉冊册
wake effect „ S Ä0…
T w ⫺T i h 2␣ t h 冑␣ t
⫽ 1⫺exp erfc (1)
T m ⫺T i k 2
k
where T w is the wall temperature when liquid crystals change to span兲. The tip leakage originates from the pressure side near the
red from green 共32.7°C兲 at time t, T i is the initial surface tem- leading edge, sweeps across from pressure to suction side over the
perature, T m is the oncoming mainstream flow total temperature, central part of the tip surface, and then exits through the suction
and ␣ and k are the thermal diffusivity and conductivity of the side. This area of tip leakage flow corresponds to a high heat
insulation blade material, respectively. The heat transfer coeffi- transfer coefficient region as shows in Fig. 7共b兲. The heat transfer
cient was obtained from Eq. 共1兲. Please note that here the oncom- coefficient along this tip leakage flow path is higher toward the
ing mainstream flow total temperature T m 共instead of the local pressure side than the suction side. This happens because of the
bulk flow temperature inside the tip flow passage兲 is used as a flow entrance effect. The leakage flow enters through the pressure
reference temperature to calculate h. The local heat transfer rate is side and exits through the suction side. A separation vortex is
q⫽h(T w ⫺T m ). The experimental uncertainty in the measurement generated at the pressure side as the incoming flow separates from
of the local heat transfer coefficient, based on Kline and McClin- the inner edge of the pressure side rim and exits near the trailing
tocks’ 关28兴 methodology, is about ⫾ 8.3 percent. The uncertainty edge of the blade. The lower heat transfer coefficient toward the
in the region of blade tip edge could be higher due to two- suction side may be the result of a boundary layer that develops at
dimensional conduction effect. the pressure side edge and grows toward the suction side. In ad-
dition, although the static pressure of the suction surface is lower
Heat Transfer Measurements than that of the pressure surface, it is higher than the lowest static
pressure spot in the central blade region. After the leakage flow
Experiments are performed at a cascade exit Reynolds number passes the lowest pressure spot, it actually encounters an adverse
of 5.3⫻105 . The corresponding flow velocity at the cascade exit pressure gradient and is forced to change its original flow direc-
is 50 m/s. For the 3 tip gaps, tests were run for no rod, no wake tion from towards the suction surface to towards the trailing edge.
effect cases (T̄u⫽0.7 percent) and cases with wake Strouhal num- This could also contribute to the comparatively lower heat transfer
ber of 0.1(T̄u⫽10.4 percent). coefficient in the tip region toward the suction side. In Fig. 7共b兲,
Figure 7 presents the detailed Nusselt number distributions for we also notice a triangular region of significantly low heat transfer
the 3 tip gaps under the no wake effect condition. The general coefficients between the leading edge and the front suction sur-
experimental trends observed are consistent with the findings of face. An explanation for this occurrence can be seen in Fig. 6, the
previous investigations. However, the local surface heat transfer static pressure distribution plot. The pressure gradient is small
distributions are varied among studies because the blade shapes from the leading edge to X/C x ⫽0.2, where this low heat transfer
and flow conditions are different. In Fig. 7, regions of high and triangular region exists, while the pressure gradient is doubled
low heat transfer coefficients are shown very clearly in the de- from X/C x ⫽0.2 to 0.4, where the leakage flow goes through and
tailed distribution contour. The magnitude of the Nusselt number accelerates. Figure 7共b兲 also shows a thin high heat transfer trace
varies from about 200 to over 1000. There is no data for the starting from the leading edge portion and lasts for a short dis-
trailing edge portion 共about 25 percent of the chord length兲 due to tance along the suction side edge, this is because of a pressure
the difficulty in instrumenting heating wire in this area. gradient which exists at the leading edge portion. This pressure
Figure 7共b兲 shows the tip local heat transfer contour plots for a gradient drives another small leakage flow over the other side of
typical tip clearance gap of 5.4 mm 共2.1 percent of the blade the low heat transfer triangular region.
Fig. 9 Blade tip averaged Nusselt number distribution based
on local blade thickness
Figure 9 presents the averaged Nusselt number (N̄u) along the

blade tip width direction for the three different tip gaps. It shows
that, since the leading edge region has a very high heat transfer
coefficient, the averaged Nusselt number decreases at first from
Fig. 8 Detailed Nusselt number distributions without unsteady X/C x ⫽0 to 0.1. It then increases along the chord length and de-
wake effect „ S Ä0.1… creases towards the trailing edge. The highest averaged Nusselt
number is shown to be at around X/C x ⫽0.5 for all six cases
studied. Differences mainly exist in the mid-chord region from
X/C x ⫽0.2 to X/C x ⫽0.6. The largest tip gap, 3 percent, has the
Figure 7共a兲 and 7共c兲 present the tip local heat transfer contour
highest averaged Nusselt number in comparison to the other two
plots for another two typical tip clearance gaps of 7.62 mm and
tip gaps. Unsteady wake effect slightly enhances the averaged
2.86 mm 共3 percent and 1.1 percent of the guide blade span兲,
Nusselt number for the tip gap of 3 percent, but diminishes with a
respectively. These figures show that the tip gap has a great effect
decrease of tip gaps.
on tip heat transfer coefficient distribution. A wider tip gap results
in an overall higher heat transfer coefficient distribution, while a
narrower tip gap results in an overall lower heat transfer coeffi- Conclusions
cient distribution. This is because a wider tip gap allows more Heat transfer coefficients and static pressure distributions on a
leakage flow, while a narrower tip gap increases the flow resis- turbine blade tip were measured in a large-scale low-speed wind
tance and decreases the leakage flow. Tip gaps also have an effect tunnel facility. Tests were performed on a five-blade linear cas-
on the leakage flow path direction. The narrower tip gap results in cade. The mainstream Reynolds number based on cascade exit
the forward shifting of the leakage flow towards the leading edge. velocity was 5.3⫻105 . The blade was a large-scale two-
The low heat transfer triangular region is reduced and the heat dimensional model of an aircraft turbine blade. The blade configu-
transfer coefficient in this region increases. In Fig. 6 it is observed ration was scaled up to five times and produced a velocity distri-
that, for the smallest tip gap, the pressure gradient is slightly in- bution typical of a turbine blade row. Measurements were made at
creased in the front tip portion and greatly reduced in the central three different tip gap clearances of about 1.1 percent, 2.1 percent,
portion, which causes the leakage flow to shift slightly forward and 3 percent of the blade span. The wake Strouhal number was
and be weakened on the central tip surface. The other small leak- kept at 0 or 0.1. Detailed heat transfer coefficient distributions
age flow over the suction side edge and leading edge portion is were measured using a transient liquid crystal technique. The con-
also weakened with a decreased tip gap. clusions based on the experimental results are:
Figure 8 presents the detailed Nusselt number distributions for
the 3 tip gaps with unsteady wake effect. Unsteady wake results in 1 The detailed heat transfer measurements combined with the
an overall, though slightly, increased heat transfer coefficient dis- pressure distribution measurements provide comprehensive infor-
tribution for the wider tip gap 3 percent. For the tip gap of 2.1 mation and a better understanding of local heat transfer behavior
percent, the high heat transfer area in the mid-chord region is on blade tip surface.
slightly reduced due to the unsteady wake effect, which results in 2 A major pressure-driven leakage flow exists between the
slightly lower averaged Nusselt numbers in that region than the pressure and suction surfaces in the mid-chord region, which re-
case without unsteady wake effect. Unsteady wake does not show sults in a critical region of very high heat transfer coefficients. A
much effect on the smallest tip gap of 1.1 percent. This is because minor pressure-driven leakage flow also exists from the leading
the small tip gap has the effect of preventing the unsteady wake edge portion and lasts for a short distance along the suction side
from getting into the tip gap and enhancing the tip surface heat edge.
transfer. Unsteady wake has the greatest effect on the tip gap of 3 3 A triangular region of low heat transfer exists between the
percent: the pressure driven leakage flows are disturbed and the leading edge and the front suction surface.
heat transfer coefficient in the low heat transfer triangular region 4 Because of the entrance effect on the pressure side tip surface
is increased. as well as the adverse pressure gradient on the suction side tip
surface in the mid-chord region, a higher heat transfer coefficient ␣ ⫽ thermal diffusivity of blade material
exists on the tip surface toward the pressure side as opposed to the (0.135⫻10⫺6 m2/s)
suction side. ␯ ⫽ kinematic viscosity of cascade inlet mainstream air
5 Larger tip gaps result in a higher overall heat transfer coef-
ficient, while smaller tip gaps result in a lower overall heat trans-
fer coefficient. Decreased tip gaps slightly push the tip leakage References
flow towards the leading edge suction side.
关1兴 Bindon, J. P., and Morphus, G., 1988, ‘‘The Effect of Relative Motion, Blade
6 Unsteady wake effect slightly enhances the averaged Nusselt Edge Radius and Gap Size on the Blade Tip Pressure Distribution in An
number for a tip gap of 3 percent, but diminishes with the de- Annular Turbine Cascade With Clearance,’’ ASME Paper 88-GT-256.
crease of tip gaps. 关2兴 Bindon, J. P., 1989, ‘‘The Measurement and Formation of Tip Clearance
7 Comparison with previous studies shows that local heat Loss,’’ ASME J. Turbomach., 111, pp. 258–263.
关3兴 Moore, J., Moore, J. G., Henry, G. S., and Chaudhury, U., 1989, ‘‘Flow and
transfer coefficients are sensitive to the blade tip geometry and Heat Transfer in Turbine Tip Gaps,’’ ASME J. Turbomach., 111, pp. 301–
flow conditions. 309.
关4兴 Yamamoto, A., 1989, ‘‘Endwall Flow/Loss Mechanisms in a Linear Turbine
The observations and conclusions made out of this work are Cascade With Blade Tip Clearance,’’ ASME J. Turbomach., 111, pp. 264–
still limited in the sense that the test was conducted with station- 275.
ary blades. Cautions need to be taken in extending the findings to 关5兴 Yaras, M. I., and Sjolander, S. A., 1991, ‘‘Effects of Simulated Rotation on
Tip Leakage in a Planar Cascade of Turbine Blades: Part I—Tip Gap Flow,’’
rotating systems. In addition, the tests were run in a low speed ASME Paper 91-GT-127.
wind tunnel and there is a difference to the real Mach number 关6兴 Sjolander, S. A., and Cao, D., 1995, ‘‘Measurements of the Flow in an Ideal-
distribution of the engine design. Local flow field measurements ized Turbine Tip Gap,’’ ASME J. Turbomach., 117, pp. 578–584.
in the clearance gap region may be needed for a more complete 关7兴 Kaiser, I., and Bindon, J. P., 1997, ‘‘The Effect of Tip Clearance on the
Development of Loss Behind a Rotor and a Subsequent Nozzle,’’ ASME
characterization/description of the heat transfer results. Paper 97-GT-53.
关8兴 Mayle, R. E., and Metzger, D. E., 1982, ‘‘Heat Transfer at the Tip of An
Unshrouded Turbine Blade,’’ Proc. Seventh Int. Heat Transfer Conf., Hemi-
Acknowledgment sphere Pub., New York, pp. 87–92.
关9兴 Metzger, D. E., and Rued, K., 1989, ‘‘The Influence of Turbine Clearance Gap
This paper is prepared with the support of the NASA Lewis Leakage on Passage Velocity and Heat Transfer Near Blade Tips: Part I—Sink
Research Center under grant number NAG3-2002. The NASA Flow Effects on Blade Pressure Side,’’ ASME J. Turbomach., 111, pp. 284–
292.
technical team is Mr. Robert Boyle and Dr. Raymond Gaugler. 关10兴 Rued, K., and Metzger, D. E., 1989, ‘‘The Influence of Turbine Clearance Gap
This work was also supported by the Texas Higher Education Leakage on Passage Velocity and Heat Transfer Near Blade Tips: Part II—
Coordinating Board–Advanced Technology Program under grant Source Flow Effects on Blade Suction Side,’’ ASME J. Turbomach., 111, pp.
number 999903-165. Their support is greatly appreciated. 293–300.
关11兴 Chyu, M. K., Moon, H. K., and Metzger, D. E., 1989, ‘‘Heat Transfer in the
Tip Region of Grooved Turbine Blades,’’ ASME J. Turbomach., 111, pp.
131–138.
Nomenclature 关12兴 Metzger, D. E., Bunker, R. S., and Chyu, M. K., 1989, ‘‘Cavity Heat Transfer
on a Transverse Grooved Wall in a Narrow Flow Channel,’’ ASME J. Heat
C ⫽ tip clearance gap 共mm兲 Transfer, 111, pp. 73–39.
C ⫽ blade chord length 共23.2 cm, distance between the 关13兴 Metzger, D. E., Dunn, M. G., and Hah, C., 1991, ‘‘Turbine Tip and Shroud
Heat Transfer,’’ ASME J. Turbomach., 113, pp. 502–507.
blade leading edge and trailing edge兲关14兴 Ameri, A. A., and Steinthorsson, E., 1995, ‘‘Prediction of Unshrouded Rotor
C x ⫽ blade axial chord length 共17 cm兲 Blade Tip Heat Transfer,’’ ASME Paper 95-GT-142.
d ⫽ wake generator rod section diameter 共mm兲关15兴 Ameri, A. A., and Steinthorsson, E., 1996, ‘‘Analysis of Gas Turbine Rotor
h ⫽ local blade tip heat transfer coefficient 共W/m2•°C兲 Blade Tip and Shroud Heat Transfer,’’ ASME Paper 96-GT-189.
关16兴 Ameri, A. A., Steinthorsson, E., and Rigby, L. D., 1997, ‘‘Effect of Squealer
h̄ ⫽ averaged blade tip heat transfer coefficient along Tip on Rotor Heat Transfer and Efficiency,’’ ASME Paper 97-GT-128.
blade tip width direction 关17兴 Ameri, A. A., Steinthorsson, E., and Rigby, L. D., 1998, ‘‘Effect of Tip Clear-
k ⫽ thermal conductivity of blade material 共0.159 ance and Casing Recess on Heat Transfer and Stage Efficiency in Axial Tur-
W/m•°C兲 bines,’’ ASME Paper 98-GT-369.
关18兴 Ameri, A. A., and Bunker, R. S., 1999, ‘‘Heat Transfer and Flow on the First
k air ⫽ thermal conductivity of mainstream air Stage Blade Tip of a Power Generation Gas Turbine: Part 2—Simulation Re-
LE ⫽ leading edge of the blade sults,’’ ASME Paper 99-GT-283.
n ⫽ number of rods on wake generator 关19兴 Bunker, R. S., Baily, J. C., and Ameri, A. A., 1999, ‘‘Heat Transfer and Flow
N ⫽ speed of rotating rods on the First Stage Blade Tip of a Power Generation Gas Turbine: Part
1—Experimental Results,’’ ASME Paper 99-GT-169.
Nu ⫽ local blade tip Nusselt number based on axial chord, 关20兴 Yang, T. T., and Diller, T. E., 1995, ‘‘Heat Transfer and Flow for a Grooved
hC x /k air Turbine Blade Tip in a Transonic Cascade,’’ ASME Paper 95-WA/HT-29.
N̄u ⫽ averaged blade tip Nusselt number, h̄C x /k air 关21兴 Azad, G. M. S., Han, J. C., Teng, S., and Boyle, R., 2000, ‘‘Heat Transfer and
Pressure Distributions on a Gas Turbine Blade Tip,’’ ASME J. Turbomach.,
P ⫽ local static pressure 122, pp. 717–724.
P 0 ⫽ total pressure at the inlet 共atmospheric pressure兲关22兴 Azad, G. M. S., Han, J. C., and Boyle, R., 2000, ‘‘Heat Transfer and Pressure
PL ⫽ streamwise length on the pressure surface 共25.6 cm兲 Distributions on the Squealer Tip of a Gas Turbine Blade,’’ ASME J. Turbo-
q ⫽ local blade tip surface heat transfer rate mach., 122, pp. 725–732.
关23兴 Ou, S., Han, J. C., Mehendale, A. G., and Lee, C. P., 1994, ‘‘Unsteady Wake
Re ⫽ Reynolds number based on exit velocity and axial Over a Linear Turbine Blade Cascade with air and CO2 Film Injection: Part
chord, V 2 C x / v I—Effect on Heat Transfer Coefficients,’’ ASME J. Turbomach., 116, pp.
S ⫽ wake Strouhal number, 2 ␲ Ndn/(60V 1 ) 721–729.
SL ⫽ streamwise length on the suction surface 共33.1 cm兲关24兴 Du, H., Han, J. C., and Ekkard, S. V., 1998, ‘‘Effect of Unsteady Wake on
Detailed Heat Transfer Coefficient and Film Effectiveness Distributions for a
t ⫽ liquid crystal color change time Gas Turbine Blade,’’ ASME J. Turbomach., 120, pp. 808–817.
TE ⫽ trailing edge of the blade 关25兴 Vedula, R. J., and Metzger, D. E., 1991, ‘‘A Method for the Simultaneous
T i ⫽ initial temperature of blade surface Determination of Local Effectiveness and Heat Transfer Distributions in
T m ⫽ mainstream total temperature Three-Temperature Convection Situations,’’ ASME Paper No. 91-GT-345.
关26兴 Ireland, P. T., and Jones, T. V., 1987, ‘‘The Response Time of a Surface
T̄u ⫽ free stream mean turbulence intensity at the inlet Thermometer Employing Encapsulated Thermochromic Liquid Crystals,’’ J.
T w ⫽ liquid crystal color change from green to red Phys. E, 10, pp. 1195–1199.
V 1 ⫽ cascade inlet velocity 关27兴 Teng, S., Sohn, D. K., and Han, J. C., 1999, ‘‘Unsteady Wake Effect of Film
Temperature and Effectiveness Distributions for a Gas Turbine Blade,’’
V 2 ⫽ cascade exit velocity ASME J. Turbomach., 122, pp. 340–347.
X ⫽ streamwise distance from blade leading edge 共Fig. 5兲, 关28兴 Kline, S. J., and McClintock, F. A., 1953, ‘‘Describing Uncertainties in Single
also the distance along chord length 共Fig. 9兲 Sample Experiments,’’ Mech. Eng. 共Am. Soc. Mech. Eng.兲, 75, pp. 3–8.
Transport Phenomena of Developing mixed convection flow. Morcos et al. 关6兴 performed an experi-
mental study on laminar mixed convection heat transfer in the
Laminar Mixed Convection in entrance region of inclined rectangular ducts. A numerical study
of combined forced and free convection for laminar flow in the
Inclined Rectangular Ducts With Wall entrance region of isothermal, inclined tubes was carried out by
Transpiration Choudhury and Patankar 关7兴. Their results revealed that the buoy-
ancy effects have a considerable influence on the fluid flow and
heat transfer characteristics.
Wei-Mon Yan The heat transfer in porous tube in the region of fully developed
e-mail: wmyan@huafan.hfu.edu.tw flow has received considerable attention. Olson and Eckert 关8兴
Department of Mechanical Engineering, Huafan investigated experimentally the effects of injection on mixing
length and eddy diffusivity of turbulent flow. Their results showed
University, Shih Ting, Taipei, TAIWAN 22305, ROC the normalized friction factors agreed with those measured in ex-
ternal flow and the normalized velocity defect profiles with injec-
Pei-Yuan Tzeng tion agreed with those for flow without injection. The studies of
Department of Aeronautical Engineering, Chung Cheng heat transfer in porous tube in the developing region are available
Institute of Technology, Taoyuan, TAIWAN in the literature 关9–12兴. Yuan and Finkelstein 关9兴 presented some
entrance region heat transfer results that were restricted to asymp-
33509, ROC totically small values of the wall injection. Pederson and Kinney
关10兴 extended the work of Kinney 关13兴 to the thermal entrance
region in a porous channel for various thermal boundary condi-
A numerical calculation has been carried out to investigate the tions. Raithby 关11兴 solved the energy equation by the Fourier
mixed convection heat transfer in inclined rectangular ducts with methods. He also obtained a number of asymptotic expressions of
wall transpiration. The vorticity-velocity method is employed to the solution of strong injection. The experimental results of Hirata
solve the governing equations. The present paper particularly ad- et al. 关12兴 revealed that the flow becomes fully developed up to
dresses the effects of the independent parameters, namely, mixed the tube center at the axial location greater than 13 times of tube
convection parameter ⌬, modified Rayleigh number Ra*, wall diameter. They also proposed a simple expression of defect law
Reynolds number Re w and aspect ratio ␥. The predicted results for velocity distribution in the core region.
show that either wall injection or wall suction has a considerable The fluid flow and heat transfer in rectangular duct with porous
impact on the flow structure and heat transfer performance. Ad- walls has received attention recently. Hwang et al. 关14兴 presented
ditionally, it was found that for injection case 共 Re w ⬍0 兲 , the Nus- a numerical forced convection heat transfer and fluid flow in the
selt number Nu is retarded with an increase in the wall Reynolds entrance region of a square duct. Cheng and Hwang 关15兴 con-
number Re w , but the trend is reverse for the suction flow 共 Re w ducted an experimental study to investigate the same problem.
⬎0 兲 . 关DOI: 10.1115/1.1374439兴 They found that the wall transpiration has a significant effect on
the heat transfer and fluid flow in the entrance region of porous
Keywords: Heat Transfer, Laminar, Mixed Convection, Three- duct, but the buoyancy effect still remains untreated. Very re-
Dimensional cently, the mixed convection heat transfer in a horizontal rectan-
gular duct with wall transpiration was investigated by Lee and
Yan 关16兴. They showed that the buoyancy effect has a significant
impact on the characteristics of flow structure and heat transfer
Introduction performance. The objective of this work is to examine the wall
The heat transfer and fluid flow in porous-walled passages have transpiration effects on mixed convection in inclined ducts.
received great attention in the past decades due to their wide ap-
plications. The porous-walled ducts are used in the transpiration
cooling of high temperature thermal systems, gas-turbine blades,
combustion chambers, exhaust nozzles, porous-walled flow reac-
Analysis
tors and solar energy collectors. Mixed convection heat transfer Consideration is given to a steady laminar upward flow in the
influenced by thermal buoyancy force has received much atten- entrance region of a rectangular duct inclined at angle ␦ to the
tion. But the effects of wall transpiration on mixed convection horizontal, as shown in Fig. 1. The u, v , and w are the velocity
heat transfer in inclined ducts have not been widely studied al- components in the x, y, and z directions, respectively. The fluid
though they are always encountered in many practical situations. enters the channel with a uniform velocity w o and a constant inlet
For vertical ducts, the gravity force acts in the main flow direc- temperature T o . The porous-duct walls are maintained at a con-
tion, and there is no secondary flow in the cross section. For stant temperature T w . In addition, the porous wall may be sub-
horizontal ducts, the combined buoyancy forces are normal to the jected to a suction or blowing velocity v in . The temperature of
main flow direction and they induce secondary flows in the cross the injected or suctioned fluid is the same as that of the duct wall.
plane. For inclined ducts, however, buoyancy forces act in both The flow is assumed to be steady and have constant thermo-
main flow and the cross-stream directions. physical properties except for the density change with temperature
Numerical and experimental studies of fully-developed mixed that leads to the buoyancy terms in the momentum equations.
convection heat transfer in inclined ducts have been examined by Boussinesq approximation is invoked for the consideration of
many investigators 关1–3兴. However, the assumption of fully- de- buoyancy. With the above assumptions, the present problem can
veloped flow can only be established if the duct is very long. A be formulated by the following equations 关16兴:
numerical solution for the problem of mixed convection heat
transfer in the entrance region of inclined rectangular ducts was ⳵ 2 U/ ⳵ X 2 ⫹ ⳵ 2 U/ ⳵ Y 2 ⫽ ⳵␰ / ⳵ Y ⫺ ⳵ 2 W/ ⳵ X ⳵ Z (1)
studied by Morcos and Abou-Ellail 关4兴. Cheng and Yuen 关5兴 pre-
⳵ 2 V/ ⳵ X 2 ⫹ ⳵ 2 V/ ⳵ Y 2 ⫽⫺ ⳵␰ / ⳵ X⫺ ⳵ 2 W/ ⳵ Y ⳵ Z (2)
sented a series of photographs of developing secondary flow pat-
terns in the thermal entrance region of inclined tubes for laminar U ⳵␰ / ⳵ X⫹V ⳵␰ / ⳵ Y ⫹W ⳵␰ / ⳵ Z⫹ ␰ 共 ⳵ U/ ⳵ X⫹ ⳵ V/ ⳵ Y 兲
Contributed by the Heat Transfer Division for publication in the JOURNAL OF ⫹ 共 ⳵ W/ ⳵ Y • ⳵ U/ ⳵ Z⫺ ⳵ W/ ⳵ X• ⳵ V/ ⳵ Z 兲
HEAT TRANSFER. Manuscript received by the Heat Transfer Division May 23, 2000;
revision received January 20, 2001. Associate Editor: F. Cheung. ⫽ ⳵ 2 ␰ / ⳵ X 2 ⫹ ⳵ 2 ␰ / ⳵ Y 2 ⫺ 共 Ra* /Pr兲 • ⳵ ␪ / ⳵ X (3)
U ⳵ W/ ⳵ X⫹V ⳵ W/ ⳵ Y ⫹W ⳵ W/ ⳵ Z The boundary conditions for this problem are given by the fol-
lowing:
⫽⫺d P̄/dZ⫹ ⳵ 2 W/ ⳵ X 2 ⫹ ⳵ 2 W/ ⳵ Y 2 ⫹⌬/Pr ␪ (4) at the entrance, Z⫽0:
U ⳵ ␪ / ⳵ X⫹V ⳵ ␪ / ⳵ Y ⫹W ⳵ ␪ / ⳵ Z⫽ 共 ⳵ ␪ / ⳵ X ⫹ ⳵ ␪ / ⳵ Y 兲 /Pr.
2 2 2 2
(5) W⫽1, U⫽V⫽ ␰ ⫽0, ␪ ⫽0 (8a)
In writing the above equations, the axial diffusion terms of the on the wall 1, Y ⫽0:
momentum equations and the energy equation are neglected due
to the assumptions of high Reynolds and Peclect numbers 关16– W⫽0, U⫽0, V⫽Rew , ␪ ⫽1 (8b)
18兴. In nondimensionalizing the governing equations, the follow-
ing dimensionless variables are introduced: on the wall 2, X⫽0:
X⫽x/D e Y ⫽y/D e W⫽0, U⫽Rew , V⫽0, ␪ ⫽1 (8c)
Z⫽z/ 共 Re D e 兲 Z * ⫽Z/Pr on the wall 3, Y ⫽(1⫹ ␥ )/(2 ␥ ):

U⫽uD e / ␯ V⫽ v D e / ␯ W⫽0, U⫽0, V⫽⫺Rew , ␪ ⫽1 (8d)
W⫽w/w o P̄⫽p̄/ 共 ␳ w 2o 兲 and on the symmetric plane, X⫽(1⫹ ␥ )/4:
(6)
␪ ⫽ 共 T⫺T o 兲 / 共 T w ⫺T o 兲 Gr⫽g ␤ 共 T w ⫺T o 兲 D 3e / ␯ 2 U⫽ ⳵ V/ ⳵ X⫽ ⳵ W/ ⳵ X⫽ ⳵ ␪ / ⳵ X⫽0. (8e)
Rew ⫽ v in D e / ␯ Ra⫽PrGr In this work, there are one geometry parameter, ␥, and four
flow/thermal parameters, ⌬, Ra*, Rew , and Pr. ⌬ is a mixed con-
⌬⫽ 共 Ra/Re兲 sin ␦ Ra*⫽Ra•cos ␦
vection parameter defined by Eq. 共6兲 that contains the ratio of the
Pr⫽ ␯ / ␣ ␥ ⫽a/b. Rayleigh number Ra to the Reynolds number Re, modified by an
inclination angle factor sin ␦. For positive ⌬, the thermal buoy-
An additional constraint which is used to deduce the axial pres- ancy aids the main flow, vice versa for negative ⌬. Ra* is a
sure gradient in axial momentum equation is that global mass modified Rayleigh number given by Eq. 共6兲. It is seen that with
conservation at any axial location must be satisfied. This con- the introduction of the independent parameters ⌬ and Ra*, the
straint is expressed as inclination angle from the horizontal direction ␦ does not appear
explicitly in the formulation. Moreover, from Eq. 共6兲, it is obvious
W̄⫽1⫺Rew 共 1⫹ ␥ 兲 2 / ␥ •Z, (7)
that as
where Rew is the wall Reynolds number based on the injection or
suction velocity v in . For injection case, the Rew is negative, while ␦ →0: sin ␦ →0, Ra*→Ra, ⌬→0 (9)
for suction case the Rew is positive.
and Eqs. 共1兲–共5兲 become the governing equations for mixed con-
vection in a horizontal rectangular duct. Also, as
␦ → ␲ /2: sin ␦ →1, Ra*→0, ⌬→Ra/Re (10)
and the buoyancy term in the axial momentum equation 共4兲

becomes
␦ → ␲ /2: 共 ⌬/Pr兲•␪→共Gr/Re兲 • ␪ , (11)
which is the correct buoyancy parameter for laminar mixed con-

vection in the axial momentum equation in a vertical duct. Thus, it
is clear that the introduction of ⌬ and Ra* as independent param-
eters has not only reduced the number of governing parameters of
the problem but has permitted a compact formulation in which all
inclination angles, including the limiting cases of horizontal and
vertical rectangular ducts, can be recovered. To reduce the com-
putational efforts, the Prandtl number Pr⫽0.7 for air is used in
this work. Effects of the rest four parameters are investigated. ⌬
was set to be 100 and ⫺100 and Ra* was taken between 0 and
105 . The wall Reynolds number Rew ranges from 1.5 to ⫺2 with
aspect ratio ␥ being 2, 1, and 0.5.
After the velocity and temperature fields are obtained, the com-
putations of the local friction factor and Nusselt number are of
practical interest. Following the usual definitions, the expression
for the product of peripherally averaged friction factor and Rey-
nolds number, f Re, and Nusselt number, Nu, can be written based
on the axial velocity gradient and temperature gradient on the duct
walls,
f Re⫽⫺2 共 ⳵ W/ ⳵ n 兲 (12)
Nu⫽ 共 ⳵ ␪ / ⳵ n 兲 / 共 1⫺ ␪ b 兲 , (13)
where the overbar represents average around the perimeter. ␪ b is

Fig. 1 Physical configuration and coordinate system the bulk fluid temperature.
Solution Method Results and Discussion
The governing equations are numerically solved by the The effects of mixed convection parameter ⌬ and modified
vorticity-velocity method for three-dimensional parabolic flow Rayleigh number Ra* on the axial variations of f Re and Nu at
关17兴. Details of the solution procedure have been described else- Rew⫽⫺1 and ␥⫽1 are presented in Fig. 2. From the defining
where 关16兴, and not repeated herein. equations of Ra* and ⌬ 共Eq. 共6兲兲, it is seen that holding ⌬ fixed
To ensure the independence of the numerical results, a grid- while increasing Ra* is like decreasing the inclination angle ␦ and
dependence test was performed before the main course of the simultaneously increasing the Rayleigh number Ra. Alternatively,
computations. The grids were uniformly arranged in the cross- the same effect can be obtained at a fixed inclination angle ␦ by
sectional plane but nonuniformly distributed in the axial direction increasing Ra and Re simultaneously such that the value of ⌬
for the uneven variations of W and ␪ at the region near the inlet. It remains the same. In any case, the effect of increasing Ra* at a
was found in the separate numerical runs that, for a typical case fixed ⌬ is to increase the combined buoyancy effects. In Fig. 2,
with Rew⫽⫺1, Ra* ⫽5 ⫻ 104 , ⌬⫽100, and ␥⫽1, the deviations the effect of Ra* is practically negligible when Ra* ⭐103 . A
in Nu and f Re calculated with I ⫻ J⫽25 ⫻ 50 and 40 ⫻ 80 monotonic decrease in Nu near the inlet is caused by the entrance
(⌬Z * ⫽1 ⫻ 10⫺5 ⬃5 ⫻ 10⫺4 ) are always less than 2 percent. effect. The onset of buoyancy effect occurs at some axial location
Furthermore, the deviations in Nu and f Re calculated on the grids from the inlet depending mainly upon the value of Ra*. A mini-
of using I ⫻ J(⌬Z * )⫽25 ⫻ 50(2 ⫻ 10⫺6 ⬃5 ⫻ 10⫺4 ), and 25 mum local Nu will not appear unless the entrance effect is bal-
⫻ 50(1 ⫻ 10⫺5 ⬃5⫻10⫺4 ) are estimated to be within 1 percent. anced out by the buoyancy effects. After reaching the local mini-
In addition, the computations involving I ⫻ J⫽50 ⫻ 50( ␥ ⫽2) mum point, the variation of local Nu generally shows an increase
and 100 ⫻ 25共␥⫽0.5兲 grids are also used and found to be suffi- to a maximum local Nu. At large downstream distance, the local
ciently accurate to describe the mixed convection heat transfer in Nu gradually approaches the asymptotic value given by the Graetz
inclined rectangular ducts. As an additional verification of the solution. Comparing the solid and dashed lines indicates that, at a
computational procedure, the hydrodynamically developing flow fixed Ra*, larger f Re and Nu are noted for a system with a
was computed without buoyancy effects. The results were com- positive buoyancy parameter ⌬ due to the aiding buoyancy force
pared with those of Shah and London 关18兴. The apparent friction for ⌬⬎0. It is worth noting in Fig. 2共a兲 that the effects of Ra* on
factor was found to agree within 2 percent at all axial locations. the f Re is relatively insignificant for the system with a positive ⌬.
Besides, the limiting results of forced convection in an one- Figure 3 presents the wall Reynolds number Rew on the axial
porous-wall square duct with various injected rates were obtained variations of the circumferentially averaged friction factor f Re
and compared with those of Cheng and Hwang 关15兴. Excellent and Nusselt number Nu. Relative to the result of impermeable
agreement was found. These program tests confirm the accuracy duct system (Rew⫽0), the effects of wall transpiration on the local
and convergence of the present numerical solution. f Re is slight near the inlet. The decay of f Re with Z * is due to
Fig. 2 The axial variations of the f Re and Nu for RewÄÀ1 and Fig. 3 Effects of wall Reynolds number Rew on the axial varia-
␥Ä1 with ⌬ and Ra* as parameters tions of the f Re and Nu for Ra* Ä5Ã104 , ⌬Ä100 and ␥Ä1
Fig. 4 The axial variations of the f Re and Nu for RewÄÀ1 and Fig. 5 The axial variations of the f Re and Nu for RewÄÀ1 and
␥Ä2 with ⌬ and Ra* as parameters ␥Ä0.5 with ⌬ and Ra* as parameters
the entrance effect. In addition, a slightly larger 共smaller兲 f Re is those of ␥⫽1. However, the results show that the Nu for aspect
noted for a system with a suction 共blowing兲 case. At about Z * ratio ␥⫽0.5 共narrow and tall channel兲 is superior to that for ␥⫽2
⫽8 ⫻ 10⫺3 , the curves for all blowing and suction cases con- 共thin and flat channel兲. This is because that the relatively longer
verge. But after this location, the f Re increases with increasing vertical side wall has a stronger buoyancy effect on the flow and
axial location for the case of fluid injection (Rew⬍0), and the heat transfer for a channel with a smaller ␥. Additionally, in the
enhancement in f Re due to the wall injection effects increases absence of buoyancy, the Nusselt number would not depend on
with an increase in the injection rate. This can be explained by the the aspect ratio, with ␥⫽0.5 and 2 yielding one curve.
steeper axial velocity gradient near the duct walls for the stronger
injection rate. The mass extraction (Rew⬎0) will decelerate the
main stream velocity. Hence the velocity gradient at the duct wall Conclusion
diminishes. This is projected in Fig. 3共a兲 that the f Re decreases The mixed convection heat transfer in inclined rectangular
with increasing suction rate Rew . It is also found in Fig. 3共b兲 that ducts with wall transpiration was numerically investigated by us-
a smaller Nu results for a system with a greater injection rate ing a vorticity-velocity method. The effects of the mixed convec-
(Rew⬍0). This is owing to the fact that the wall injection in- tion parameter ⌬, modified Rayleigh number Ra*, wall Reynolds
creases the thermal boundary layer thickness and stimulates the number Rew , and aspect ratio ␥ on the momentum and heat trans-
fully developed temperature profiles being formed. That is, the fer were examined in detail. What follows is brief summary of the
temperature difference of the heated wall and the bulk fluid tem- major results.
perature is smaller for injection flow (Rew⬍0). Hence the Nusselt
number is smaller for a higher negative Rew . On the contrary, the 1 Within the range of parameters under investigation, the
temperature difference is increased by the fluid extraction in the buoyancy effects are practically negligible when Ra* ⭐1
case of suction flow (Rew⬎0). Hence a higher Nu is obtained with ⫻103 .
greater suction rate. 2 For injection case (Rew⬍0), the Nusselt number Nu is re-
The effects of the aspect ratio of a rectangular duct on the flow tarded with an increase in the wall Reynolds number Rew ,
and heat transfer results are of practical interest. The local f Re but the trend is reverse for the suction flow (Rew⬎0).
and Nu for the aspect ratios ␥⫽2 and 0.5 are shown in Figs. 4 and 3 The variations of the local Nusselt number are characterized
5 with mixed convection parameter ⌬ and modified Rayleigh by a decay near the inlet in which the forced-convection
number Ra* as parameters. In Figs. 4 and 5, the lowest curve can entrance effect dominates; but the decay is attenuated by the
be regarded as a limiting case for purely forced convection. Com- onset of buoyancy-driven secondary flows.
paring Figs. 2, 4, and 5, it is found that the general behaviors of 4 A stronger buoyancy effect is noted for a system with a
the f Re and Nu for aspect ratios ␥⫽2 and 0.5 are quite similar to smaller ␥ 共␥⫽0.5兲.
Acknowledgment 关15兴 Cheng, Y. C., and Hwang, G. J., 1995, ‘‘Experimental Study of Laminar Flow
and Heat Transfer in a One-Porous-Wall Square Duct With Wall Injection,’’
The financial support of this research by the National Science Int. J. Heat Mass Transf., 38, pp. 3475–3484.
Council under the contract NSC 87-2212-E-211-006 is greatly 关16兴 Lee, K. T., and Yan, W. M., 1998, ‘‘Mixed Convection Heat Transfer in
Horizontal Rectangular Ducts With Wall Transpiration Effects,’’ Int. J. Heat
appreciated. Mass Transf., 41, pp. 411–423.
关17兴 Ramakrishna, K., Rubin, S. G., and Khosla, P. K., 1982, ‘‘Laminar Natural
Nomenclature Convection Along Vertical Square Ducts,’’ Numer. Heat Transfer, 5, pp. 59–
79.
a, b ⫽ width and height of a rectangular duct, respectively 关18兴 Shah, R. A., and London, A. L., 1978, Laminar Flow Forced Convection in
D e ⫽ equivalent hydraulic diameter, 2ab/(a⫹b), m Ducts, Sppl. 1 to Adv. Heat Transfer, Academic Press, New York, pp. 196–
222.
I, J ⫽ number of finite difference divisions in the X and Y
directions, respectively
n ⫽ dimensionless outward normal direction to the wall
Ra* ⫽ modified Rayleigh number, Racos ␦
Rew ⫽ wall Reynolds number, v in D e / ␯ Clouds Over Soot Evaporation: Errors
v in ⫽ magnitude of fluid velocity injected or suctioned
through the porous walls in Modeling Laser-Induced
X,Y ,Z ⫽ dimensionless rectangular coordinate, X⫽x/D e , Y Incandescence of Soot
⫽y/D e , Z⫽z/(Re De)
Z * ⫽ dimensionless z-direction coordinate, z/(Pr Re De)
⫽Z/Pr G. J. Smallwood
Greek Symbols Senior Research Officer
␦ ⫽ duct inclination angle e-mail: greg.smallwood@nrc.ca
⌬ ⫽ mixed convection parameter, 共Ra/Re兲sin ␦
␥ ⫽ aspect ratio of a rectangular duct, a/b D. R. Snelling
␰ ⫽ dimensionless vorticity in axial direction Principal Research Officer
Subscripts
b ⫽ bulk fluid quantity F. Liu
o ⫽ value at inlet Associate Research Officer
w ⫽ value at wall
Superscripts Ö. L. Gülder
⫺ ⫽ average value
Senior Research Officer,
Group Leader
References
关1兴 Cheng, K. C., and Hong, S. W., 1972, ‘‘Effect of Tube Inclination on Laminar
Combustion Research Group,
Convection in Uniformly Heated Tubes for Flat-Plate Solar Collectors,’’ Sol. Institute for Chemical Process & Environmental
Energy, 13, pp. 363–371.
关2兴 Sabbagh, J. A., Aziz, A., El-Aring, A. S., and Hamad, G., 1976, ‘‘Combined
Technology,
Free and Forced Convection Heat Transfer in Inclined Circular Tubes,’’ National Research Council Canada,
ASME J. Heat Transfer, 98, pp. 322–324.
关3兴 Abdelmeguid, A. M., and Spalding, D. B., 1979, ‘‘Turbulent Flow and Heat
1200 Montreal Road, Ottawa,
Transfer in Pipes With Buoyancy Effects,’’ J. Fluid Mech., 94, Part 2, pp. Ontario, Canada K1A 0R6
383–400.
关4兴 Morcos, E. M., and Abou-Ellail, M. M. M., 1983, ‘‘Buoyancy Effects in the
Entrance Region of an Inclined Multirectangular Channel Solar Collector,’’
ASME J. Sol. Energy Eng., 105, pp. 157–162. The ambiguity and incorrect treatment of the evaporation term
关5兴 Cheng, K. C., and Yuen, F. P., 1985, ‘‘Flow Visualization Studies on Second-
ary Flow Pattern for Mixed Convection in the Entrance Region of Isothermally among some LII models in the literature are discussed. This study
Heated Inclined Pipes, Fundamentals of Forced and Mixed Convection,’’ does not suggest that the correct formulation presented for the
ASME HTD, 42, pp. 121–130. evaporation model is adequate, or that it reflects the soot evapo-
关6兴 Morcos, S. M., Hilal, M. M., Kamel, M. M., and Soliman, M. S., 1986, ‘‘Ex- ration process under intense evaporation. The emphasis is that the
perimental Investigation of Mixed Laminar Convection in the Entrance Region
of Inclined Rectangular Channels,’’ ASME J. Heat Transfer, 108, pp. 574– current evaporation model must be used correctly in the evalua-
579. tion of the LII model against experimental data. Numerical results
关7兴 Choudhury, D., and Patankar, S. V., 1988, ‘‘Combined Forced and Free Lami- are presented to demonstrate the significance of the molecular
nar Convection in the Entrance Region of an Inclined Isothermal Tube,’’ weight associated with the heat of evaporation and the thermal
ASME J. Heat Transfer, 110, pp. 901–909.
关8兴 Olson, R. M., and Eckert, E. R. G., 1966, ‘‘Experimental Studies of Turbulent velocity of carbon vapor on the results obtained with the evapo-
Flow in a Porous Circular Tube With Uniform Fluid Injection Through the ration model. Other errors frequently repeated in the literature
Tube Wall,’’ ASME J. Appl. Mech., 33, pp. 7–17. are also identified. 关DOI: 10.1115/1.1370507兴
关9兴 Yuan, S. W., and Finkelstein, A. B., 1958, ‘‘Heat Transfer in Laminar Pipe
Flow With Uniform Coolant Injection,’’ Jet Propul., 28, pp. 178–181.
关10兴 Pederson, R. J., and Kinney, R. B., 1971, ‘‘Entrance-Region Heat Transfer for
Laminar Flow in Porous Tubes,’’ Int. J. Heat Mass Transf., 14, pp. 159–161.
关11兴 Raithby, G. D., 1971, ‘‘Laminar Heat Transfer in the Thermal Entrance of Introduction
Circular Tubes and Two-Dimensional Rectangular Ducts With Wall Suction Laser-induced incandescence 共LII兲 has experienced rapid devel-
and Injection,’’ Int. J. Heat Mass Transf., 14, pp. 224–243.
关12兴 Hirata, Y. S., Hirata, Y. S., and Ito, R., 1982, ‘‘Experimental Study of Flow opment as a powerful diagnostic tool for spatially and temporally
Development in a Porous Tube With Injection or Suction,’’ J. Chem. Eng. resolved measurements of soot volume fraction and primary par-
Jpn., 15, pp. 455–451. ticle size in various applications such as diffusion flames and die-
关13兴 Kinney, R. B., 1968, ‘‘Fully Development Frictional and Heat Transfer Char- sel engine exhaust. The LII technique is based on the detection
acteristics of Laminar Flow in Porous Tubes,’’ Int. J. Heat Mass Transf., 11,
pp. 1393–1401.
关14兴 Hwang, G. J., Cheng, Y. C., and Ng, M. L., 1993, ‘‘Developing Laminar Flow Contributed by the Heat Transfer Division for publication in the Journal of Heat
and Heat Transfer in a Square Duct With One-Walled Injection and Suction,’’ Transfer. Manuscript received by the Heat Transfer Division April 3, 2000; revision-
Int. J. Heat Mass Transf., 36, pp. 2429–2440. received December 7, 2000. Associate Editor: R. Skocypec.
and analysis of the incandescence signals of the enhanced thermal The terms in Eq. 共1兲 represent, in order, the laser energy ab-
radiation from soot particles subjected to an intense laser pulse. sorption by soot particle, heat conduction loss from the soot par-
Development of the mathematical model describing the heat and ticle to the surrounding gas in the transition regime, heat loss due
mass transfer processes of LII is not only essential in understand- to soot evaporation, heat loss through the mechanism of thermal
ing some aspects of experimental results but it is also useful to radiation, and finally the rate of soot particle internal energy
improve the capabilities of the technique. change. The absorbed laser energy is determined by the laser in-
The first effort to model the nanoscale heat and mass transfer tensity q, and the absorption cross section C a ⫽( ␲ 2 d 3p E(m))/␭ ,
processes of soot in LII was made by Eckbreth 关1兴. Subsequent where d p is the diameter of the primary particles, E(m) is a re-
improvement and application of this model have been presented fractive index dependent function, and ␭ is the wavelength of the
by Melton 关2兴, Dasch 关3兴, Tait and Greenhalgh 关4兴, Hofeldt 关5兴, laser. The parameters in the conduction term include: k a , the heat
and recently by Mewes and Seitzman 关6兴, Snelling et al. 关7兴, Mc- conduction coefficient of air; T, the soot temperature; T g , the gas
Manus et al. 关8兴, Will et al. 关9兴, Snelling et al. 关10兴, and Schraml temperature; d p , the diameter of the primary particles; G, a
et al. 关11兴. Given the assumptions of the current LII model 关10兴, it geometry-dependent heat transfer factor; and ␭ MFP , the mean free
is not surprising to see some discrepancies between the model path. The parameters in the evaporation term are: ⌬H v (T), the
predicted and the experimental soot temperatures 关11兴. The dis- particle temperature dependent heat of evaporation of graphite;
crepancies found in the study of Schraml et al. are partially attrib- M v (T), the particle temperature dependent molecular weight of
uted to their incorrect treatment of the evaporation heat loss term. soot vapor; M, the mass of soot particle; and t is time. q rad is the
Further research and evaluation of the LII model are clearly reheat loss term due to radiation. Additional parameters presented in
quired. Nevertheless, the usefulness of the current LII model has the internal energy term are: ␳ s , the density of soot; and c s , the
been demonstrated in several studies 关5,8–11兴. Central to the LII specific heat of soot.
process is soot evaporation, which reduces the soot particle size To illustrate the differences in the treatment of the evaporation
and provides an effective cooling mechanism that limits the fur- heat loss term between the present LII model and the models
ther rise of soot particle temperature. An adequate treatment of the presented and/or employed by others, these evaporation models
evaporation term is the key to the success of an LII model to are compared in Table 1, where ␤ is the evaporation coefficient,
predict the time-resolved soot particle size, soot temperature, and P v (T) is the particle temperature dependent vapor pressure of
the excitation curve. Unfortunately, significant differences exist in soot, and R is the universal gas constant.
the treatment of the vaporization heat loss term among the LII Except for a minor difference in the introduction of the evapo-
models found in the literature. Our research revealed that the am- ration coefficient ␤ between Model I and Models II and III, two
biguity, confusion, and incorrect treatment of the soot evaporation important differences are observed. First, Model I employs a dif-
term originated in the first LII modeling paper by Eckbreth 关1兴. ferent molecular weight associated with the heat of vaporization
The incorrect formulation has been widely spread in the growing from Models II and III: Model I uses the molecular weight of the
community of LII modeling 关2,4,9兴, although the correct formula- soot vapor (M v ) while Models II and III use the molecular weight
tion of the soot evaporation model was given by Dasch 关3兴 and of the solid carbon (M s ). Secondly, a factor of ␲ is missing from
Hofeldt 关5兴. The present paper is motivated by the pressing need the denominator of the term inside the square root in Model II. It
to clarify the confusion and incorrect treatment of the soot evapo- is not clear which expression was used in the studies of Will et al.
ration model among some researchers working on LII modeling. 关9兴 and Schraml et al. 关11兴 since formulation for dM /dt was not
The objectives of this study are 共1兲 to clarify the confusion sur- given in their papers. Here we assume that Model III employs the
rounding the soot evaporation model and establish the correct for- same dM /dt formulation as Model I.
mulation in order to prevent further use of the incorrect one, and It is understandable to easily make incorrect use of the molecu-
共2兲 to numerically evaluate the impact of using the incorrect soot lar weight associated with the heat of vaporization if careful
evaporation model. thought is not given to the physical situation of soot 共assuming
graphite兲 evaporation. It is well known that multiple species of
carbon coexist in the vapor with C 3 as the most abundant species
Evaporation Models based on the thermodynamic equilibrium calculations of Leider
A detailed description of the LII model employed in this study et al. 关12兴. In addition, the heat of evaporation, ⌬H v , is the en-
has been given by Snelling et al. 关10兴 and therefore is not repeated ergy required to evaporate unit mole solid carbon into multiple
here. The energy balance equation is given here for the conve- gaseous carbon species. Therefore, the mean molecular weight of
nience of the reader. the vapor should be used, not the molecular weight of the solid
carbon, so that the heat of evaporation is consistent with the mo-
2k a 共 T⫺T g 兲 ␲ d 2p ⌬H v 共 T 兲 dM 1 dT lecular weight. Certainly, it would be correct to use the molecular
C a q⫺ ⫹ ⫹q rad⫽ ␲ d 3p ␳ s c s
共 d p ⫹G␭ MFP兲 M v 共 T 兲 dt 6 dt weight of solid carbon if only a single species (C 1 ) exists in the
(1) vapor, or if ⌬H v corresponding to C 1 only is used. A temperature
Table 1 Comparison of the evaporation models in the literature
a
Dasch 关3兴 did not use the evaporation coefficient ␤.
b
The square root term was given as 冑 Mv
2
RT, which is likely to be a typo.
c
Will et al. 关9兴 and Schraml et al. 关11兴 did not present their expressions for dM /dt.
dependent mean molecular weight is used in the numerical model, Results and Discussion
rising from 24 at 2000 K to 36 at 3600 K to 48 at 4700 K. Thus
Characteristics of the excitation laser were given in detail in
the error in using C 1 , with a molecular weight of 12, increases
Snelling et al. 关10兴. Briefly, the wavelength of the laser is 532 nm
from overestimating the enthalpy loss from soot mass evaporation
and it has a typical Q-switched temporal profile of 7 ns FWHM
by a factor of 2 at 2000 K to a factor of 4 at 4700 K.
duration. The spatial profile of the laser intensity is top-hat 共uni-
As for the second difference associated with the thermal veloc-
form兲. The primary soot particle size assumed in the calculations
ity of vapor between Model I and Model II, it is emphasized here
is 32 nm. Unless otherwise stated, the evaporation coefficient ␤ is
that the expression of Model I is correct. A detailed derivation of
assumed to be 1.0. E(m) based on the refractive index of soot due
the mass flux of vapor that crosses unit area toward one side was
to Lee and Tien 关15兴共0.176 at 532 nm兲 was used in all the calcu-
given by Kennard 关13兴 and is written as
lations. The numerical solver uses an adaptive fourth-order
⌽⫽ P v 冑 Mv
2 ␲ RT
. (2)
Runge-Kutta algorithm that optimizes the step size to minimize
the truncation error. The numerical results discussed below were
obtained using a time step of 0.25 ns. Investigations were per-
Application of Eq. 共2兲 to calculate the evaporation heat loss term formed to ensure that this time step was sufficiently small to
leads to the expression of Model I given in Table 1. The rationale achieve results independent of the time step, with maximum de-
of introducing an evaporation coefficient, ␤, has been discussed viation of less than 0.1 percent. Effects of using the three different
by Kennard 关13兴. The evaporation heat loss term of Model II evaporation models given in Table 1 on the model predictions
given in Table 1 implies that the mass flux of vapor would read were investigated.
冑
The physical properties used in the present calculations are
Mv specified as follows. The local gas temperature T g ⫽1800 K, a
⌽ ⬘⫽ P v (3)
2RT typical value found in flames. The corresponding values of k a , G,
and ␭ M F P are 0.12 W/m K, 22.064, and 0.5665 ␮m, respectively.
resulting in a predicted enthalpy loss that is 1.8 times too high.
The soot density and specific heat are taken at the values of local
The combination of the inconsistent use of molecular weights and
the incorrect formulation for the mass flux of vapor presented in gas temperature, i.e., ␳ s ⫽2.2 g/cm3 关12兴 and c s ⫽2.1 J/g K 关16兴
Model II produces an overall predicted enthalpy loss that ranges since they only vary mildly with temperature. Properties related to
from 3.6 times too high at 2000 K to 7.2 times too high at 4700 K. the soot evaporation rate are obtained from the equilibrium calcu-
Such overestimations of the enthalpy losses are anticipated to lations of Leider et al. 关12兴 and are given as fifth-order polyno-
cause the particle temperatures to be much cooler and the evapo- mial expressions as P v ⫽exp(兺i⫽0 5
piTi) atm, M v ⫽ 兺 i⫽0
5
m iT i
rated mass loss to be much less with Models II and III in com- kg/mole, and ⌬H v ⫽ 兺 i⫽0 h i T J/K mole. The fitting coefficients
5 i
parison with Model I. are given in Table 2.

In addition to the confusion and incorrect treatment regarding When conducting numerical calculations of soot particle size
the evaporation term of the LII models in the literature, it is also and temperature using the present LII model, it is worth pointing
worth pointing out that Melton 关2兴 gave an incorrect expression out that there are uncertainties in the predictions due to uncertain-
for the thermal conductivity of air, which has been previously ties associated with the thermal properties of soot such as the
identified 关6兴. According to Melton, the thermal conductivity of absorption function E(m), the mean molecular weight of soot
air was calculated as K a ⫽5.83⫻10⫺5 (T/273) 0.82 W cm⫺1 K⫺1 vapor M v , the heat of vaporization ⌬H v , the vapor pressure P v ,
based on the expression given by Tsederberg 关14兴. It was noticed and input parameters for the heat conduction rate. Based on our
by Mewes and Seitzman 关6兴 that this expression yields air thermal best knowledge it was estimated that uncertainties of these prop-
conductivity that is only about one fourth of well-established val- erties are well within 30 percent. The effects of the uncertainty in
ues. After consulting the expression given by Tsederberg, it was these properties on the numerical predictions were investigated by
realized that Melton 关2兴 neglected to convert calorie to joule in the perturbing the values of the properties discussed above by ⫾30
expression shown above, thus accounting for the discrepancy. In percent using Model I at a laser fluence of 0.725 J/cm2. At this
addition, the analytical expression for the thermal conductivity of laser fluence, there is significant soot vaporization and properties
air given in Tsederberg is applicable only up to 1273 K. There- associated with the heat conduction loss have almost no effect on
fore, neither the expression given in Melton nor the expression the solution during and shortly after the laser pulse 共the laser pulse
given in Tsederberg should be employed in LII modeling since duration is 30 ns兲. Effects of uncertainty of E(m), M v , ⌬H v , and
temperatures in a flame are well above the upper temperature limit P v on the predictions are shown in Fig. 1. Uncertainties in E(m),
of the expression given in Tsederberg. On the other hand, one M v , and ⌬H v have a significant effect on particle size but have
should be aware that use of the thermal conductivity of air in less impact on particle temperature, Figs. 1共a兲, 共b兲, and 共c兲. Varia-
modeling the heat conduction term is only an approximation, tion of the vapor pressure by 30 percent has a modest effect on the
since the local thermal conductivity of combustion gas varies sig- particle temperature and minimal effect on particle size, Fig. 1共d兲.
nificantly in a flame and in the exhaust gases due to variations in Overall, the uncertainty in the predicted particle temperature and
the composition. particle size is about 2 percent and 20 percent, respectively.
Furthermore, there is a typographical error in Melton 关2兴 in the
sign of the exponent for the heat of vaporization of carbon, which Effects on the History of Soot Particle Size and Tempera-
should have read 7.78⫻10⫹5 J/mol. ture. Calculations were carried out using the three different
Table 2 Fitting coefficients for the vapor pressure, mean molecular weight, and the heat of vaporization
Fig. 2 Effects of some evaporation model parameters and the
thermal conductivity on the predicted soot particle size „a… and
temperature „b…. The temporal profile of the laser intensity is
shown for reference.
the incorrect K a affects the soot temperature only near the end of
the laser pulse 共30 ns兲, Fig. 2共b兲, when heat conduction becomes
important. Although not demonstrated in Fig. 2, use of Melton’s
K a results in an incorrect prediction of the temperature decay at
longer times. Results obtained with Model I using ␤ ⫽0.7 are also
shown in the figure to demonstrate the effect of the evaporation
coefficient. To present a thorough picture about the effect of the
molecular weight 共either M s or M v 兲 and the thermal velocity of
carbon vapor, equivalent to the mass flux expressions given in
Eqs. 共2兲 and 共3兲, results based on M v and Eq. 共3兲 are also shown
in Fig. 2.
Results shown in Fig. 2共a兲 reveal that it is very crucial to em-
Fig. 1 Uncertainty analysis for specified thermal properties of
ploy a molecular weight consistent with the heat of evaporation
soot: „a… refractive index dependent function; „b… mean molecu-
lar weight of soot vapor; „c… heat of vaporization; and „d… vapor for the prediction of the soot particle size, while missing a factor
pressure. of ␲ in the thermal velocity of soot vapor has only slight effect on
the particle size 共see difference of results between Model II and
Model III兲. Model I employing the incorrect thermal conductivity
results in a slightly bigger particle size at the end of the soot
evaporation models shown in Table 1 for a laser peak fluence of evaporation.
0.725 J/cm2, which is equivalent to a laser pulse energy of 6 mJ Effects of different evaporation models and the evaporation co-
for the spatial profile. These results are compared in Fig. 2. Model efficient on the predicted soot temperature are shown in Fig. 2共b兲.
I predicts that significant soot evaporation occurs under the com- Model I predicts that the peak soot temperature is about 4530 K,
putational conditions with the particle size reduced by about 26 while Models II and Model III predict about 4177 and 4277 K
percent. In contrast, Models II and III predict that the reduction in respectively. The reduced evaporation coefficient, ␤ ⫽0.7, reduces
particle size is only about 8 percent. It is also interesting to ob- the mass loss rate during the first 10 ns of the laser pulse, which
serve that Model II predicts the particle size that is only very can be observed from the slightly slower decreases of soot particle
slightly lower than that from Model III. A slightly smaller particle size, Fig. 2共a兲. This reduced mass loss in turn reduces the heat
size starting at the end of laser pulse is noticed as a result of using loss through evaporation, resulting in a higher peak soot particle
Melton’s K a since heat conduction loss is underestimated. Use of temperature, Fig. 2共b兲. At later times, this increased maximum
used, the LII model predicts the excitation curve for prompt LII
signal collected at a fixed duration in qualitative agreement with
experimental observation.
2 The incorrect use of the molecular weight for solid carbon
associated with the heat of evaporation drastically affects the re-
sults of the model prediction. It results in lower soot temperature
and significantly less soot evaporation. It also fails to predict the
correct shape of the excitation curve.
3 The incorrect expression for the thermal velocity of carbon
vapor affects the model results significantly less than the effect of
the incorrect molecular weight.
The present study by no means attempts to claim that the cor-
rect formulation presented for the state-of-the-art evaporation
model is adequate and truly reflects the soot evaporation process
under intense evaporation. The emphasis of this work is that the
current evaporation model must be used correctly in the evalua-
tion of the LII model against experimental data.
Fig. 3 Variation of the predicted and the experimental normal-
ized prompt LII signals with laser fluence References
关1兴 Eckbreth, A. C., 1977, ‘‘Effects of Laser-Modulated Particulate Incandescence
on Raman Scattering Diagnostics,’’ J. Appl. Phys., 48, pp. 4473–4479.
temperature increases the mass loss rate due to the strong depen- 关2兴 Melton, L. A., 1984, ‘‘Soot Diagnostics Based on Laser Heating,’’ Appl. Opt.,
23, pp. 2201–2208.
dence of the vapor pressure on temperature, resulting in a slightly 关3兴 Dasch, C. J., 1984, ‘‘New Soot Diagnostics in Flames Based on Laser Vapor-
smaller final soot particle size. ization of Soot,’’ Twentieth Symposium (International) on Combustion, The
It is evident from the results shown in Fig. 2 that the inconsis- Combustion Institute, pp. 1231–1237.
tent use of M s in the evaporation heat loss term, ⌬H v /M s , has a 关4兴 Tait, N. P., and Greenhalgh, D. A., 1993, ‘‘PLIF Imaging of Fuel Fraction in
Practical Devices and LII Imaging of Soot,’’ Berichte der Bunsengesellschaft
far more significant impact on the performance of the model than fuer Physikalische Chemie, 97, pp. 1619–1625.
use of an incorrect expression for the thermal velocity or an inad- 关5兴 Hofeldt, D. L., 1993, ‘‘Real-Time Soot Concentration Measurement Tech-
equate evaporation coefficient. It is interesting to observe from the nique for Engine Exhaust Streams,’’ Society of Automotive Engineers, SAE
results shown in Fig. 2共a兲 that parameters that directly alter the Paper No. 930079.
关6兴 Mewes, B., and Seitzman, J. M., 1997, ‘‘Soot Volume Fraction and Particle
mass loss rate of the soot particle, such as the evaporation coeffi- Size Measurements with Laser-Induced Incandescence,’’ Appl. Opt. 36, pp.
cient ␤ and the thermal velocity, ultimately have an insignificant 709–717.
effect on the soot particle size. Instead, they have a greater effect 关7兴 Snelling, D. R., Smallwood, G. J., Campbell, I. G., Medlock, J. E., and Gülder,
on the soot temperature. Parameters that directly affect the heat Ö. L., 1997, ‘‘Development and Application of Laser-Induced Incandescence
共LII兲 as a Diagnostic for Soot Particulate Measurements,’’ Advanced Non-
loss rate, such as the molecular weight associated with the heat of Intrusive Instrumentation for Propulsion Engines, AGARD Conference Pro-
evaporation, have far more impact on the results, in particular the ceedings 598, pp. 23-21 to 23-29.
soot particle size. The complex interaction of these parameters is 关8兴 McManus, K. R., Frank, J. H., Allen, M. G., and Rawlins, W. T., 1998,
due in part to their strong temperature dependencies. The results ‘‘Characterization of Laser-Heated Soot Particles Using Optical Pyrometry,’’
AIAA Paper No. 98-0159.
are indeed a manifestation of the physics of LII: soot temperature 关9兴 Will, S., Schraml, S., Bader, K., and Leipertz, A., 1998, ‘‘Performance Char-
is the driving force of soot particle evaporation. acteristics of Soot Primary Particle Size Measurements by Time-Resolved
Laser-Induced Incandescence,’’ Appl. Opt., 37, pp. 5647–5658.
Effects on Excitation Curve. The predicted normalized LII 关10兴 Snelling, D. R., Liu, F., Smallwood, G. J., and Gülder, Ö. L., 2000, ‘‘Evalu-
signal at 400 nm 共collected at 20 ns after the start of the laser ation of the Nanoscale Heat and Mass Transfer Model of the Laser-Induced
pulse for a gate width 18 ns兲 are compared in Fig. 3 with the Incandescence: Prediction of the Excitation Intensity,’’ Thirty Fourth National
Heat Transfer Conference, NHTC2000-12132.
experimental data of Ni et al. 关17兴. The excitation curve predicted 关11兴 Schraml, S., Dankers, S., Bader, K., Will, S., and Leipertz, A., 2000, ‘‘Soot
by Model I agrees qualitatively well to the experimental curve, Temperature Measurements and Implications for Time-Resolved Laser-
while Models II and III fail to capture the correct shape of the Induced Incandescence 共Tire-LII兲,’’ Combust. Flame, 120, pp. 439–450.
excitation curve. 关12兴 Leider, H. R., Krikorian, O. H., and Young, D. A., 1973, ‘‘Thermodynamic
Properties of Carbon up to the Critical Point,’’ Carbon, 11, pp. 555–563.
关13兴 Kennard, E. H., 1938, Kinetic Theory of Gases, McGraw Hill Book Company,
Conclusions New York, pp. 63–69.
关14兴 Tsederburg, N. V., 1965, Thermal Conductivity of Gases and Liquids, The
Numerical results obtained using different evaporation models M.I.T Press, Cambridge, MA, p. 89.
are presented and compared to demonstrate the importance of the 关15兴 Lee, S. C., and Tien, C. L., 1981, ‘‘Optical Constants of Soot in Hydrocarbon
evaporation model to the overall performance of the LII model. Flames,’’ Eighteenth Symposium (International) on Combustion, The Combus-
The following conclusions are reached based on the numerical tion Institute, pp. 1159–1166.
关16兴 Chase, M. W., Jr., Davies, C. A., Downey, J. R., Jr., Frurip, D. J., McDonald,
results of this study: R. A., and Syverud, A. N., 1985, ‘‘JANAF Thermochemical Tables,’’ Third
Edition, Journal of Physical and Chemical Reference Data, 14, Suppl. 1.
1 Use of a molecular weight consistent with the heat of evapo- 关17兴 Ni, T., Pinson, J. A., Gupta, S., and Santoro, R. J., 1995, ‘‘Two-Dimensional
ration is crucial to the overall performance of the LII model. Imaging of Soot Volume Fraction by the Use of Laser-Induced Incandes-
When a correct molecular weight based on the carbon vapor is cence,’’ Appl. Opt., 34, pp. 7083–7091.

J Heat Transfer 2001 Vol 123 N4

Uploaded by

Copyright:

Available Formats

J Heat Transfer 2001 Vol 123 N4

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 2001 Vol 123 N4

Uploaded by

Copyright:

Available Formats

Transactions Journal of

of the ASME Heat Transfer ®

Journal of Heat Transfer Volume 123, Number 4 AUGUST 2001

Natural and Mixed Convection

EVAPORATING CAPILLARY PORE FLOWS RANGING FROM 10-mm TO 10-␮m DIAMETER

H. J. Kim and K. D. Kihm

Fig. 4 Image of the atomic normal stress „axial… forces

Fig. 2 Instantaneous snapshot of polymer molecular motion

Fig. 5 Image of the atomic shear stress „friction… forces

VISUALIZATION OF ATOMIC FORCE MICROSCOPY FROM MOLECULAR DYNAMICS SIMULATIONS

SUBCOOLED POOL BOILING HEAT TRANSFER IN MICROGRAVITY AND Hi-g

John McQuillen and Mary Vickerman

SHEAR LAYER INSTABILITY AND MIXING IN MICRO HEAT SPREADERS

OSCILLATORY ENTRAINED DROPLET EHD TWO-PHASE FLOW

LOCAL FIN-SURFACE HEAT TRANSFER FOR FLOW AROUND A CIRCULAR CYLINDER

Keywords: Conduction, Contact Resistance, Heat Transfer, Modeling, Tribology

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 625

Thus if Pe→⬁, the penetration ␦ is small, and may be taken to

Review and Solution of Stationary and Moving Heat

for the maximum contact temperature.

626 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

for the dimensionless thermal resistance based upon the maximum

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 627

In the next section, the results will applied to develop new

Analysis of Real Contacts

628 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 629

630 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 631

632 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

Keywords: Finite Element, Heat Transfer, Inverse, Numerical Methods

Introduction order Tikhonov regularization on two two-dimensional test geom-

Contributed by the Heat Transfer Division for publication in the JOURNAL OF

using the Combined Residual and Smoothing Operator 共CRESO兲 where

Moo ⫽ 冕 NT Nd⌫ o , (3) where x T

where N is the vector of shape functions for the outer boundary

634 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

共 AT A⫹tRT R兲 x̂⫽AT f. (12)

where Xo contains the part of the basis vectors used to represent

We assume that our estimate for the temperature solution T̂ and

where ␣ is the vector of expansion coefficients 共containing N modes

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 635

636 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

trial forward solutions, and to order them carefully so that the

solutions which are smooth on the boundaries, and which are

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 637

particular way to estimate the location of the bend: we choose t to

638 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

Fig. 7 Geometry and finite element mesh for annulus test

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 639

inner boundary, a constant temperature on the outer boundary, and

640 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

represented exactly in terms of the first GESL cluster—those cor-

Table 5 Inverse relative error in temperature for the first tem-

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 641

642 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

Journal of Heat Transfer AUGUST 2001, Vol. 123 Õ 643

644 Õ Vol. 123, AUGUST 2001 Transactions of the ASME

Keywords: Computational, Film Cooling, Heat Transfer, Turbines, Turbulence

冑冉冊 ⫺0.1178 primary vortices.

冑冉冊 ⫺0.3123 cal structures is quite distinct from that by turbulence gener-